Bibliographic Latent Semantic Analysis

In some cases, when users conduct a search, they not only want the retrieved documents to be relevant to the query, but also the documents to be relevant to each other. To address such expectation, we propose a new retrieval method--Bibliographic Latent Semantic Analysis (BLSA)-- to automatically capture and build inter-links between two documents sharing the same bibliographic information such as authorship. BLSA expands the Latent Semantic Analysis (LSA)’s term-document matrix by including bibliographic information before the Singular Value Decomposition (SVD) processing. The primary advantage of applying BLSA in information retrieval is that, under certain circumstances, BLSA enhances the correlations between documents sharing the same bibliographic information. As a result, these documents will appear closer on the retrieval return list. Two case studies are given to illustrate the implementation of BLSA in real applications and to compare its performances with that of LSA and Vector Space Model (VSM)