Automatically Creating Reading Lists with Topical PageRank
Add to your list(s)
Download to your calendar using vCal
If you have a question about this talk, please contact Ekaterina Kochmar.
We present an algorithm for creating reading lists – lists of papers given
by an expert to a novice, designed to bring the novice up to speed in a
certain area. Our algorithm uses a variant of PageRank that is age-corrected
and sensitive to the mixture of papers’ topics as determined by the LDA
topic model.
When compared to a gold standard of reading lists which we collected from
experts, our algorithm outperforms three currently used keyword-based search
engines: Lucene, Google Scholar and the Google-indexed ACL Anthology. As
evaluation metrics we use F-measure, as well as a new evaluation metric
specific to reading lists which we introduce here. It estimates the degree
of substitutability of expert papers by system-found ones by the number of
links in the citation network between them. We also evaluate on the task of
reference list reintroduction. When reintroducing the reference list of
thousands of papers, our unsupervised algorithm performs on a par with the
current state-of-the-art method, which is supervised.
This talk is part of the NLIP Seminar Series series.
This talk is included in these lists:
Note that ex-directory lists are not shown.
|