School of Electronic Engineering and Computer Science

Reflecting User Information Needs Through Query Biased Summaries

Abstract

It is often the case that information retrieval (IR) systems do not present users with enough information in order to assist them in judging the relevance of documents retrieved in response to a query. Typical IR systems usually output the title and the first few sentences of the source text as ‘relevance clues’. This usually leads to the users having to refer to the full text of the documents in order to locate the relevant to the query content that each document conveys.

The aim of the work reported in this thesis, is to investigate the effectiveness of automatically generated summaries customised to a specific query, in assisting users to judge the relevance of documents retrieved in response to that query. More specifically, a summary will automatically be generated and presented for each retrieved document, aiming to provide users with enough evidence about the relevance (or non-relevance) of each document to the query.

In order to examine the effectiveness of this approach, a comparative evaluation against a typical IR system response is performed. The results from this evaluation indicate that the presence of the automatically generated summaries as accompanying information for each retrieved document, improves the accuracy of the relevance judgements while at the same time reduces to a minimum the need to refer to the full text of the documents.

The Thesis is available as a Technical Report (TR-1997-35) of the Computing Science Department, University Of Glasgow. You can get an electronic copy from here ...

Home

Teaching and supervision