A financial anomaly prediction approach using semantic space of news flow on twitter

This study represents an initial endeavor to harness the potential of the semantic space within the Twitter news flow to forecast financial anomalies. In pursuit of this objective, approximately two million entities were extracted from the news text disseminated by the most widely followed news channels on Twitter. These entities were scrutinized over 12 years to explore potential correlations between their evolution and future stock market anomalies. The examination focused on the centrality measures of these entities within their daily semantic graphs, with particular emphasis on identifying the most correlated entities. Subsequently, these entities were employed to construct a logistic regression model capable of predicting the presence of future anomalies and their direction whether indicative of an upward trajectory associated with a rise in stock prices or a downward trajectory associated with a decline in prices. The evaluation results demonstrate a remarkable level of accuracy for the prediction model, thereby holding promise for further advancements in this interdisciplinary research domain that encompasses natural language processing, complex networks, and artificial intelligence. Lastly, the findings are discussed in light of pertinent theories that furnish a robust foundation for future investigations.