Adapting Information Presentation and Retrieval through User Modelling N. Bogonikolos 2 , C. Makris 1 , A. Tsakalidis 1 , B. Vassiliadis 1,2 1 Department of Computer Engineering and Informatics 26500 Patras, Greece 2 ZEUS Consulting S.A., 93 Riga Feraiou st., 26221, Patras, Greece bb@zeusnet.gr Abstract The amount of structured information available in Internet sources is rapidly increasing. This information includes commercial databases on product information and information on e-services forming the so-called e- shops. However, the process of using this information has become more complicated, and can sometimes be tedious for users with different goals, interests, levels of expertise, abilities and preferences. This paper deals with the definition of an architectural framework for intelligent, adaptive and personalised navigation within large hypertext Electronic Commerce environments. 1. Introduction As business transitions into the new economy, system usability has become a strategic goal. Especially in business to consumer applications, users highly evaluate the quality of their interactive shopping experience. On the other hand, the explosion of available information, in the form of services and products, has created an information stress problem similar to that of the Internet. The process of using e-commerce services has become more complicated, and can sometimes be tedious for users with different goals, interests, levels of expertise, abilities and preferences. Boston Consulting Group recently announced that a full of 28% of online purchasing transactions failed and 1/3 of them stopped shopping on- line due to usability difficulties [1]. This paper expands the model-based architecture presented in [2] to promote adaptivity and increases performance at all levels: information presentation, processing and storage. More specifically, the architecture uses both the decision-making and the rule-based paradigm to achieve a consensus between system performance and adaptivity through user modelling. This hybrid approach enables the efficient use of all kinds of user information (static and dynamic). The rule-based approach constitutes mainly a static approach, in the sense that adaptations are pre-determined by rules and cannot be modified at run-time. This approach has a relatively low cognitive load and is efficient for providing adaptability through user characteristics that are assumed to remain relatively static. Decision-making theory is based on simpler representation formalisms for expressing assumptions about the users and thus is more flexible enabling adaptation decisions at run time [3]. Systems that use either of these approaches have shown great promise in theory but have not yet convinced about their usability. Commercial applications are much more demanding, and in the case of business to consumer commerce, a choice has to be made. Dynamic and less adaptable, or fully dynamic and slower [4]. The proposed server side architecture uses metadata for personalisation, a metaphor that is not just about using adaptation in some ad hoc manner, but understanding service provision. Part of this notion of service provision is based on analysis of metadata created through observation of user behaviour at run time and modelling of user characteristics. The remainder of the paper is as follows: in section 2 an overview of related techniques and applications is provided. In section 3 the architecture is examined in detail. In section 4 algorithms for managing user metadata are presented and finally, section 5 discusses future directions are discussed. 2. Overview of related work Adaptable hypertext systems have been considered in a wide range of recent research efforts. The relevant literature offers numerous examples illustrating tools for constructing adaptive interaction [5], and case studies in which adaptive interface technology has improved, or has the potential to improve, the usability of an interactive system [6,7]. Decision-theoretic techniques and Bayesian networks for managing the complexity of information