AI-Driven Analysis of Illegal Online Lending in Indonesia

Joko Sriwidodo

Edi Saputra Hasibuan

Anjar Kususiyanah

 

Abstract: Illegal online lending promotions in Indonesia use claims of legitimacy, speed, price, and financial simulations that may obscure licensing, total cost, data practices, and responsibility. This study analysed promotional-risk narratives and evaluated an auditable Legal NLP workflow for preliminary regulatory review. The archive contains 313 source rows, representing 285 Uniform Resource Locators (URLs) after cross-channel normalisation; 103 analytical rows are linked to 102 unique URLs. The collective human–AI review standardised model-assisted phrase proposals into 36 final labels with 245 occurrences. False legality and credibility claims and extreme speed-and-ease claims were the most frequent categories (30 each), followed by misleading financial simulations (28). For legal retrieval, the authors verified 39 current provisions and 4,017 query–provision relevance judgments. At k=5, Indo-LegalBERT achieved Hit@5 of 90.29% and MRR@5 of 56.54% under the strict relevance standard; the extended standard produced Hit@5 of 100% and MRR@5 of 65.24%. A provider-reported Gemini 3.7 Flash run then drafted provisional syllogisms from the reviewed provisions. Of the 103 records, 30 were approved as generated, 71 were approved after author revision, and two were excluded because the source text was not assessable. This study contributes a traceable human–AI framework for prioritising the regulatory review of illegal lending promotions.