عنوان مقاله [English]
Objectives: Due to the increasing share of consumption and watching videos - especially movies and series - in the basket of Iranian households, several systems have been set up to facilitate people's access to these videos. One of the most important types of these systems is the video-on-demand system which has taken unprecedented growth in attracting audiences in recent years. Just as the multiplicity of content in these systems causes users to be diverse and satisfied, this multiplicity can be more confusing for them to find interesting content. Therefore, the need for recommendation systems to further predict user interests and provide consistent content is felt more and more day by day. The purpose of this study is to provide an efficient method of recommending videos based on user viewing data in the video-on-demand system.
Methods: In this research, a new bidding algorithm based on users' tastes and video watching data in the video-on-demand system is presented. This algorithm is based on the concepts and indicators of social network analysis. How this algorithm works is that first the similarity of the videos is calculated based on the percentage of movies viewed by the user and based on that the similarity matrix of the movies is formed. In the next step, based on the similarity matrix of the films, the communication graph of the films is formed, and in the next step, while discovering the communities in the graph, the centrality indicators of each film are calculated.
Then, based on the viewing data of each user and the history of his favorite and non-favorite movies, the members of the community with the most favorite movies are selected as candidates for the user and after calculating their distance from the user's favorite and non-favorite movies, and if the final index is positive, Bids are offered to the user.
Resultds: The performance of the proposed algorithm of this research is evaluated on the data of 50, 100, and 200 users of the CinemaMarket site, the results of which show the better performance of the proposed algorithm of this research in comparison with the algorithms of Naive Bayes, k-nearest neighbors and ID3.
Conclusions: Recommendation systems offer content by having a lot of information from users and their behavioral records. The proposed algorithm of this research can perform the task of suggesting the desired content of users with the least possible information, ie information related to users' observations with desirable and acceptable performance.