Research On Geographic Data Recommendation Methods Based On Knowledge Graphs

With the rapid accumulation of geographic data,it has become increasingly difficult for researchers to find the data they need in the vast amount of data available.The established geographic data sharing platform only provides data services based on keyword search,which is suitable for users who already have a clear demand for data.However,non-geographic information users are often faced with problems or scenarios,such as "soil erosion research in Guizhou",and existing data search engines cannot return query results based on such scenarios,so we need to build correlations between user scenarios and existing data categories to bridge this semantic gap.Because of its strong knowledge representation and semantic reasoning capabilities,the knowledge graph has become a necessary research content and technical tool for the deepening development of big data and artificial intelligence in the field of geology,and provides new ideas for intelligent recommendation of geographic data.The current generic knowledge graph does not meet the requirements of the geographic domain,while the existing recommendation methods are not suitable for geographic data recommendation.Thus,this study aims to construct a domain-specific knowledge graph that centers on user scenario requirements to address the special requirements of geographic data recommendation.This is achieved by linking semantic networks of user scenarios,domain concepts,geographical names,and geographic data.Moreover,this study designs data recommendation methods to enhance the accuracy and efficiency of geographic data recommendation.The construction of the domain knowledge graph involves two levels: ontology layer construction and entity layer construction.For the ontology layer construction,two methods are used: top-down design of the domain ontology adapted to this study,and bottom-up improvement of the existing attribute normalization method.The ontology construction includes extracting concept relationships,extracting document knowledge,and storing knowledge.We incorporate a contextual relationship mining algorithm during the concept relationship extraction process to ensure data integrity.The final results of the extraction process are then stored in the Neo4j database.After the construction of the knowledge graph in the field of geology is completed,this study optimizes and improves the traditional data recommendation scheme.The related research improvements are mainly: first,for the recommendation of complex research scenarios,this study mines the strong association relationships between research areas,research images and research contents from the above knowledge graph based on association rule mining algorithms,and combines them to form typical research scenario templates for researchers to select;when a researcher selects a research scenario,it is further parsed and analyzed to show all research data that match the target.Rather,the meta-path-based recommendation model is designed in this study based on the meta-path relevance in the above knowledge graph,which is centered on article nodes and derives relevant data based on the information associated with article nodes;in addition,we use the similarity between entities as its connection basis and calculate relevant data based on inter-graph connectivity.Finally,in order to fully prove the effectiveness of the design,the optimization improvements in this study were verified by meticulous experiments,and the following conclusions were obtained:(1)For the improvement of the attribute normalization algorithm,we use the "landscape,forest,field,lake and grass" dataset to conduct experiments and verify the accuracy of the experimental results by using manual discrimination.The results show that the average precision(Presition),recall(Re-call),and F1 values of the improved algorithm are 91.0%,95.9%,and 93.4%,respectively.Compared with the traditional attribute normalization algorithm which only focuses on similarity,the optimized method can not only identify synonymous attributes better,but also identify near-synonymous attributes.Applying this method to attribute normalization research can significantly improve the efficiency and accuracy of its processing.(2)For the validation of the effectiveness of the incorporated contextual relationship mining algorithm,this study uses the extracted data of the Mapping Narrative Table as the training set,and divides some of them into test sets to test the algorithm.The final average accuracy of 82.9% was obtained,which fully proved the reliability of the algorithm in this study.(3)For the association rule mining algorithm,the association rule set obtained from the output of the results with the number of supported items greater than 10 and the confidence degree greater than 30% and they are filtered and organized;the meta-paths are queried and visualized using the Neo4j database,which fully proves the feasibility of the algorithm and the interpretability of the recommendations.