Research On Sentiment Analysis Of Weibo Stock Review Based On Word2vec And LSTM

With the continuous expansion of the scale of netizen and the continuous innovation of the Internet model,unstructured online review text data is increasing with each passing day.These data contain abundant information and can generate a huge value space.However,facing enormous amounts of unstructured data resources,traditional methods that rely too much on manual processing show some limitations in the use of data.Storage,processing,analysis,and use of massive unstructured text data are both opportunities and challenges for intelligence scholars.Sentiment analysis is an effective way to extract valuable information fron unstructured data and use the data to solve practical problems.Deep learning is a series of complex algorithms that simulate information processed by the human brain.It can automatically map raw data to higher-level,more abstract feature spaces through multi-layer nonlinear transformation.Sentiment analysis based on deep learning can process large-scale text data which only needs a small amount of manual intervention.It enable the disclosure,description and knowledge discovery of massive unstructured data resources.The research focus of this paper is adopting the sentinent analysis method based on deep learning to conduct sentiment analysis of large-scale online stock review texts,finding the relationship between sentiment tendencies of the stock analysts and the trend of the stock market from the interpretation of the analysis results and providing valuable intelligence services for investors or related regulatory authorities.This article focuses on two issues.One is whether the method based on deep learning is suitable for the sentiment classification of the Weibo stock review text;the other is the relationship between the sentiment tendencies about the future stock market of Weibo big V securities analysts and the real stock market trend.In view of the first question,based on the relevant theories and methods of deep learning,natural language processing and sentiment analysis,this paper divides the sentiment classification of stock review texts into three sub-tasks,namely(1)extracting predictive Weibo reviews;(2)Training word vectors;(3)Training classification models.Then compare and analyze the models that can be used in each part and then determine the model that is suitable for this study.For the second question,this paper proposes a method of correlation analysis and Granger causality test.One of the results of this paper is to propose a fusion model based on word2vec and LSTM for the sentiment analysis of Weibo stock review.The fusion model is divided into six steps:(1)Data acquisition.Using web crawlers to capture Weibo big V securities analysts' review text containing the keyword "big market";(2)Data preprocessing.Exclude the contents irrelevant to stock review;(3)Extract predictive Weibo review.Tagging part weibo reviews contain future forecasting content in advance,and using the similarity calculation method to extract other predictive tweets;(4)Trains word vectors using the word2vec tool;(5)Trains the classifiers using the LSTM model and applies the classifiers on the other Weibo stock review texts;(6)Construct sentiment index time series and trend index time series.Then use correlation analysis and Granger causality test to quantify the statistical relationship between sentiment index and trend index in different market segments.This paper conducts an empirical study on the text sentiment analysis fusion model based on word2vec and LSTM.With the empirical analysis,this paper finds:(1)The accuracy rate of the sentiment classification of Weibo stocks based on word2vec and LSTM is 75.2%.It is significantly better than the method based on lexicon and the method based on classical machine learning.This shows that the nethod based on deep learning is suitable for the sentiment classification task of the Weibo stock review text;(2)The Weibo big Vs are more inclined to publish positive or neutral remarks.They will be more cautious when expressing remarks that express a negative attitude;(3)When the broader market fluctuates violently,there is no correlation between the sentiment tendencies of the Weibo big V securities analysts and the trend of the real market,but during the period when the broader market moves steadily,emotional attitude of the Weibo big V security analysts for the future market has a certain predictive effect on the broader market.