Research On Chinese Negation Identification Based On Deep Learning

Negation in natural language is a widespread and complex language phenomenon,which tends to reverse the truth of proposition in the text expression,the position of opinion,the polarity of emotion and attitude.In the process of negative information identification,if the negative expression is distinguished from the factual information,the reliability and the value of the obtained information will be greatly improved.At present,negative information identification has become a research hotspot in natural language processing.As a basic task,it is not only important for information retrieval,emotion analysis,text mining and information extraction,but also has a positive effect on the deep semantic understanding of text.Most of the existing research on the identification of negation,which has been achieved preliminary results,is English oriented.While the work in Chinese is relatively less.In the related research,the machine learning method is usually used to convert the negative information into a sequence annotation problem.This method relies on heavy feature engineering,and the performance obtained in Chinese negation identification is not high.In recent years,an increasing number of deep learning technologies have been successfully applied in natural language processing,and good performance has been achieved especially in sequence tasks.Therefore,this paper mainly researches on the method of Chinese negation identification based on deep learning.The main contents are as follows:First of all,comprehensive study of both English and Chinese negation identification method is conducted,and then the solutions to subtasks in the negation identification are analyzed and classified.By summarizing the advantages and disadvantages of these methods,a new idea using the technology of deep learning to solve the problem of Chinese negation identification is proposed.Then,after analyzing the characteristics of deep learning technologies related to solve the problem of sequence,a model of Chinese negation cue detection based on long-short term memory neural network is proposed.This model labels the sentence sequence based on character-based and word-based framework respectively.At the same time,the word embedding is used to capture words semantic information.The character-based and word-based vectors,obtained by pre-training,are used as the input characteristics of the model.Experiments show that the effect of this model is better than traditional sequence labeling method based on the Conditional Random Field(CRF).Finally,in view of the large features space of the scope of negation and the complexity of designing and obtaining features,the method based on long-short memory neural network is still used to resolution the scope of negation.According to the characteristics of tasks,the additional information related to negation cue is introduced by word embedding to improve the model.The experimental results show that the model also has advantages compared with the traditional method based on single classifier,and the improved method with negation cue-embedding also improves the performance of the model.This paper tried to use deep learning technologies to solve the problem of Chinese negation identification.With no need of highly-engineered artificial features,which means less dependence on domain knowledge,this method is effective for the task and there is still a large room for improvement.