Abstract:

The research on movie recommendation algorithms developed rapidly because of the generation of new technologies such as Web 2.0， Big data， and cloud computing. Most recommendation systems operate based on preference similarity or social distance. However， current movie recommendations still face sparse issues on item ratings and social relationships. According to sociological theory， preference similarity is one of the important factors in the formation of social relationships， and social relationships also contribute to the emergence of similar preferences. Thus， considering multiple association relationships between users and the interaction between these relationships is the key to accurately capturing user needs and improving recommendation accuracy. A double-layer network-based recommendation model is proposed that simultaneously considers users’ social relationships and preference similarity. The social network is obtained based on the "follow" relationship and the preference similarity network is built based on similar reviews. Based on the single-layer processing of the double-layer network， the Louvain method is used to reduce the dimension of a large group of users. Then， the influence of intra-class users is calculated based on multi-level relationships and the personalized recommendation is provided for target users. Three main contributions are made by this paper. Firstly， the sparsity problem of users’ evaluation of the target movie is improved by extending the target movie to a collection of similar movies. Secondly， the sparsity of users’ social relationships is alleviated by preference similarity. Thirdly， considering the double-layer relationships， the needs of target users and recommendation lists can be accurately predicted. This paper takes Douban data （including 384548 users， 269469 movies， and the Top 250 movie list in 2022） as an example to verify the effectiveness of the recommendation model. Through experimental analysis， the sparsity problem of user relationships is effectively alleviated， the communities that users are interested in are efficiently identified， and more novel items are recommended to target users. Additionally， this method is also suitable for predicting missing information and identifying similar preference communities， providing data support for a more comprehensive understanding of user preferences.

Key words: double-layer network, social relationships, preference similarity, movie recommendation