An Integrated System for Filtering News and Managing Distributed Data

With the development and diiusion of the Internet worldwide connection, a large amount of information can be delivered to the users. To avoid their being overrowed by the incoming data, methods of information ltering are required. Thus, there is the problem of determining what information is relevant to the user and how this decision can be taken by a supporting system. Parametric and qualitative descriptors of user's interest must be generated. This paper presents two approaches. The rst concerns an information ltering system based on an adaptation of the generalized probabilistic model of information retrieval. The user proole is a vector of weighted terms which are learned from the relevance assessment values given by the user on the training set. Positive terms are considered relevant to the informative need of the user, negative ones irrelevant. The relevance values are interpreted as subjective probabilities and hence are mapped into the real interval 0; 1]. ProFile is a ltering system for the netnews which uses this model with a scale of 11 predeened values of relevance. ProFile allows the user to update on{line his proole and to check the discrepancy between his assessment and the prediction of relevance of the system. The second concerns the InfoAgent, a system for supporting users in retrieving data in distributed and heterogeneous archives and repositories. The architecture is based on the metaphor of the software agents and incorporates innovative hints from other elds: distributed architectures, relevance feedback and active interfaces. The system has a cooperative and supportive role: it understands the user's needs and learns from his behavior. Its aim is to disengage the user from learning complex tools and from performing tedious and repetitive actions.