L14 - Wildcard Queries

Wildcard Queries
J. Pei: Information Retrieval and Web Search -- Wildcard Queries 2

Inverted Indexes
Query Brutus AND Calpurnia
Vocabulary Lookup
Given an inverted index and a query, we need to
determine whether each query term exists in the
vocabulary
If so, identify the pointer to the corresponding postings
Hashing or search trees?
How many keys (terms)?
Is the number of keys static or changing a lot?
Operations on the keys, insertions only or insertions +
deletions?
Relative frequencies of key accesses?
Hashing
No easy way to find minor variants of a
query term
Minor variants could be hashed to very different
buckets
Cannot find all terms with the same prefix
For web search, the vocabulary size keeps
growing
A hash function may become insufficient after
several years
Search Trees
Easy to find all terms with the same prefix
Balancing search trees
Logarithmic search time
Cost: rebalancing
B-trees
Every internal node has a number of
children in interval [a, b]
Good for disk-based data storage
When Are Wildcard Queries Useful?
A user is uncertain about the spelling of a query term
S*dney ! uncertain about Sydney or Sidney
A user is aware of multiple variants of spelling a term and
(consciously) seeks documents containing any of the
variants
Color versus colour
A user searches documents containing variants of a term
that would be caught by stemming, but is unsure whether
the search engine conducts stemming
judicia* ! judicial versus judiciary
A user is uncertain about the correct rendition of a foreign
word or phrase
Universit* Stuttgart
Trailing Wildcard Queries
A trailing wildcard query has only one * symbol at
the end of the search string
Example: mon*
Trailing wildcard queries can be answered
efficiently using a search tree
Walk down the tree following the symbols m, o, and n in
turn
Enumerate the set W of terms in the dictionary with the
prefix mon
Use |W| lookups on the inverted index to retrieve all
documents containing any term in W
Leading Wildcard Queries
A leading wildcard query has only one *
symbol at the beginning of the query
Example: *mon
A leading wildard query can be answered
efficiently using a reverse search tree
Each root-to-leaf path corresponds to a term in
the dictionary written backwards
The term lemon is represented by a path
root-n-o-m-e-l
A Little More General Case
How to answer queries containing only one
* symbol but can be in any position
Example: se*mon?
Rewrite the query to se* AND *mon
Use two search trees
A search tree to answer query se*, find the set
W of terms
A reverse search tree to answer query *mon,
find the set R of terms
W ! R is the set of terms satisfying the query
General Wildcard Queries
A general wildcard query can have any number of
* symbol at any position
Framework
Rewrite a given wildcard query q as a Boolean query Q
on a specially constructed index, such that the answer
to Q is a superset of the set of vocabulary terms
matching q
Check each term in the answer to Q against q,
discarding those vocabulary terms that do not match q
Two methods: permuterm indexes and k-gram
indexes
Permuterm Indexes
Use a special symbol $ to
mark the end of a term
Term hello is represented as
hello$
A permuterm index contains
various rotations of each term
augmented with $ all linked to
the original vocabulary term
The permuterm vocabulary: the
set of rotated terms in the
permuterm index
Query Answering One * Symbol
Rotate a wildcard query so that the * symbol
appears at the end of the string
Example: rotate m*n to n$m*
Look up the string in the permuterm index
Find terms n$ma and n$moro ! man and
moron are the answers
Query Answering Multiple *s
Example query: q = fi*mo*er
Conduct query Q = er$fi
Check each term returned from Q against q,
only search the inverted index for those
terms satisfying q
Cost: the permuterm index is quite large
since it contains all rotations of each term
On average 10 times for English documents
Discussion
For query q = f*mo*er, we can run queries
Q1 = er$f and Q2=mo and obtain the
intersection of the answers
Is the method good? Why?
For query q = b*etro*t
Run query Q1 = t$b*
Run query Q2 = etro*
Which way is better? Why?
K-gram Indexes
A k-gram is a sequence of k characters
Use symbol $ to denote the beginning and end
of a term
3-grams of castle: $ca, cas, ast, stl, tle, le$
A k-gram index contains all k-grams that
occur in any term in the vocabulary
Each postings list points from a k-gram to all
vocabulary terms containing that k-gram
Query Answering
Example query re*ve
Run the Boolean query $re AND ve$
False positive may happen
Query red*
Run Boolean query $re AND red
Term retired is an answer
Postfiltering: check terms returned from the
Boolean query against the original query
More on Wildcard Queries
Wildcard queries can be quite expensive
The added lookups in the special index, filtering
Most commonly, the capability of wildcard
queries is hidden behind an advanced
query interface
Most users never use
Do not encourage users to invoke wildcard
queries when they do not require it
Reduce the processing load on a search engine
Summary
Vocabulary lookup: hashing versus search
trees
Wildcard queries are powerful in search
Permuterm indexes
K-gram indexes

L14 - Wildcard Queries

Uploaded by

Copyright:

Available Formats

L14 - Wildcard Queries

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

L14 - Wildcard Queries

Uploaded by

Copyright:

Available Formats

Wildcard Queries

J. Pei: Information Retrieval and Web Search -- Wildcard Queries 2

You might also like