Combining User-based and Item-based Collaborative Filtering Techniques to Improve Recommendation Diversity

Nowadays collaborative filtering technologies are widely used in many websites, while the majority research literatures focused on improving recommendation accuracy. However, it had been recognized that improving recommendation accuracy was not the only requirement for achieving user satisfaction. One important aspect of recommendation quality, recommendation diversity gained focus recently. It was important that recommending a diverse set of items for improving user satisfaction since it provided each user with a richer set of items to choose from and increased the chance of discovering potential interest. In this study, a synthetically collaborative filtering model was proposed, which combined the user-based and item-based collaborative filtering techniques. This model gave each user an option to adjust the diversity of their own recommendation list by using the prevalence rate and novelty rate parameters. Experiments using real-world rating datasets indicated the proposed model had effectively increased the recommendation diversity with little decrease in accuracy and surpassed the traditional collaborative filtering techniques.