Sequence recommendation algorithm based on cross domain user preference migration

Abstract

Data sparsity is a problem faced by most recommendation systems. Sequence recommendation is a recommendation method to predict user preferences based on user behavior sequence. There is also the problem of data sparsity. User behavior sequences are scattered in different fields, and user interests are related in different fields. User behavior sequence data in some fields are very sparse, and user behavior sequence data in some fields are relatively dense. From a single domain, when the user behavior sequence data is sparse, the user preference model can not be accurately constructed based on the user behavior sequence. In order to solve this problem, starting from the relevance of domain user interests, this paper migrates the user preference information of related domains to build the user preference model of the target domain, and proposes a sequence recommendation method based on cross domain user preference migration. Firstly, this paper studies the feasibility of knowledge map guiding preference migration, constructs cross domain knowledge map according to the connection relationship between items in different fields, and embeds the relationship path between items into item representation, so that the relationship path can guide user preference migration between different fields. Then, the domain preference distribution and general preference distribution are established. The MMD distance is used to align the general preference distribution, and the anti transfer learning is used to align the domain preference distribution and general preference distribution. According to the user's general preference distribution in the source domain, the user's preference model in the target domain is constructed. Finally, experiments verify the effectiveness and correctness of the proposed model.