Understanding user search intentions

Abstract

With the current boom surrounding Web 2.0, Search engine optimization is becoming increasingly important and there has been a noticeable interest in understanding the intention of a user's query to improve search engine performance. User queries on the web can be classified into three types according to users intention: informational query, navigational query and transactional query. In this project we study how web search engine query intentions can be identified and to which extent by constructing an automatic intention classification scheme based on the maximum entropy model. To get maximum performance out of the classifier we exploit several features extracted from a major search engine's results, past user click data, web search query logs, and related web document content. To evaluate our classification scheme we carried out a human subject study and compared the automatic classification results to manually annotated results and observed considerable high agreement. Finally, by evaluating the classifier with one query log and then demonstrating its classification performance other query logs, we establish its robustness. Our evaluation shows that while statistical methods perform best for navigational query identification, text classification techniques are best at identifying transactional queries.