Research And Implementation Of Multi-modal Recommen-Dation Algorithm Based On Social Relationship

In recent years,the rapid development of Internet technology has been accompanied by the generation and dissemination of massive information.Rapidly and accurately conveying appropriate information has become the common demand of Internet users and information providers,and recommendation algorithms have emerged as the times require.In real life,multiple modal information such as pictures,text,and sounds of items will affect the results,and social activities will also affect the user’s decisionmaking.Therefore,we believe that introducing the multimodal information of the recommended object and the user’s social relationship into the recommendation algorithm can extract more features,improve the recommendation accuracy,and help solve the cold start problem and improve the recommendation efficiency.Therefore,we propose a multimodal recommendation algorithm based on social relations and implement a recommendation system.The main contents are as follows:(1)Extraction and fusion of multimodal features.This paper uses natural language processing,image processing and other technologies to process the text description and picture information of the recommended object respectively,uses the deep learning network to achieve feature extraction and fusion,and uses the generated feature vector to represent the recommended object in the recommendation model.In this way,the characteristics of the recommended object can be fully utilized,the accuracy of the recommendation can be improved,and the cold start problem can be improved.(2)Personalized recommendation based on social relationship.This paper uses the graph attention network combined with the latent semantic model to construct a social network graph based on the user’s social relationship,aggregates the characteristics of the user and his social friends,and simulates the impact of social relationships on user preferences.On the premise of protecting users’ personal privacy,we can improve the understanding of user preferences and improve the accuracy of recommendations.(3)Research and implementation of recommendation system.Based on the social relationship-based multimodal recommendation algorithm proposed above,a recommendation scheme is proposed,which is combined with the database and web page display interface to realize a recommendation system with functions such as data management,model training,prediction,and recommendation,etc.To sum up,this paper analyzes the deficiencies of existing recommendation models in the field of content recommendation,proposes a multimodal recommendation algorithm based on social relationships,designs and implements a recommendation system,and implements the above models in the system.After model verification experiments and system tests,the system has good performance and can well meet the requirements and adapt to realistic recommendation scenarios.