The Design And Implementation Of Chinese Texts' Sentiment Analysis Program Based On An Improved LSTM Framework

With the development of the Internet,most users leave comments on the Imternet to communicate with each other.The comments can be speared quickly in a quite large scale to make influence on many people.Making more use of these comments datas can be useful both under commercial consideration and social consideration.The current research in academia focuses on making classification of these comments datas using machine learning techniques,such as deep learning,which is always a binary classification task.The answer of the classification is either positive or negative.The research on Chinese sentiment analysis is relatively slow.After studying basic information on related technolohgy,following jobs have been done:Firstly,aiming at the fact that the sentiment of a comment has been not only affected by the early information of the text,but also the later one,an improved scheme has been proposed.The new scheme basing on LSTM can proceed data in binary directions in a parallel way of connecting two LSTM networks.Then a third-party data has been used to verify that the BiLSTM model can perform better than traditional LSTM method in text-processing tasks.Secondly,because the main job of this paper is to process sentiment analysis on Chinese comments in this paper,the method to analyze words has also been discussed in this paper,and a new method to analyze words in maximum entropy model basing on dictionary-wordssegment algorithm has been proposed.Comparing with other different words-segment algorithms,this new method has been proved efficient.Then a new word embedding method has been applied in this paper to train Chinese word embedding,and Chinese comments data have been used to verify this method useful or not.Finally,the whole comments data have been divided into training set,validating set and test set to measure performance.The training set and validating set have been used in training the two different models,the test set has been used to analyze results,and a third-party sentiment analyze program has been used to compare whis the new method.The new method has been proved efficient in Chinese comments'sentiment analysis.