Alexa TaskBot Dataset

该数据集包含了在Alexa Prize TaskBot挑战赛中收集的真实人类与智能助手对话以及评分，专注于开发在烹饪和DIY领域的对话式任务助手（CTA）。值得注意的是，只有约10%的用户提供了评分，这使得数据分析面临较大的挑战。平均而言，每个对话包含8到9个来回，对话长度存在较大差异。该数据集的规模为1681个对话，按照90/10/10的比例分为训练集、验证集和测试集。任务目标是根据用户对话预测评分。

This dataset contains real human-assistant conversations and corresponding ratings collected from the Alexa Prize TaskBot Challenge, which focuses on developing conversational task assistants (CTA) in the cooking and DIY domains. Notably, only approximately 10% of users provided ratings, posing significant challenges for data analysis. On average, each conversation consists of 8 to 9 conversational turns, with substantial variations in conversation length. The dataset comprises 1,681 total conversations, and is split into training, validation, and test sets at a ratio of 90/10/10. The task objective is to predict ratings based on user conversations.