Jimmy Lin
The iSchool
University of Maryland

8th floor conference room, 38A
Tuesday, March 11, 2008, 2pm

Abstract

In this talk, I will describe attempts at enhancing text retrieval in the biomedical domain through a better understanding of how PubMed users behave.  
Analysis of transaction logs from PubMed reveals that there is some level of predictability in users' interactions with the search engine, which suggests 
that accurate user modeling may be possible.  In addition, log data shows that related article links in PubMed are useful for information-seeking tasks.

Visual and statistical characterizations of related article networks are able to explain why related links are useful, in relationship to well-known ideas 
in information retrieval such as the cluster hypothesis.  This analysis suggests that standard graph algorithms may be applied to enhance text retrieval. 
I will demonstrate that this is indeed so: PageRank scores computed on related article networks can yield statistically-significant gains in retrieval effectiveness, 
as measured by standard metrics using TREC test collections.


About the Speaker

Jimmy Lin is an assistant professor in the iSchool at the University of Maryland.  He received his Ph.D. in computer science from MIT in 2004.  
Jimmy's research lies at the intersection between information retrieval and natural language processing, with a special interest in the biomedical domain.  
He leads the IBM/Google "Cloud Computing"
initiative at Maryland.