Research On The Construction Of Geographic Entity Relationship Based On Knowledge Map

Currently,there are massive geographic data in the network,but it is difficult to extract geographic entity relationships or structured geographic information from these large and diverse types of data on the Internet,so it has great research value.The knowledge graph is a semantic network,its role is to describe the relationship between entities in reality,and it is used in many fields.Extracting geographic information from the network,and then acquiring geographic information knowledge through knowledge graph related technologies,which has become a common method in this field.In view of the current situation of massive and complex geographic information data on the Internet and the difficulty of using data organization,this paper designs a method for constructing network text entity relationships in the geographic field based on knowledge graphs.Its main contents include:(1)In the process of personal editing and public review of online encyclopedia data,errors in text labels are inevitable.Aiming at the problem that the network text is too complicated after collection,this article establishes a text classification method based on the TF-IDF algorithm,adds a geographic dictionary to the text preprocessing stage,and then corrects the TF-IDF geographic features by establishing a text vector space model For the weight calculation method,the K-nearest algorithm is finally selected to realize text classification.The geographic dictionary is used to process the feature dimensions to further reduce the dimension,and the modified feature weight calculation method can optimize the classification results and further clear the network text that does not belong to the geography.(2)Generally,only when there is a large amount of artificially labeled corpus can relationship construction be achieved.At the same time,this construction model based on weakly supervised back marking often has a large amount of noise during the process of obtaining training corpus,and it will also appear.The defect of insufficient corpus.For the above phenomenon,this paper establishes a triple expansion algorithm at the beginning,which is based on the synonymous expansion of relational feature words,and further realizes the retrieval and matching of more training corpus,so as to obtain more training corpus.Then,the relational feature words are used to filter the corpus for the purpose of optimizing the corpus,and then through the triple-tagged text retrieval process,the sentences are obtained to obtain the training corpus,and these sentences must contain entity pairs and their relational feature words to reduce noise.Through experiments,it is shown that the method adopted in this paper can greatly reduce the noise of the corpus,and at the same time solve the problem of insufficient corpus,which is of great significance for the future extraction of entity relationships.(3)For the process of constructing entity relationships with different types,this paper has designed two construction methods.One is to use the maximum entropy model to construct the entity relationship.This method uses the relationship classification theory and uses the n-pattern feature extraction mode to characterize the difference of each relationship text to achieve a limited type of relationship construction purpose;the second is Syntactic analysis tree and CRF are combined to form a construction method.This method is based on syntactic analysis and sequence annotation,and can construct all types of relationships.Experiments have proved that the results of these two methods established in this paper are more accurate.Compared with the current model of artificial entity relationship construction,the efficiency is greatly improved.