Using Deep Learning to Develop a Stock Price Prediction Model Based on Individual Investor Emotions

Abstract

The general purpose of stock price prediction is to help stock analysts design a strategy to increase stock returns. We present the conceptual framework of an emotion-based stock prediction system (ESPS) focused on considering the multidimensional emotions of individual investors. To implement and evaluate the proposed ESPS, emotion indicators (EIs) are generated using emotion term frequency–inverse emotion document frequency ($\mathit{etf}-\mathit{iedf}$), which modifies term frequency–inverse document frequency ($tf-\mathit{idf}$). Stock price is predicted using a deep neural network (DNN). To compare the performance of the ESPS, sentiment analysis and a naïve method are employed. The prediction accuracy of the experiments using EIs was the highest at 95.24%, 96.67%, 94.44%, and 95.31% for each training period. The accuracy of prediction using EIs was better than the accuracy of prediction using other methods.

Keywords:

• Stock price prediction
• Multidimensional emotions
• Emotion indicator
• Deep neural network

Additional information

Funding

This research has been conducted a research grant of Kwangwoon University in 2019.