FamilyTool
收藏资源简介:
FamilyTool是一个基于家庭知识图谱构建的多跳个性化工具使用基准数据集。该数据集由复旦大学计算机科学系和复旦大学现代语言学研究所创建,包含了多个家庭关系的样本,用于模拟真实世界中的个性化多跳工具使用场景。数据集涵盖了1到3个关系跳的查询,难度不同,可以展示不同的复杂程度。FamilyTool还包含了用于评估大型语言模型工具使用能力的KGETool评估流程。
FamilyTool is a benchmark dataset for multi-hop personalized tool use constructed based on household knowledge graphs. This dataset was developed by the Department of Computer Science and the Institute of Modern Linguistics of Fudan University. It contains multiple samples of family relationships, designed to simulate real-world personalized multi-hop tool use scenarios. The dataset covers queries with 1 to 3 relation hops, varying in difficulty to demonstrate different levels of complexity. FamilyTool also includes the KGETool evaluation pipeline for assessing the tool use capabilities of large language models.

- 1FamilyTool: A Multi-hop Personalized Tool Use Benchmark复旦大学计算机科学系,复旦大学现代语言学研究所 · 2025年



