Combination of Evidence for Effective Web Search

摘要

In this paper we describe Carnegie Mellon University's sub-mission to the TREC 2010 Web Track. Our baseline run combines different methods, of which in particular the spam prior and mixture model were found the most effective. We also experimented with expansion over the Wikipedia corpus and found that picking the right Wikipedia articles for expansion can improve performance substantially. Furthermore, we did preliminary experiments with combining expansion over the Wikipedia corpus with expansion over the top ranked web pages.