Research On The Construction Of Weapon Ontology For Military Field

The construction of military weapon ontology refers to the identification of weapon entities in the military field and the extraction of semantic relations between weapon entities.Reliable data support can be provided for intelligence personnel by building a large-scale and high-quality military weapon ontology.There is a wide range of application value in strategic information retrieval and weaponry equipment research and development.At present,the existing ontology construction methods are mainly oriented to general domains,and there are no good methods for specific domain ontology construction,especially in the military field.This thesis focuses on the issue of weapon entity recognition and entity semantic relationship extraction in the construction of military weapon ontology.The specific research content of the thesis is as follows:(1)To solve the problem of low recognition accuracy due to the complex composition of military weapon entities,a military weapon entity recognition model combining double-layer multi-head self-attention and Bi LSTM-CRF is proposed.First,the character feature,location feature and label feature are used to vectorize one character;Double-layer multi-head self-attention mechanism are then used to weigh and fuse the three input features and extract the key character features;Finally,The recognition result is further filtered by template matching when combining the word formation features of military weapons,and the weapon entities are obtained.To verify the effectiveness of the proposed model,a military weapon entity recognition data set containing 2053 pieces of data was constructed,in which there are 6 types of weapon entities,including "aircraft","ships","tank and armored vehicles","missiles","artillery",and "firearms".The experimental results show that the method proposed in this thesis improves the accuracy rate,recall rate and F1 value by 1.15%,0.97%,and0.97%,respectively,compared with the optimal results of the classic deep learning method(Bi LSTM-ATT-CRF).(2)Comprehensively considering the attribute parameter information of the weapon entity and the relationship between the entities,the semantic relationship extraction of the weapon entity is divided into three categories: attribute extraction,classification relationship extraction and non-classification relationship extraction.Template matching and military encyclopedia data filling are used to complete the attribute extraction;for the classification relationship extraction,K-Means clustering algorithm are first used to divide the semantically similar weapon entities into the same cluster,and TF-IDF algorithm are then used to select the typical weapon entity name from the cluster as the category name of the cluster to obtain the classification relationship of the weapon entity;The method of fusing semantic role and dependency parsing is used to extract the non-classification relationship of the weapon entity.The semantic relationship of weapon entities is extracted based on the task of weapon entity recognition using the above three methods,and the extraction results verify the effectiveness of the above methods.(3)The Neo4 j graph database is used to store and visualize the constructed military weapon ontology,and the function of querying weapon entity attribute parameters and weapon entity relationship query are also realized.Finally,the work of this paper is summarized and prospected.