Research On The Construction Of Weapon And Equipment Knowledge Graph

Military weaponry is an indicator of whether a country is strong.The development of informationize military weaponry is not only the basic content of military reform,but also the key to achieving the goal of informatization construction of our army.The knowledge graph can be regarded as a triplet of countless entity relationships,linking the world's knowledge to form a huge graph network and storing it in the computer,helping people to make correct decisions on a specific problem.This subject builds knowledge maps in the field of military weaponry,collects information on military weaponry and analyzes the correlation between military weaponry to achieve an assessment of the overall combat effectiveness of the military.In the process of constructing the knowledge map of military weaponry,it mainly studies the aspects of military weaponry entity extraction,military weaponry entity triplet extraction and Knowledge fusion of weapon and equipment knowledge map.In terms of military weaponry entity extraction,the fusion model of domain BERT model and BILSTM model with embedded word vector and word conversion vector is proposed to identify the weapon entity.First,the BERT model is used to pre-train military weaponry corpus.Secondly,the word vectors are trained through the Word2 vec model to provide a priori semantic information,and at the same time,more priori information is input into the model by embedding the word conversion vector.Finally,the hierarchical entity extractor extracts entities of different categories.Experiments show that the model has strong coding ability and sufficient prior knowledge,and the F1 value on the Global Military Network corpus reaches 91.436%,which has a good effect on the extraction of weapons and equipment entities.In terms of entity relationship triplet extraction of military weaponry,an entity relationship extraction model(HSL)based on hierarchical sequence annotation is proposed.HSL transforms the entity relationship triplet extraction task into a subject sequence labeling task and an object relationship sequence labeling task.First,HSL uses GLU dilated convolutional coding with residual links to generate intermediate vectors.The intermediate vectors are passed through the Self Attention mechanism to obtain the subject encoding vector,and the subject's label sequence is decoded to extract the subject.Secondly,the subject is used as a priori feature and the previous intermediate vector passes through the Self Attention mechanism again to obtain the object coding vector.Finally,a fully connected layer representing different subject-object relationships is used to extract an encoding vector of a certain subject-object relationship,and a tag sequence of the object is decoded to extract the object.Experiments show that the model effectively solves the problem of triplet overlap,and the effect exceeds the mainstream entity relationship triplet extraction model.The F1 value on the military corpus data set reaches 79.17%.In terms of Knowledge fusion of weapon and equipment knowledge map,a synonym expansion method based on the fusion of Glove and Word2 vec models is proposed.Use the Glove model and Word2 vec model to train the military weaponry corpus for word vectors,and calculate the similarity between the two words through the Euclidean distance to obtain the synonym expansion results,the final extended results of the two models are obtained by taking the intersection of the two models.Experiments show that the model is effective in synonym expansion,and the F1 value reaches 62.72% under 38 synonym corpus of military weaponry attribute.At the same time,the synonym of the military weaponry knowledge map was compared by using the synonyms extended by different models.This model has the most attribute fusion times,and it has a better effect on weapon equipment knowledge fusion.