A Sentiment Index Study Based On Social Platform Investor Sentiment Analysis

With the popularity of the Internet,stockholders are more and more inclined to use financial social media platforms to learn about financial information in the stock market and to express their views on the future trend of the stock market in social media platforms.Therefore,the analysis of the sentiments of stockholders in financial social platforms can reflect the views and attitudes of users.Stock commentary not only affects the investment judgment of the general public,but also,to a certain extent,the trend of individual stocks and the broader market.In order to promote the health of financial and securities markets,stock reviews are categorized into sentiment categories to quantify investor sentiment and,in combination with other sentiment proxies for the stock market,to study the impact of investor sentiment on broad market trends.In the past,scholars have mainly used indirect stockholder sentiment indicators to quantify stockholder sentiment,but lacked research on the subjective emotions of specific stockholders.Some scholars combine the information in social networks that can reflect the subjective emotions of stockholders,but only use the emotion dictionary to quantify stockholder sentiment.Due to the problems of colloquialism and short,concise content,the use of sentiment dictionaries for sentiment classification cannot fully consider the semantics of the text,and is prone to generalization.To solve the above problems in the categorization of sentiment in financial social networks and the construction of an index of stockholder sentiment,this paper focuses on the following aspects:(1)This paper uses the Bi-LSTM(Bi-directional Long Short-Term Memory)affective classification model,which is able to integrate the semantic features of the text,to perform word vector transformation using word2 vec based on the fusion of forward LSTM and backward LSTM memory networks,which integrates the word and word The model is able to focus on key words and fully consider text semantics when classifying sentiments.Finally,a comparison experiment is conducted in this paper,and the experimental results show that the Bi-LSTM affective classification model is better than the CNN model.(2)In this paper,a method to calculate the sentiment value of a stock comment text,and construct an investor sentiment index using the sentiment value of the text and an indirect investor sentiment proxy.Firstly,the weight of each comment text is calculated,and secondly,the sentiment value of the comment text is obtained by multiplying the sentiment classification result of each comment with the corresponding weight.Combined with indirect investor sentiment proxies,an investor sentiment index is constructed by principal component analysis.Through an experimental comparison,it is proved that the inclusion of the sentiment value of the comment text is more realistic than the investor sentiment index constructed using the indirect investor sentiment proxy alone.(3)This paper uses the BP neural network prediction model to verify that adding the investor sentiment index predicts the results more accurately than not adding the investor sentiment index.