Centrality based Document Ranking

Abstract

In this paper, we address the problem of ranking clinical documents using centrality based approach. We model the documents to be ranked as nodes in a graph and place edges between documents based on their similarity. Given a query, we compute similarity of the query with respect to every document in the graph. Based on these similarity values, documents are ranked for a given query. Initially, Lucene1 is used to retrieve top fifty documents that are relevant to the query and then our proposed approach is applied on these retrieved documents to re-rank them. Experimental results show that our approach did not perform well as the documents retrieved by Lucene are not among the top 50 documents in the Gold Standard. © 2014 23rd Text REtrieval Conference, TREC 2014 - Proceedings. All rights reserved.