Adaptable inheritance-based prediction model for multi-criteria recommender system

A recommender system is an emerging personalization strategy in web applications to deal with information overload. Most recommender systems suggest items to users based on a single criterion, i.e., overall ratings. However, multi-criteria ratings are used for modeling more complex preferences of users. The incorporation of criteria ratings can lead to generating more reliable recommendations for users. Despite the success of multi-criteria methods, there is a need to be further optimized to handle popular issues, e.g., sparsity and cold-start. This research study presents a novel multi-criteria collaborative filtering recommender system based on optimization algorithms. The proposed method consists of an adaptable predictive model called multi-criteria inheritance-based prediction (MC-INH-BP). MC-INH-BP allows the customizing of the predictive model to suit the user context. Also, we propose a user profiling method called dynamic user interest print (D-UIP). The D-UIP stores the dynamic preferences of the users. The use of D-UIP reduces the impact of four challenges, critical users, tolerant users, dynamic opinion of the recurring users, and dynamic quality of the item. A set of experiments are conducted to compare MC-INH-BP with other single-criterion and multi-criteria collaborative filtering methods. The benchmark dataset, HotelExpedia, is used. The results prove the capability of MC-INH-BP to achieve better prediction accuracy regardless of the current context of the user. Besides, the results reveal that MC-INH-BP mitigates the cold start and sparsity issues.