Query Reformulation Type Classification Using Query Log

概要

論文の詳細を見る
Most web search engines today recommend more specific queries to users based on their current query. We think important query search information can come from looking at previous queries and seeing how the user reformulates their next query to find the results they are looking for. Our work aims to predict one of five given reformulation types between sequential query pairs. We use a support vector machine algorithm which utilizes features that analyze each query and the correlation between previous and post queries. In addition, we add characteristic classifiers which focus on only two reformulation types and finding distinct differences between them to aid the SVM algorithm predictions. Using both of these methods together produces 83% precision. The ability to predict the type of query a user will input next based on their searching history will lead the way for better query recommendation in web search engines.
2010-05-13