1. Introduction

With the explosive growth of online information, such as news articles, email messages, scientific literature, government documents, and information about all kinds of products on the Web, search engines have now become essential tools in all aspects of our life; clearly, their effectiveness would directly affect our productivity and quality of life.

The current generation search engines are very useful, but they tend to work well only for easy queries (e.g., finding homepages or finding popular/known topics) and would perform poorly for difficult queries (e.g., queries with ambiguous words or queries about unknown topics). It is quite common that a user would have to reformulate a query many times, yet still would not get satisfactory results. In such cases, the user has to either spend a significant amount of time in searching or simply give up. Thus improving search accuracy for such difficult queries can bring significant benefits to users.

This project aims to develop an interactive search environment (called QueryClinic) to better support a user in finding such hard-to-find information. The main idea behind QueryClinic is to make the search process a collaborative process in which QueryClinic and a user would interact with each other and work together toward improving search results. This is in contrast with the current search process in which a search engine passively responds to a user's query with some search results. Specifically, QueryClinic would actively involve a user in the search process so that the user can give more input to the search process and also receive more guidance and assistance in reformulating queries. In this way, the system would be able to collect more informative feedback information and analyze the entire interaction session as a whole (rather than just a few words in the query) to understand more precisely the user's information need, which in turn helps the system better direct the search process and improve search accuracy.

2. Team members

• ChengXiang Zhai (Professor)
• Alexander Kotov (Ph.D. student)
• Yuanhua Lv (Ph.D. student)
• Parikshit Sondhi (Ph.D. student)
• V.G.Vinod Vydiswaran (Ph.D. student)
• Xuanhui Wang (Ph.D. student)

• Negative feedback: When the initial search results are extremely poor, all the top-ranked documents may be non-relevant. To help users in such a situation, we studied how to learn from the top-ranked non-relevant documents (i.e., negative feedback) to improve the ranking of the unseen documents and proposed effective methods for negative feedback [Wang et al. SIGIR 08].
  
  
• Browsing with multi-resolution topic map (Systematic Query Suggestion): When querying doesn't work well, browsing can be very useful. Current search engines support browsing mainly in two ways, i.e., through hyperlinks and manually fixed categories. We proposed to build multi-resolution topic maps to enable a user to navigate into relevant information without needing to formulate queries, thus users can benefit from browsing when they cannot formulate effective queries. A topic map can also be regarded as a way to systematically suggest queries to users. We further propose to construct a multi-resolution topic map based on search logs, so that future users can follow the "footprints" left by previous users in the information space, achieving social surfing. Evaluation shows that users can find relevant information more effectively with navigation than with query reformulation, especially when the queries are difficult [Wang et al. CIKM 2008, Wang et al. CIKM 2009]. Click here to see more information about social surfing.
  
  
• Natural question guided search: We studied how to engage users in an interactive retrieval system naturally through presenting clarification questions to users as supplementary information to regular search results. Such a question-guided search paradigm not only enables a user to reach directly the answers to questions related to the information need, but also serves as a natural way to refine the user's interests. Accurate determination of these questions may substantially improve the quality of search results and usability of search interfaces. We proposed a new framework for question-guided search, in which a retrieval system would automatically generate potentially interesting questions to the users. Since the answers to such questions are known to exist in search results, these questions can potentially guide the users directly to the answers they are looking for, eliminating the need to scan the documents in the results list. Moreover, in case of imprecise or ambiguous queries, automatically generated questions can naturally engage the users into feedback cycles to refine their information need and guide them towards their search goals. Implementation of the proposed strategy raises new challenges in content indexing, question generation, ranking and feedback. We proposed new methods to address these challenges and evaluated them with a prototype system on a subset of Wikipedia. We evaluated these methods with a user study. Experiment results show that the proposed method for question-based query refinement allows the users to more easily navigate in search results and effectively explore the results space in an interactive and natural way [Kotov & Zhai WWW 2010].

4. Selected Publications (See all publications)

---

Text Information Management Group Home | About| People | Projects| Publications | Downloads| Links| Wiki