| In the ubiquitous information environment,the relationship between human,objects and events in the real world and the network world can be organized,described and understood by using location as the link,and dynamic information fusion with multi-temporal,multi-scale,multi-level and multi-granularity can be realized.Due to the fuzziness in the process of human cognition and conceptualization of the real world,there exist a large number of fuzzy place names without definite spatial scopes except national boundaries and administrative divisions.Location information is not only partially stored in the form of structured basic geographic data,but also contained in unstructured data such as video,text and images to a large extent.In particular,text form is the most commonly used form of location information expression(such as place name,address,place name + space relationship,etc.),and its expression has obvious characteristics of qualitativeness,blur,diversity and scale.In this paper,information extraction and knowledge graph construction techniques are used to construct location knowledge graph based on multi-source structured and unstructured data,and qualitative location spatial area calculation based on knowledge graph is realized,which provides reference for geographic knowledge service and intelligent perception of location information.The specific research contents and achievements of this paper are as follows:(1)Construction of location information expression model and description feature analysis.By summarizing and analyzing the composition of location information,concluding the location information expression model,analyzing the description characteristics of location information in Chinese text,such as toponym entity,time information,attribute information and spatial relations,enriching the existing location information annotation corpus based on multi-source text data sources.(2)Location information extraction.A BERT+Bi GRU+CRF model combining toponym features was proposed to realize toponym entity recognition,and attribute information recognition based on BERT+Bi GRU+CRF model.Based on the combination of rule model and PCNN,the spatial relations between toponym entities and the association between toponym entities and attributes were extracted.The experiment results show that the F1 value of toponym entity recognition are 92.77%,time information recognition is 91.2%,area,population,rainfall attribute information recognition are 89.84%,90.07%,and 90.92%,respectively.Based on the combination of rule model and PCNN,the F1 value of the spatial relationship extraction between toponym entities is 83.91%,the F1 value of the association extraction between toponym entities and time information is 86.35%,and the F1 value of the association extraction between toponym entities and area,population and rainfall information are 85.37%,83.38% and 84.18% respectively.(3)Construction and visualization of location knowledge graph.Based on the location information extraction results,structured toponym database and POI data source,through the disambiguation and fusion of toponym entities,the location knowledge graph was constructed,and the storage and visual expression of location knowledge was realized based on Neo4 j.Based on the knowledge graph,the location knowledge query and inference were carried out,and the qualitative location spatial area calculation based on the compound spatial relation calculation model was realized.The prototype system of qualitative location spatial area calculation based on knowledge graph was designed and implemented,and the functions of multi-source text data query,location knowledge graph visualization and query,spatial area calculation were realized. |
