Adaptive Generation of Medical Education Animations for Enhanced Health Literacy: ddddA Personalization Approach for Diabetes, Vaccination, and Mental Health Communication

Abstract

The increasing interconnection of modern financial markets has led securities violations-such as insider trading, market manipulation, and disclosure misconduct-to exhibit cross-market, multi-entity, and temporally progressive characteristics, posing significant challenges to traditional rule-based and post-event regulatory frameworks. In response to the growing demand for proactive and risk-oriented supervision, this paper addresses the task of early identification of securities violations using AI-driven analysis of cross-market multi-source data, with particular relevance to regulatory authorities, brokerage compliance departments, and merger and acquisition funds. We propose CRG-Former (Causal Relational Graph Transformer), a deep learning framework that integrates cross-market financial time-series data, heterogeneous relational graphs among market participants, and causality-aware attention mechanisms to detect potential violations at an early stage. The model employs Transformer-based temporal encoders to capture evolving abnormal trading patterns, heterogeneous graph attention networks to model complex relational dependencies, and causal attention constraints to align model inference with legal notions of behavioral causation. To enhance regulatory usability, CRG-Former further incorporates uncertainty-aware risk prediction, enabling probabilistic early warning rather than deterministic judgments. Experiments on a multi-market dataset integrating equity transactions, derivatives activity, corporate disclosures, and regulatory enforcement records show that CRG-Former achieves an AUC of 0.912, outperforming strong baseline models by over 6%. Moreover, the proposed framework provides an average early warning lead time of 18 trading days before confirmed violations, demonstrating its effectiveness in delivering timely, risk-based, and operationally meaningful signals for AI-empowered securities supervision.