Adaptive Information Retrieval and Its Application to Executive Information Systems


Rey-Long Liu



As more and more information become available, the quality of retrieving and filtering the information will be a key concern of information users. Therefore, many previous studies on Information Retrieval (IR) focused on how relevant information may be retrieved for general users. However, for those users like executives of an organization, relevant information may be useful only if each executives individual (and even unique) ways of working and thinking may be considered. In order to retrieve useful information efficiently and effectively, an IR system should be able to adapt itself to each different executives behaviors of requesting information.

In this project, we will study IR techniques from the perspective of individual information needs of executives. We plan to design an Adaptive Information Retrieval system (AIR) and then apply the system to the design of Executive Information Systems (EIS). In the study, AIR may serve as an adaptive personal IR agent for executives. It may dynamically capture the executives information needs by observing his/her preferences of information. The preferences include the hyper-links, keywords, concepts, and natural language queries that are frequently and recently issued by the executive. By recording both the contents and the shifts of the user preferences, AIR may know the current trend of information needs of the executive and thus accordingly provide suitable information services to the executive.

AIR contains four main modules: the Rule-Based Reasoning module (RBR), the Case-Based Reasoning module (CBR), the Thesaurus Browsing module (TB), and the Document Browsing module (DB). RBR accepts from the user natural language queries and outputs the critical features for specifying the case of information needs. CBR then uses the features to identify suitable cases for setting the search parameters and strategies. TB then allows the user to browse the set of thesaurus entries selected according to the search parameters and strategies. DB allows the user to browse the corresponding documents of the thesaurus entries clicked by the user. When the user is browsing thesaurus entries (in TB) and documents (in DB), AIR will record the users preference for refining the performance of RBR which may be extended by learning new rules using machine learning techniques. After a few transactions, AIR may become more personalized for the user.

AIR will be applied to the design EIS, because providing personalized information services is a critical issue of developing EIS as well. In the application, we will develop a set of cases (i.e. types) of information needs for CBR and explore the performance of AIR in the context of EIS and the Internet.

Keywords: Adaptive Information Retrieval, Machine Learning, Executive Information System

     Back to Research Project