Research And Application Of Knowledge Graph Construction And Recommendation Based On Bidding Dat

Under the general trend of enterprise informationization,bidding,as a major form of enterprise procurement materials,recruitment talents and project outsourcing,has been more widely used electronically.However,most of the existing electronic bidding websites are limited to the retrieval and publication of bidding data.As an intelligent product in the era of big data,recommendation algorithm can obtain potential relationships among data entities through training and calculation of data,and these relationships can provide people with scientific reference opinions to help them make decisions.Therefore,it is of deep research significance to apply recommendation algorithms to the field of electronic bidding.In addition,Neo4 j,as a non-relational graph database,can store and visualize the bidding data in the form of knowledge graph,and can calculate the similarity between nodes to alleviate the cold start problem arising from the recommendation based on bidding data.The work content of this article is as follows.(1)Knowledge graph construction based on bidding data.The bidding data is crawled on public bidding information websites using crawler technology,and the bidding project names are processed using Jieba word splitting,and the bidding subject labels and project labels are extracted by word frequency sorting after removing redundant words.The top-down approach is used to design the knowledge graph,and the Cypher statement is used to construct the knowledge graph in Neo4 j.The similarity between bidding subjects and bidding projects is calculated using the node similarity function in the Neo4 j graph database,and the similarity calculation method between irrelevant bidding subjects and projects is designed,and finally the bidding Finally,we generate the bidding dataset and subject preference features based on the obtained similarity calculation results.(2)Improvements based on deep learning interest recommendation model.A deep learning interest recommendation model is proposed that combines a multi-headed attention mechanism and a residual network structure,using a multi-headed attention module to calculate the weights of the assigned subject preference dataset and the currently selected items.The obtained feature results are connected with user features and item features and then passed to the output training layer via the residual network structure for training.The experimental results on the bidding dataset show that the proposed recommendation model meets the expectations and has some validity and feasibility.(3)Design and implementation of the bidding recommendation system.Conduct system requirements analysis through investigation to determine the functional and non functional requirements to be realized by the system.The system was designed and implemented using Pycharm tool and Python programming language,and the front-end of the system used Echarts tool to visualize user portraits and bidding portraits,and designed relevant functional test cases to test the system.