Research And Application Of Key Technologies For Building Knowledge Graphs In Software Knowledge Domain

Knowledge graph provides an effective data organization mode for mining useful information.Knowledge graphs have been well applied in various industries,such as law,medical care,finance and other fields.However,the software knowledge field has not yet established a corresponding knowledge graph.The key to the construction of the knowledge graph in the software knowledge domain is to extract information from massive data.This article focuses on the key technologies for the construction of the knowledge graph in the software knowledge domain.The main innovations of the work content are as follows:Summarize the key technologies of the knowledge graph in software knowledge domain.Due to the short of relevant work,this article summarizes the key technologies of mapping technology based on specific domain by fully investigating a large number of domestic and foreign documents.At first,it outlines the source of specific domain knowledge graph data and the corresponding format of these data.Secondly,it introduces the key technologies and related algorithm models.Collect and label the data in the software knowledge field.We uses web crawler technology to obtain the data we need since there is no existing data set available for our work,and then construct entity tables after preprocessing the raw data.Finally,we mark out the features of these text data.Propose a Bi LSTM-CRF Entity Recognition Model based on position coding and multi-head attention.We find that the introduction of position encoding and multi-head mechanism will further improve the accuracy of entity recognition through the research of the traditional entity recognition algorithm LSTM-CRF.The multi-head attention model can process the interconnection between any two words at the same time,not just by sequence extraction,and this mechanism can also solve the long-distance dependence problem of traditional recurrent neural networks.Propose a Bert-based entity relationship classification method.In order to solve the problems of traditional text feature spareness and strong context dependence,this paper proposes an enhanced semantic network model Bert-Bi GRU combined with enhanced semantic network using multi-head attention mechanism.The generated vector is used as the word representation of the training text for semantic enhancement,and then input it into Bi GRU to extract further contextual features,and then use the multi-head attention mechanism to adjust the semantic relationship weights,and finally classify these relations through softmax.Experiments show that compared with other mainstream methods,the enhanced semantic network model proposed in this paper has a significant improvement in classification effect.Designed and implemented a visualization system for the knowledge graph of the software knowledge domain.After two experiments of entity recognition and relation extraction,some relationship modifications are made based on the triple relationship group obtained.Visualization is used for the display of final knowledge graph.This paper sorts out the key technologies for the construction of knowledge graphs in the software knowledge domain,and takes the improving of the accuracy of entity recognition relationship extraction as the main research goal.By studying the application of deep learning in knowledge graph construction in software knowledge domain,the accuracy of entity recognition and relationship extraction is further improved,and the problems of some existing key technologies of domain knowledge mapping construction are solved,which has high application value and practical significance.