TED-FinRisk: a hybrid topic-enhanced framework for typologizing illegal fundraising risk from multi-source case texts

Abstract

Illegal fundraising is increasingly conducted through dispersed digital channels, cross-regional operations, and rapidly changing promotional narratives, which makes static rule lists and purely expert-driven typologies difficult to maintain. This paper presents TED-FinRisk, a hybrid framework for constructing an interpretable risk taxonomy from multi-source case texts. The framework combines term frequency–inverse document frequency (TF-IDF) feature construction, topic exploration assisted by t-distributed stochastic neighbor embedding (t-SNE), latent Dirichlet allocation (LDA)-based topic extraction, and Analytic Hierarchy Process (AHP)-based expert priority calibration under the Financial Action Task Force (FATF) threat–vulnerability–consequence perspective. After the taxonomy is finalized, FinBERT is used for a reclassification consistency check aligned with the final expert labels, examining whether the taxonomy corresponds to recognizable text-semantic patterns within the study corpus. The study is based on a corpus of more than 1,000 illegal fundraising cases and nearly 6,000 associated risk-related records covering scenarios such as elder care, virtual currencies, supply-chain finance, agricultural ventures, film investment, social e-commerce, and overseas investment. The resulting taxonomy contains six primary categories and a set of secondary labels linked to operational indicators for downstream monitoring. These results suggest that combining data-driven topic discovery with expert calibration can produce a risk taxonomy with regulatory semantics and monitoring relevance. The paper contributes a methodology-oriented case study, a reusable taxonomy design process, and a structured basis for future benchmarking and early-warning applications.

The authors' abstract, as published at the source. Journal Of Big Data, 2026 · DOI ↗