Word Sense Disambiguation In Natural Language Understanding And Its Application In Instrument Industry

With the rapid development of artificial intelligence,the development level of deep learning has been significantly improved.The need for computers to understand natural language is more and more urgent.In machine understanding of natural language,it is inevitable to solve the problem of ambiguity.Word sense disambiguation aims to determine the specific meaning of an ambiguous word in the text.In recent years,it has been attracted the attention of researchers.In the field of instruments,there will be many tasks that require computers to understand text well,such as requirements analysis,human-computer interaction.An essential part of computer understanding of texts is word sense disambiguation.There is still a lot of room for improvement in the accuracy of the existing word sense disambiguation model.An effective way to improve the accuracy of the word sense disambiguation model is to inject prior knowledge into the model.The integration method is relatively simple,and there is no corresponding knowledge base in the Chinese word sense disambiguation.This thesis proposes a word sense disambiguation model that integrates knowledge,obtains the complementary knowledge through the encyclopedia knowledge base,and combines it into the word sense disambiguation model.The accuracy of the word sense disambiguation model is obtained,and then it is applied to reading comprehension in the field of instrumentation to prove the effectiveness of the model.First,this thesis designs a word sense disambiguation model that integrates sense terms.The crawler obtains the semantic items of the words in the encyclopedia,and the semantic knowledge features are injected into the embedding layer of the pre-training language model.Then the embedding layer is applied to the pre-training language model to improve the correct rate of word sense disambiguation.Second,a word sense disambiguation model that integrates hypernym and knowledge is proposed.The knowledge incorporated in the model mainly includes epistatic relation and semantic knowledge.Firstly,a definition of the upper and lower relationship is improved,and a deep learning model for the extraction of the upper relationship is designed to judge the upper relationship of the ambiguous words in the input text.At the same time,the middle layer of the word sense disambiguation model that integrates the semantic terms is injected into the model.Output features and the two models’ final output features can be fused to obtain a word sense disambiguation model that integrates knowledge.Finally,the comparison experiment proves the critical role of the sense terms and the upper relationship in the word sense disambiguation model.Third,based on the word sense disambiguation model in the general domain,this thesis proposes a domain-oriented word sense disambiguation model and applies it to the reading comprehension task in the instrument domain.The domain-oriented word sense disambiguation model adds a domain knowledge description module to the word sense disambiguation model that incorporates knowledge.By establishing an instrument domain knowledge base,the output features of the module and the output features of the original model are organically integrated to eliminate the word sense.The domain knowledge is injected into the discriminative model.Then a domain-oriented reading comprehension model is designed based on this,and the validity is proved through reading comprehension experiments in the instrument domain.