SHERLOCK: Towards Dynamic Knowledge Adaptation in LLM-enhanced E-commerce Risk Management
Researchers have developed SHERLOCK, a new framework designed to enhance e-commerce risk management by integrating Large Language Models (LLMs) with structured domain knowledge. The system addresses the limitations of LLMs in handling complex fraud patterns and sparse domain knowledge through a three-module approach. It constructs a knowledge base, employs a specialized retrieval-augmented generation strategy, and includes a self-evolving platform for continuous improvement. Initial tests at JD.com showed significant improvements in investigation efficiency and the system's ability to adapt to evolving adversarial tactics. AI
IMPACT Introduces a novel framework for dynamic knowledge adaptation in LLMs, potentially improving AI applications in specialized domains like fraud detection.