| When studying the volatility of the stock market,in addition to comprehensively analyzing the relevant financial data affecting the stock market,it can also be considered with the help of short text data of stock reviews.However,due to the limitation of sentiment analysis technology,the current analysis of stock review text data is mainly based on manual annotation of sentiment polarity and then stock market volatility prediction,which has problems such as long time,high cost,and insufficient use of actual stock data.Based on this,this paper proposes a statistical analysis method that combines historical stock data and stock review text data organically,and establishes a "five poles" sentiment analysis model and an "expert" weight function model for stock ups and downs to predict the ups and downs of individual stocks..First,use the seed dictionary and the emotional tendency point mutual information algorithm to establish an emotional dictionary to mark the sentiment of stock reviews as three polarities: up,neutral and down;The feature words of the stock text data and the sentiment dictionary database are further marked,and the three types of sentiment are further marked as five polarities: big up,small up,neutral,small down and big down,and then use the marked stock evaluation data for training.The classifier model implements the five-pole sentiment classification for new stock reviews.Secondly,use the review classification results corresponding to the reviewers under each stock,the classification prediction accuracy of each reviewer,and the frequency of classification results to calculate the percentage of each classification result;"Typical sorting",select the "experts" among the reviewers and normalize the sorting results,establish a weight function model,and calculate the weight of each "expert" in the predicted score;By combining the weights of "experts",a prediction model of "experts" was established,and the prediction scores and prediction accuracy of individual stocks by "experts" were calculated.The changing trend of the forecast accuracy of individual stocks in different time periods.Finally,combining the sentiment polarity classification results with the actual stock ups and downs data,using the method of empirical analysis,the construction of the "five poles" sentiment analysis model and the "expert" weight function model are tested.The results show that the accuracy rate of "five-pole emotional labeling" is greater than 75%,and the accuracy rate of the classifier model generated by this training also exceeds 70%,and the prediction accuracy rate of the established "expert" weight function model is in It fluctuates between 60% and 100% and stabilizes at around 80%.The "five poles" labeling model constructed in this paper realizes the automatic labeling of emotional labels to a certain extent,which helps to solve the complex and cumbersome problems of manual labeling;Predicting the ups and downs trend is a new attempt,which is innovative to a certain extent. |
