Recommendation Of Multi Domain Scientific And Technology News Based On Knowledge Graph And User Portrait

Science and technology news is a central reflection of the scientific and technological power of society,reflecting the current development of science and technology and the focus of scientific and technological development.With the continuous efforts of researchers and the development of the Internet,many large academic organizations and related data service companies have opened up a large amount of academic data to the public,and tens of thousands of new academic theses are published every day,and the data related to scientific and technological information is growing.Due to the characteristics of huge quantity,wide range of fields,and complex classification of scientific and technical information data,it makes researchers get only a large amount of invalid information when they read it,even if they spend a lot of effort.In the massive data environment,how to make accurate recommendations for science and technology news has become an urgent problem in current scientific research.The work done in this thesis has the following aspects.(1)A knowledge extraction method for science and technology news based on BERT combined with attention mechanism of local features(BALCF)is proposed,which fuses bi-directional LSTM recurrent neural network for deep extraction of entity-entity relationship of multi-domain science and technology news.Combined with the generic knowledge graph to supplement the inter-entity relationships and avoid the incomplete information of the graph.The user interaction relationship is added to lay the foundation for the subsequent user portrait construction,and the multidomain science and technology news knowledge graph is constructed and persisted by Neo4j graph data.(2)A tagging and community-based user portrait construction method for science and technology news is proposed,which combines user behavior and uses the knowledge graph structure to represent user information and science and technology news tags to improve the authenticity of the portrait and the accuracy of user interest acquisition.Based on incremental update and knowledge distillation for multi-domain science and technology news classification user portrait update,the realtime,consistency and completeness of multi-domain technology information user portrait are maintained.(3)A new multi-domain science and technology news recommendation method,i.e.,multi-domain science and technology news recommendation method based on knowledge graph and user portrait(KGUPN)is proposed.Based on the multi-domain science and technology news interest propagation embedding representation algorithm of knowledge graph,the implicit semantic similarity calculation is introduced to generate multidomain science and technology news recommendation candidate results.Based on the multi-domain science and technology news classification user portrait and label matching method,the recommendation candidate results are secondly filtered and sorted to complete the accurate recommendation of multi-domain science and technology news.(4)A multi-domain science and technology news recommendation system based on knowledge graph and user portrait is designed and implemented.The system mainly consists of the following modules:feature extraction and knowledge graph construction module for multidomain science and technology news,user portrait construction module for multi-domain science and technology news,and accurate recommendation module for multi-domain science and technology news.The following functions are mainly implemented:knowledge graph of science and technology news,user portrait of science and technology news,and recommendation and display of multi-domain science and technology news,and the system is tested and verified.