Military Text Detection And Analysis Based On Domain Knowledge Bas

Modern military building is changing into information construction due to the fast growth of emerging technologies.Traditionally,military writings were frequently stored on paper books and newspaper articles,without the formation of a hierarchical knowledge system.Moreover,these discontinuous and unstructured military texts not only make fast analysis of military information in the field impossible,but also severely limit the effectiveness of military professionals in investigating and analyzing military occurrences.It not only causes plans to be delayed,but also increases the likelihood of errors.As a result,the goal of this project is to apply natural language processing technology to create a military text knowledge base,complete structural analysis of military text,and increase military text analysis efficiency.Faced with complicated text information in the military field,this research will begin with the construction algorithm of the domain knowledge base,design and achieve the military text detection and analysis algorithm,analyze the military text structurally,and provide algorithms and technical support to the military text detection and analysis system.The main content of this paper involves the construction of military knowledge base system and the detection and analysis of military text.The main work and innovations are as follows:(1)A military text detection model based on Trans-CNN was proposed.Facing the disadvantages of the existing text classification models’ poor computation speed,high storage space.Meanwhile,these models are unable to consider the overall semantic information and local characteristics of sentences.In this study,a hybrid neural network model called Trans-CNN is suggested for military text detection without the help of a pre-trained language model.This model enhances the expression of the key sentence features by using the multi-head attention mechanism,uses gated neural units and convolutional neural networks to focus on the global and local feature information of the sentence.Firstly,the word embedding layer is used to obtain the vector representation of the military sentence.Secondly,the gated reluctant unit is used to focus on the temporal correlation information of the text to strengthen the temporal feature expression.Thirdly,the multi-head attention mechanism and convolutional neural network are used to focus on the global and local feature information,respectively.Meanwhile,the global features and local feature are fused.Finally,the Soft Max classifier is constructed to obtain the label of the text.The experimental results show that the method proposed in this paper can obtain the category information of the text effectively,which is without the help of a pre-trained language model.Excellent results have been achieved on IMDB and the military text data set(JSWBD)which is constructed in this paper,with F1 values of 89.88% and 96.46%,respectively.(2)A military named entity recognition method based on adversarial learning and global pointer was proposed.The MNER-ALGP model,which is built on adversarial learning and global pointer,is suggested as a solution to the issues with current Chinese entity recognition models’ bad anti-interference ability and inaccurate entity boundary identification.Firstly,the Ro BERTa-WWM model is used to optimize the semantic representation of military text.Meanwhile,the adversarial learning strategy is introduced to disturb the word embedding layer of the model,which will enhance the anti-interference ability of the model.Then,the bidirectional gated reluctant unit is used to mine the timing characters information,which will enhance the temporal feature expression among military texts.Finally,the global pointer is constructed in the decoding layer to obtain more accurate entity boundary recognition results.In order to verify the effectiveness of the model,this paper constructs a military entity recognition data set which is named JSSTD,and conducts abundant experiments on the Resume and MSRA data sets.The experimental results show that the F1 values of the MNER-ALGP model are 93.89%,96.10% and95.82% on the JSSTD,Resume and MSRA data sets,respectively,all of which are better than the models compared in this paper.(3)A subject-object interactive military relation extraction algorithm based on enhanced key information was proposed.In order to solve the problem that the current relation extraction model does not pay enough attention to the local feature information of the military text,while the performance of relation extraction is poor.A subject-object interactive military relation extraction model based on enhanced key information named SORECI is proposed.Firstly,the pre-trained language model BERT is used to obtain the global semantic representation of the sentence.Meanwhile,LSTM and CNN are used to extract the character criticality in the sentence.Secondly,the character criticality is sued to weight the global semantic representation of the sentence,which will obtain the character vector enhanced by key information.Thirdly,the double-pointer network is used to mark the position where the subject entity appears,to decode the subject entity information.Finally,the subject entity information and the character vector are fused to predict the position and relation type of the object entities.Meanwhile,the relation extraction result is obtained.The experimental results show that the relation extraction algorithm proposed in this paper achieve good performance on the NYT data set and the military relation extraction data set JSGXD.The F1 values are 93.2% and 91.9%,respectively,which are better than the models compared in this paper.(4)A military text detection and analysis system based on domain knowledge data was constructed.From the design and deployment of military text detection and analysis algorithm to the construction the military knowledge base system,this paper describes the implementation route of algorithm and system in detail.In the process of algorithm construction,text classification technology is used to complete the detection of military text,and the entity recognition and relation extraction algorithms are used to complete the construction of military text analysis function.In terms of system construction and deployment,Py Torch,Fast-API,VUE,et al.are used to implement system function construction,complete the development of system function pages,and provide convenience for system users.