Online Topic Detection and Tracking System and Its Application on Stock Market in China

Financial markets are very sensitive to emerging news related to stock because investors need to continuously monitor financial events when deciding buying and selling stocks. Tracking important events has done mostly using rule-based methods in the past, which is time-consuming in the face of huge online news data. To track this issue, in this paper, a novel document embedding technology based on TF-IDF and BERT incorporating online text cluster algorithm to form an automated event detection system is proposed. Embedding technology is first used to encode text to vectors and then an online text cluster algorithm - SinglePass is implemented to accomplish topic tracking. Experiment results show that the proposed algorithms can effectively detect and track online topics. In addition, both domestic and international events such as the outbreak of novel coronavirus (COVID-19) and Sino-U.S. trade war and their impact on capital market in China are analyzed, which demonstrate the practical and economic value of proposed system.