Core Concepts
SimuCourt introduces a benchmark for evaluating judicial decision-making agents and proposes a multi-agent framework, AgentsCourt, to simulate court processes.
Abstract
Introduction: Discusses the impact of deep learning on the legal domain and the need for cross-stage collaboration in judicial decision-making.
SimuCourt: Introduces the benchmark encompassing 420 real-world judgment documents and the AgentsCourt framework.
Data Extraction: Includes key metrics on the performance improvement of the framework in legal grounds generation.
Experiments: Details the evaluation metrics, human assessment, and comparison with baselines.
Discussion: Analyzes the limitations, judicial knowledge of LLMs, and the impact of multi-agent court simulation.
Conclusion: Summarizes the contributions of SimuCourt and AgentsCourt.
Stats
대부분의 노력이 개별 사법 단계에만 집중하고 교차 단계 협력을 간과한다.
모델이 첫 번째 및 두 번째 인스턴스 설정에서 8.6% 및 9.1% F1 점수 향상을 달성했다.
법적 근거 생성에서 프레임워크의 성능 향상.
법적 근거 평가에서 AgentCourt의 성능 평가.
Quotes
"The court holds that the accused, John Doe, has repeatedly stolen citizens' property, ... , and should be severely punished."
"Our framework outperforms the existing advanced methods in various aspects, especially in generating legal grounds."