Interactions Between Stereotypes

Despite the fact that stereotyping has been used many times in recommender systems, little is known about why stereotyping is successful for some users but unsuccessful for others. To begin to address this issue, we conducted experiments in which stereotype-based user models were automatically constructed and the performance of overall user models and individual stereotypes observed. We have shown how concepts from data fusion, a previously unconnected field, can be applied to illustrate why the performance of stereotyping varies between users. Our study illustrates clearly that the interactions between stereotypes, in terms of their ratings of items, is a major factor in overall user model performance and that poor performance on the part of an individual stereotype need not directly cause poor overall user model performance.