Chinese Text Sentiment Classification Based On LSTM

With the development of intelligence and digitization of the Internet,various online social and shopping platforms have developed rapidly,and people are more enthusiastic about expressing their opinions through these platforms.Every day,the Internet generates massive unstructured text information,which contains users' views and attitudes about people,things,and events.The effective management of these text data and the discovery of valuable emotional information about it can help promote the development of Internet public opinion analysis,medical applications,enterprise management,and product marketing.In text sentiment classification tasks,methods based on sentiment dictionaries and machine learning are suitable for situations where the amount of text corpus data is small and text semantics are simple.With the explosive growth of text information on the web and the continuous enrichment of expressions,many researchers have gradually applied deep learning methods to text sentiment classification,and have made breakthrough progress.Convolutional neural networks(CNN)can effectively capture local feature information on spatial structures,but lack the ability to learn the contextual relevance to words.Recurrent Neural Network(RNN)can better solve the problem of semantic context,but gradient explosion and gradient disappearance are easy to occur during training.Aiming at these problems,an LSTM based Chinese text sentiment classification model is designed for text representation and sentiment classification.Based on the analysis of text representation methods and deep learning models,this paper makes in-depth researches on how to effectively express the emotional information on text and reasonably construct a network model.The main research work is as follows:1.Review the research on text sentiment classificationBased on the in-depth study of related concepts and theories of textsentiment classification in the field of NLP,the basic process of text sentiment classification is combed,and the basic theories and related technologies used in the text sentiment classification process are explained in detail.2.Improve text sentiment classification model with single feature inputConstruct a text sentiment classification model based on multi-feature representation to the study of text sentiment classification.Aiming at the particularity of Chinese text and the limitation of word expression method of emotion expression,the emotion vector was introduced into the basis of obtaining the word sense vector.The construction of multi-feature representations of words containing sentiment vectors as the input of the Chinese text sentiment classification model based on LSTM,to a certain extent,makes up for the shortcomings of the word sense vector for the lack of expression of emotional information.3.A text sentiment classification model based on the influence of words with weightsIn order to further to emphasize the expression of emotional information on the text,a word attention layer is added on the basis of constructing a multi-feature representation model to highlight the words with a large emotional contribution and improve the impact on keyword differentiation on text classification.In addition,a word-tag relationship network model is constructed to obtain deeper tag dependencies and enrich the feature information about the text.4.Emotion classification experiments on Chinese textIn order to verify the effectiveness of the method in this paper,comparative experiments were conducted on different data sets,and the experimental results were analyzed and summarized.The experimental results show that the LSTM + DW + Attention + relationship model proposed in this paper can effectively distinguish texts of different emotion categories,which improves the classification accuracy to a certain extent.