Can Automatic Abstracting Improve on Current Extracting Techniques in Aiding Users to Judge the Relevance of Pages in Search Engine Results?

摘要

Current search engines use sentence extraction techniques to produce snippet result summaries, which users may find less than ideal for determining the relevance of pages. Unlike extracting, abstracting programs analyse the context of documents and rewrite them into informative summaries. Our project aims to produce abstracting summaries which are coherent and easy to read thereby lessening users’ time in judging the relevance of pages. However, automatic abstracting technique has its domain restriction. For solving this problem we propose to employ text classification techniques. We propose a new approach to initially classify whole web documents into sixteen top level ODP categories by using machine learning and a Bayesian classifier. We then manually create sixteen templates for each category. The summarisation techniques we use include a natural language processing techniques to weight words and analyse lexical chains to identify salient phrases and place them into relevant template slots to produce summaries.