Research On The Constructing Rules And Automatic Recognition Of English Military Equipment Names

Automatic processing of military texts is an important part of military informatization.Recognition of military named entities is the foundation of the automatic processing of military texts.English military equipment names are an important kind of military named entities.Due to the complexity of the components of military equipment names,the large number of constructive patterns and the small-scale tagged corpus in the military field,it is difficult to identify English military equipment names.At the same time,English military equipment names contain obvious domain features,which can support its recognition.Taking the identification of English military equipment names as a specific task,this paper aims to argue the validity of domain features in the domain-specific entity recognition tasks.The main contents of this paper are as follows:(1)Firstly,this paper studies the constructive rules of English military equipment names.While revealing the naming rules of English military equipment names,two major domain knowledge bases,namely,Military Equipment Name Component Dictionary and Military Equipment Name Pattern Set,are obtained.Supported by the two domain knowledge bases,this paper designs a rule-based recognition algorithm to identify English military equipment terms and English military equipment names in texts respectively.The purpose of the former is to verify the descriptive ability of the military equipment name component classification system constructed in this paper,so as to illustrate that the research on the constructive rules of military equipment names in this paper is scientific and reasonable.The purpose of the latter is to analyze the difficulties and key points of military equipment name recognition through rule-based recognition algorithm.(2)With the two domain knowledge bases as language resources,this paper incorporates the domain features of military equipment names into CRF(Conditional Random Fields)model and Bi-LSTM(Bidirectional Long-Short-Term Memory)+Multi-Head-Attention+CRF model respectively.Besides,this paper compares the recognition results of domain features,POS(Part of Speech)features and dependency syntactic features in the two models respectively,which argues the validity of domain features in the domain-specific entity recognition further.In the specific research progress,this paper improves CRF model and Bi-LSTM+Multi-Head-Attention+CRF model according to the features of military equipment names.(3)After arguing the validity of domain features in domain-specific entity recognition tasks through experiments,this paper designs seven evaluation indicators and investigates the distribution of POS features,dependency syntactic features and domain features of English military equipment names in the corpus.On this basis,this paper provides a scientific and reasonable explanation for the phenomenon that domain features support domain-specific entity recognition tasks better than general linguistic features,which further demonstrates the validity of domain features in domain-specific entity recognition tasks from a theoretical level.(4)Through conducting experiments on different corpus scales,this paper reveals the characteristics of CRF model and Bi-LSTM+Multi-Head-Attention+CRF model in utilizing features.It is worth mentioning that this paper finds that POS features and dependency syntactic features have requirements for the size of labeled corpus to have a positive supporting effect on Bi-LSTM+Multi-Head-Attention+CRF model while the domain features don't have such requirements.(5)By comparing the recognition results of CRF model with Bi-LSTM+Multi-Head-Attention+CRF model,this paper finds that CRF model outperforms Bi-LSTM+Multi-Head-Attention+CRF model under certain conditions.This discovery could provide guidance for researchers to select models and features according to different tasks(especially in engineering practice).(6)Based on the research on the English military equipment name recognition,this paper evaluates domain-specific entity recognition tasks generally and puts forward some specific methodological suggestions,hoping to provide more references for researchers in related fields.In this paper,researches on the constructive rules and recognition of English military equipment names are conducted comprehensively,which can provide effective technical ideas and methods for the construction of practical English military equipment name recognition system in future.In addition,this paper can provide references for the recognition of other military named entities and other domain-specific entities.