Connectionist Data Mining Techniques for Deriving Characteristics of System Use

摘要

This paper describes two separate research projects, which utilise a similar methodological approach to data mining for user-specific characterisation analysis. One project is concerned with authorship characterisation using a large database of email message traffic and the other relates to profiles of users and their use of a digital library. Both projects utilise standard statistical techniques as part of their analytical framework, but both have employed connectionist methods using fuzzy labelling and clustering in conjunction with artificial neural networks, (in this case Kohonen self-organising maps often referred to as SOM's), in order to assess the potential of this approach to data mining for deriving system evaluation characteristies. Each project is summarised in this paper and some results of the work to-date described. It is suggested that connectionist data mining techniques are at least of high quality supplemental value when considering alternative methodological approaches to system evaluation studies.