Scholarlink

Health insurance fraud, as a complex and widespread crime, poses a serious threat to healthcare funds. The most critical phase of the fraud chain is detecting drug trafficking among suspects. However, procuratorial officials face significant challenges in this task, due to the complexity of multi-modal digital forensic data, diverse suspects’ communication styles, and intricate relationships among them. To address these issues, LISPER, a procuratorial service to detect drug trafficking is proposed for procuratorates, which fully leverages multiple large language models(LLMs). Firstly, LISPER extracts multi-source heterogeneous digital forensic data into standardized ones. Then, it accurately tags suspicious entities for individuals of suspects. Finally, LISPER incrementally maintains a relationship network for all the suspects, among which heterogeneous entities like suspects, drugs, and locations are contained. Successfully run in a district-level procuratorate of Beijing, LISPER has presented more than 10K judicial evidences from hundreds of detected suspects. A demo video of LISPER is available at https://youtu.be/24N3p3BYDlQ .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A LLMs-Based Procuratorial Service for Detecting Drug Trafficking on Digital Forensics Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Haoyang He,
  • Weilong Ding,
  • Hao Zhang,
  • Zhuofeng Zhao

摘要

Health insurance fraud, as a complex and widespread crime, poses a serious threat to healthcare funds. The most critical phase of the fraud chain is detecting drug trafficking among suspects. However, procuratorial officials face significant challenges in this task, due to the complexity of multi-modal digital forensic data, diverse suspects’ communication styles, and intricate relationships among them. To address these issues, LISPER, a procuratorial service to detect drug trafficking is proposed for procuratorates, which fully leverages multiple large language models(LLMs). Firstly, LISPER extracts multi-source heterogeneous digital forensic data into standardized ones. Then, it accurately tags suspicious entities for individuals of suspects. Finally, LISPER incrementally maintains a relationship network for all the suspects, among which heterogeneous entities like suspects, drugs, and locations are contained. Successfully run in a district-level procuratorate of Beijing, LISPER has presented more than 10K judicial evidences from hundreds of detected suspects. A demo video of LISPER is available at https://youtu.be/24N3p3BYDlQ .