Explainable Argumentation Computation Benchmark
收藏资源简介:
Explainable Argumentation Computation Benchmark是由浙江大学团队开发的一个用于训练和评估大语言模型(LLMs)在抽象论证框架(AAF)中计算扩展能力的基准数据集。该数据集包含60,000个训练样本和2,000个测试样本,涵盖了从6到25个论证的多种难度级别的抽象论证框架。每个样本包括指令、问题、解释和答案四个部分,旨在通过详细的算法解释来增强模型的可解释性和泛化能力。数据集的创建过程包括随机生成论证框架、去除重复样本并计算真实语义标签。该数据集主要应用于法律决策、政策合规分析和伦理决策等需要可废止推理的实际场景,旨在提升LLMs在论证计算中的准确性和透明度。
Explainable Argumentation Computation Benchmark is a benchmark dataset developed by the Zhejiang University team for training and evaluating Large Language Models (LLMs) on their capacity to compute extensions in Abstract Argumentation Frameworks (AAF). This dataset includes 60,000 training samples and 2,000 test samples, covering Abstract Argumentation Frameworks of varying difficulty levels with 6 to 25 arguments each. Each sample consists of four parts: instruction, question, explanation, and answer, aiming to enhance the model's interpretability and generalization ability through detailed algorithmic explanations. The dataset creation process involves randomly generating argumentation frameworks, removing duplicate samples, and calculating ground-truth semantic labels. This benchmark is primarily applied in real-world scenarios requiring defeasible reasoning, such as legal decision-making, policy compliance analysis and ethical decision-making, with the goal of improving the accuracy and transparency of LLMs in argumentation computation.




