Mining Navigation Behaviors for Term Suggestion of Search Engines

Abstract

Query expansion is extensively applied in information retrieval systems, such as search engines. Most conventional approaches to query expansion have been developed based on textual analysis of documents. However, different issues such as segmentation and feature selection must be addressed, which might influence performance seriously. This work focuses mainly on avoiding the above problems of textual analysis and thus proposes a collaborative method of applying access logs in the search engines to term suggestion (i.e., query expansion). A co-clicked behavior-based term suggestion is presented to suggest user-oriented terms. Analyzing the co-clicked behaviors of users in the access logs for term suggestion eliminates the need to perform textual analysis and provides some positive characteristics that previous approaches neglected, such as content independent, adaptability, and extensibility. Furthermore, limitations of current search engines, including problems of word mismatch and partial match, can also be overcome. Here, a search engine prototype is also developed to demonstrate the results of term suggestion. Experimental results demonstrate that the precision rates of information retrieval systems can be improved by using the suggested terms of the proposal presented herein. The proposed term suggestion also performs better than another famous Chinese term suggestion system, namely OPENFIND.