Research And Application Of Explainable Recommendation Model Based On Attributes And Reviews

In today's information society,people are faced with massive amounts of homogeneous information and screened important information every day.How to find the most important information in complex data has become an important issue.Personalized recommendation system is the core technology for screening information,which has been closely focused by scholars and researchers.However,though the traditional personalized recommendation technology has achieved very significant success in terms of accuracy,these traditional recommendation algorithms have the unexplainability of the black box model.Moreover,deep learning is particularly obvious in this regard.The basic question "how should the system explain these recommendations to the user" has not received enough attention.The lack of transparency confuses users: Users can only assess the quality of recommendations by adopting suggested measures,such as buying top-ranked products.But the recommender system should first let users establish trust in the system.For these reasons,the explainability of recommendations has become as important as accuracy.How to develop a recommendation model with the same accuracy and interpretability has become a hot issue at present.This paper unifies the two common information of attributes and reviews,and focuses on the accuracy and interpretability of the recommendation system,and does the following:In view of the above problems,the main research contents of this paper are as follows:(1)A new neural explainable rating prediction recommendation model(NERAR)that unifies attributes and reviews is proposed.This model unifies the information of attribute features and review features,which uses a tree-based model to learn attribute features from auxiliary information.Besides,the paper then uses time-aware gated recurrent units(T-GRU)to model the user's current review features.Convolutional neural network(CNN)method is leveraged to process item review features.Finally,the factorization machine(FM)is used to fuse user and item vectors to derive the final result.In the fusion process of multiple reviews and attributes,this paper uses the currently popular attention mechanism to fuse a variety of information.Compared with a simple average summing algorithm,the attention mechanism is more consistent with the information for users.Different attractive characteristics and different information also play different roles in the user's final rating of the item.Finally,the corresponding information is selected as the recommended explanation based on the learned attention weights.A large number of experiments on the Amazon dataset demonstrate that our model is better than the latest recommendation models in terms of accuracy.Experimental examples also show that our model can provide sufficient explanations.(2)Applying the explainable recommendation model to the key scientific and technological research and development project of Jilin Province,"Research,development and application of rapid knowledge Sharing System in the Era of Big Data and mobile internet".The model uses the abstract information in the paper as the review text information in the model,and at the same time uses the attribute information of the user and the product itself to obtain the recommendation results and corresponding explanations,which has achieved good results.