Aspect Level Sentiment Analysis For Fusion Of Bi-LSTM And Text Information

Sentiment analysis is a fascinating task in natural language processing,aiming at extracting and organizing sentiment information from text in an automatic way.With the analysis of granularity in recent years,aspect level sentiment analysis becomes one of the most important research directions of sentiment analysis.It has a very important application and research value.The aspect level sentiment analysis based on neural network is a popular method of aspect level sentiment analysis in recent analysis.It is different from the method based on sentiment lexicon or machine learning algorithm,which has modularized structure,the aspect level sentiment analysis based on neural network realizes an end-to-end neural network model by building a neural network that can learn and adjust all parameters at the same time.Now,the aspect level sentiment analysis based on neural network has occupied the leading position,but there are still many problems,the most important problem is to distinguish the importance of different context words to aspect word.Context words are very important for inferring the sentiment polarity of the aspect word,and different context words have different effects on the sentiment polarity of the aspect word.Therefore,how to represent the semantic relations between context words and aspect word in an effective way is a key and difficult point in current research.In addition,the word embedding used in the aspect level sentiment analysis based on neural network is learning by context which ignore the sentiment of the text,it may not be significant in other natural language processing tasks,but it can lead to larger problems in sentiment analysis.So as the research goes on,how to learn the word embedding that have both context's information and sentiment's information can also be a challenging task.Therefore,this article innovative in the aspect level sentiment analysis based on neural network learning each context word's importance as the breakthrough point,the application of the concept of sentiment word embedding in aspect level sentiment analysis task.The main innovations and contributions of this paper are as follows:(1)The algorithm of learning word embedding which based on context is extended to add the sentiment of text,so that the word embedding will have some sentiment information when it is initialized.(2)Join the position information of context words and aspect word.Assign positional weights according to the distance between context words and the aspect word,the closer the distance,the greater positional weight,otherwise the positional weight will be smaller,and the positional weight will be adjusted flexibility according to the part of speech.(3)Join the semantic information of context words and aspect.The semantic weight of each context word and aspect word obtained by an attention mechanism network model,which indicates that the different context word has different influence on the object word.The experimental results show that the method proposed in this paper can not only enhance the aspect level of sentiment analysis based on neural network performance effectively,but also have advantages when compared to the traditional machine learning methods.