Event-driven sentiment analysis for stock prediction using constructed domain-specific Chinese financial sentiment lexicon Take the stock price of Haitian Flavouring & Food Company Ltd. as an example

Sentiment analysis is essential in predicting fluctuations in the stock market. Public emergencies could lead to various discussions and opinions that significantly impact corporate image and stock prices. The recent incident involving Haitian Flavouring & Food Company Ltd. is an example of how market sentiment can influence stock prices, with the company's market value plummeting by nearly RMB 33 billion. Accurately capturing market sentiments is crucial for predicting stock trends and reducing investors' losses. However, previous approaches rely on general lexicons, which are not tailored to the finance domain. To address this, we construct a domain-specific Chinese financial sentiment lexicon (DSCFSL) using a BERT model and a corpus extracted from newspapers and stock forum posts. Following the framework of "decomposition and synthesis", we integrate the EEMD (Ensemble Empirical Mode Decomposition) method and the LSTM (Long Short-Term Memory) to predict stock prices using the extracted sentiment indicators. The experiments show that this approach has comparable performance to state-of-the-art methods.