基于多源数据深度融合的金融时间序列预测

Financial Time Series Prediction Based on Deep Fusion of Multi-source Data

文章提出一种双阶段深度学习的金融时间序列预测模型,研究股民评论、金融新闻资讯与股票指标多源数据对股票市场波动的影响。该模型运用word2vec并结合卷积神经网络对非结构化文本数据进行情感分析,计算情感权重并与股票指数联合;通过双向长短时记忆网络结合注意力机制关注文本重点语义分布,提升全局时序信息敏感度,从而完成非线性、时变性的股指预测。所提模型相比于单一使用股票指数,其均方误差降低0.264,比BiLSTM股票预测模型降低了0.186。实证结果表明,端对端的多源数据融合情感分析模型能够有效解决因多级因素导致的股票市场波动性与不规律性,从而对股票指数进行预测。

This paper proposes a two-stage deep learning financial time series forecasting model to study the impact of stockholder comments, financial news and stock index multi-source data on stock market volatility. Firstly, this model uses word2vec combined with Convolutional Neural Networks(CNN) for sentiment analysis of unstructured text data, and calculates the emotional weight and combines with the stock index, then focuses on the semantic distribution of text and improves the sensitivity of global time series information by combining Bi-directional Long and Short Term Memory(BiLSTM) with Attention mechanism, so as to complete the nonlinear and time-dependent stock index prediction. Compared with the single stock index, the Mean Square Error(MSE) of the proposed model is reduced by0.264, and is 0.186 lower than that of the BilSTM stock prediction model. The empirical results show that the end-to-end multi-source data fusion sentiment analysis model can effectively solve the volatility and irregularity of the stock market caused by multi-level factors, thus predicting the stock index.