Read like a human: the role of quantified knowledge in market predictions

In this study, I propose a novel entity-specific sentiment index to examine how massive general knowledge can be quantified and used to extract better financial inferences from media outlets following a similar reasoning process as human news readers. With the advent of graph representation techniques, external knowledge can be represented by knowledge graphs, and then quantified through graph embedding processes. By modifying traditional sentiment analysis using quantified knowledge, I find that the introduction of external knowledge significantly and consistently improves the predictive power of sentiment indexes as indicators of stock market activity.