The Design And Implementation Of Domain-Konwledge-Graph-Based Meteorological Disaster Text Recommendation System

The popularity of computers and the Internet has made information easily accessible to people.In recent years,especially with the rapid development of mobile devices,we are surrounded by the massive amounts of information all the time.However,most of the information is invalid or worthless.The existence of worthless information consumes people’s precious energy and time when people find their interesting content.Search Engines and Recommender Systems are two technologies that help us to extract interesting content from massive information.At present,recommender systems are widely used in recommending interesting content to users.Compared with the active retrieval method,recommendation is a passive reception process,which omits various problems caused by users’ active retrieval.However,the sparsity problem in the recommendation system still seriously reduces its accuracy.The knowledge graph has the powerful ability of knowledge interconnection,and it can be used as an aid at the knowledge level for recommendation algorithm,which plays a very good role in improving the recommendation system.Based on the above background,this paper studies the application of meteorological disaster text recommendation system based on domain knowledge graph.The specific work includes:(1)The construction of the meteorological disaster knowledge graph.The Request mechanism,the Selenium(an automated testing tool)and the BeautifulSoup(a web page parser)are used to collect news texts related to meteorological disasters from the open Internet platform to complete the construction of the data set.For the application,the data set is annotated and the ontology of the knowledge graph is constructed,and the triple information is stored in the Neo4j graph database through technologies such as knowledge extraction and knowledge fusion.Finally,complete the construction of the knowledge graph and visualize it.(2)Knowledge representation learning with translation models.To apply the knowledge graph to the text recommendation algorithm,the knowledge representation learning is performed based on the knowledge graph and the performance of the TransE,TransH,TransR and TransD translation models is evaluated on the domain knowledge graph.(3)Research on algorithms about calculating text similarity for recommendation algorithm.Based on the highly condensed feature expression of news headlines,the entity feature vector obtained by knowledge representation learning and the title feature vector are fused to enhance the document-level feature expression,and the document-level feature expression of the news text is calculated in pairs to generate the result of recommendations.(4)The design and implementation of the demonstration system.Based on the above algorithms,the text collection module,knowledge extraction module,knowledge update module,knowledge graph visualization module and text similarity calculation recommendation module are integrated into the system through the Django framework to complete the design and implementation of the system.