Research Of Representation Learning Based On Deep Learning Model In Recommendation System

With the advent of the era of big data,recommendation systems have been applied in many areas of the Internet,and are gradually changing people's lives.The recommendation system essentially connects users and projects in a certain way,and there are many ways in this way.Different ways constitute different recommendation systems.The cross-platform method formed by the influence of different platforms on user behavior can be developed into cross-domain recommendation;the method of friend recommendation can be developed into social recommendation.Deep learning methods are often used to learn effective low-dimensional representations such as images and text.Such representations can be used to supplement or replace traditional recommendation algorithms,such as collaborative filtering.This article takes deep learning methods as the core and proposes different solutions for cross-domain recommendation and social recommendation respectively.The main innovations of this article are as follows:(1)This paper proposes a representation learning algorithm based on semantic feature confrontation learning in cross-domain recommendation.This algorithm combines content-based and transfer-learning-based methods,and does not require user overlap or item overlap between the auxiliary domain and the target domain..The algorithm proposed in this paper solves the three challenges faced by current cross-domain recommendation: the common cross-domain recommendation systems(CDRSs)are difficult to find the connection between heterogeneous information domains,so most of the focus is on homogeneous information domains,or It is assumed that the users are overlapping;the usual CDRSs are developed based on historical feedback or based on recommending different items in the target domain to explore uncertain user interests,but it is difficult for them to strike a balance between the two;the usual CDRSs because there is no Historical feedback or association proofs for users or projects,so cold start issues will reduce their recommended performance.In order to meet these challenges,this paper extracts the migratable features between related domains and obtains the association,and then conducts experiments on the two data sets.The results show that it has good performance.(2)This paper proposes a representation learning algorithm based on dual attention graph model in social recommendation,which not only considers the retrieval of user preference features and product features from user comments,but also considers the user's social perception network and product Association-aware network impact,which improves recommendation performance.The main contribution of the algorithm proposed in this paper has three aspects: for the first time,the comment-based hierarchical attention model is combined with two graph models(user social model and product association model),and then used for rating prediction;the proposed algorithm not only improves the rating The interpretability of the prediction also shows the effective influence of the social network on the rating prediction;the results on the real social data set show that the algorithm proposed in this paper is superior to other latest algorithms based on commentary social rating prediction.