Scribd
Upload
27 views2 pages

Overview of The Project

This document provides an overview of a recommender system project. The project aims to recommend items like research papers to users based on their past feedback. It will use Poisson factorization, which enjoys more efficient inference and better handles sparse data compared to Gaussian factorization. The system will also integrate item relations to further improve performance. Two real-world datasets from CiteULike will be used to test the system, with one containing citation relations between articles collected from Google Scholar.

Uploaded by

kks
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
27 views2 pages

Overview of The Project

This document provides an overview of a recommender system project. The project aims to recommend items like research papers to users based on their past feedback. It will use Poisson factorization, which enjoys more efficient inference and better handles sparse data compared to Gaussian factorization. The system will also integrate item relations to further improve performance. Two real-world datasets from CiteULike will be used to test the system, with one containing citation relations between articles collected from Google Scholar.

Uploaded by

kks
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 2

1.

1 OVERVIEW OF THE PROJECT


Th CTPF is to recommend items to users based on their past feedback of the
users. For example, we can deploy a recommender system to recommend
papers (references) to researchers in CiteULike Poisson factorization is a form
of probabilistic matrix Compared to Gaussian factorization, Poisson
factorization enjoys more efficient inference, better handling of sparse data, and
significantly improved predictive performance
To avoid clutter and better demonstrate the differences
between other stronger models, we choose to drop the
corresponding lines for baselines Markov Chain CTPF can
further improve the performance by effectively integrating the
item relations into modeling User-oriented recommendation
tries to recommend items to target users To avoid clutter and
better demonstrate the differences between other stronger
models, we choose to drop the corresponding lines for
baselines MC and MP in the following experiments.
. We use two real-world datasets to conduct our
experiments.Both of them are from CiteULike,2 but they are
collected in different ways with different scales and degrees of
sparsity. For the feedback matrix in the datasets, if a user reads
(or posts) a paper, the corresponding feedback is1. Otherwise,
if a user has not read (or posted) a paper, the corresponding
feedback is missing (denoted by 0). The first dataset, citeulikea, is from [45]. Note that the original dataset in [45] does not
contain relations between items.
. We also crawl the citations between the articles from
Google Scholar. Note that the final number of tags associated
with all the collected articles is far more than the number

(273)of seed tags. we remove any users with fewer than three
articles.

You might also like