Extract-biased pseudo-revelance feedback

Abstract

Successfully retrieving a web document is a twofold problem: having an adequate query that can usefully and properly help filtering relevant documents from huge collections, and presenting the user those that may indeed fulfill his/her needs. In this paper, we focus on the first issue -the problem of having a misleading user query. The aim of the work is to refine a query by using extracts instead of full documents. Extracts, in our context, are actually summaries of documents of a hitlist produced by an extractive automatic summarizer. Automatic summarization of single and multi-documents is explored through GistSumm, our Gist Summarizer, which is based on the gist of a document, hence its name. Results on pseudo-relevance feedback for the Portuguese CHAVE collection show that gist-based extracts may improve information retrieval.

Summarizing Scientific Texts: Experiments with Extractive Summarizers

“…Summaries are also important tools for other automatic tasks. In information retrieval, for instance, it was shown that indexing summaries may be better than indexing the complete documents (see, e.g., [10]) and that summaries are useful for refining queries (e.g., [1]).…”