Research On Personalized Recommendation And Interpretability Of User Intention And Interest Perception

With the development of cloud computing,big data,Internet of things and other technologies,the massive amounts of applications in the Internet have triggered the explosive growth of data scale.Big data contains rich scientific research value and commercial value,but it also brings serious information overload to users.As an effective method to address information overload problem,recommendation system has been successfully applied in many fields,such as social network,e-commerce,streaming media recommendation and so on.In the field of personalized recommendation,users’ behavior is very complex and sparse.Most of the existing works adopt the sequence model based on deep learning to model the user’s behavior.However,there are still several problems in the existing methods of user interest transformation.First,the sequence models with onedimensional structure cannot capture the complex branching structure and the nonlinear interest transformation in user behavior.Secondly,existing sequence models usually lose part of the key information of the long user behavior sequence.Finally,the methods based on deep learning are usually difficult to explain the model,which limits the application of deep learning in real-world scenes that need to be explained.This paper addresses the above issues in the following aspects.Firstly,this paper captures the preferences and intentions in user behavior sequences by using Graph Neural Networks,hierarchical attention,user intent untangling and other mechanisms to recommend the points of interest that the user may enj oy.Secondly,the Graph Neural Network has poor interpretability and is not good at handling heterogeneous data.It becomes an important and challenging task how to make full use of the advantages of Graph Neural Network to satisfy the requirements of scene with explanatory and complex data form.Specifically,the main work of this paper consists of three parts.(1)This paper presents an Attentive Sequential Model based on Graph Neural Network(ASGNN),to effectively model user behavior data,accurately capture user preferences,and use them adaptively to improve the performance of the recommendation system.Especially,ASGNN uses a gated graph neural network to model the user’s behavioral data,and a personalized hierarchical attention network is designed to capture and adaptively use the user’s preferences,thus providing accurate recommendations for the user.(2)In order to explore users’intentions in a fine-grained way and to further improve the performance of the recommendation system,this paper presents an IntentAware recommendation model based on Graph Neural Network(IAGNN).This model uses a gated graph neural network to model user behavior,uses the attention mechanism to capture user preferences,and designs a disentangling module based on the Graph Neural Network to extract user intentions.Especially,IAGNN combines user preferences with user intentions to further improve the performance of the recommender system.(3)In order to meet the complex data structure and interpretability requirements of real game store recommendation scenarios,this paper presents a model named Heterogenous Graph Attention Network for Explainable Recommendation(MHANER).Specifically,MHANER uses a heterogeneous graph attention network to model complex user behavior data,and learn the representations of users and items.Then,a personalized and precise recommendation model based on attention mechanism is proposed.In addition,MHANER uses a multi-level subgraph pattern mining method to mine key subgraph structures that influence user behavior and provide users with personalized and diverse explanations of recommendation.