User models: theory, method, and practice

The prospects and implications of automatic filtering of information, and the results suggest that the prediction of preferences can be straightforward when general categories for news articles are used; however, prediction for specific news reports is much more difficult.

Abstract

While the technology of new information services is rapidly advancing, it is not clear how this technology can be best adapted to people's needs and interests. One possibility is that user models may select and filter information sources for readers. This paper examines the prospects and implications of automatic filtering of information, and focuses on predicting preferences for news articles presented electronically. The results suggest that the prediction of preferences can be straightforward when general categories for news articles are used; however, prediction for specific news reports is much more difficult. In addition, an effort is made to establish a systematic study of the effectiveness of information interfaces and user models. Fundamental issues are raised such as techniques for evaluating user models, their essential components, their relationship to information retrieval models, and the limits of using them to predict user behavior at various levels of granularity. For instance, prediction and evaluation methodology may be adopted from personality psychology. Finally, several directions for research are discussed such as treating news as hypertext and integration of news with other information sources.