结合情感分析的股票预测研究

发布时间：2018-04-10 03:02

本文选题：股票预测　切入点：情感分析　出处：《内蒙古大学》2017年硕士论文

【摘要】：股票投资是一种非常活跃的投资理财方式。投资者在股票市场上的交易行为都以盈利为目的。目前,股票预测大多仅基于股票交易的历史数据。本文研究结合股票评论文本情感分析的股票预测模型。模型设计中分析情感倾向、股票交易指标、时间序列等方面的数据。情感分析:本文以活跃股评论坛特定股票的股评文本作为分析数据。这些论坛数据是大量含噪音的短文本,反映的是中小股票投资者的观点。使用SVM分类器,利用JAVA版本的LIBSVM工具包进行文本分类,计算分析得到情感倾向指数Bs。工作改进:分析多时段数据建立BP网络预测模型,并根据MIV算法求解不同时段数据的不同影响值;在计算文本情感倾向指数Bs时,加入文本作者影响权重。模型设计:只有五个股票交易指标作为输入量的BP神经网络模型,作为参考模型,记为模型一;结合情感指数Bs和五个股票交易指标的多指标BP网络模型,记为模型二;分析预测日之前五个交易日收盘价的多时段BP网络预测模型,记为模型三。结论:包括情感指数的预测模型二要比模型一的准确性高;模型三结合MIV算法得出了预测日前五个交易日的影响权重值,结果符合越靠近预测日的数据影响权重越大的趋势。
[Abstract]:The stock investment is one kind of very active investment finance way.Investors in the stock market trading behavior with the purpose of profit.At present, stock forecasts are mostly based on historical data of stock trading.This paper studies the stock prediction model based on the emotion analysis of stock review text.In the design of the model, we analyze the data of emotion tendency, stock trading index, time series and so on.Affective Analysis: this paper uses the stock review text of the active Stock Review Forum as the analysis data.These forum data are a lot of noisy short-text, reflecting the views of small and medium-sized stock investors.By using SVM classifier and LIBSVM toolkit of JAVA version, text classification is carried out, and the affective tendency index (Bs.) is obtained by calculation and analysis.Work improvement: the BP neural network prediction model is established by analyzing the multi-period data, and the different influence values of the data in different periods are solved according to the MIV algorithm, and the influence weight of the text author is added in the calculation of the text affective tendency index Bs.Model design: there are only five stock trading indicators as input BP neural network model, as a reference model, as model one, combined with emotion index Bs and five stock trading indicators of multi-index BP network model, as model two;The BP neural network forecasting model of five trading days before the forecast date is described as model 3.Conclusion: the accuracy of the prediction model 2 including emotion index is higher than that of model 1. Model 3 combined with MIV algorithm has obtained the influence weight of the first five trading days of the forecast day, and the result accords with the trend that the influence weight of the data closer to the forecast day is greater.
【学位授予单位】：内蒙古大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1

【参考文献】