Contextualization from the Bibliographic Structure

Abstract

Bibliographic or citation structure in a document contains a wealth of useful but implicit information. This rich source of information should be exploited not only to understand what and where to nd the important documents, but also as a contextual evidence surrounding the important and not so important documents. This paper measures the e ects of contextual evidences accumulated from the bibliographic structure of documents on retrieval e ectiveness. We propose a re-weighting model to contextualize bibliographic evidences in a query-independent and query-dependent fashion (based on Markovian random walks). The in-links and out-links of a node in the citation graph could be used as a context. Here we hypothesize that the document in a good context (having strong contextual evidences) should be a good candidate to be relevant to the posed query and vice versa. The proposed models are experimentally evaluated using the iSearch Collection and assessed using standard evaluation methodologies. We have tested several variants of contextualization, and the results are signi cantly better than the baseline (indri run).