Server Selection with Clustering and Term Correlation

Dyce Jing Zhao

HKUST

A meta-search engine answers a user query by sending the query to selected participant search servers. However, most existing server selection methods are ineffective for multi-term queries, because they assume that the correlation of query terms follows some pre-defined probability function. In order to address this limitation, we propose a server selection method that considers not only the clustering of terms in each search server but also the correlation of terms in each cluster. Our experiments demonstrate that combining clustering and term correlation improves meta-search quality significantly.