SIGNATURE WORK
CONFERENCE & EXHIBITION 2022

The real-time stock price is determined by the company value, public information, and the inherent stochasticity in the stock market. This interdisciplinary work aims at examining the causal link between stock price and market sentiment by adding sentiment scores as a feature to deep neural networks. We use a model based on existing financial lexicons, a valuable comment filter, Long Short-Term Memory (LSTM), and herd instinct in behavioral finance. Based on the financial lexicons, we assign sentiment scores with polarity to comments on the financial platforms. This model is trained by a text dataset about news titles and comments of consumer goods ETF collected from Weibo (微博) and annotated manually from the project. We compared the performance of different prediction methods. We found that involving the sentiment score of each comment improves the accuracy and applicability of the machine learning prediction models.