基于微博搜索和SVM的股市时间序列预测研究

Stock Market Time-series Prediction Based on Weibo Search and SVM

近年来,随着微博的快速发展,其海量信息的挖掘已经成为一个热门的学术焦点。本文针对微博数据挖掘在金融领域的应用提出一种基于微博搜索和SVM的股市时间序列预测方法。以微博搜索功能为基础,进行主题、未来倾向、情感三级分类,实现对微博平台上投资者情绪进行侦测,并计算相应的投资者看涨情绪指标和看跌情绪指标。将两个指标数据引入传统的基于股市历史数据的时序预测方法,形成基于SVM的多变量时序预测模型MTPH&BSI。经过样本训练、参数寻优、测试样本预测等过程,实验结果表明本文所构造的预测模型比传统基于历史数据的单变量时序预测模型具有更佳的预测性能和更好的泛化能力。本文对于研究微博等社会化媒体的服务能力具有借鉴意义。

With the rapid growth of Weibo, its vast data mining technology has become a major academic topic in recent years. This paper provides a method of time-series prediction on stock market based on Weibo search and support vector machines （SVM）, aiming at the application of Weibo data mining in the financial area. Topics, future trends and sentiments can be a- chieved three-level-classification on the basis of Weibo search, which can monitor investors＇ sentiments on Weibo and calculate related sentiment indexes, namely bullish sentiment index （BSI （1） ） and bearish sentiment index （BSI （2））. The multi-varia- ble time-series prediction based on history data and BSI （MTPH＆BSI） has formed, attributed to the introduction of the two inde- xes to the single-variable time-series prediction based on history data. After training patterns, optimizing parameters and testing patterns of prediction, the experimental results show that the proposed prediction model is better than the traditional model in pre- dicting performance and generalizing abilities. This paper is significant to study the service capabilities of socialized media as Weibo.