A hybrid recommender system for recommending relevant movies using an expert system. (15th November 2020)

• Record Type: Journal Article
• Authors: Walek, Bogdan; Fojtik, Vladimir
• Is Part Of: Expert systems with applications. Volume 158(2020)
• Publication Date: 2020-11-15
• Subjects: Recommender system -- Hybrid recommender system -- Expert system -- Collaborative-filtering -- Content-based filtering -- Movies -- Rating; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2020.113452
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: A hybrid recommender system using an expert system is proposed. The system combines a CF system, a CBF system and an expert system. The expert system evaluates the importance of each recommended movie. The system is compared with other typical recommender approaches. The system is validated on a group of users with promising results. Abstract: Currently, the Internet contains a large amount of information, which must then be filtered to determine suitability for certain users. Recommender systems are a very suitable tool for this purpose. In this paper, we propose a monolithic hybrid recommender system called Predictory, which combines a recommender module composed of a collaborative filtering system (using the SVD algorithm), a content-based system, and a fuzzy expert system. The proposed system serves to recommend suitable movies. The system works with favorite and unpopular genres of the user, while the final list of recommended movies is determined using a fuzzy expert system, which evaluates the importance of the movies. The expert system works with several parameters – average movie rating, number of ratings, and the level of similarity between already rated movies. Therefore, our system achieves better results than traditional approaches, such as collaborative filtering systems, content-based systems, and weighted hybrid systems. The system verification based on standard metrics (precision, recall, F1-measure) achieves results over 80%. The main contribution is the creation of a complex hybrid system in the area of movie recommendation, which has been verified on a group of users using the MovieLens dataset and compared with other traditional recommender systems.