Labeling algorithm using fingerprint similarity and query distance based on click-through data for optimizing search

Historical click-through data is very effective to improve performance of search engine. In this paper, labeling algorithm (LA) using fingerprint similarity and query distance based on click-through data is proposed. This strategy obtains relevant pages of query by two ways. One uses labels between query and page produced in bipartite graph. Another one treats relevant pages of similar queries as its own. Finally we calculate relevancy between queries and pages using fingerprint similarity and query distance. Based on AOL query logs, we conduct experiments on LA. Results show LA can greatly improve search with low computational complexity. Compared with other query log mining algorithms, LA achieves higher precision and lower hit location.