Improving search quality of the Google search appliance

Abstract

In this thesis, we describe various experiments on the ranking function of the Google Search Appliance to improve search quality. An evolutionary computation framework is implemented and applied to optimize various parameter settings of the ranking function. We evaluate the importance of IDF in the ranking function and achieve small improvements in performance. We also examine many ways to combining the query-independent and query-dependent scores. Lastly, we perform various experiments with signals based on the positions of the query terms in the document.