A Commonsense Knowledge-Enabled Textual Analysis Approach for Financial Market Surveillance

Abstract

Market surveillance systems (MSSs) are increasingly used to monitor trading activities in financial markets to maintain market integrity. Existing MSSs primarily focus on statistical analysis of market activity data and largely ignore textual market information, including, but not limited to, news reports and various social media. As suggested by both theoretical explorations in finance and prevailing market surveillance practice, unstructured market information holds major yet underexplored opportunities for surveillance. In this paper, we propose a news analysis approach with the help of commonsense knowledge to assess the risk of suspicious transactions identified in market activity analysis. Our approach explicitly models semantic relations between transactions and news articles and provides semantic references to words in news articles. We conducted experiments using data collected from a real-world market and found that our proposed approach significantly outperforms the existing methods, which are based on transaction characteristics or traditional textual analysis methods. Experiments also show that the performance advantage of the proposed approach mainly comes from the modeling of news-transaction relationships. The research contributes to the market surveillance literature and has significant practical implications.