Evaluating the effectiveness of sentiment-based models for stock price prediction | |
| Author | Tachapong Panpoonsup |
| Call Number | AIT RSPR no.DSAI-22-04 |
| Subject(s) | Stock price forecasting Neural networks (Computer science) |
| Note | A research study submitted in partial fulfillment of the requirements for the degree of Master of Science in Data Science and Artificial Intelligence |
| Publisher | Asian Institute of Technology |
| Abstract | Stock price prediction has attracted a lot of interest from investors, analysts, and re searchers. However, it can be insufficient to predict stock prices based solely on his torical prices. Indeed, news can provide crucial information that can help explain stock price fluctuations. This study evaluated the effectiveness of sentiment-based models by extracting news content from Market Watch with historical price information from Yahoo Finance to predict stock prices. Two small cap companies (Gaps and Urban Outfitters) and two large cap companies (Tesla and JP Morgan Chase) were chosen. FinBERT, a transformer-based neural network that has been pretrained and finetuned with financial textual data, was compared to BERT for calculating the sentiment polar ity of news articles. Then, these sentiment values were combined with historical price data for prediction purposes. Various temporal deep learning based models, including 1- dimensional convolutional neural network (1DCNN), long short-term memory (LSTM), attention-based LSTM and temporal convolutional neural network (TCN), were com pared. Key findings validated the usefulness of incorporating sentiment to enhance prediction performance but not for all conditions, especially for small cap firms when news are relatively scarce. In addition, the ability to model long-term dependencies were found to be essential in achieving consistent and robust prediction performance across window sizes as seen in the superiority of LSTM and attention-based LSTM. Last, using Latent Dirichlet Allocation, we identified specific news topics with a signif icant impact on stock price. This work contributes to promising evidence for predicting stock prices by combining sentiments and historical price data using transformer-based model such as BERT and FinBERT combined with typical temporal network such as attention-based LSTM1 . |
| Year | 2022 |
| Type | Research Study Project Report (RSPR) |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Data Science and Artificial Intelligence (DSAI) |
| Chairperson(s) | Chaklam Silpasuwanchai |
| Examination Committee(s) | Mongkol Ekpanyapong;Dailey, Matthew N. |
| Scholarship Donor(s) | AIT Scholarship |
| Degree | Research Studies Project Report (M. Sc.) - Asian Institute of Technology, 2022 |