TOOL-ED
收藏资源简介:
TOOL-ED数据集是由东北大学计算机科学与工程学院的研究团队基于EMPATHETICDIALOGUE(ED)数据集构建的,旨在通过工具调用能力增强大型语言模型(LLM)的共情响应生成。该数据集包含对话历史和工具使用轨迹,用于训练模型在生成共情响应时灵活整合外部知识。数据集的创建过程涉及将共情工具的使用实例插入到原始对话数据中,并通过LLM进行标注和筛选。TOOL-ED数据集主要应用于共情对话生成领域,旨在提高模型在理解用户情感和生成适当响应方面的能力。
The TOOL-ED dataset was constructed by the research team from the School of Computer Science and Engineering, Northeastern University, based on the EMPATHETICDIALOGUE (ED) dataset, with the objective of enhancing the empathetic response generation capabilities of large language models (LLMs) through tool calling abilities. This dataset comprises dialogue histories and tool usage trajectories, which are utilized to train models to flexibly integrate external knowledge when generating empathetic responses. The dataset construction process involves inserting instances of empathetic tool usage into the original dialogue data, followed by annotation and filtering via LLMs. The TOOL-ED dataset is primarily applied in the domain of empathetic dialogue generation, aiming to improve models' abilities to comprehend user emotions and generate appropriate responses.

- 1TOOL-ED: Enhancing Empathetic Response Generation with the Tool Calling Capability of LLM东北大学计算机科学与工程学院 · 2024年



