1 Collaborative Maintenance Maria-Angela Ferrario & Barry Smyth Department of Computer Science, National University of Ireland, Dublin Belfield, Dublin 4, Ireland 1 INTRODUCTION In this paper we examine a classical AI problem (knowledge maintenance) and propose an innovative solution (collaborative maintenance) that has been inspired by the recommendation technique of collaborative filtering and the concept of web communities. In particular, we will look at Ulysses, a database driven, entertainment Web site that uses user profiling, case-based reasoning, and automated planning techniques to provide users with personalised entertainment guides for Dublin. Specifically, we will describe the collaborative maintenance approach, which facilitates a distributed maintenance strategy, whereby Ulysses users are encouraged to maintained the back-end information repository. 2 BACKGROUND Ulysses was originally conceived of as a CBR-based on-line entertainment planner – the objective was to develop an Internet-based system, capable of monitoring and profiling the needs of individual users, and of producing personalised entertainment guides (plans) for these users. Ulysses is implemented as a database-driven web site with a layer of artificial intelligence technology to take care of the user profiling, planning and content recommendation. Since the system has been developed the issue of maintenance has come sharply into focus. Specifically, the nature of Ulysses’ domain is such that the source information is constantly changing and that this information is very subjective in nature. This presents two problems: 1) how can we maintain a current information repository; and 2) how can we characterise and rate the information items so that they can be correctly recommended to the right users at the right time. For reasons of limited resources the idea of a dedicated content editor was a non- starter, and instead we decided to look to artificial intelligence techniques for a potential solution. 2.1 Core Technologies The success of the World-Wide Web has helped to shape the development of a number of important concepts. Firstly there is the idea of virtual communities [3]; that is, the idea that online users can be automatically clustered into groups with similar tastes and preferences. Closely tied with this concept is the technique of collaborative filtering, which is an approach to information filtering and recommendation that uses the virtual community idea [1, 4, 5, 7]. Very simply, information for a