Clustering and Case-Based Reasoning for User Stereotypes

A user stereotype represents a certain kind of user who exhibits a set of specific characteristics - an abstraction of a group of similar users. Looking at statistical data gathered from a population of users, these groups can be identified either manually or using automated clustering %%@ techniques, and constructed by generalizing the most significant features of the identified groups. Case-Based Reasoning (CBR) is an Artificial %%@ Intelligence (AI) method based on the idea of reusing past experience in a domain-specific library of problem-solution descriptions, known as cases. %%@ By representing a solution to the problem of supplying a typical kind of user with appropriate information, it is natural to see user stereotype cases %%@ as part of a CBR process. This thesis describes the usage and creation of user stereotypes in novel domains, aided by the use of clustering %%@ techniques. The first application domain is personalization on the World Wide Web (WWW), where user stereotypes and a filtering technique called %%@ category-based filtering are combined to handle a frequently occurring problem on WWW sites attempting to automatically recommend items of interest to %%@ site visitors. In the second application domain, psychophysiological medicine, clustering is utilized to identify recurring patterns of previously classified time-series of Respiratory Sinus Arrhythmia (RSA). Using a combination of expert knowledge and repeated clustering, the aim %%@ is to incrementally build a case library of stereotypes which can be used in a CBR system for automated classification of RSA sequences.