Research On Negation Identification

Negation is a complex linguistic phenomenon that exists in all languages.It can reverse the semantic information contained in sentences or words.More and more natural language processing tasks need to mine valuable information from text information.Negation existence will lead to the mining of useless or even completely opposite information,which has a negative impact on the performance of the task.Therefore,it is necessary to identify the negative information,separate the positive information from the negative information in the text information,identify the negation scope of action.Thus obtaining the correct meaning of sentence expression and improving the performance of tasks such as sentiment analysis,information retrieval,and information extraction.This paper focuses on the following aspects of Chinese negative information recognition and English negative information recognition:1)Using the Bidirectional Long Short Term Memory Network combined with the Conditional Random Fields model(BiLSTM-CRF)to identify negation cue and negation scope for Chinese and English respectively.The model can overcome the shortcomings of the conditional random field relying on the manual extraction feature and the shortcoming of the Bidirectional Long Short Term Memory Network without considering the label sequence globally.The pre-trained word embedding is the input feature to detect the cue.Based on this,add the known cue features to define the scope.The experiment proves that in the Chinese and English corpus,the experimental results of negative cue detection and negative scope recognition based on the model are better,and the model has strong generalization ability among different texts in the biomedical field of English corpus.2)Add a self-attention mechanism for negation scope identification.For the problem that the negation scope is too long difficult to identify complete,the added self-attention mechanism layer can shorten the distance between any two words by calculating the similarity of them inside the sentence,which improves the long-term context dependence of model learning.The experimental results show that the method is important for improving the performance of negation scope recognition.3)Integrate the dependency syntax information to identify the negation scope.Negative coverage field is the scope of the negation cue.There is a dominant and dominant relationship in the dependency syntax.Therefore,based on the BiLSTM-CRF model,the distance between dependent words is improved by incorporating dependent syntax information to improve the ability of the bidirectional LSTM layer to learn long-term context dependencies.The experimental results show that this method can improve the effect of negation scope recognition on the expression of more standardized corpus.