Investigation And Implementation Of Social-group-based Video Recommendation Algorithm

With the gradual maturity of Internet technology, social networking servers have been greatly prevalent in our daily lives. Video recommendation has become an essential part of online video service.The current mainstream recommendation algorithms include content-based filtering, collaborative filtering, and social network-based recommendation. The common drawback of these algorithms is their deficiency to deal with cold start and data sparsity. Without adequate historical viewing records, traditional video recommendation algorithms often cannot satisfy the needs of users. In this work, the author proposed a social-group-based recommendation algorithm, which considers both user preference and social group affiliation in social networks. Based on the implicit feedback of individual users, his or her friends, and group mates, the algorithm recommends videos that may be of users' interest.The recommendation results are approved by the users, which achieved higher click-through-rate. The main contributions of this work are listed as follows.1. This work investigated the research directions as well as relevant techniques of personalized recommendation and social recommendation at home and abroad. The author compared several representative algorithms, analyzed the features of social group, and identified the difference of candidate video based on analysis of social-group affiliation and friends' preference.2. This work designed and implemented a social-group-based personalized recommendation model and algorithm. The author elaborated on the workflow of the algorithm through three key parts:intra-group video ranking, group scoring, and result aggregation. The proposed algorithm could not only achieve recommendation based on information within a single-group, but also aggregate multiple groups'output to obtain the optimal results.3. This work compared the social-group-based video recommendation algorithm with traditional collaborative filtering and content-based filtering algorithm. Compared with traditional video recommendation algorithms, social-group-based algorithm improved the click-through rate and increased the diversity of recommended videos.The experiments also proved that social-group-based algorithm could effectively reduce the influence of cold start and data sparsity, and improve the recommendation accuracy, diversity, serendipity,4. This work further explored the respective advantages of collaborative filtering and social-group-based algorithm in terms of recommendation click-through-rate, and further explored potential hybrid strategy.