AUTOMATIC ROUTING AND RETRIEVAL USING SMART - TREC-2

The Smart information retrieval project emphasizes completely automati c approaches to the understanding and retrieval of large quantities of text. We continue our work in the TREC 2 environment, performing both routing and ad-hoc experiments. The ad-hoc work extends our investiga tions into combining global similarities, giving an overall indication of how a document matches a query, with local similarities identifyin g a smaller part of the document that matches the query. The performan ce of the ad-hoc runs is good, but it is clear we are not yet taking f ull advantage of the available local information. Our routing experime nts use conventional relevance feedback approaches to routing, but wit h a much greater degree of query expansion than was previously done. T he length of a query vector is increased by a factor of 5 to 10 by add ing terms found in previously seen relevant documents. This approach i mproves effectiveness by 30-40% over the original query.