| As a kind of filtering technology,a recommendation system is used to analyze the user’s historical behavior and selects the contents they may like from the massive data.However,the previous recommendation systems only focus on the accuracy of the recommendation and could not provide the reasons,which would cause users to question the recommendation result and reduce the frequency of using it.An explainable recommendation system,which can explain the results of recommendation has been favored by scholars in recent years.However,it still has some shortcomings.For example,most studies weigh explainability and accuracy,and ignore users’ demands for diversity and novelty.However,there are still some deficiencies,such as most of the studies weighing explainability and accuracy,and ignoring the user’s demand for diversity and novelty.In this paper,matrix factorization,tensor decomposition,and many-objective optimization are used to construct a two-stage explainable recommendation model to meet the overall needs of users.The work of this paper is as follows:(1)View of the previous studies on explainable recommendation systems are mostly exploring the accuracy and explainability,ignoring the user’s demand for diversity and novelty.as well as the previous problems in the quantification of explainability.This paper proposes a two-stage matrix factorization explainable recommendation model to ensure the overall performance of recommendations.Firstly,the first stage of this model improves the previous explainability definition and provides a new explainability constraint for matrix factorization,which makes the explainability definition more consistent with the user’s original intention of scoring items.Then,to avoid annoying users with a single recommendation result,a many-objective optimization module is constructed to meet users’ requirements for accuracy,explainability,diversity,and novelty in the explainable recommendation system.(2)Since reviews can contain users’ preferences in specific aspects and have strong explainability,this paper extends the idea of explainability quantification in ratings to reviews,and designs an explainable recommendation model based on tensor decomposition.Firstly,the tensor is used to store the four tuples extracted from the reviews,and the explainability is quantified from the perspective of reviews by calculating the user’s preference for aspects and the inclusion of items.Considering the same preference behavior between user groups,an accuracy constraint is designed.On the premise of satisfying the overall tensor performance,it can also consider the interaction among similar users,so that the generated results can better meet the users’ preferences.Similarly,the model follows a two-stage model and uses the many-objective optimization module to provide users with more accurate,explainable,diverse,and novel results from the perspective of reviews.(3)Individually using user ratings or reviews for the recommendation,the recommended results do not reflect users’ overall habits and preferences.Ratings-reviews recommendation model combines the advantages of both,which can not only get the overall evaluation of the user but also reflect the user’s preference for specific features.Therefore,this paper establishes a two-stage hybrid explainable recommendation model.The joint improved matrix decomposition and tensor decomposition techniques are used to generate accurate and explainable candidate recommendation items from two aspects.Combined with the respective advantages of ratings and reviews,a hybrid many-objective optimization module is constructed to provide users with more accurate and realistic recommendation results from both ratings and reviews. |
