Research On Deep Learning Method For Cross Domain Recommendation Via Knowledge Transfer

Recommendation algorithms in a single domain have been widely developed,and the most widely used of which is the collaborative filtering recommendation algorithm.With the dramatic increase of users and projects,traditional collaborative filtering algorithms have faced the huge challenges of data sparsity.Cross domain recommendation technology is one of the effective methods to solve this problem.The goal of cross domain recommendation is to obtain user preferences from the auxiliary domain and transfer them to the target domain,which can alleviate the problem of sparse data in the target domain and then improve the performance of recommendation.Mature fields have accumulated a large amount of user behavior data,it not only can effectively alleviate the data sparsity and cold-start problems faced by recommendation systems with low user access,but also it can improve user's satisfaction and experience by sharing information resources between fields.This paper proposes two deep methods for cross domain recommendation via knowledge transfer,and the main work of this article is as follows:(1)We analyze the background,the significance and the current state of cross domain recommendation,providing solid theoretical support for later research.(2)We propose a deep method cross domain recommendation,which incorporates aesthetic characteristics based on the fact that a user has consistent aesthetic preferences between different domains.We use aesthetic preferences as a bridge and transfer user's aesthetic preferences from auxiliary domain to target domain,which can alleviate the data sparsity issues in target domain and improve the recommendation performance of the target domain.Compared with current the-state-of-art methods,our method outperforms them with multiple different criterions,verifying the effectiveness and reliability of our method.(3)We propose a deep cross domain recommendation method combined with unstructured text.This method transfers user preferences from the auxiliary domain to the target domain and mixes the unstructured text information in the target domain to alleviate the data sparsity issues in the target domain from two aspects.So that it can improve recommendation performance in target domain.Compared with current the-state-of-art methods,our method outperforms them with multiple different criterions,verifying the effectiveness and reliability of our method.(4)We design a clothing cross domain recommender system based on the above proposed two methods.This system analyzes users' behavior records and combines the proposed methods to provide personalized recommendations for user.