Research On Explainable Conversational Recommendation System Based On Product Relation

With the rapid development of Internet techniques,users have entered the era of information overload from the era of information shortage.It is increasingly difficult for people to find products or information they are interested in from a vast amount of information.Recommendation system is proposed to solve the problem of information overload when users have no clear requirements.It accurately recommends interested products to potential users.In general,users leave a series of behavioral information on the platform,such as clicking on products they are interested in.This time-related information becomes a very important factor in the recommendation system.Therefore,session-based recommendation(SR)is proposed.Session-based recommendations rely on the user’s historical sequence,and recommend products that the user is likely to be interested in at the next moment.In addition,platforms and users are not only satisfied with recommending product lists with high accuracy,but also put forward certain requirements on the explainability of models.For session-based recommendation,this work aims to explore the relationship between products through graph neural networks to further improve the accuracy and explainability of the recommendation system.By reviewing related work,it can be found that the current research has the following problems.Firstly,most of methods model a session by treating it as involving only one behavior type.However,in most scenarios,a session is mixed with different behavior types.And most of them ignore the substitutable and complementary relationship between products.Secondly,most graph neural network methods focus on modeling the relationship between products that appears together,ignoring the distinction between directed(causality)and undirected relationships(correlation).Thirdly,session-based recommendation mainly focuses on improving the recommendation accuracy,but the internal mechanism of the model is not clear.The lack of transparency and explainability makes it difficult for users to understand the recommendation results and thus reluctant to accept the recommended products.Therefore,based on the graph model(graph neural network and knowledge graph),this paper aims to improve the accuracy and explainability of session-based recommendation.The research content and the main conclusions of this paper are as follows:Firstly,session-based recommendation by exploiting substitutable and complementary relationships from multi-behavior data(SCSR).In most scenarios,a session in e-commerce is mixed with different behavior types.Users may conduct a series of different actions(such as click,add to the shopping cart and purchase)simultaneously on different products.There are few works in session-based recommendation to directly analyze the relationship between products through the multi-behavior data.Therefore,based on the multi-behavior data,this chapter learns the substitutable and complementary relations of products and forms the substitutable and complementary graph.Specifically,we first construct first-order substitutable relationships and three second-order complementary relationships through predefined rules.After that,we further leverage the denoising network to autonomously remove the noise.In addition,the substitutable and complementary relations are restricted in the loss function.Finally,we evaluate our model on two real-world datasets Tmall and Yoochoose.The experimental results show that the substitutable and complementary relationship between products is helpful to improve the effectiveness of session-based recommendation.Secondly,causality and correlation graph modeling for effective and explainable session-based recommendation(CGSR).In order to further explore the relationship between products for more accurate session-based recommendations.We mine the direct causal relationship between products through data rules,and then construct the causality and correlation graphs among products.We then choose three real-world datasets to evaluate the performance of different approaches.The experimental results show that the performance of CGSR is significantly better than other baseline methods,validating its effectiveness of distinguishing causality relationship between items from correlation relationship.In addition to improving the accuracy of SR,our model also provides explainability.We design an explainable framework on CGSR to clarify why a specific item is recommended on both session and item levels.Thirdly,a generic reinforced explainable framework with knowledge graph for session-based recommendation(REKS).To solve the explainability problem in SR,we want to design an explainability recommendation framework,which has certain generalization ability and can be combined with other non-explainable session-based recommendation models.Therefore,we construct a knowledge graph containing session information and product information,and integrate the non-explainable sessionbased recommendation model into the state vector,reward strategy and loss function of the Markov decision process.We then conduct extensive experiments on four datasets to validate the effectiveness of our model.With REKS framework,the performance of every baseline can be significantly improved in almost all cases,validating the effectiveness of our framework for non-explainable session-based recommendation methods on recommendation task.With regard to the explainability of SR,this chapter conduct a survey to evaluate user satisfaction towards the explanations generated by our framework.We also give three explanation cases to intuitively elaborate how REKS interprets the recommendation results.In summary,this study wants to improve the accuracy and explainability for SR.Firstly,by utilizing user’s multi-behavior data,it explores the substitutable and complementary relationship between users,as well as the causal and correlation relationship among products.This provides theoretical reference for mining the relationships among products and learning the feature representation of products in SR.Secondly,this paper proposes a session-based recommendation explainable framework,aiming to solve the "black box" problem.The framework optimizes the accuracy and explainability,and provides theoretical inspiration for model-based explainable recommendation systems.In addition,improving the explainability of recommendation results can enhance users’ online experience,further improving their overall satisfaction with the platform and ultimately bringing significant economic benefits and comprehensive benefits in actual application environments.