Flex-TravelPlanner
收藏资源简介:
Flex-TravelPlanner是一个用于评估语言模型在动态、多轮规划场景中灵活推理能力的基准数据集。该数据集基于TravelPlanner数据集,重点关注对增量变化和优先级约束的灵活推理。数据集包含120个多轮场景和134个具有冲突软硬约束的场景。数据集的构建过程涉及对TravelPlanner数据集的约束添加和修订,以及对软硬约束的引入。该数据集旨在解决实际规划问题中不断变化的需求和竞争约束的平衡问题,为LLM的动态规划能力提供评估框架。
Flex-TravelPlanner is a benchmark dataset developed to assess the flexible reasoning abilities of language models within dynamic, multi-turn planning scenarios. Built upon the original TravelPlanner dataset, this benchmark focuses on flexible reasoning regarding incremental changes and priority constraints. It comprises 120 multi-turn scenarios and 134 scenarios involving conflicting hard and soft constraints. The dataset construction process entails adding and revising constraints on the base TravelPlanner dataset, as well as introducing hard and soft constraint types. This work targets resolving the trade-off between evolving demands and competing constraints in real-world planning tasks, thereby providing an evaluation framework for the dynamic planning capabilities of Large Language Models (LLMs).

- 1Flex-TravelPlanner: A Benchmark for Flexible Planning with Language Agents韩国科学技术院 · 2025年



