I got a suggestbot thingy, but it says it belongs to RyanVG. Is there a way to make one for me? Thanks. ELH50 05:34, 11 January 2006 (UTC)
Thanks for the note, I had some trouble with the posting to talk pages with the bot and did a few manually. Apparently, I made some mistakes, but now I know to go back and check the others. Thanks again. -- ForteTuba 20:56, 11 January 2006 (UTC)
Thanks for the fix. I like what you did with it, and will definitely put it to use. ELH50 20:09, 12 January 2006 (UTC)
Wow... great idea!
I'm not a fan of this new bot. It suggested I go to community portal to look for pages to work on, but that's just an annoyance. I can find pages to work on on my own, and unless it's giving me specific recommendations, I'd rather stick with my interests. Since I bothered to sign up for an account, it might be safe to assume that I'll go looking if I want something to work on, and it's probably safe to assume that I can find the community portal eventually. --Keflavich 03:04, 19 January 2006 (UTC)
I recieved an request, but I have editted most of theses articles (recently to be exact). -ZeroTalk 16:38, 7 March 2006 (UTC)
Apparently it's not as good at throwing away already-edited articles as I thought, I'll look into it, thanks. -- ForteTuba 16:57, 7 March 2006 (UTC)
Nice one! I just noted it on wikien-l and IRC, so expect a bunch of beta testers :-) My list is entirely Scientology-related stuff - it's seen me working around WP:SCN ... Would it be possible to add a feature to request the focus of the search? Scattergun/tight? - David Gerard 16:45, 7 March 2006 (UTC)
Right now I'm working on a version that only recommends pages that have a work tag on them (e.g., stub, cleanup, etc.) That'll make the search a lot wider, especially for work types other than stub, just because there are many fewer articles that have work tags. As for having it be part of the request, I can imagine ways to do it (like preferring to recommend only from/only not from a certain category). No promises, but thanks for the feedback (and the beta-testers). It'll be interesting to see what power editors think about it... -- ForteTuba 21:49, 7 March 2006 (UTC)
I see it includes a few red links, is it possible to include more? As a way of crossing the most requested pages with an editors own interests? So you'd get a list of the most linked to red links that conforms with someones interests. Rx StrangeLove 17:12, 7 March 2006 (UTC)
Quite. Red links are my forte'. I tend to make articles and construct large edits rather than minor. Throwing red links at me would keep me busy. -ZeroTalk 17:32, 7 March 2006 (UTC)
This is a good idea. On my list, btw, half the articles thrown at me are ones I'm not interested in editing at all, and one or two are those I'm barely interested in. I don't care for all those individual "19xx in Singapore" articles, really. Maybe year articles should be excluded, although perhaps there are a few year-editing fanatics around. Johnleemk | Talk 18:24, 7 March 2006 (UTC)
One of the recommenders is link-based; apparently, the YEAR in Singapore articles are well-linked to something you've edited andso you got a bunch of those. Sorry... -- ForteTuba 21:49, 7 March 2006 (UTC)
It has to get pretty lucky for red links. There are three recommenders, one based on article text, one based on linking, and one based on patterns of similar edits between pages. Only the linking-based one can get at the red links, and then it has to be something that's relatively oft-linked from the set of pages that you've edited. Hopefully the version that has articles that need work tags will do better at finding articles to make a major contribution to. The current version is set to make about 1/3 of its recommendations to be stubs.
If people thought that red vs. tagged-needing-work was really important, I could try to make it work harder to find red-linked articles. -- ForteTuba 21:49, 7 March 2006 (UTC)
Personally, I'm OK with it; there may be some editors who feel that it's spammy. Not a surprising list of related topics that I hit, so....good job! ;) RasputinAXP c 15:03, 8 March 2006 (UTC)
Would be better if it only grabs last 500 major edits.
can it be any 500 instead of the last 500? I've been so deep in project edits, some of which may be less relevant, that going back 300 or so may be a more accurate pic... just wondering. It would need to be optional since that's not true for everyone! ++Lar: t/c 06:05, 9 March 2006 (UTC)
What would you think about something that would let you give it arbitrary wikitext and/or mediawiki-generated HTML and extracts the links from it? More work to use, but it could be optional (and useful for people who want to target suggestions). -- ForteTuba 14:30, 9 March 2006 (UTC)
Sounds neat but is it a way to do what I suggested or just some other neat idea? ++Lar: t/c 22:31, 9 March 2006 (UTC)
What I heard you say was that you want to be able to tell it which 500 edits to use; the question is how best to tell SuggestBot which 500 edits (or more generally, which edits). It'd be easy to have a flag for the first vs last 500, but to make any arbitrary 500 would be harder -- you'd have to be able to say what date to start or stop getting edits at because of the way Special:Contributions works. Still not _hard_, but more work for people. It's also not very general... I was trying to come up with a reasonably easy, general solution that would let people pick whatever articles they want. For example, lots of people have sections on their User pages where they list their major contributions, articles created, etc. Those people could just paste that part of the page that represents their main interests in order to get better suggestions.
In your case you'd page through your contributions until you found what you wanted to use, then paste that page into a box on some SuggestBot request page. It is more work, alas... -- ForteTuba 16:30, 10 March 2006 (UTC)
I and possibly others at Wikiversity am interested (extremely) in reviewing the results with SuggestBot periodically (say 3-6 months) with a view towards eventually building tailored guidebots (mine would mirbot, I have reserved his account at advogato if I can find a computer that remembers the password or hack it myself) Mirbot might be a special conspirator buddy for me and my buddies when somebody calls us lying trolls or make us wear a punitive kick me sign in the hall on wall vulnerable to lynching by four or five other classes cause we put one on somebody else as a joke and nobody is apparently watching the kiddies. We would divise appropriately fandangos to resort the local kiddy gangs and the adult hiearchy to our satisfaction. Semi secret polling and public gambling pools at local, regional, virtual, real, etc. Las Vegas could be used to finance the operation as long as we could prove consistently that we did ourselves and others more good than harm at a profit while management risks and liabilities while healing ourselves and patenting the data and using it for future calibration checks and possibly real conspiracy management inside the big federal institutions alleged to be incorruptable. Who wants to be a secret agent when one can be a puppeteer or puppet master designers inc. ala space jamborees? When the government insists on their right to steal our data for the war effort maybe we have enough data to prove we are winning the war while the generals run off with the goodies for highest body counts. An anti-kelly war movement which would make me and others anti oddballs or airport operations seeking space beenies and weenies for glory roads as free agents only when Malibu surfing gets inappropriate. Seriously? No silly.....anti spin ward. Up or Down Periscope? Choices have consequences. gork that as a tautology yet? No but I will eventually. Thank you Lord. Amen. Lazyquasar 11:44, 12 March 2006 (UTC)
Not sure what to say. -- ForteTuba 17:57, 14 March 2006 (UTC)
I really like the idea of the suggest bot and the way I was informed was a good and effective idea. Since then I have worked on some of the articles you have suggested. --TD 12:34, 10 February 2006 (UTC)
Sweet, good to know and glad you found it useful. We were afraid it might be on the spammy side to post on talk pages but, no complaints so far. -- ForteTuba 15:53, 10 February 2006 (UTC)
(in response to comment on my talk page) It's cool, I understand the need for checking effectiveness. Still, I think it might be more effective to have a form than a bot. If a user wants to contribute, but doesn't have a good idea of what they would best contribute to, it might help for them to be able to ask a bot to figure out what they're likely to want to work on based on past edits, or something like that. --Keflavich 18:56, 19 January 2006 (UTC)
I agree something that lets a person explicitly say "I'm interested in X, Y, and Z" is probably going to be better than trying to just guess based on past edits. We're trying recommendations with and without filtering out minor edits, but people are surely smarter. The only bad thing is that it requires more work. I think that people who are already pretty committed would be willing to put in work (with you, AllyUnion suggested this would be better as a tool than a bot and long-term I agree) but newer members we're targeting won't be as willing. I may be wrong -- we plan to look at how people use it in tool form after this first go-round. Thanks again for feedback, it helps. -- ForteTuba 00:52, 20 January 2006 (UTC)
Dan, I discovered your Suggestbot project from another user's talk page that happened to be in my watchlist. That user Yally Bammy hasn't edited recently because he vandalized pages, was subsequently blocked for it, and is very unlikely to return. In your selection criteria, you should probably run a check for blocks via the block log, or warnings on the talk page before your bot wastes cycles on your full routine for suggesting articles. Although I think the intent of your bot is great, it is too trusting in assuming good faith for everyone who registers for Wikipedia. Unfortunately, some users have proven their intent otherwise. Other than that, keep up the good work and good luck with your effort. AUTiger ʃ talk/work 17:42, 26 February 2006 (UTC)
Thanks for the note, replied on your talk. -- ForteTuba 21:43, 26 February 2006 (UTC)
I haven't been through all the suggestions yet, but a few things were there that needed tuning-up. Not a total loss (especially with the entertainment value I've gotten out of Charmmy Kitty, to which there must be about ten new links this morning due to me and the "bot"). Lol. Anyway, I'd be interested in helping with suggestions and will give it some thought.
In fact, we may be on to something controversial, as someone has suggested on the talk page of Charmmy Kitty that it may be a front for endoctrinating youth about the benefits of slavery. Cheers! --DanielCD 20:03, 8 March 2006 (UTC)
I noticed you were amused, glad I could help. -- ForteTuba 22:21, 8 March 2006 (UTC)
I came here to thank you for the suggestions and somebody beat me here. I wanted to make sure that you knew we were just kidding around. Suggesting that I edit CIA, Pres. Bush 41 and 43, Osama bin Laden, Abortion struck me as funny since I have been trying to decrease my wikistress. Coupled with the priceless suggestion of Charmmy Kitty to DanielCD, we had a great laugh. Charmmy Kitty will not be neglected for sure. The reproductive organ stubs were great suggestions, by the way. thanks FloNight 15:29, 9 March 2006 (UTC)
Knowing that it recommends controversial things was actually really useful to me, I'm going to look at the part of the algorithm that caused that and try to tone it down. Why recommend things that everyone's editing? And, I really am glad that Charmmy Kitty is causing great amusement. Fun is good. -- ForteTuba 18:20, 9 March 2006 (UTC)
It also found some Psych articles I wasn;t aware of that need some tweaking. I didn't mean to imply that there was no use here. Some of the suggestions are good, like some of the disorders and such. Thanks again. --DanielCD 22:30, 9 March 2006 (UTC)
No such implication taken, and you're welcome. -- ForteTuba 22:42, 9 March 2006 (UTC)
An creative bot that will undoubtly become an asset to the community and wikipedia's tool cache. Here's hoping it improves to its highest potential and brings me more articles to edit! -ZeroTalk 16:10, 13 March 2006 (UTC)
Heh, thanks. -- ForteTuba 18:14, 13 March 2006 (UTC)
Thanks for the suggestions. One issue that was pretty apparent was that 11 of the 34 articles suggested where actually started by me, and I have edited one of the others as well. Sure, it's worth being reminded "you should flesh out the stubs you made", but I would think that it might be more useful to exclude articles that one had already edited. Guettarda 06:06, 15 March 2006 (UTC)
Thanks for the note. Right now it only grabs your most recent 500 edits in an effort to be nice to the Wikipedia servers. I'll arrange for it to look at a dump as well, to get rid of many older edits. It still, alas, won't be perfect -- dumps go out of date fast -- but it'll help. Thanks for letting me know it was a big enough problem to be worth doing this. -- ForteTuba 13:04, 15 March 2006 (UTC)
Also, I'm posting a second set of suggestions over your way. If you could, let me know if these feel better. -- ForteTuba 16:21, 15 March 2006 (UTC)
Much better. Thank you very much. Guettarda 19:31, 15 March 2006 (UTC)
Cool! It found an article that needs to be merged with one I already worked on, so I'm a perfect candidate for that! It put it on the general stub list rather than merge suggestions, for whatever reason (probablhy didn't realise that they were the same subject)... but I might not have spotted that one on my one. I slapped merge tags on both and will get to them tonite if I have time. Thanks. SuggestBot! ++Lar: t/c 14:14, 16 March 2006 (UTC) (I watch, replies can go here...)
Sweet, it's good to know it does some good things. :) Thanks for the positive vibe. Feel free to tell friends, I'm trying to get enough people to use it to figure out which recommendation strategies are better. -- SuggestBot 16:13, 16 March 2006 (UTC)
Can you make it smarter about suggesting merge candidates? I dunno, that may be a hard problem. Scattered bits that are stubby and need merging are one of the big areas that humans can fix better than bots IF they know they are there... ++Lar: t/c 20:20, 16 March 2006 (UTC)
Not sure what you mean. If you mean, could it find articles that should be merged, that'd be cool but it's not really the mission. The goal is to recommend pages that someone has already put a cleanup-type tag on (like marge). And since there aren't that many articles with a merge tag, it probably makes mediocre recommendations in that category. Does that answer your thought, or can you clarify? -- ForteTuba 21:39, 16 March 2006 (UTC)
That's OK, I am not sure what I mean either. Smile. Suppose that you could construct a big list of potential merges (yes, not quite in your mission exactly) but not yet marked as such. But *then* use that list against articles that particular people are interested in to suggest maybe there are merge candidates (first list) that a particular person would be well suited (second list) to judge. Most of the time when I see a merge candidate already tagged, I have no idea whether it is or isn't because it's out of my area of expertise. Most of the time when I see a stub, I have no idea there are other stubs out there that might be good merge candidates. intersect those two things together and??? But ya, maybe too much intelligence to ask for. ++Lar: t/c 22:44, 16 March 2006 (UTC)
The general problem of automatically finding articles that might be (stubs, to be merged, needing cleanup) and then asking people to verify them is pretty interesting. Nickj's linkbot does something like that for suggesting links to add to pages; something like Special:Shortpages could be used to identify stubs. At one point I thought that a way to filter recent changes so that it only showed things you were more likely to be interested in could be a killer app... but maybe watchlists are close enough for that? -- ForteTuba
You probably already know this, but your bot doesn't seem to handle redirects well- one of my suggestions was the Lost Twenty, which is a redirect to Jedi. --maru(talk)contribs 17:48, 16 March 2006 (UTC)
Yep, it's totally stupid about redirects for now. I should probably look at the recs and see how bad the problem is. Thanks for the note. -- SuggestBot 18:18, 16 March 2006 (UTC)
Just doing a preliminary lookover of the list I got and it looks perfect. I'll get to it when I'm done with homework and my head doesn't feel like it's throbbing. Keep up the good work :) — Ilyanep(Talk) 22:26, 16 March 2006 (UTC)
Heh, thanks. Good luck with your head. -- ForteTuba 14:15, 19 March 2006 (UTC)
Your requests page says that I recieved recommendations, but I can't find them. Can you help me out? WriterFromAfar755 01:13, 19 March 2006 (UTC)
Sigh, I mistyped your name when I ran the bot script. I'll get you next time I run it. Sorry, thanks for the note. -- ForteTuba 14:27, 19 March 2006 (UTC)
Just wanted to come on and thank you for the suggestions from suggest bot looking forward to moseying through some of the pages. As I said before its a great idea and I'm sure I won't be the only one who finds it useful. Best of luck in your endeavour -- Shimirel(Talk) 00:54, 15 March 2006 (UTC)
make it an rss feed. i can knock most of the stuff suggested here out today if i have the time. will i have to wait a couple weeks for the next list o' stuff? why not just make it a live list. e.g., if i just finished one of the merges on my list (i did, thanks), why not suggest another? seems like algorithms here are probably cheap. ... aa:talk 01:01, 17 March 2006 (UTC)
Eventually, when it's slicker, I'd probably give people an option to get a list updated every so often. RSS would be cool for some versions (i.e., something that monitored recent changes and would tell you about articles that were in your top N recommended articles might be a lot of fun), not sure it's the right tech right now. I do like the idea of making it easy to keep getting recommendations, though. Thanks for the feedback. -- ForteTuba 14:14, 19 March 2006 (UTC)
In the meantime, if an RSS feed is a pain in the butt, you could simply make it a template. That way when you updated the links, they would be automatically updated for me on my page. And mind you by "template" I mean "transcluded material". So {{User:SuggestBot/Suggestions/Avriette}} and the like. Thanks again. And, I'm mostly out of articles to edit from my suggestions. :) ... aa:talk 03:45, 26 March 2006 (UTC)
You know, I didn't even think about that as a way to put suggestions on someone's talk page. I'm substing in a template (it seemed like a lot of descriptions of templates were down on transclusion because of server load. One thing, I'm not sure I'd want to automatically update suggestions... people might want to keep them around until they're done with them. *ponder* Could automatically update once a month, and then have some way to force a re-update for the frequent editor such as yourself. The only trick is reminding people the suggestions are there... wouldn't want to post even a transcluded template once a month? I don't know how possessive people are of their talk pages, and where it would start feeling like Suggestion Spam. Still, an interesting question: how to notify and keep suggestions fresh. Congrats on finishing off your set, by the way. If you want more and haven't already, feel free to put your name back on the request list. For now, it'll dump another load of suggestions on your page, but that's okay with me if it's okay with you. -- ForteTuba 03:13, 27 March 2006 (UTC)
Great job on the bot and thanks for the great idea and hard work implementing it! I hope the following suggestions/comments on the bot's recent suggestions for me will be helpful and constructive.
Generally speaking, I found 11 of the 34 recommendations useful.
That's useful feedback (though the gold standard is how many people actually work on!)
Why pick articles at random for clean-up? Surely that duplicates already existing listings/tools? Would it not be more useful to refer people to articles that need improvement but about which they have knowledge/an interest? In other words, SuggestBot could become an extremely useful alternative to existing methods of raising the status of articles needing work. Finally, some recs will look random even though the recommenders are trying hard because there aren't that many articles tagged, e.g., style.
Partly because random helps tap into parts of the work-tagged pool that would otherwise never get recommended. I also need to see whether the smarter algorithms do any better than random: is the effect good article-picking, or just making personal assignments?
I'm not familiar with the technical difficulties here, but perhaps the number of recent edits scrutinised by the bot should be increased? Specifically, if someone does RC patrol for a while, the bot would recommend things they're not interested in or knowledgeable about. Increasing the number of edits looked at above 500 would possibly provide a better sample. Moreover, is it possible to exclude minor edits from consideration (i.e. to exclude rvv edits)?
Minor edits turned out to be unreliable in offline testing -- not enough people mark them. I do throw away any edits that have rv and revert in the comment, which helps I think. 500 is the number because I have to fetch them live from the server (dumps go out of date fast) and I'm trying to be a little bit friendly on resource consumption. In a perfect world, I figure out a better way figure out what people are interested in. Maybe looking at talk page edits, or extracting any links from their user page -- people often list their favorite contributions -- or letting people specify specific articles in some way. More work for other people, though, equals harder-to-use tool.
I'm far from being technically proficient so I'm not exactly sure what technology the bot uses, but is it not possible to use collaborative filtering like Amazon.com does? If so, an option to tell the bot what you are and are not interested in would aid the bot's work immensely. (Don't know whether this could be done on our servers though or what the issues are. Just a thought...)
One step ahead of you here, one of the algorithms is exactly collaborative filtering based on co-editing: find people who edit the same articles you do, then suggest articles they've edited and you haven't. (Aside: I work in one of the labs that co-invented collaborative filtering, see http://movielens.umn.edu . This happened a long time before I got there, but we do tend to use it a tool... it's one of the reasons I thought I could do something useful on Wikipedia.)
Nice feedback, thanks for the thoughts and ideas. -- ForteTuba 14:26, 19 March 2006 (UTC)
Great! Thanks for your response. I'm glad you found my comments helpful. Please keep up the good work! Regards, Mikker... 15:53, 19 March 2006 (UTC)
Sorry to butt in on this discussion: I don't think users' inconsistency in marking their minor edits should deter you from getting your bot to ignore them (or what about optionally ignore them)? In particular, admins using the rollback tool will always have rollback edits marked as minor! I know my rollback edits, which are pretty much scattered over the whole wiki, will skew any suggestions your bot makes. Nevertheless I will still be fascinated to see what SuggestBot comes up with! Best wishes - keep up the good work! --RobertG ♬ talk 16:57, 28 March 2006 (UTC)
Butting in is fine. Optionally ignore them seems like a win in the long run, when there's a nicer interface for getting suggestions. I kind of hope people would try them both. :) -- ForteTuba 01:39, 29 March 2006 (UTC)
Just a quick note to say thank you for the suggestions! I've left some quick Yes/No feedback on the individual suggestions on my talk page; In my particular case you might want to give less weighting to minor edits, but I'm an extremely atypical person to make suggestions for, so I'm impressed that it did as well as it did! -- All the best, Nickj(t) 01:15, 21 March 2006 (UTC)
It is a poser, what to do about the minor edits. A lot of people are inconsistent in marking them, and both offline and eyeball tests suggested that whacking them didn't help. But... maybe I'll make a run with minor edit filtering turned on anyways and see what people think. Thanks for the feedback. -- ForteTuba 15:25, 21 March 2006 (UTC)
Nifty tool. I'd like to echo the request for major edits only. Thanks for the suggestions! :-) Mindspillage(spill yours?) 02:33, 21 March 2006 (UTC)
You're welcome. Maybe I'll send along a set with minor edit filtering in a week or so and you can see if you like them better? -- ForteTuba 15:26, 21 March 2006 (UTC)
This is a very interesting idea, not particularly useful to me but I'm sure there are many people out there who will benefit from it. It's nice to see that it suggests a range of tasks rather than just stub-expanding, for which there will inevitably be more articles than anything else. I requested a set of suggestions mainly out of curiosity – my last 500 edits (and indeed most of my previous edits) were spelling corrections to what was essentially a random selection of pages.
The algorithm works well, however, as many of the suggested articles are things like biographies, military articles and fiction-related articles – looking back through my corrections, they seem to be the most common types of article I have edited – while on the other hand, there are some articles that appear to be more or less random. (Whether contributors to these articles have bad spelling or whether there are just more of them, I'm not sure) -- Gurch 10:15, 22 March 2006 (UTC)
I'm hoping there are people who will benefit. What I'm really hoping is that once it's working well we'll be able to advertise it to newer members as a way of hooking them in more tightly. As for the random, you're right -- there are some random recommendations thrown in, both for coverage and for testing that the good algorithms work better. Thanks for the note, glad it seems to work plausibly for you. One question: if you were able to specify a set of articles that better represented who you are, would you take the time to do it? -- ForteTuba 19:49, 23 March 2006 (UTC)
Hi, ForteTuba. Thanks for the suggestions. Most were good, though they are really heavy on the nuclear stuff. I'm not sure where floristry, purdah, sex segregation and battle dress uniform came from. Geraniol is the only other non-nuclear article and is an okay suggestion. Also, SuggestBot suggested klystron tube for expansion, but the expansion tag was removed on 3/21, well before the suggestions were made. To improve the results, I think it would help to review more edits. I was cleaning up a lot of nuclear related articles recently, so they dominate my results, but they are only a small portion of the total articles I have worked on. Another way to improve results would be for editors to tell the bot what they are interested in or give it some keywords. Thanks, Kjkolb 07:38, 28 March 2006 (UTC)
Thanks for the feedback. I took a peek at what the recommenders were doing, just for fun. You're right: they totally picked up that most of your recent edits are nuclear-related and that dominated the results. The ones you mention near the top were all suggested by a link-following algorithm (but then, so was Geraniol. I've been hesitating to download more edits because a) it's a load on the server, and b) the recommenders slow down if they get too many items. I wonder if it would make sense to get more edits, then take a sample, every Nth edit, over a longer time, in an effort to get a more balanced profile. The tag removed, I believe: I downloaded the lists of tagged articles back on March 15. That's one of the perils of running off of dumps and old data, away from the live servers. But, I'll mark it down as a bug. Finally, with keywords, there's an effort/usefulness tradeoff. I think people could get really nice recommendations if they were able to manually specify their profiles... but it's more work for them. That's in the long-term plans, though, a way to send in a set of articles that represent your interests. Good feedback, thanks again. -- ForteTuba 13:59, 28 March 2006 (UTC)
Wow, very scarily on-target! I had expected some oddities because when I am bored I wikify, clean up syntax, linkrot, and random typos, and I don't always remember to mark minor edits...
The recommenders are influenced by things you do more often.
Reading these makes me realise how parochial I must be in my overall editing: four suggestions are within about five miles of me geographically.
Remember, this means that _other_ people are parochial, and you're getting these suggestions because of patterns of editing and linking in articles.
It would be quite cool to be able to say "SuggestBot, I am fed up of editing the same old stuff: take out anything connected with This Topic (whatever it is), and tell me what else I can do". I am curious whether it would then find me more on the area which (I think) I go to for a change of scene.
Heh, funny. We've talked about doing something like this for research papers, using algorithms that are good for recommending for specific tasks, like changing fields. It would certainly be technically possible to specify a list of articles that represents one's interests, only a little bit of work for me, but quite a bit more work for customers.
Another thing I wonder is whether it can suggest things for groups of people. The WikiProject sort of things. Many of those and many country-based Wikipedia noticeboards have lists of (purportedly!) relevant stubs and requests for expansion which have to be updated by hand. Would it be possible to take the contributions of the people who sign up as members of those projects and produce a "Members of this noticeboard may be interested in.." list?
That's pretty clever as well. User:Pearle automatically updates the [{Template:Opentask]] list. One drag with SuggestBot is that it would be liable to re-recommend articles over and over. Again, fixable.
Anyway, overall, very interesting, and I think the bot is right: I may well enjoy editing some of those articles. I shall go and find out.
Cool. Thanks for the feedback, a couple of good ideas here. I really like the groups idea, I'll have to put it in the back of my head. -- ForteTuba 14:08, 28 March 2006 (UTC)
This bot is AMAZING! I am a space exploration geek and a lot of the suggestions are in that topic! --ExirKamalabadiJoin Esperanza! 10:35, 8 April 2006 (UTC)
It tries. Glad it looks reasonable, hope it turns out to be useful, thanks for the feedback. -- ForteTuba 12:38, 8 April 2006 (UTC)
The suggestions given to me too are very accurate, thanks a lot! Keep up the good work --lightdarkness (talk) 14:15, 11 April 2006 (UTC)
The Bot says it has made the suggestions, but it hasn't at all. No contributions for me appear in its contributions list and I have received no message. MyNameIsNotBob 21:36, 10 April 2006 (UTC)
Well, that's a good question. I tried to run a big batch overnight and it apparently conked out about halfway through (though, without the normal error messages I look for). Thanks for the note, I'll look into it. -- ForteTuba 00:34, 11 April 2006 (UTC)
I don't believe it is categorising the articles according to their appropriate maitenance tags somehow. It has given me a list of stubs, but at least two of the stubs are rather large articles without stub tags. MyNameIsNotBob 21:54, 11 April 2006 (UTC)
It makes mistakes, especially on stubs, because it works from out of date info. I download categories other than stubs a couple of times a month (to reduce load). On stubs, it's worse, because the tool that puts that list together right now only works from full dumps, and the last successful such dump was Feb 19. Sigh. It would not be too hard to get it to extract stubs from the most recent "current" dump, but even that's a couple of weeks out of date. Sorry that's not working well for you. -- ForteTuba 23:34, 11 April 2006 (UTC)
A very nice bot indeed now that I've tried it, kudos. I do have a few suggestions for improving it:
Make the bot look at info on a user's user page (if it doesn't do that already), in particular the language or location category a user is in. I'm suggesting this because the bot suggested a lot of U.S. related stuff for me (probably because articles about Russian equipment are generally edited by Americans so their editing pattern somewhat matches with mine).
Not sure where it would look, could you give me more info? Are these the "Wikipedia Babel" language-level things? I'm also not sure how I'd use that info to limit the articles recommended, any suggestions?
Yeah, I mean the Wikipedia Babel boxes. The bot could perhaps use those to group users like it groups users by editing pattern. That is, assuming there's a correlation between the editing of country related articles and native language in the Babel box.
I'm not sure I'm a believer -- if there's a correlation between those two things, there'll be a correlation between those users anyways, without the Babel box. There is a more general case of what you're saying, though, using Wikipedia categories to help out with recommendations. That could be powerful.
Possibly add suggestions for articles to create. These can be gathered by looking at redlinks.
That's a poser, but you're not the first person to suggest this. Only one of the algorithms (at least, as of now) has a shot at finding those -- the other two look at existing pages. I can imagine ways to do it, but the best ones require an up-to-date database and some fiddling with that. How popular do you think this would be, assuming it could be done in a reasonable way?
I think it could be popular, since after all.. what's more fun than creating an article? But then again, that might just be me.
Good enough for me. Next time I get to do some coding I'll see if there's an easy way to hack that in. If I can download a list of all uncategorized pages I can at least let the one recommender take a crack at them.
Would it be possible to program SuggestBot to ignore minor edits I've made and pages I've only edited once or twice? Most of its suggestions were good, but some of them related to topics on things based apparently on the fact that I've corrected a spelling or removed a dead link in article otherwise of no interest to me. Thanks! Angr (talk • contribs) 14:27, 12 April 2006 (UTC)
The general idea of trying to figure out what really matters to someone is pretty interesting (and your suggestions are ways to get at that). I may try running it for a while with/without minor edits to see which one does better. In offline testing against a dump removing minor didn't help, but it's low cost to add. Long-run it would be nice to make all these kind of suggestions as options that people can play with in order to tune things the way they want. As for articles not edited often, it's not as clear that's a win, especially because it only picks up 500 edits (in an effort to reduce server load). Thanks for the suggestions. I think I'll have to try the minor removal soon just because so many people think it'll help. Hopefully they're right. -- ForteTuba 20:09, 12 April 2006 (UTC)
Fantastic recommendations. Minor edits notwithstanding (and they are the overwhelming majority, for I am an omnivorous reader and compulsive spelling-and-punctuation WikiGnome), SuggestBot homed right in on two or three of my most outstanding interests and nailed the parts that need work. But people keep asking you if ignoring minor edits wouldn't improve things. My informed intuition is that it won't, and it may even hurt.
Tell me: If you plot, for some individual, degree of interest versus rank of interest, over a whole bunch of topics (whether those are encyclopedia articles, terms used in their library research, or books they've looked at on Amazon), what will the curve look like? I strongly suspect it's a "long-tail"-type power law. So yeah, most of their minor edits are part of the tail—so are most of their nonminor edits. Both contain extractable information.
Or suppose you're examining a corpus of recorded behaviors, in which a certain pattern is crudely evident—crudely, because it's rare. Does increasing the corpus size by a modest factor (less than an order of magnitude) improve the resolution with which you can see patterns? Even if the added material is complicated, noisy, and sometimes irrelevant and/or just plain irregular? If the behavior in question is language, heck yeah; that's within my specialty. But I'd guess it's also true for other behaviors you might study, including editing preferences.
Or, put a third way, you are trying to model the process of choosing to edit an article, based on actual behavior, but to provide the model with vastly more information than a human can use. Exemplar-based modeling, with explicit storage of and reference to bulk data, is also within my specialty, and here again performance is predicted as much by the breadth as by the depth of the data, and not so much by its messiness or cleanness. Moreover, the fewer intermediate levels of explicit smarts you try and build in, the better the model performs. Which is to say, an exemplar-based model is an empirical automated pattern abstractor, so apply it to your data and let it automatically abstract; and the less you impose your preconceptions, the more helpfully empirical it's gonna be.
Wow, that was a bit of a long spiel. Bibliography available upon request, but be warned that (a) it's mostly linguistic in character, (b) it won't be as complete on the corpus linguistics stuff as on the modeling stuff, and (c) a few of the items consist of my name or my advisor's name, followed by (unpublished fooling around), and that's it. Anyway, I'll be keeping an eye out to see when you do conduct a minor-ignoring run, and see what becomes of my predictions.
(Ooh, ooh! Idea for a blinded side-by-side test! During a test period, run the bot both ways, and give users both sets of suggestions, privately recording the order you presented them in and the date. Also present a short notice asking users to determine which set looks more helpful, click through to a feedback page, and sign either the left or the right list accordingly. There you've got everything you need to determine it objectively. Er, polysubjectively. Whatever.) eritain 07:07, 13 April 2006 (UTC)
Sweet that the recommendations worked out for you. As for the minor edits issue: as far as I know, the jury is still out on the question of figuring out what's important, but I agree that relevant algorithms are pretty good at ignoring irrelevant data. In collaborative filtering, at least, I've done a little unpublished fooling around on trying to weight people's ratings by various statistical criteria of the items they've rated. No joy, at least for the obvious ones like popularity or disagreement on an item. The metric a pretty standard CF metric, mean absolute error between held-out ratings and predictions for those. It's not a great metric, but I haven't hade time to fool around recently with a dissertation on the horizon. It seems like it might be more useful to weight/ignore certain events when you're trying to use implicit ratings (e.g., does editing an article mean you're interested in it? Does viewing the web page for a book mean you'd probably like it?) or segment someone's interests (e.g., this set of books are gifts I've bought for my parents, they're not really my interests).
The nice thing, as you say, is that you can do empirical studies by running various configurations and testing them. That's how I decided that SuggestBot was worth pushing on: the recommendations, primitive as they are, did a lot better than random. That suggested it can be useful. I've already been comparing various versions of the recommendation algorithms, and a few simple ways of filtering the edits that are chosen. I'm not asking people to do A-B tests on the lists. I think that has some value, especially for poking around a big design space to find promising areas, but I'm partial to real outcomes, in this case helping people edit articles. I'm hoping once I feel pure about the dissertation that I can invest a good amount of time into this.
Thanks for the nice note. I'd surely take pointers you have to interesting related work. -- ForteTuba 13:25, 13 April 2006 (UTC)
I know you're a bot, but i'd still like to thank you! Most of your suggestions were spot on, but all of them in the general vicinity of my interests. Good job! The Minister of War(Peace) 16:34, 15 April 2006 (UTC)
Great, glad you like it. Good feedback is always good to hear. -- ForteTuba 12:44, 18 April 2006 (UTC)
Thanks all the same. Radagast 18:53, 25 April 2006 (UTC)
Sure thing. The suggestions get generated by algorithms that don't know what the heck is going on, so you get some clunkers. This happens especially in the categories that have a lot fewer articles -- less chance of seeing anything that looks relevant. Hope you like them and you find some useful things to do there; thanks for the feedback. -- ForteTuba 17:00, 28 April 2006 (UTC)
Thanks for the list most of the articles mentioned will get some attention from me, though two stubs initially seem out of place predominately I work on geo-articles of/for Western Australia, yet I get one on a New South Wales road stub, and Marla, South Australia geostub curiously I have physiclally been on that road and through Marla though I hadnt visited either article, does Suggestbot know more than we are led to believe. Gnangarra 15:51, 27 April 2006 (UTC)
Its spies are, in fact, everywhere. :) It tries to pick up on patterns, so if people who are like you had worked on those articles, there is some change SB would figure it out for you, too. So, glad it's working for you, and that you'll be making these things better. -- ForteTuba 17:03, 28 April 2006 (UTC)
Useful to know. It doesn't update its lists often (especially for stubs -- that's based on a February dump) but enough people are sad about it that I just have to do something. Thanks for the note. -- ForteTuba 23:18, 29 April 2006 (UTC)
For some reason, SuggestBot thinks I really love cryptography. While I edited a few crypto articles in the past, I think at least half of the suggested articles were on crytography and I don't know enough about the subject to make many meaningful edits to them. Perhaps more variety would be helpful. Peyna 17:11, 29 April 2006 (UTC)
That's interesting. It's not easy to enforce variety -- it's based on looking for patterns and, for some reason, there were a lot of cipher-related links attached to some of the articles you had edited. (The cipher articles were all recommended by a link-following algorithm.) You can imagine forcing it to pick articles from different categories, etc. Thanks for the note. -- ForteTuba 23:27, 29 April 2006 (UTC)
Thanks for the suggestions, it really helps to know which Pokémon articles need work without sifting. A couple of things though -
I seem to have at least 5 articles on "business executives".. no clue really, I was watching Hustle last night, but I don't think Suggestbot is that good.
I was linked to a minor social liberal political party in Austria? Austrian politics isn't exactly my cup of tea.
I went ahead and moved the name over, and tried to make it clear that requests should go there not here, thanks for the note. -- ForteTuba 02:19, 8 May 2006 (UTC)
Would it be possible to find out *why* something specific has been suggested? Or would that be too complicated? I'm frankly intrigued by some of the suggestions I recieved. UnDeadGoat 23:00, 1 May 2006 (UTC)
We don't have great explanations for specific recommendations. I can go look at the log to see which algorithm generated which recommendations. Exactly what caused the recommendations is a lot harder because the algortihms combine info based on a lot of edited items. Which specific item caused you to be similar to some other Wikipedian, and get suggestions for other articles they edited, hard to say without a lot more logging than we do now. -- ForteTuba 01:01, 2 May 2006 (UTC)
Would it be possible for SuggestBot to recommend certain types of work depending on the user's preference? For example, when I requested a list, I got lots of stubs, and a few cleanups, merges, wikifies, unreferenceds and non-stub expansions, which is obviously intentional. Thing is, I'm not the best at writing or researching, so most of those stubs are going to get crossed off with my list due to lack of interest and/or expertise to expand them. I would quite like it if you could make a request for a suggestions list that consisted mostly of copyedits, cleanups or even NPOV problems. --Sam Blanning(talk) 13:41, 2 May 2006 (UTC)
The main problem with copyedits and NPOVs is that there are "only" a couple of thousand articles in each category, at least last I looked -- and recommendations looked pretty skanky when I was trying to do them on a list that small. I also didn't want to throw giant lists of suggestions onto people's pages. Maybe next time I do a little coding on the bot I'll try adding them just for fun.
As for preferences, it would be a moderate amount of work. I want to make the tool more useful, and an interface that let people specify what their interests are more accurately than just taking edit history, how many suggestions to get, etc., is on my list of nice-to-do. Adding requests for specific kinds of work is a pretty good idea too, thanks. Unfortunately, I'm not going to be doing any major work on the bot until either I'm done with this pesky dissertation, I get an account on the toolserver (it's a giant pain to work with a standalone copy of WP), or most probably, both. Sigh. Good idea from you, though, and I'll keep it on the list for when I do more active development. -- ForteTuba 00:40, 5 May 2006 (UTC)
Thanks for your suggestions, they are quite fun. There were some suggestions that surprised me a bit, but I would have expected an even wilder mix given my random editing style. However, Bastille Linux sticks out a bit, because this is an article that I already repeatedly contributed to a year ago, nevertheless it's still a stub that needs work, and it's a topic I'm interested in. Thanks again! Nikai 11:48, 5 May 2006 (UTC)
Cool, glad you like them. Did you do the work on B.L. under a different name or an IP, or is there maybe a bug lurking about? SuggestBot tries to be sensitive about not suggesting already-edited pages (though as the last full dump goes further and further out of date, it gets harder.) -- ForteTuba 02:17, 8 May 2006 (UTC)
No, I edited under the same account, in February/May 2005. So either your dump is older, or SuggestBot didn't see me. Cheers, Nikai 15:06, 8 May 2006 (UTC)
Useful to know, thanks for that info. -- ForteTuba 21:25, 11 May 2006 (UTC)
I've been told to Wikify Farmers ditty twice, despite its nonexistence, and I have no idea what it is or why it was recommended . . . UnDeadGoat 19:57, 7 May 2006 (UTC)
Useful to know, thanks. I haven't had suggestbot update its list of pages-in-categories for quite a while now, I'll run it overnight. -- ForteTuba 02:14, 8 May 2006 (UTC)
Now, this sounds cheesy, but SuggestBot helped reunite me with an old friend, by suggesting they edit an article, which they did edit, which I saw the history of, and saw their name. Thanks! --Bobadot 20:51, 8 May 2006 (UTC)
Very cool. It would be fun to think about tools for bringing people in WP closer together, there's so much social going on under the veneer of what looks to casual outsiders as a big encyclopedia-building project. -- ForteTuba 21:26, 11 May 2006 (UTC)
10 for 34 on articles I'd have worked on. Since I had only two weeks of history at the time, that seems solid to me. 4 I destubbed, 2 I improved but not beyond a stub, 2 I couldn't Google anything to add, 1 already non-stub, and one merge that I decided was inappropriate and eliminated the suggestion for. Details are on my talk page if you care to look. I'll probably wait a week or two before asking for another hit, so my edit pattern won't be so largely based on the last round of suggestions. GRBerry 01:31, 12 May 2006 (UTC)
Glad they were helpful to you. I hope to check out the details sometime in the not-too-distant future, it should be useful. Thanks for the feedback and for doing some that were good for you. -- ForteTuba 00:09, 14 May 2006 (UTC)
Hi
Thanks for sending the SuggestBot stuff - Yes, it was fun.
I've prepared some feedback for you, in case it's helpful. Please visit User:Ballista/SuggestBot. I would be grateful for feedback on the feedback, if you have time - just stick it on User talk:Ballista or on the User:Ballista/SuggestBotpage. Good health and fortune. Ballista 06:34, 10 May 2006 (UTC)
Just came back to correct misprint & to say 'Hi!'. Ballista 17:08, 11 May 2006 (UTC)
Come back again and copied this lot over from your 'Talk' page; I'd forgotten you had a specific page for the job! Ballista 03:01, 12 May 2006 (UTC)
It'll probably be a little while before I go into detailed feedback, but I'll check it out next time I have some real time to work with SB. Glad they were fun for you, and thanks for the feedback. It will help in the long run. -- ForteTuba 00:11, 14 May 2006 (UTC)
Hope this helps improve SuggestBot. Cynical 22:03, 13 May 2006 (UTC)
Those are useful. I thought I had duplicates removed, but perhaps not--you're the second person to report that recently. Next time I've coding time I'll look at that. As for the redirect, that's unfortunately a known issue that I just haven't prioritized enough to deal with. It should know better, though. Thanks for the feedback, hope some of the suggestions were useful. -- ForteTuba 00:11, 14 May 2006 (UTC)
The bot gave me some good links to look at, although about 1/3rd were brain-related, which I can only assume had something to do with the fact that several of my recent edits were to change a bunch of redirects based on the Nintendo DS "Brain Training" game. Besides that, it did a pretty good job given it didn't have a large list of contributions to work from. Confusing Manifestation 12:14, 26 May 2006 (UTC)
One of its recommendation strategies is text-based matching, very search-engine like. I can easily imagine how it would get overwhelmed by seeing the word "Brain" in multiple article titles. Sometimes this works great at finding useful similar articles, other times not so much. Thanks for the feedback. -- ForteTuba 15:57, 26 May 2006 (UTC)
First off, SuggestBot is one of the most innovative Wiki Bots I've seen. Second, I enjoyed fixing some of the articles suggested. One suggestion, though, is that (because I do a lot of vandalism reversion) perhaps minor edits, small edits, and edits marked with "rv" should be ignored when determining articles. Otherwise, you end up with some crazy suggestions. Porphyric Hemophiliac§ 18:39, 26 May 2006 (UTC)
Thanks for the nice words, it felt like an interesting thing to try and I'm glad you got some value out of it. As for edits being ignored, it tries to ignore "rv" edits. Minor edits didn't seem to help in offline testing, but enough people have asked for them that when I build the new, improved SuggestBot, I'm going to have an option for filtering them that people can try with and without. It's a good idea--I've noticed that people who do vandalism reversion get a lot of bogus stuff (in particular, they tend to get popular/controversial articles). Thanks again. -- ForteTuba 17:08, 27 May 2006 (UTC)
I notcied that you've left messages for Cool3 on his talk page, so I thought you mihgt want to vote in his RfA at WP:RFA. ShortJason 15:13, 27 May 2006 (UTC)
This is a bot account that posts suggested articles to edit; I don't generally have special insight into the people who use the bot and don't feel plugged-in enough to the admin process to have useful opinions to contribute. Thanks for the invite, though. -- ForteTuba 17:18, 27 May 2006 (UTC)
Thanks for the suggestions - mostly inpiring, even if somewhat haphazardly. Presumably because I've done so much of the work on the Handcolouring article, I ended up with a bizarre array of articles with the word "hand" in them: :Hand tool, Hand jive, First hand, Right Hand of Doom, I Kiss Your Hand, Madame,...! But I find this slightly random element fun, and it leads me to some subjects that I otherwise might not have looked at. Thanks again. Pinkville 20:40, 30 May 2006 (UTC)
Yeah, the text-based recommender has a bad habit of getting wrapped around an axle, but I'm glad you got something out of it. Thanks for the feedback. -- ForteTuba 02:42, 31 May 2006 (UTC)
What language is this written in? I'm a basic-level programmer, and I'd be interested in looking at the source if that's okay with you. --Mathwizard1232 04:39, 1 June 2006 (UTC)
Hi
Enjoyed the experience of working with SuggestBot - very challenging in places. Some suggestions very appropriate, one or two a bit wide of the mark. Detailed feedback on User:Ballista/SuggestBot - Thanks - Ballista 20:18, 3 June 2006 (UTC) (P.S.: I'm queuing for another go!) - Ballista 20:20, 3 June 2006 (UTC)
Sweet. I am still intending to follow up on your comments, I just have finishing a dissertation by July 1 as a #1 priority right now. -- ForteTuba 13:19, 20 June 2006 (UTC)
Just had a look at the latest bunch of suggestions - mostly good stuff, thanks. My own fault - it sent me Uusikaupunki back again! I'll rootle through these, as I go, interleaving it with my other editing agenda. As before, I'll provide feedback when finished but I'll try to do a running feedback, this time, on User:Ballista/SuggestBot. - Ballista 07:58, 11 June 2006 (UTC)
...Because I do a lot of work on disambiguation and RC patrol, but always worth a try :o) E Asterionu talking to me? 10:02, 11 June 2006 (UTC)
True enough. It tries to be smart about rv edits, but not very smart. Someday I want to let people give a list of articles that truly represents them, rather than just what they've edited -- but it's so easy for potential users to just give their name... -- ForteTuba 13:20, 20 June 2006 (UTC)
Thanks very much for this - for the most part, my first batch were absolutely perfect. Is there any reason it doesn't do non-existent articles that need writing, though? Rebecca 03:06, 12 June 2006 (UTC)
The recommender looks at article text and edit histories as ways to find things you might like. Problematic with nonexistent articles, although I can imagine doing a different kind of text matching for that. Several people have asked, enough that it might be fun to do. Thanks for the postitive feedback and the suggestion. -- ForteTuba 13:18, 20 June 2006 (UTC)
Very good bot! very good idea! Suggestions for improvement, have a filter for articles that should not be used to figure out what I'm editing. Filter on Vandalproof, rv, Revert to revision and all other simmilar texts that is given when doing vandal reverts. Second suggestion, make me be able to run it, I.e. I can click a link that will generate a new page now when I want it, not sure what you need to run this thing, it would be best if it was run on one of the wikiservers but and not sure how slow it is, but still I think it should be part of wikipedia to have a [[special:Suggest/user]] . Stefan 11:43, 12 June 2006 (UTC)
It tries to be smart about rv, but not too smart. (Better, probably, to just let peope paste in some text with wikilinks if they want to be more precise.) I also really want it to run live -- I tried to get a toolserver account but was ignored for months. So, for now it is what it is. But I agree that giving suggestions right when people are interested would be killer. Thanks for the ideas. -- ForteTuba 13:22, 20 June 2006 (UTC)
Mine were a real mixed bag... But then again, I usually don't spend much time in one area except for my translations. That's what's so odd about it: It was mostly correct in giving me a lot of political articles, especially relating to Brazil and the U.S., but it also gave me Mermaid Man (I rarely edit fictional characters, and certainly not Spongebob Squarepants. I don't know what word could have triggered it) and a few other odd choices. It didn't give me anything for Portugal, a major area of my concentration, nor any kooky conspiracy theories (though I work on those pages much less often.) How does it use subject matter or categories to suggest articles, I wonder?
Thanks for providing such a creative bot, by the way. It was helpful. Perhaps I'll try it again in a month. Grandmasterka 07:13, 13 June 2006 (UTC)
Glad you liked it. It uses three algorithms:
matching text of article titles you've edited against text of all articles needing work.
comparing your editing patterns to those of other users and suggesting things similar others have edited that you haven't yet.
following links from pages you've edited and looking "in that neighborhood" for pages that are often linked to (but not too often).
The algos are not perfect, and that would be a fine next direction to go in developing it. Not sure what caused the anomalies you saw above, it doesn't log quite enough detail about the suggestion process to say. Still, glad it was of some use and feel free to come back and be a repeat customer. -- ForteTuba 13:16, 20 June 2006 (UTC)
How long do I usually have to wait before SuggestBot suggest some articles for me? Wizrdwarts (T|C) 22:10, 16 June 2006 (UTC)
These days, while I'm trying to finish a thesis, I only think about SB once a week. I'd like it to be in an automatic mode, where people get suggestions right away (when they're most interested!) but it's not polished enough for that yet. Sorry about the delay. -- ForteTuba 13:13, 20 June 2006 (UTC)
I received a message from SuggestBot July 2, 2006. It was very helpful and I would like to continue using it, but have not received a message since then. Should I do something to get another message, or is there something wrong with SuggestBot?P.L.A.R. 20:36, 25 July 2006 (UTC)
Thanks for the suggestions! Many of them were excellent and I am now happily editing away. However, I am not sure how I got Shlomo Moussaieff as I am not Jewish, haven't looked at any Jewish articles, and rarely edit biography pages. Most of the suggestions were very good, however. cøøkiə Ξ (talk) 16:57, 20 June 2006 (UTC)
Cool, glad you found some of them useful. For some reason it's currently picking only random suggestions for the WIKIFY articles, I may have messed up last time I grabbed a list of them. -- ForteTuba 19:21, 20 June 2006 (UTC)
Thanks for this useful tool. I was familiar with some of my suggestions, but there are some new things there to occupy my time -- and it's just nice to be given a list of articles as something to focus on. I wonder if you might consider separating true stubs from section stubs, though. For example, under stubs SuggestBot recommended speciation, which is at a significantly different stage of development than Ophiomorus tridactylus. Just a suggestion. Thanks again. -- bcasterline • talk 19:24, 23 June 2006 (UTC)
You're welcome. And throwing out section stubs (or, making them a separate category) would probably be smart, that's a good idea. -- ForteTuba 12:51, 26 June 2006 (UTC)
Thanks for the suggestions. Some of them make sense to me, others don't. All in all, the list of suggestions seems a little bizzare. There are a couple of articles on the list about places in Kent, England; I know virtually nothing about Kent or anything in it, and, to my knowledge, have never edited an article about Kent (maybe reverting vandalism or cleaning up grammer). My interests are fairly broad though (which might explain the bizzare list), so I suppose I'll learn alot about Kent while editing the articles on the list. ;) ONUnicorn 19:41, 23 June 2006 (UTC)
Thanks for the feedback. I've been wondering if it gets confused by "wide" profiles where people work on a lot of different kinds of things, and wondering what a good way to deal with it would be. No answers yet, though. -- ForteTuba 12:50, 26 June 2006 (UTC)
I just got my recommended articles, and just have to say "wow": SuggestBot picked a great mix of articles for me- gaming, history, music, military, all in the range of what I feel capable of editing and all of interest to me.
If you need any help with the project(graphics, mass notifications, etc), just ask: I'll do what I can!
If Wikimoney was still around, I'd donate all of mine to this!
EvocativeIntrigueTALK | EMAIL has smiled at you! Smiles promote WikiLove and hopefully this one has made your day better. Spread the WikiLove by smiling to someone else, whether it be someone you have had disagreements with in the past or a good friend. Smile to others by adding {{subst:smile}}, {{subst:smile2}} or {{subst:smile3}} to their talk page with a friendly message. Happy editing!
Heh, thanks for the entuhsiastic support. For now, it's staying low and slow until I can put some time into implementing suggestions I've gotten over the past two months, but thanks for the offer of help. :) -- ForteTuba 12:53, 26 June 2006 (UTC)
I'd like to say how good this bot is, it handed me suggestions on ALL the topics i'm in to. Programming, gaming, electronics, data compression and more! i'm very happy with your bot, it didn't suggest anything irrelevant either. As for improvements: i have none, your bot is perfect as-is. keep up the good work. MichaelBillington 01:48, 24 June 2006 (UTC)
I have to add my $0.02 and say that I really enjoyed your bot's suggestions. Thanks for your hard work. Ifnord 17:12, 25 June 2006 (UTC)
Thanks to both of you, glad they were useful. -- ForteTuba 12:53, 26 June 2006 (UTC)
well it seems the Stubs list is pretty close. I give the Merge list 2/3.
Unfortunately on the Add Sources list I'm already on record for one of the suggestions saying "If you know any sources please add them!" so that's a bit of a downer....
The Expand section is a total puzzle, I don't see how SuggestBot came up with me expanding any of those articles. If it uses different rules for this section they aren't very good at measuring my interests/ability to expand articles (if it followed the same rules as it used to pick stubs it would have been a more worthwhile list).
It's a shame there are so many railway stations on the stubs side I really am only interested in the ones near where I live, as part of being interested in that area (rather than the topic of railway infrastructure). I guess this comes down to the railway articles being categorised as railway articles, but not being categorised for where they are geographically. That is a categorisation issue not a suggestbot issue I guess.
Garrie 22:29, 25 June 2006 (UTC)
It's not perfect as far as ruling out articles that people have edited in the past (although it tries). The expand and verify lists are both troublesome because there are relatively few articles with those tags compared to others---so it has less to choose from when making suggestions. Finally, with the railways, it looks like there was probably a very prolific editor of Aus railway articles. One of the algorithms looks for people who have edited similar things to you, and recommends articles they have edited, so you got a bunch of those recommended. Thanks for the feedback, this is useful to know. -- ForteTuba 12:58, 26 June 2006 (UTC)
Thanks for the latest round of suggestions. With a lot more edit history to work from, these look like better suggestions than the last round was. I've been doing a bunch of special:random article copyediting lately, so getting mostly topics based on my more serious editing is a nice tribute. Of course, the test will be whether I actually go edit them.
True enough -- it's easy to say "look, people like it" from comments, but actual behavior is more of a gold standard of being useful.
Could you add a way for the bot to recognize that my talk page already had a suggestion list from it and discard duplicates. It might be able to do this by just discarding anything wikilinked in other talk page sections with its auto-generated header. I got resuggested three articles that I turned down last time. GRBerry 21:55, 27 June 2006 (UTC)
Yep, after my first pass review this looks like a set more in line with my editing tastes. Having more edit history evidence for the bot to work with is helpful. GRBerry 02:24, 28 June 2006 (UTC)
Cool that it was useful. As for not re-suggesting articles, that's probably a good idea and not hard to implement. I didn't picture repeat customers, but it definitely has some. -- ForteTuba 00:37, 2 July 2006 (UTC)
I'm back; I just got my second SB list after going through most of the articles it listed last time. I noticed this time, though, that I got quite a few pages that were in the last list but I hadn't edited in the newer list- is this a known fault/purposeful logic?
Seeing this immediately after my feedback is certainly intriguing. So here is some more detailed analysis. I was about 6 weeks between suggestion sets, with a large chunk of unsuggested editing in between, and got three repeat suggestions. User:EvocativeIntrigue was about 4 days between suggestion sets, with what looks like primarily doing SuggestBot suggestions in between, and got six repeat suggestions. GRBerry 00:31, 28 June 2006 (UTC)
That was a bad pun (sorry, had to)! It was odd, but not really major: I'll probably actually work on those articles now! EVOCATIVEINTRIGUETALKTOME | EMAILME | IMPROVEME 02:42, 1 July 2006 (UTC)
Having two people point removing dups in short order is pretty compelling. I worry a little that it will run out of things to suggest, but that's easy enough to fix by doing some arbitrary penalizing of previously suggested items without throwing them out of the list entirely. I'm just glad it was mostly able not to re-suggested articles you did edit, I get occasional reports that it suggests edited articles. Not useful. -- ForteTuba 00:39, 2 July 2006 (UTC)
This bot found some articles that I can really improve! There is one article listed that doesn't exist, though. Nate 13:11, 2 July 2006 (UTC)
Cool, glad you found some useful suggestions out of it. It sometimes will recommend nonexistent articles -- the links-based recommender will sometimes spit out an article that's not there, and because it runs off of relatively old DB dumps, sometimes articles get deleted in the meantime. FWIW, some people really like the idea of getting suggestions for articles that don't yet exist. I haven't implemented that yet but I hope to someday. -- ForteTuba 16:38, 5 July 2006 (UTC)
I got suggestions from SuggestBot once after my request. Can I get regular suggestions, say monthly? It would be a hassle to make a request everytime. --theorb 10:08, 4 July 2006 (UTC)
For now, you need to request when you want more -- but that's a good suggestion, and one that wouldn't be too hard to implement. Would it be okay if it just appended them (rather than trying to fiddle through the text on user talk pages trying to replace old ones)? -- ForteTuba 16:41, 5 July 2006 (UTC)
Hate to say this, but .. please be careful about recommending merges and splits. I've recently had to deal with a crew of well-intentioned but otherwise clueless editors who threw themselves into several article splits, one rather large. Unfortunately, the articles were incorrectly tagged by yet someone else who was not an expert in the field. A fair amount of damage was done before the error was caught; I fixed one case, but the other one (scalar) remains a disaster. I first heard about suggestbot when I noticed it had listed an article from quantum field theory (good lord!) I'd just cleaned up as a suggested task for a non-physicist (when pigs have wings!) I know these points should be "obvious", but appearently, some editors have more energy than common sense. linas 02:04, 7 July 2006 (UTC)
Heh, one thing SuggestBot can't do is ensure that people do a great job of editing -- it can only find articles it thinks a given person is interested in, and it can only do that by looking at articles people edit. If someone does some cleanup on physics articles, SB might decide that quantum field theory is related to their interests. The long-term answer for SB is probably letting people do a better job of specifiying what they're interested in. Just grabbing their edit history is simple (and reasonably effective) but has known problems with picking up edits that aren't directly in a person's field of expertise. There are worlds where you try to match ability, authority, and responsibility, but that's not so much Wikipedia's game right now as far as I can see. -- ForteTuba 14:27, 16 July 2006 (UTC)
Useful suggestion, thanks. I'm hoping to do some coding in a couple of weeks and one of the primary goals is to do a better job of ignoring edits that aren't indicators of interest. It's useful to know about particular problems like this because they can easily be put in in an ad hoc way. -- ForteTuba 14:28, 16 July 2006 (UTC)
In fact, I think that any edit summary with the substring "disambig" should probably be ignored. This covers a number of forms a user may use when disambiguating. CRGreathouse (talk • contribs) 04:23, 29 July 2006 (UTC)
I specifically do disambig on topics that interest me, so I can read about them more. If I hadn't have done this then I think suggestbot would have suggested some other (less relevant) edits. But for most cases, disambigs should be ignored IMHO. MichaelBillington 07:42, 29 July 2006 (UTC)
Thanks for the feedback. This is a #1 problem for SuggestBot that people mention so I will add disambig to the current, pretty ad-hoc way of throwing away edits. -- ForteTuba 21:16, 6 August 2006 (UTC)
Currently the SuggestBot creates a list of suggestions and posts this to the User:talk:pages
Would it be possible to create a page on first request User:Username/Suggestbot, then leave a message on the talk page linking to the list. SuggestBot could then post updates monthly or quarterly to the list, leave another message when it happens. Gnangarra 08:54, 11 July 2006 (UTC)
It would. In fact, the first version was "user:SuggestBot/Suggestions for Username", and it left notes on people's talk pages. People almost never used the suggestions, but that's probably because it was picking people to make suggestions to randomly instead of letting people ask. That's probably not as high a priority for me as making the suggestions better, but it's worth doing at some point. -- ForteTuba 14:31, 16 July 2006 (UTC)
Hey there. I am currently (very slowly) rewriting and expanding the David Tonkin article and, to my surprise, find that SuggestBot has suggested expanding that very article. Not that SuggestBot has politely but firmly suggested that I get my backside into gear and complete the rewrite I better do it. Cheers --Roisterer 11:38, 11 July 2006 (UTC)
Heh, sweet. But don't do it for SuggestBot, do it for you. :) -- ForteTuba 14:31, 16 July 2006 (UTC)
Some curious users of it.wiki are wondering whether SuggestBot can work on wikis in languages different from English. Is there any chance? How difficult would be setting it up? Let me join the congratulation chorus. Thank you. --Paginazero 08:42, 12 July 2006 (UTC)
It probably could, although right now it still has some problems with characters not often used in English. These are hopefully going to be fixed in a future rewrite of the software. I'm going to be chatting with some people about what the most interesting ways to grow SuggestBot would be, and doing it cross-lingually and cross-culturally would be pretty cool. Thanks for the suggestion, and the congrats. -- ForteTuba 14:34, 16 July 2006 (UTC)
How long does SuggestBot usually take to respond to requests? An hour? 24 hours? I'm not talking about how long it takes to generate the requests once it checks them.
P.S. I think that SuggestBot might include categories in its analysis, if it doesn't already. So if I edit a lot of articles with the category Google, it would suggest some Google-related articles I haven't edited yet.
These days it's running once to twice a week as I finish my dissertation and pack for moving. It would be about 10,000 times better if it ran more often, I agree. I'm hoping that once things settle down (end of month?) I can run it more often. I also agree that thinking abou categories might be useful. There's been some research around using categories for product recommendations by Cai Ziegler that suggest it's helpful. The links recommender probably often uses category information when following links (e.g., from an article to its category page). But, explicit intelligent use of categories might be better. -- ForteTuba 13:03, 17 July 2006 (UTC)
Is SuggestBot thrown off-scent by work on disambiguation pages?
Just had a fascinating collection of articles back from this most innovative of bots! However, for me the success rate was rather low. In case you want to analyse further, here's my assessment of each suggestion...
Thus only about 20% of the suggestions are articles I might edit - so 80% failure I'm afraid.
I think that what may have confused the issue was that I have done some work on disambiguation pages, particularly when these contained a link to a page I had done work on. Some of the bot's totally uninteresting suggestions were other links from these disambiguation pages. Hebrides 12:56, 18 July 2006 (UTC)
Could be. I'll put throwing away disambig pages on the list of cheap ways to maybe improve its performance. It does use a link-following algorithm for about 1/3 of its recommendations. I'd really like to set it up to let people specify interests a little more precisely than "grab their edits" and now that I've graduated and moved to the new job place maybe I'll have some time to work on it. Thanks for the suggestion. -- ForteTuba 13:23, 30 July 2006 (UTC)
Since you ask (I'm always reluctant to suggest somebody else does some work ;-), here's an idea for an enhancement to SuggestBot. I would be interested to know why some of the articles were suggested - any chance of having the option to see an explanation? (possibly via popups when you hover over a link, or whatever mechanism you like). Hebrides 13:08, 18 July 2006 (UTC)
For now, probably not. One of SuggestBot's missions is to help me study better ways of recommending things, and one of the ways I tell how it does is to see which recommenders get their suggestions worked on more often. Showing why an item gets recommended might bias that. Plus, the explanations right now would be pretty boring: "The article text was similar to the titles of articles you have edited"; "People who have edited the same things as you in the past have also edited this article"; "This article has a fair number of links from articles you've edited". In principle the recommenders could be made to give a more detailed explanation of which people, which article titles were important, how the links were found, but that seems like more work for relatively low value. (Though, I agree that getting explanations of recommendations is a pretty interesting idea.) Thanks for the suggestion. -- ForteTuba 13:27, 30 July 2006 (UTC)
This section really needs to be reconsidered. It has usggest that I add sources to article that I have never edited before and would therefore be unlikely to know what the sources are for the article. Suggest that this be limited to articles that the user a previously made edits too. All the other suggestions were fine, with expected anomolies considering that I have reviewed a lot of GA articles and done some editing of them in the process. Gnangarra 14:55, 20 July 2006 (UTC)
Not sure what "GA" articles is, but let me know so I can consider trying to remove them from people's profiles if I can find an algorithmic way to do that. Interesting point about the add sources section, though ideally SuggestBot finds things that you do know something about (or are interested enough in to find out). Add sources is one of the work types that people are less willing to do. Not sure what to do about it yet, but something to ponder. Thanks for the note. -- ForteTuba 13:35, 30 July 2006 (UTC)
About 75% of the articles you suggested were really good; in fact, I was looking for one of them yesterday but I didn't know the spelling. You obviously picked up on my interest in Pompeii, as well as Ancient history and the humanities. However, I did get a curious amount of stubs and articles on Spanish provinces and I'm not sure why. Also, I got suggestions for Bual and Desmond Boal; I presume that it comes from my work on the article on Augusto Boal, but these articles really have nothing to do with the theatre practitioner. I'm pretty sure that they came from those disambiguation things at the top of the articles. Apart from that, I'm really happy with the suggestions, and I'm amazed by Suggestbot. Good work! --Lord Pheasant 03:31, 28 November 2006 (UTC)
Sweet, glad it was of some use. As for the Boal confusion, one of SuggestBot's strategies is to use text similarity to make recommendations and I'm pretty sure that's what happened there. The Spanish provinces were found by the link-following strategy, it looks like. Both strategies get tempted by words/links that appear sometimes but not too often, and I'm guessing that's what happened here. Still, glad you liked it. -- ForteTuba 22:49, 5 December 2006 (UTC)
I really can't criticise you since i can't do any better.
but i looked up Hose (like medievil clothing) and then edited it cos it had a broken hyperlink. and now many of my recommendation are about hoses, but of types i know nothing about, (uretha, gardern hose, kelly hose etc.) the amount of these may have been due to me not haveing edited many articals, perhaps you could set it up not to base it on edits taged minor (though i might have forgot)
Yeah, it gets confused, especially the text-similarity strategy, which I'm sure is what happened here. Sorry about that. As for minor, it doesn't appear to help in general based on some experimenting I've done, although I should go back and look at it now that I have more people who've used it. -- ForteTuba 22:51, 5 December 2006 (UTC)
The suggestions SuggestBot gave me were amazingly ALL my interests, every single one of them I would love editing. Thank you so much, you're doing a great job. --*katespeak 21:11, 27 July 2006 (UTC)
Glad they were helpful -- sometimes SB loses and sometimes it wins and I'm glad it won for you. Enjoy the editing, and feel free to come back for more later. -- ForteTuba 13:28, 30 July 2006 (UTC)
Yeah, it's really incredible, it recommended 2 articles that I had looked at and considered changing before. It really is clever what you have done with it, thanks. --James086 13:28, 27 September 2006 (UTC)
On the whole the selection of articles was spot-on! Somehow (I presume through categories or comparing with other users) you identified my interest in the historic county of Yorkshire.
What SuggestBot missed entirely, is that I'm a casual editor, rather than a writer. I correct inaccuracies, spelling mistakes, etc, but don't have the time or resources to expand articles or find sources.
On the whole, however, I was very impressed with the selection of articles. Yorkshire Phoenix (talk * contribs) 14:30, 28 July 2006 (UTC)
Yeah, SuggestBot doesn't try to determine editing styles, although figuring out what kind of work people like to do would be pretty interesting. It picks a predetermined mix of work types, mostly stubs because there are so many of those compared to others. Would you want to be able to tell it "hey, no stubs, but lots of cleanup and wikify"? -- ForteTuba 13:38, 30 July 2006 (UTC)
That sounds like an idea: but it was mostly the stubs it suggested that I could get my teeth into (adding templates, infoboxes, etc) and when I got onto the right hand column I was finding the articles were still relevent but there was nothing I could do with them. Yorkshire Phoenix (talk * contribs) 08:06, 31 July 2006 (UTC)
Ah, gotcha. Not sure what to do about it, any ideas? -- ForteTuba 14:05, 31 July 2006 (UTC)
Perhaps it could somehow work out the difference between editors (in this context to mean people who correct typos, add categories, etc, like myself) and contributors (who write lots of original/sourced content). I expect this could be done by looking at the comparative size of the article before and after our edits(?) Yorkshire Phoenix (talk * contribs) 14:11, 31 July 2006 (UTC)
That would be one way to go -- we thought about trying to classify editors and edits, and edit size was one of the criteria we talked about. Probably not in the short term, though, sorry... hopefully it will come up with at least some useful stubs for you if you come back for seconds and thirds. -- ForteTuba 21:21, 6 August 2006 (UTC)
It seems that the simpler way would be the one you've already discussed elsewhere, of weighting against the edits marked as "minor edits." Once that's working, you could also try including in the original suggestions message a line like, "SuggestBot pays less attention to the edits you've marked as 'minor,' so ..." as a way to remind people to di8stinguish minor edits. Lawikitejana 02:46, 7 September 2006 (UTC)
Right now it pays no attention to them (well, technically, for some people it does and some it doesn't, while I test whether it helps people get better suggestions) at all. How valuable is it to WP for people to mark minor edits as minor? -- ForteTuba 12:24, 7 September 2006 (UTC)
Just to point out SuggestBot seems to be bringing up a few quirks. For example, it's telling me to wikify Uploading and downloading and merge Poland in Eurovision Song Contest, when none of these articles actually need these. The changes weren't made after SuggestBot had already suggested on my talk page, so there must be something wrong with it. Madder 18:46, 28 July 2006 (UTC)
It downloads a list of which articles need which kind of work "every so often", which hasn't been so often the last 6 weeks (last update of these lists was June 10). I am, in fact, running that part of the bot right now, so hopefully it'll get better. Thanks for the note. -- ForteTuba 13:39, 30 July 2006 (UTC)
Why not ask people whear they live and then associate articles to them in that area, whear they would have more common knowledge???
PS
What does "Wikify" mean???
Chris5897 11:39, 2 August 2006 (UTC)
Here's a little description of [[1]], basically it means to fix the markup so that it's in Mediawiki (instead of HTML) markup, add links to other articles, etc. As for the other idea, hadn't thought about that... it might be useful, especially for things like people and geography. Not sure how to implement it though, except through using categories. Categories I want to explore in the longer term but for now I'm going to be making some incremental improvements. -- ForteTuba 21:24, 6 August 2006 (UTC)
Thank you for making the bot give me suggestions. Most of the suggestions reflect my editing interests... although I don't know why SuggestBot would recommend expanding musical bios to me, when the only musical bio I've really contributed to is Fasy. --Gray Porpoise 14:55, 18 August 2006 (UTC)
Glad it mostly worked out. As for the musical bios... it just depends on the link structure around the things you edit, and the things that people like you edit. -- ForteTuba 12:40, 28 August 2006 (UTC)
SuggestBot did a pretty good job of pulling up some articles based on my most recent edits. For those who aren't getting results they like, perhaps they should try editing a whole string of articles that they find really interesting and then submitting their name to SuggestBot.
One idea I have that might make SuggestBot even more attuned to a persons taste in articles is to include the user's Watch List in the algorithm. I would think most people's watch list is a snapshot of the stuff that they are really into. That might even make it easier for people to tune SuggestBot to their interests. They could Watch a whole bunch of articles that they are into and SuggestBot would really nail their interests.
Well, in retrospect, the watchlist would have been a good idea but I just discovered that no one else can see my watchlist. So, scratch that idea. The tool works great anyway!
Good ideas. There was some discussion about using watchlists when I first created the bot... the other problem with them is that people sometimes put on pages they'd like to see deleted, etc. But I agree, watchlists would probably be better than edit history. I keep meaning to write an interface where someone can paste in some WikiText with links, and instead of fetching edits it'll just read the links the person specifies. That would let people use watchlists (or parts of watchlists), categories, contribution lists (a lot of people make lists of articles they've made major contributions to), etc. More work, but maybe more promising. The idea of having people edit articles that really define them right before using SB is good, but since there's some delay between posting your name and getting suggestions, probably wouldn't work as well as you'd hope. That's something else I'd like to fix: semi-automatic suggestions right when you want them. So many things to fix, so little time. Sigh. -- ForteTuba 12:44, 28 August 2006 (UTC)
Wow! How very cool. My recommendations were right up my alley... One of them turned out to be a tiny island off the coast of Massachusetts less than 5 miles from my home!
It's so cool, that my suggestion is that you offer two types of suggestions. An editor can request suggestions for articles to edit - like you're doing now. The other request would be for suggested articles for reading. "Based on your editing preferences, SuggestBot thinks you may enjoy reading the following articles:" Of course instead of articles that need help, you would instead provide articles that are "done."
A lot of people enjoy clicking on "Random Article" links. This might satisfy the same curiousity - but with recommended articles instead based on your cool algorithim.
Your algorithim is so good. It would be wonderful to be able to share it with more editors in this other direction as well.
Thanks again. I have worked on a number of the articles suggested so far... I want to do a bunch of them before I request my next batch. --AStanhope 04:54, 24 August 2006 (UTC)
Nice, I'm glad they were useful (and, useful enough to work on -- that's the gold standard). It would be easy to just recommend any article in Wikipedia whether it needed work or not, more as a reading list in your sense. Maybe I'll throw that in next time I do some coding on it. -- ForteTuba 12:47, 28 August 2006 (UTC)
I have one suggestion for SuggestBot. I'd like to know if it's possible for the bot to see what languages you know (useing BabelBoxes) and have a section with "articles needing translation".
Other than that, I really love SuggestBot! Thanks! Alohasoy 15:33, 27 August 2006 (UTC)
I think that's a great idea, but I'd recommend checking for categories instead of babelboxes (e.g. Category:User de-N instead of {{User de-N}}). It would be more efficient because a couple of babelboxes have female versions (Template:User de-f for example), so female German speakers' userpages don't show up as "transclusion" in the "what links here" page of the (main) male version, but everything shows up properly in the categories. --Zoz(t) 16:51, 27 August 2006 (UTC)
Interesting suggestion. What's the easiest way to figure out whether a page needs translation? Are there tags, etc.? I've been anglo-centric and don't know a lot about how the various languages of wikipedia cross over. -- ForteTuba 12:49, 28 August 2006 (UTC)
There is Wikipedia:Translation into English which seems to be readable by a bot: source languages and target articles are clearly labelled. It's a bit problematic though, because many times the article is already translated (but it isn't removed from the list) so this would give a lot of false-positives. Comparing the length of the source article and that of the English article would probably help sorting out already translated pages, but coding this would be relatively difficult I think (you'd have to parse articles in other lang editions). Also, many times there is no English language article yet and that means you can't figure out whether the user would be interested in that article. And easier and probably more efficient alternative would be using {{FAOL}}: fortunately you can figure out the source language by checking the language parameters on the talk page to which this template transcludes. Compare that with the lang categories the user placed themselves in, check if they match and check if they'd be interested in that article. --Zoz(t) 22:05, 28 August 2006 (UTC)
Could it scan the language links at the bottom of the page? It would at least give a start of an idea as to what needs translation. And if it's a false positive (the translation does exist, but hasn't been linked up to the other language), it means the user can add the translation link. --Carl(talk|contribs) 14:29, 29 August 2006 (UTC)
I suppose you could just grab page lengths and look for ones that are, say, 10x or more different as a way to reduce false positives. Are there lots of articles that have these tags or are on that list? If there aren't many articles, SuggestBot won't be very useful probably; it's hard to find an interesting article for someone given a relatively small set of choices. -- ForteTuba 12:30, 7 September 2006 (UTC)
I think you should take talk pages into account. Imo it would definitely be an improvement - users post a lot to talk pages of articles of their interest. I don't think people post that much on "talk pages of articles they'd like to see deleted", but if they do then they're clearly interested in that topic. Also, I think it would be useful to check more than the last 500 edits — e.g. RC patrol, mass typo fixes with AWB or XfD participation give no information to SuggestBot but takes up a lot of edits from that 500. And the less informative edits read the less accuracy SuggestBot has. (By the way, my recommendations were quite accurate on average, although it was a bit unbalanced as it suggested a crazy amount of chess- and Scientology-related stubs, even though all my edits to Scientology-related articles were nothing more than general tidying.) --Zoz(t) 17:07, 27 August 2006 (UTC)
True on the talk page edits. It would be some re-coding and would eat up some more memory on the server, but in principle doable and perhaps worth it. As for the 500, that's partly to minimize SB's being a server hog, although in the grand scheme of things I think it's pretty tame on using WP resources. I agree that losing edits to reverts and disambigs isn't helpful, although I'm not sure having a large number of edits helps it that much. In fact, for the text-based recommender, I think it might hurt -- it grabs a medium-rare word and says "wow, that's an important word". In your case I think that's what happened with chess. More article titles = more chances to grab a rare word. If you have any suggestions on how to do a better job of filtering out reverts, disambigs, typo fixes, and other mechanical edits, based on comment lines, that would be great. I don't know all the abbreviations that experienced editors use. Thanks for the feedback, it makes me want to start coding except for having to prepare lecture for tomorrow. :) -- ForteTuba 12:55, 28 August 2006 (UTC)
In my case grabbing more than 500 edits would have definitely helped. I made about 10x more edits to lucid dreaming than to all chess-related articles altogether, but SBot didn't notice it because I haven't edited it my last 500. But maybe that's not the case with most editors. Anyway, as for filtering out reverts and such, I think I could code a quite efficient heuristic for that. By the way, it occured me a couple of weeks ago that the ratio of reverts to all edits would probably give a better and more objective estimate on the "vandalism threat level" template. I'd like to write a bot which updates a similar template, but it would need 2 things: a filter which recognizes reverts (I think I will be able to do that) and a frame which handles i/o (I have never written a WP bot before, so I don't know how to do that). If I write the filter, I'd give it to you if you want to use it with SBot, naturally. But I don't want to start coding the filter while I don't have a proper frame for my bot. Help me out, and I'll help you. Deal? :) --Zoz(t) 22:58, 28 August 2006 (UTC)
In general, I have posted a lot to the talk pages I'm interested in. If it's just a revert or cleanup, I usually don't. So, looking at my talk contributions would have been, in my opinion, more on target, especially if a categorization-based heuristic is used. That may have given me a more useful stub list than David Wm. Sims and Bill Sims, which are most likely based on a strict word matchup (I edit The Sims 2 a lot). --Carl(talk|contribs) 14:26, 29 August 2006 (UTC)
It surely did find them based on text. I am hoping to be doing some re-working over holiday break to make it better able to represent profiles, and thinking about talk page edits is one strategy I might pursue. Thanks for the feedback, sorry for the delayed response. -- ForteTuba 22:53, 5 December 2006 (UTC)
Your suggestions definetly helped the encyclopedia. It led me to Stoned where I gave it a {{prod}} tag. HighInBC 15:52, 28 August 2006 (UTC)
This is a dumb question from someone who's more techie than wiki, but how valuable is deleting articles in WP? -- ForteTuba 12:21, 7 September 2006 (UTC)
My opinion is that deleting articles
Makes it easier to find useful information on wikipedia (especially when using categories or lists)
Gives us a common goal of creating an encyclopedia instead of allowing it to devolve into an information repository. This creates more of a community.
Gives wikipedia users a common expectation that when they find an article, it is something worth reading. johnpseudo 22:17, 2 November 2006 (UTC)
Yeah, I'm pretty curious about the community side of Wikipedia in general, how people become avid community members and all. I do think a big part of why people become passionately attached is because they think it is _right_ to do it this way: open-access, and specifically an encyclopedia. Now what might be cool would be to spin off a separate site that _is_ more about the less notable bits: everyone and everything that people want to talk about. That might be fun for people to work on. Of course, if you go too far in that direction you bump into MySpace. -- ForteTuba 22:56, 5 December 2006 (UTC)
Anyway, I realised that I hadn't edited very many pages up until that point and it probably didn't have a lot to go on. Since then, I've edited a lot more pages that I take an interest in so hopefully my second request will come up with improved suggestions - I'll come back and let you know! Curiousbadger 14:01, 30 August 2006 (UTC)
The text-based recommender gets captured by things like that. Sometimes it works, sometimes not -- let me know if the second set feels better. -- ForteTuba 12:22, 7 September 2006 (UTC)
Hi, and thanks for SuggestBot! I only just got my list today, but i didn't feel that some of the suggestions were quite accurate.
It gave me lots of suggestions about places i have never heard about, the radio, football, and various other things i have nothing to do with, and i think there was only one suggestion about cattle- when most of my edits are to do with cattle! Is this a problem? CattleGirl 09:44, 4 September 2006 (UTC)
I'm looking at the recommendations. The text-based recommender got interested in the word "Grammar" because you edited a couple of school-related things. The mysterious places in the Netherlands were recommended by a links-based recommender. It got confused because there's a template that puts a bunch of links at the bottom of the page for Netherlands provinces, which made them look very attractive. It happens -- the algorithms look for certain features in the way articles are related to each other, and sometimes that works well and sometimes not. Sorry it's not so well for you in this case. -- ForteTuba 12:35, 7 September 2006 (UTC)
Hi -- The SuggestBot page says that it posted my recommendations "Storkk -- 17:57, 24 August 2006 (UTC)". I'm looking at the history of my talk page, and see no edits to it around that time. Do you know what might have gone wrong? The recommendations are supposed to go on the user's talk page, right? Thanks! --Storkk 23:37, 4 September 2006 (UTC)
My name was taken off the list by a vandal, and there wasn't even a mention in the 'suggestions posted' section - Blood red sandman 13:04, 5 September 2006 (UTC)
I've checked the edit history. Blood red sandman's name was removed in this edit by ForteTuba, the maintainer of SuggestBot. I notice that neither Blood red sandman nor JakGd1 received suggestions and both decapitalized the "u" in "User" when submitting the request. Is this a coincidence, or does it demonstrate a bug in SuggestBot? Michael Slone (talk) 21:49, 5 September 2006 (UTC)
That would explain why I couldn't work out who the hell took my name out from the history - since I thought it was vandalism I naturally never bothered to check the edit by the bot's maintainer. You're right, it does look like a bug, doesn't it? Could easily be sorted temporarily by telling users to ensure they capitalise their Us while a patch is worked out by ForteTuba, or someone else who actually knows anything about how you do this sort of thing - Blood red sandman 22:19, 5 September 2006 (UTC)
Just to say, I've now received mine after relisting myself, and this thing's brilliant! I do some random article work and newpage patrolling etc, so I hadn't expected much relevant stuff, but this thing really did well at filtering out the cr*p! Good stuff! - Blood red sandman 12:20, 6 September 2006 (UTC)
Yeah, it's a bug. The thing that parses the list of names assumed a very pristine format. I made it a little more robust, thanks for the notes. Also, glad that it was able to find some useful work for you. -- ForteTuba 12:41, 7 September 2006 (UTC)
Yes, I think it does. See, the great thing is, it seems to realize when you're editing an article in a field you don't normally edit...so it doesn't think you're "interested" in an article that you reverted vandalism on! Genius. However...it suggested some really whacked-out articles for me... Nice work Forty! SoaP 16:44, 6 September 2006 (UTC)
Sometimes you win, sometimes you lose -- glad some of them were helpful. -- ForteTuba 12:41, 7 September 2006 (UTC)
I was pretty impressed with my suggestions, although my tendency to get involved in disputes that require many edits about topics I don't really care about just by chance understandably skewed some of them. The only problem I noticed was that it suggested Michael Szymczyk, a deleted/protected page which I AfD'd. Great work on this, though! -Elmer Clark 23:31, 6 September 2006 (UTC)
Cool, glad you liked it/found it useful, and thanks for the compliment. It tries to ignore some kinds of edits (reverts, disambigs) but still is easy to confuse for people with diverse activity. -- ForteTuba 12:42, 7 September 2006 (UTC)
Would it be possible to work a screen for biography articles into the algorithm? I'm a regular contributor at Joan of Arc and about a quarter of SuggestBot's suggestions were to random biographies of people whose first name was Joan. Interesting selection on the whole, but not much where I could add more than a typo fix or an odd sentence. Durova 14:04, 8 September 2006 (UTC)
To be more exact, 11 out of 34 suggestions were random instances of people whose first name was Joan. Durova 15:10, 10 September 2006 (UTC)
Yeah, the text-based recommender sometimes really makes stupid decisions. Sorry about that. I'm pondering throwing it away, or at least doing something that doesn't let one word dominate the results. -- ForteTuba 22:57, 5 December 2006 (UTC)
The system appears to work pretty well. The vast majority of the suggestions were clearly within areas I consider myself knowledgable in and are the sorts of things I'd contribute to. On the down side, I didn't act on a single suggestion. Many of the suggestions were fixed by the time I got there. Others are labelled as being a stub, but I don't really see anything to add (perhaps I should track down the guidelines for a stub, and if I'm confident the existing article is good enough and delete the stub tag). Of the remainder, some were off topic (I did some simple knowledge free editing of a bunch of comic book articles, as a result SuggestBot suggested I work on a few comic articles. But I'm actually fairly ignorant in the area). The last few were relevant, but subareas I don't know much about.
Still, I found it interesting and will likely give it another spin at some point.
A suggestion: perhaps allow people requesting a SuggestBot to specify a focus? "I'm interested in working on articles like these", followed by a list of articles. The more general idea would to say, "I'm interested in working on this topic", but that seems a much harder problem. If I list a bunch of role-playing pages I've worked on, I suspect SuggestBot can figure out which subgrouping I'm interested in. If I say the general topic of "role-playing games", that's harder.
Thanks for the work. Alan De Smet | Talk 23:10, 14 September 2006 (UTC)
To be more specific (in the hope it helps refine the algorithm): I was going through comic articles deleting some unintentional link spam. My edits were almost entirely deleting a single line from the External links section. The edits weren't straight reversions as the person who added them also added useful content. Perhaps edits which consist of deleting a single line that begins with "* " and contains an external link should be weighted less on the theory they're probably just deleting linkspam?
On that note, I get the impression that SuggestBot is at least partially run by hand. I'd be interested in getting updated reports whenever SuggestBot's copy of the database updates, perhaps as often as weekly, but if you're doing this by hand I don't want to bother you for something I'll typically just briefly glance at. Alan De Smet | Talk 23:14, 14 September 2006 (UTC)
Sorry about the delay getting back, I've been coping with classes. I wish it were able to tell articles that people have some knowledge of automatically. Empirically, removing edits marked minor makes no difference to how often people accept suggestions. The idea of letting people specify articles is good and one I've wanted to do for a while, just never carved out the time to make it happen. I think it would help a lot, though. And you're right, it's almost entirely run by hand; it's a useful prototype, but not robust enough to be let into the world on its own. :) -- ForteTuba 12:44, 7 October 2006 (UTC)
I recently edited Team Galaxy (Pokémon) and because of that, I got some suggestions for editing Astronomy-related articles, which I have no clue of. Maybe the bot could check the talk page for WikiProject templates? Also, I got some seemingly random suggestions. But I look forward to editing the other ones! -De Nam 10:59, 15 September 2006 (UTC)
Thanks for the comment. You did, indeed, get some suggestions for astronomy because the text-based recommender got interested in the word "Galaxy". Using WikiProject (or even category) templates would probably help out. At some point I'm going to get excited about coding again... -- ForteTuba 12:48, 7 October 2006 (UTC)
Hi! I'm a contributor of Italian wikipedia! Is it possible for SuggestBot to suggest articles in other wikis? Or, maybe, can bots running on other wikis use directly its scripts? They have to be greatly modified to run correctly? I'm not such an expert on that matter, as you can guess, I just can use my simple GiacoBot with pywikipedia framework scripts... Thanks for your patience! --Gia.cossa 09:51, 24 September 2006 (UTC)
It in principle could, but I'd need to set up a version for each language. It's also got a few issues with internationalization and character sets that would need to be resolved. I may be getting some other people interested in working on Wikipedia around my new school and if so there's more of a chance that we'll think about going international, but for now, English-only. Very self-centered, I know, and sorry I can't help out for now. -- ForteTuba 13:03, 7 October 2006 (UTC)
Hello, and thanks for the suggestions. :) Unfortunately the database dump the bot is using appears to be quite a lot out of date. The merge/cleanup/wikify articles I clicked, none of them had the tag anymore. Another point: even though I've done a batch run with AutoWikiBrowser on Category:Chess players, it doesn't necessarily mean that I know anything more about them. The number of, well, numbers in the suggestions is also a bit disappointing. I'd have expected a bit more variance. Now I got recommendations about nine numbers, such as 181. And I don't know if your bot already does it, but maybe it should put less weight on minor edits. Generally the bot however did do a very decent job, so keep up the good work. :) --ZeroOne 19:10, 28 September 2006 (UTC)
Thanks for the note. The tags are a little out of date, and I need to make it do a better job with that. The chess players is just a poser... if it looks like you do a lot of editing on a topic, it looks like you know and/or care about it. Minor edit removal doesn't materially affect how often people edit suggestions based on empirical results. The numbers is an awkward poser... the Category:Integers tag includes a moderate number of links to other numbers, which makes the links-based recommender very excited. This has caused problems for other people as well in a number of categories (on the other hand, it also sometimes leads to stunningly good recommendations). Maybe I should penalizing links that are included through category tags. Thanks for the feedback, glad that at least some of its output was useful to you. -- ForteTuba 13:17, 7 October 2006 (UTC)
SuggestBot's suggestions were all pretty good, yet it suggested wikifying Retro Gamer, when besides correcting a typo in the Perfect Dark article, I've never done anything in that area. LWF 20:53, 7 October 2006 (UTC)
Sorry, I'm way behind on replying to these. Best guess is that Retro Gamer was the highest-ranked suggestion it could make for you that needed Wikifying, there aren't that many articles in the Wikify category so it does what it can. Not sure that's the case, but that's the guess. -- ForteTuba 21:25, 15 November 2006 (UTC)
I think Suggestbot is great, personally. My suggestion is that I would prefer to have things like this or the signpost on sub-pages, for point and click reference, than mixed up on my talk page with everything else.
That's how I posted it to the request page, but it posted to my talk page instead. Is there any way to change this? - jc37 03:57, 13 October 2006 (UTC)
It's currently not configured for anything but posting to talk pages (and it's probably handy to do so because you get the "new messages" banner and think to look for the suggestions), but you're not the only one who's requested this. It looks like I may do some developing on SuggestBot over winter break and that will be one of the things I may do, make posting style more flexible. Thanks for the feedback. -- ForteTuba 21:26, 15 November 2006 (UTC)
I have my sub-pages on my watchlist : ) - If it helps, Wikipedia:Wikipedia Signpost/Tools/Spamlist seems to be able to do the sub-pages, so perhaps someone there may have some pointers : ) - jc37 15:15, 26 November 2006 (UTC)
That makes sense, that people watch their subpages. A little awkward if SuggestBot has to create the page, though -- not able to put it on the watchlist (or can you watchlist a nonexistent page? That might be cool.) I suppose a warning that you should create and watch the page if you want suggestions posted to one would go a long way. I have started a little bit of recoding and specifying a post location is part of the new framework. Thanks for the feedback. -- ForteTuba 23:00, 5 December 2006 (UTC)
I had mostly good results, although several of the stubs had been expanded (and one had been deleted) by the time I got to it. I had two near misses, though, that you could potentially use as examples for something to improve. I frequently edit articles related to Bleach (manga). SuggestBot suggested that I edit Static (Bleach album), which is related to Bleach (band) and White King, which is related to Bleach. So I can see why it did that, but maybe it should look at where a link pipes to, instead of the actual link text (or maybe it already does that, idk). Also there were a few suggestions that I simply have no clue where they came from, for instance Black people and Neopet Faeries. Overall though, very very good. I was surprised that my results were as consistent as they were!--Anaraug 13:53, 17 October 2006 (UTC)
Glad it mostly worked out for you. The Bleach thing looks like it may have found those pages through following links on disambiguation pages, and it would be pretty clever if SuggestBot ignored them. Thanks for the note on that. -- ForteTuba 21:29, 15 November 2006 (UTC)
Portrait de l'artiste sans barbe has been moved to Self-Portraits by Vincent van Gogh, and Salon des Indépendants to Société des Artistes Indépendants. For Les Alyscamps it is up to you to resolve the conflict you see; from my place I cannot see whether your remark is pointing to the horizontal or to the upright versions. And for Neo-impressionism, I would like to be in touch with you directly. |RPD 01:31, 25 October 2006 (UTC)
(Are you or were you "Stumps"?) Alas, SuggestBot is stupid about redirects for now. If it's low-cost to fix it when I do the re-implementation of it, I will. You're welcome to ask me about Neo-impressionism if you wish, but I don't know anything about it except that SuggestBot recommended it to you for some reason. If there's a reason that it shouldn't have that might point out ways to improve things, I'm all ears though. -- ForteTuba 21:33, 15 November 2006 (UTC)
Thanks for creating SuggestBot, it's a great idea. I've got one little problem though, I've been taking part in disambiguation link repair (see Wikipedia:Disambiguation pages with links) and have made lots of minor edits to a wide range of articles, which has given me a set of suggestions for topics that I don't really know much about. Would it be possible, when calculating what is interesting to us, to either give less weight to articles where we've only made minor edits, or to ignore any changes we've made where the description contains the phrase 'Diambiguation link repair'? Sorry to be picky. Thanks again for coming up with such a useful tool. Gasheadsteve 12:32, 25 October 2006 (UTC)
No problem. I would like to build better profiles, although I think the way to do that really well is to ask people to make their own list of articles. (Perhaps, by pasting in a watchlist or a part of their userpage where they list the contributions they're proudest of.) I've been trying for some people to remove minor edits and others not, and at least in the data that I looked at so far, there's no difference in how often people take up the suggestions. I'll probably look again at the profile-building problem when I rewrite SB in the next month or two. Thanks for the feedback, sorry for the delay in response. -- ForteTuba 21:35, 15 November 2006 (UTC)
No, it's just redirect-dumb for now. There's some cost to chasing them. And no, I have no idea on the Austrian politics. One was chosen randomly, SuggestBot apparently could find any articles for you for Wikifying that were relevant to your interests. -- ForteTuba 21:39, 15 November 2006 (UTC)
Congratulations on writing this wonderful bot! Even though I am recent here (4 days!) your bot came up with some wonderful suggestions! Yuser31415 06:24, 31 October 2006 (UTC)
Sweet, I'm glad it was able to help. SuggestBot was originally meant to help people who had edited just a few articles find more they'd be interested in working on, although in practice I think most of its users are more experienced. I'm glad, however, that it's useful for you. -- ForteTuba 21:43, 15 November 2006 (UTC)
perhaps you can weigh on a user's userboxes to decide what pages (s)he would like to edit? For example, if one's userboxes says she is vehemently pro-choice, she would want to edit articles based on abortion and pro-choice politics. This would be a great thing for userbox enthusiasts in Wikipedia. mirageinred 20:57, 31 October 2006 (UTC)
Alas, parsing userboxes out of a user's page would be reasonably difficult, and unless they point to categories or some other way to link them directly to individual articles, it would be really hard to use them smartly. I am interested in better ways to establish a person's interests but not sure right now that's the way to go. (Though I do know a couple of people who are interested in studying the adoption of userboxes in WP). Thanks for the note. -- ForteTuba 21:45, 15 November 2006 (UTC)
Some good suggestions, it also took Crow literally and threw up some biography stubs on people surnamed crow after I'd been editing crow (bird) pages. cheers. Cas Liber 20:39, 10 November 2006 (UTC)
The text-based recommender strikes again! Glad some of it was helpful, though. -- ForteTuba 21:46, 15 November 2006 (UTC)
It picked out mostly the pages that I would consider editing, which was good because doing categoriesing and stub sensor work I was worried it may just pick a broad variety of articles that I am not intested in editing to add content to.
The merge and wikify suggestions did give examples where this action was no longer appropriate.
The expand seciton however was completely useless as I do not know enough or have sufficient interest to expand them.
On the whole it gave good suggestions. Ksbrown 16:08, 15 November 2006 (UTC)
Cool that it made some good choices for you. It does use lists of articles-needing-work that get a little out of date, because it runs from dumped copies of the database. As for expand, there aren't that many articles tagged for expansion so it has to go a little farther afield. I'm pondering removing the expand tag, but I want to go back and look at the data to see how often people work on each type before I go changing it. Thanks for the feedback. -- ForteTuba 23:03, 5 December 2006 (UTC)
14:58, 8 March 2006 (UTC)
15:03, 8 March 2006 (UTC)
~Linuxerist L / T 04:21, 20 March 2006 (UTC)
