Research On Explainable Recommendation Based On Deep Learning

In recent years,benefiting from the rapid development of Mobile Internet and Smart Devices,various applications have continuously emerged.But at the same time,it has also brought explosive growth of data.To solve the problem of information overload on the Internet,the recommender system was proposed and further promoted the innovation and application of various e-commerce service platforms.In order to make users better understand recommendation models and results,explainable recommendations are gradually becoming a hotspot in the research field of recommendation systems.In explainable recommender systems,the recommendation model not only gives a list of items,but also explains the recommendation results.Under deep learning technology,this article utilizes hierarchical attention networks to fully extract text features,which accurately represent users and items.Another,this paper exploits attention networks and Gated Recurrent Unit to fuse explicit and implicit features,and generate explainable natural language sentences.Specifically,the main work of this article includes:Due to exisiting recommender systems have insufficient text features extraction and use text features that contain redundant information to directly represent users and items.This article proposes a hierarchical attention model guided by crowd intelligence,namely HANCI.First,the model proposes the importance of features guided by crowd intelligence,which replaces text features in existing work with a user opinion-aware features to represent items.Specifically,the crowd intelligence features are defined as the features extracted from all reviews on an item,which are not completely equivalent to the features that a user cares about.Thus,we propose the importance of features,which is used to extract the features that a user cares about from the crowd intelligence features as the representation of the user's personalized preference on the item.Second,the model designs a hierarchical attention network based on the multi-level analysis of review text to extract richer and more precise user personalized preferences and item latent features via exploring the importance of words,the usefulness of reviews and the importance of features.To validate the effectiveness of HANCI,we performed extensive experiments on three publicly accessible datasets.The results show that this model can improve the accuracy of rating prediction,and provide word-level and review-level explanations.Due to providing review-level explainations may faces the risk of privacy exposure,and deep learning is difficult to explain high-dimensional latent features.This article proposes an explainable recommendation model for generating natural language sentences based on explicit and implicit features,namely GNLE.First,this model uses natural language processing tools to extract explicit features from all reviews of the target item.Second,this model designs a feature-aware attention module to extract user preference features and item highlight features from the explicit features.Third,this model fuses explicit features and implicit features to represent the target user and item.Finally,this model introduces a context-aware Gated Recurrent Unit to generate natual language sentence as explanations,and balances the tasks of recommendation and explaination by superparameter .To validate the effectiveness of GNLE,we performed extensive experiments on two publicly accessible datasets Electronics and Movies?and?TV.The results show that the explicit feature-aware attention and contextaware Gated Recurrent Unit sentence generator can improve the accuracy of rating prediction and generate easy-to-understand explanations.