Human Factors Based Partitioning Versus Data Clustering for Recommendations

Recently, it has been broadly recognized that valid human factors, such as demographics, emotions and personality factors, influence users´ preferences and decision making process. Researchers in the field of user modeling mostly focus on designing domain-specific questionnaires to obtain human factors explicitly. However, these factors may be not easy to collect in real-world systems because of users´ privacy concerns. In addition, inferring human factors from their implicit behaviors is still a challenging research problem. On the contrary, users´ preferences are meant to be discovered directly from their behaviors in the data driven methodology. In this paper, we attempt to compare the degree of consistency and performance between these two methodologies in identifying similar user groups. To achieve this, we first propose two different user grouping approaches in different methodologies: the user partitioning using human factors, and the data clustering based on users´ behaviors. Furthermore, we demonstrate the high degree of consistency between these two partitioning strategies on two small context-aware data sets, and elaborate the effectiveness of our proposed approach in the item recommendation problem.