Interactive information organization: Techniques and evaluation

The explosive growth of digital information available on-line and ubiquity of the Internet require development of effective techniques for information search and access. Locating interesting information on the World Wide Web is the main task of on-line search engines. Such an engine accepts a query from a user and responds with a list of documents or web pages that are considered to be relevant to the query. The pages are ranked by their likelihood of being relevant to the user\u27s request. The majority of today\u27s Web search engines follow this scenario. The ordering of documents in the ranked list is simple and intuitive. The user is expected to follow the list while examining the retrieved documents. In practice, browsing the ranked list is rather tedious and often unproductive. Existing evidence shows that users quite often stop and do not venture beyond the first screen of results or the top ten retrieved documents. In this thesis we study alternative document organization techniques which can help the user to find the relevant information in the retrieved data much more quickly than the ranked list. The performance of an interactive system is determined by both the system components and the user of the system. We introduce a novel evaluation approach that is based in part on modeling the system-user interaction. It allows us to separate the user\u27s effect on the overall performance from the system qualities. We apply this evaluation method to two different document organization techniques. The first technique exhaustively partitions the document set into well-defined groups. The second system visualizes documents as objects in space arranged in proportion to the inter-document similarity. We show that both systems can be used much more effectively than the ranked list approach. We use a reinforcement learning algorithm to build a “wizard” tool that helps the user to navigate the system. This wizard provides better support than traditional relevance feedback. In conclusion we illustrate the application of the technologies evaluated in the thesis on an example of document organization interface system for a web-based search engine