Research On Personalized Recommendation Algorithm Based On Neural Network

With the rapid progress and development of my country's mobile Internet technology,the problem of information flow overload has become very significant.Two main problems arise from this: companies cannot effectively provide users with products,and users cannot see the products they are interested in.In order to effectively solve these problems caused by the information overload,personalized recommendation algorithms came into being.The personalized recommendation algorithm analyzes related data such as item characteristics,time,address,and user needs,calculates the user's interest in different items,and forms a recommendation list to display to other users.Neural networks have many applications in the fields of computer vision,natural language processing,and speech recognition.Neural networks can use its complex network structure to calculate the internal connections of data.In the field of advancing algorithms,using neural networks to replace traditional recommendation algorithm models can improve the accuracy of recommendation results.In recent years,knowledge graphs have developed rapidly.Knowledge graphs contain complex entity semantic information,which provides additional information for the relationship between items and items.Introducing knowledge graphs into recommendation algorithms can not only solve data sparsity and cold start problems,but also improve the accuracy of the recommended results.The user's historical sequence data contains a wealth of information,such as user interests,item timing relationships,and user current needs.Research on historical sequences is of great significance in recommendation algorithms.In order to make full use of historical sequence information and improve the accuracy of recommendation results,the main research contents of this article are as follows:(1)This paper constructs a recommendation system called ACRec,which uses the self-attention mechanism and convolutional neural network to perform feature calculation and extraction on the user's historical sequence from the two aspects of the sequence relationship of the items and the user's interest features.ACRec first uses the self-attention mechanism to calculate the relationship between the items in the historical sequence,and calculates the user's attention to the items,then sends the result to the convolutional neural network for feature extraction,and finally calculates the extracted features and the target item and get the user's degree of interest to the target item.And this paper designs a product layer for ACRec to enhance the semantic relationship between users and items in the historical sequence.This paper believes that only calculating the characteristics of the historical sequence cannot clarify the semantic relationship between the user and the historical sequence,so the product layer is designed to increase the association between the item and the user in the training phase.The calculation result of the product layer will be multiplied by the impact factor and then added to the overall loss function.(2)This paper introduces the knowledge graph on the basis of ACRec,and constructs the ACRec-KG recommendation algorithm model.In this paper,a conversion layer is designed to fuse the head entity of the knowledge graph and the item information of the recommendation algorithm,and the knowledge graph representation module is designed to use the translation distance model to train the vector representation of the knowledge graph.This paper conducts a comparative experiment on ACRec on two public data sets Movie Lens-1M and Video?Games.Experimental results show that ACRec improves the accuracy of the recommendation results.The experimental results show that the introduction of knowledge graph can improve the accuracy of recommendation results.