QCRI/WASIL
收藏资源简介:
WASIL是一个基于LLM助手在真实世界中的阿拉伯语口语交互数据集。该数据集包含约9,000个轮次(具体为来自93个用户的9,304个轮次),涵盖多种阿拉伯语方言和国家,并包括用户对助手响应的明确反馈,如喜欢或不喜欢信号以及标量分数。每个交互都包含一个喜欢/不喜欢反应;对于不喜欢的响应,还进一步标注了一个或多个反馈类别,例如:未能遵循指令、缺乏事实准确性、对风格/格式不满、避免回答、不符合阿拉伯或伊斯兰文化、令人不安的内容、宗教不正确、语法错误、太简短或太长。数据集分为测试集(1,416个提示)和反馈集(988个提示),测试集包含额外的人工标注,如语言/方言标签、可回答性标签、ASR假设、后编辑转录和现代标准阿拉伯语翻译。
WASIL is a dataset of in-the-wild Arabic spoken interactions with an LLM-based assistant. The dataset contains ∼9K turns (9,304 turns from 93 users), spans multiple dialects and countries, and includes explicit user feedback on assistant responses, including like or dislike signals and scalar scores. Each interaction includes a like/dislike reaction. Disliked responses were further labeled with one or more feedback categories: failed to follow instructions, lacked factual accuracy, displeased with style/format, avoided answering, not aligned with Arabic or Islamic culture, disturbing content, religiously incorrect, grammatical errors, too brief, or too long.




