Sentiment Analysis for Predicting the Variation Trend of Stocks: A Case Study of Vanke Co., Ltd

Qiuyi Jin, Jin Zheng

Abstract: Stock prices constantly fluctuate due to changes in supply and demand. This relationship between supply and demand is highly sensitive to investor sentiment and current news. For example, when a company announces a new product, it suggests the potential for increased future earnings, typically leading to more people buying stocks and driving up stock prices. In addition to official information like government economic reports and company and industry news, the attitudes of traders are significant, as they can reflect a direct intention to sell or purchase stocks. Furthermore, interconnected companies, such as investors and investees, may share revenue and exhibit similar stock trends, making it necessary to consider relationships between listed companies. Our primary focus is on China Vanke Co., Ltd. (referred to as Vanke) and its related companies, executives, industries, and concepts. We establish information about companies and their relationships using a knowledge graph. China Securities Journal and Sina Finance are chosen as data sources for their credibility and extensive information. We compare different text classification algorithms and select the Convolutional Neural Network (CNN) model to categorize news into different industries and label them as either ‘positive’ or ‘negative.’ Additionally, we adopt the Bidirectional Long Short-Term Memory (Bi-LSTM) model integrated with a Conditional Random Field layer (CRF) to extract organizations, individuals, and locations relevant to the news to uncover key information. The organizations and individuals identified are associated with the companies and executives in the knowledge graph, linking the news to the relevant entities. This research contributes to the field of sentiment analysis in financial news, particularly with respect to stock price prediction and the identification of key factors influencing market trends.