Intelligent Recommendation Algorithm Of Textile E-commerce Based On User Behavior

At present,the recommendation algorithms of B2C(Business-to-Customer)e-commerce platforms such as Taobao and Jingdong have been relatively mature,and due to the large amount of data and rich types,they can make recommendations with high accuracy for users.However,B2B(Business-to-Business)e-commerce platform is different from B2 C e-commerce platform.The transaction between enterprises is usually only browsing and consulting on the platform,and the transaction process is mainly offline.Therefore,B2 B e-commerce platform is faced with the following challenges: First,the collected data type is single and sparse.Second,although the knowledge graph can effectively solve the sparsity problem of collaborative filtering,the combination of the association information between the item entities in the knowledge graph and the overly dense or overly sparse collaborative filtering information of the user item bipartite graph is easy to cause the learned user item vector to be interfered by irrelevant information and cannot accurately express its features.Third,the existing selfattention mechanism only considers the correlation between vectors,and lacks the more complex semantic correlation between entities.In order to solve the above problems,the main research work of this paper is as follows:(1)Aiming at the problem of sparse data faced by traditional B2 B e-commerce platforms,this paper preprocesses the user browsing data,merchant data and commodity data of textile ecommerce enterprises for one year,extracts the user browsing preferences and the correlation information between commodities,and constructs the knowledge map of textile e-commerce.(2)Aiming at the problems that the user item association vector of the traditional recommendation algorithm is interfered by irrelevant information and entity relationship is represented by a single entity relationship,this paper proposed a new recommendation model,Balanced Knowledge Attention Network,which is mainly composed of three modules: Balance heterogeneous data propagation layer,hyperplane knowledge self-attention embedding layer and prediction layer.The balanced heterogeneous data layer is composed of balanced cooperative propagation and knowledge graph propagation,which mainly solves the problem that the user item vector is interfered by irrelevant information.Hyperplane knowledge selfattention embedding layer is composed of hyperplane knowledge self-attention network and embedding,which mainly solves the problem of low correlation between entities.The prediction layer is composed of aggregation and prediction.The prediction learning model is established by taking the embedded representation vector of users and goods through aggregation as input and their interactive relationship as output.In this paper,the proposed model is applied to the real textile e-commerce data set,and the experimental results show that the performance of BKAN recommendation algorithm based on knowledge graph is improved by 1%～3% compared with that of recent years.(3)On the basis of the above studies,this paper designs and implements the textile ecommerce recommendation system.The data preprocessing program and BKAN algorithm are applied to the recommendation system.The recommendation results show that the model in this paper can provide more comprehensive recommendations for textile users.The system includes data storage module,recommendation algorithm module,system management and display module.