Using Metadata for Query Refinement and Recommendation

Abstract

Lengthy lists of search results are the fruit of both short queries and conventional Web search result displays. They are problematic for meeting user’s information needs. This paper describes topic extraction and representation from metadata, the first part of a project that will develop an interactive visual query refinement and recommendation (QRR) service to alleviate the problems due to lengthy lists of search results. The topic extraction uses the Latent Dirichlet Allocation (LDA) algorithm to mine the intra- and inter-document relations and represent them in topic and features. The paper presents how the LDA algorithm extracts topics and features from metadata records contained in NSDL search results, which will be used by an interactive visual QRR service in the next step of the project.