Abstract

This paper considers hoxv web search engines can learn front the successful searches recorded in their user logs. Document Transfor marion is a feasible approach that uses these logs to improve document representations. Existing test collections do not allow an adequate investigation of Document TYansformatiom but *ve show how a rigorous evahmtion of this method can be carried out using the referer logs kept by web servers. We also describe a new strategy tbr Document Transformation that is suitable for long-term incremental learning. Our experiments show that Document Transformation inrprovcs retrieval performance over a medium sized collection of webpages. Commercial search engines nmy be able to achieve sinfilar improvements by incorporating this approach.

Citations