| In this thesis we adopt an artificial intelligence approach in investigating the online information retrieval activity and in studying the design of online information retrieval systems. Our initial approach was to investigate the interactions between the searcher and the retrieval system, and the searcher and the reference librarian. Our objective was to develop cognitive models of these processes that will provide a sound basis for incorporating search assistance into an online system.; This research consisted of three stages: an empirical investigation stage, a system building stage, and a system testing stage. During the empirical investigation stage, we conducted two cognitive psychology based empirical studies observing the searcher/librarian and the searcher/system interactions. We have derived some important findings from these studies. Firstly, we have identified a taxonomy of misconceptions users have about the subject area, the system and the classification scheme. Secondly, we have developed five online search strategies. Thirdly, we have developed an empirically based process model for assisting searchers in refining their queries.; At the system building stage, a blackboard architecture was adopted for our prototype system where the proposed capabilities (alleviating searcher's misconceptions, selecting appropriate search strategies, and refining searchers' queries) were incorporated. This system was created using Lisp and Flavors.; After the prototype system development effort, a system testing stage followed. Our system was tested against an existing online catalog in a laboratory experiment. The test results showed that our system out-performed the existing online catalog in search recall and satisfaction level.; Our research suggests a knowledge-based information retrieval system design. Armed with the appropriate search heuristics, the knowledge about the subject area and the classification scheme, and the query refinement mechanisms, we believe, information retrieval systems can become more responsive and intelligent, capable of providing interactive support of a variety of users. |
