| The main contents of national defence mobilisation usually include political mobilisation,armed forces mobilisation,national economic mobilisation,people’s air defence mobilisation,national defence transport mobilisation and information resources mobilisation,etc.People’s air defence mobilisation occupies an important position as part of national defence mobilisation.However,the amount of data on people’s air defence mobilisation is usually huge,and it is tedious to search for the content through human checking the database,and there are problems such as scattered information construction and lack of professional human defence engineering maintenance and management personnel.The implementation of the question and answer system effectively reduces the workload of human defence project maintenance managers.This paper first constructs a Knowledge Graph automatically for structured data in the field of people’s air defence mobilisation,then designs a question and answer system for people’s air defence mobilisation based on the Knowledge Graph,and finally implements and tests the system.The contribution of this paper are the following:（1）In the knowledge graph construction stage,the automatic construction technique of ontology file is proposed in conjunction with the current bottom-up construction method,and the tedious problem that the ontology file needs to be constructed manually is solved through the method of database identification and relational pattern conversion,and the Knowledge Graph of the people’s air defence domain is further generated based on the ontology file,with a total of 14,206,523 entity nodes and 27,498,521 relations.（2）To address the problem of poor recognition when there are more entity types in the Named Entity Recognition in the people’s air defence mobilisation domain,an attention mechanism is introduced to add weights to the key information,and the BERT-BiLSTM-ATT-CRF model is proposed for the people’s air defence mobilisation domain,and the effectiveness of the model is verified by comparing the BiLSTM-CRF and BERT-BiLSTM-CRF experiments..To address the problem of poor multi-intent recognition in short text,the BERT-TextCNN model approach is used for multi-intent recognition,which effectively extracts higher-level n-gram local order features and increases the richness of the final classification features,and has better recognition effects on short text datasets compared with BiLSTM-ATTention and BERT models.（3）Design and implement a question and answer system for people’s air defence mobilisation.The system has the functions of viewing Knowledge Graph,character portraits,Graph search and intelligent question and answer.In response to the problem of slow user input,voice recognition is added to the system.Also design and implement question and answer system applets to make the system more usable. |
