GoEmotions

GoEmotions是由斯坦福语言学、谷歌研究院和亚马逊Alexa合作创建的大型人工标注数据集，包含58,009条英文Reddit评论，标注了27种情感类别或中性。该数据集旨在通过精细的情感分类，提升机器对语言中情感表达的理解，适用于构建同理心聊天机器人和检测有害在线行为等下游任务。数据集的创建过程包括从流行的英文Reddit子论坛中精心挑选评论，并设计了一个考虑心理学文献和数据覆盖的情感分类法。GoEmotions的应用领域广泛，包括客户反馈分析和聊天机器人增强等，旨在解决机器对细微情感表达理解的不足。

GoEmotions is a large-scale manually annotated dataset co-created by Stanford Linguistics, Google Research, and Amazon Alexa. It contains 58,009 English Reddit comments annotated with 27 emotion categories plus the neutral class. This dataset aims to enhance machines' comprehension of emotional expressions in language through fine-grained emotion classification, and is applicable to downstream tasks such as building empathetic chatbots and detecting harmful online behaviors. The dataset's creation process involves carefully curating comments from popular English Reddit subreddits, and designing an emotion taxonomy that considers psychological literature and data coverage. GoEmotions has a wide range of application areas, including customer feedback analysis and chatbot enhancement, and aims to address the deficiency of machines in understanding subtle emotional expressions.

搜集汇总

背景与挑战

背景概述

GoEmotions是一个细粒度情感标注数据集，包含58,009条从Reddit提取的评论，标注了27种情感类别或'Neutral'。数据经过人工标注和筛选，基于标注者一致性分为训练、测试和验证集，适用于情感分析任务，并提供了基线模型和数据分析工具，但需注意数据集可能存在的偏见和局限性。