PRODIGy

PRODIGy是由布鲁诺凯斯勒基金会创建的一个基于个人资料的对话生成数据集，包含20850条对话，旨在通过整合标准和复杂的个人资料表示，如交流风格、传记和个性，来提高对话代理的一致性和连贯性。数据集从康奈尔电影对话语料库中提取，通过与外部资源对齐，进一步丰富了角色资料。PRODIGy适用于测试多种基于生成语言模型的基准，特别是在域内和跨域设置中，通过自动和人工评估，显示了基于个人资料的模型在生成连贯和信息丰富的响应方面的优势。

PRODIGy is a profile-based conversational generation dataset created by the Bruno Kessler Foundation, which comprises 20,850 dialogues. It aims to enhance the consistency and coherence of conversational agents by integrating both standard and complex profile representations, including communication styles, biographies, and personality traits. The dataset is derived from the Cornell Movie-Dialogs Corpus, and its character profiles are further enriched through alignment with external resources. PRODIGy is applicable for benchmarking a wide range of generative language models, particularly in in-domain and cross-domain settings. Both automatic and human evaluations have demonstrated the advantages of profile-based models in generating coherent and informative responses.