Clustering Similar Query Sessions Toward Interactive Web Search

Clustering Similar Query Sessions Toward Interactive Web Search Chien-Kang Huang, Lee-Feng Chien, and Yen-Jen Oyang Department of Computer Science, National Taiwan University Page 115 ~ 134 Proceedings of Research on Computational Linguistics Conference XIII (ROCLING XIII) Taipei, Taiwan 2000-08-24/2000-08-25 115 Clustering Similar Query Sessions Toward Interactive Web Search Chien-Kang Huang, Lee-Feng Chien*, Yen-Jen Oyang Department of Computer Science, National Taiwan University, Taiwan. *Institute of Information Science, Academic Sinica, Taiwan. ckhuang@mars.csie.ntu.edu.tw, *lfchien@iis.sinica.edu.tw, yjoyang@csie.ntu.edu.tw Abstract A new effective log-based approach for interactive Web search is presented in this paper. The most important feature of the proposed approach is that the suggested terms corresponding to the user’s query are extracted from similar query sessions, rather than from the contents of the retrieved documents. The experiment results demonstrate that this approach has a great potential in developing more effective web search utilities and may inspire more studies on advanced log mining mechanisms. 1. Introduction Users' queries for Web search are usually short. For example, the average length of TREC topic description for conventional text retrieval is 15 tokens [11,12], while analyses of web search engine logs reveal that the average query length for Web search is about 2.3 tokens [6,9]. Short queries means that the information about the user’s intention provided to the search engine is very limited. To deal with the short query problem, interactive search techniques [2,7] which attempt to identify the user’s intentions and suggest more precise query terms are therefore commonly incorporated in Web search engine design. To determine more relevant query terms for each given query, the conventional 116 interactive search processes often rely on the key terms in the retrieved documents [2,7,10]. The key term set is extracted either statically from the documents during preprocessing or dynamically on-the-fly. Since the precision rates of the retrieved documents are usually not high enough, the extracted key terms are often found not relevant and not very helpful in practical Web search services. In fact, extraction of relevant terms can be carried out by analyzing users’ logs. In recent years, mining search engine logs has been obtaining more attention. Silverstein et al. [9] performed a second-order analysis on a log with a huge number of Web query terms. The results are then used to facilitate phrase recognition and query expansion [3]. In this paper, we propose a new approach based on log analysis for developing more effective interactive Web search engines. The most important feature of the proposed approach is that the suggested terms are extracted from similar query sessions, rather than from the contents of the retrieved documents. A query session is defined as a sequence of search requests issued by a user for a certain information need. The basis of the proposed approach is that two users with the same information need will issue common or related query terms. For example, in search for a subject regarding “search engine technology”, a user may submit query terms such as “search engine”, “Web search”, “Google”, “Web search and multimedia”, while another user …

Content truncated to 500 words