Research And Application Of Key Technologies Of Service Transaction Recommendation Based On Knowledge Graph

Nowadays,the total amount of data in the service transaction field is showing explosive growth.How to efficiently capture valuable information from the data has severely shackled the development of current service transaction applications,and the most effective technology that can currently break free of this shackle Is a recommendation system.At the same time,the unique graph structure of the knowledge graph has brought new ideas for improving the recommendation effect in the field of service transactions.Therefore,this paper builds a knowledge graph by using data in the field of service transactions,and uses the knowledge graph vectorization technology to extract rich semantic information,integrate it with the recommendation algorithm and improve the recommendation performance.It focuses on exploring the construction technology of service transaction knowledge graph and the design of the fusion method of recommendation algorithm and knowledge graph in top-N and score prediction recommendation scenarios.The main contents of this article are as follows:1.Propose a method of constructing knowledge graph in the field of service transaction.Introducing the concept of knowledge graph into the field of service transactions,first of all,the knowledge characteristics in the field of service transactions are analyzed to complete the division of entities and relationships in the field,thereby realizing the construction of the ontology database in the field of service transactions.Then use knowledge extraction technology to extract the specific entity and relationship knowledge of service transactions according to the definition of entities and relationships in the ontology database.Finally,by analyzing the pros and cons of storing data in a relational database,we chose to use the graph database Neo4 j for knowledge storage.2.A top-N recommendation algorithm based on knowledge graph is proposed.The algorithm can be divided into three parts,including a BiLSTM network structurewith attention mechanism,a time evolution model and an item similarity candidate pool based on knowledge graph,and a fusion training method of joint learning is used for learning training.The semantic information of items extracted from the knowledge graph adds the semantic information of items to the original recommendation framework.Experiments show that compared with other top-N benchmark models,the recommendation algorithm based on knowledge graph has improved in accuracy and recall rate respectively.3.A score prediction recommendation algorithm based on knowledge graph is proposed.In the CNN-BLSTM neural network framework,this method introduces the knowledge graph through the integration method of sequential learning,that is,the vectorization of the knowledge graph extracts the time semantic information and the comment information together as the input of the CNN-BLSTM neural network,which makes up for the original The recommendation algorithm does not consider the shortcomings of time information.Compared with other benchmark models for scoring prediction,in the four experimental data,the recommendation algorithm based on knowledge graph has an average maximum decrease of 37% in root mean square error.