Research On Recommendation Algorithm Based On Convolution Neural Network

With the development of the times and the advancement of Internet technology,e-commerce and the Internet of Things have made people's lives more convenient,while they have brought about a series of problems such as information overload and information security.In the field of e-commerce,on the one hand,users often cannot know the authenticity of the information faced with the various interactive information,and it is difficult for people to obtain reliable and satisfactory information quickly and accurately.On the other hand,the merchant cannot know the users' real feedback and interest to provide satisfactory service.This will not only lead to low information utilization,affecting the platform's credibility,but also can not guarantee the interests of users.In order to make better use of these information,find out the relationship between user behavior and merchandise,and solve the problem of users' facing various choices of many commodities.The reliability and accuracy of the recommendation system is increasingly important.Aiming at these problems,this paper puts forward the following improvements:(1)Aiming at the problem of the implicit feature of the commentary information,the backpropagation algorithm of the convolutional neural network is used deeply to mine the implicit information and obtain the hidden features of the user and the project.For the problem of the credibility of comment information,a user reputation credibility calculation method based on user behavior is proposed.According to the time decay function of user behavior data and the regression analysis of the numbers of transactions and comments,the calculation method of the credibility of the comment information is obtained,which improves data quality.(2)Aiming at the inconsistency of the evaluation criteria of the users in the traditional collaborative filtering algorithm,a user-item scoring matrix improvement method based on user attributes is proposed.According to the three basic user attributes that affect user decision-making,the users are clustered.Then,the weighted factors are added to the users after clustering.Thereafter,the proportions of users with different degrees of similarity are adjusted.The more similar the user evaluation criteria based on the population attributes are,the closer the concept is to the optimization of users.Similarity calculation method.For users who do not have a common score,theuser-to-item preference value is calculated through the item-attribute matrix,and the more common attribute preferences,the higher the user similarity.Finally,the two-part similar user set is combined to predict the project score of the target user.Experiments show that the improved recommendation algorithm is improved in both recommendation accuracy and recall rate.(3)For the problem of user cold start,a general user profile model based on association rules is proposed.The user item label is generated by the association rule.The user is clustered based on the user population attribute,and the central user item label of each user group is calculated according to the center of gravity method.The target user matches the central user based on the population attribute,obtains the general user item tag belonging to the target user,and generates a recommendation list.Experiments show that the recommended algorithm works slightly better on the user's cold start problem.