Research On Sentiment Analysis Of Stock Reviews In Liquor Sector Based On Word2vec And Ensemble Learning

Due to the rapid development of information society,many Internet models have emerged,and these models are constantly innovating.In addition,people's cultural level is generally improving,so we can find that the scale of Internet users is increasing rapidly.Moreover,due to the gradual improvement of people's living standards,more and more people invest their spare money.These investors will obtain the latest investment information on various Internet information platforms and express their personal views.Therefore,the stock evaluation data on the network is also proliferating.There is infinite information in these stock evaluation data,which is waiting for us to mine.However,because the volume of these unstructured stock evaluation text data is too large,there are some limitations if we only rely on the traditional manual processing methods for data processing and analysis.Therefore,it is very necessary to use computers to process and analyze the massive unstructured stock evaluation text data.However,the focus of this thesis is to use the ensemble learning algorithm to analyze sentiment in stock review texts,and then explore the relationship between the emotional tendency of micro-blog professional stock commentator and retail investors and the trend of the liquor sector,so as to provide evidence for regulators and investors.The stock reviews of retail investors in the liquor stock bar of Oriental Fortune and the stock reviews of micro-blog professional stock commentator on the liquor sector are collected as the research object of this thesis,and python software is used for crawling the relevant data from January 1,2020 to June 8,2021.The stock transaction data of the liquor sector is downloaded in the Tongdaxin financial terminal.The main content and empirical results of this thesis are as follows: Firstly,the sentiment analysis is carried out on the collected stock reviews of micro-blog professional stock commentator and retail investors,and the Version 2.0 Financial Sentiment Dictionary is used to segment each sentence in the corpus in a precise mode,and remove stop words.After forming a list of words,Word2 vec is used to train the word vector,and the 30-dimensional word vector of each comment is extracted.Secondly,the five integrated learning methods of GBDT,Ada Boost,XGBoost in Boosting,random forest and Stacking in Bagging are used to build a emotion classification model.The evaluation indicators are used to compare the effects of classification models.Then,the overall analysis and correlation analysis are carried out on the sentiment of stock reviews and the trend of the sector,and then the regression analysis of the three is carried out to explore the indicators that have a significant impact on the trend of the liquor sector.The dynamic relationship between retail investor sentiment and the trend is studied by constructing a vector autoregressive model and the Granger causality test is used to explore the causal relationship between the three.The research results show that:(1)The evaluation indicators of sentiment classification based on word2 vec and ensemble learning can be concluded that Stacking has a better sentiment classification effect on micro-blog professional stock commentator,and both Stacking and XGBoost have better sentiment classification effects on retail investors.It shows that Stacking is suitable for the sentiment classification task of investor comments;(2)Through the overall trend analysis,it is concluded that the trend of retail investor sentiment value and micro-blog professional stock commentator sentiment value is basically the same,but the overall trend of micro-blog professional stock commentator sentiment value is positive.The sentiment value of retail investors is in a state of fluctuation.It shows that micro-blog professional stock commentator tend to publish posts with positive emotions;(3)Through correlation analysis and regression analysis,it can be concluded that there is a significant positive correlation between the two,and compared with the well-known stock critics in Weibo,the sentiment of retail investors has a greater positive impact on the rise and fall of liquor sector;(4)By constructing a vector autoregression model,it can be concluded that there is a lag effect between the changes in retail investor sentiment and the trend of the liquor sector;(5)The results of the Granger causality test show that the emotional changes of retail investors can help predict the future rise and fall of the liquor sector.The comments of micro-blog professional stock commentator stock critics on social media can help predict the ups and downs of the sector.There is no ability to predict each other's emotional tendencies between stock reviews and stock reviews of retail investors.