somosnlp-hackathon-2025/gastronomia-hispana-dpo

该数据集名为“Gastronomía Hispana DPO”，包含关于西班牙和拉丁美洲美食领域的对话数据，用于训练专注于该领域语言模型的直接偏好优化（DPO）技术。数据集由约470个偏好对组成，涵盖471个独特的食谱，所有内容均为西班牙语。数据集的结构包括选定的对话、被拒绝的对话、食谱ID、食谱名称和内容类别。类别包括成分、烹饪技术、基本食谱和文化背景。数据集适用于烹饪聊天机器人、食谱推荐模型、烹饪教学系统和饮食领域NLP研究。

The dataset named Gastronomía Hispana DPO contains conversational data on the Spanish and Latin American culinary field, intended for training language models specializing in this domain using Direct Preference Optimization (DPO) technique. The dataset consists of approximately 470 preference pairs, covering 471 unique recipes, all in Spanish. The structure of the dataset includes chosen conversations, rejected conversations, recipe ID, recipe name, and content categories. Categories include ingredients, cooking techniques, basic recipes, and cultural context. The dataset is applicable for culinary chatbots, recipe recommendation models, cooking tutoring systems, and NLP research in the culinary domain.