Passau-Spontaneous Football Coach Humour (Passau-SFCH)
收藏资源简介:
Passau-SFCH数据集是由奥格斯堡大学的研究团队创建的,专注于自发幽默的多模态预测。该数据集包含约11小时的录音,主要来源于德国足球甲级联赛教练的非脚本化新闻发布会。数据集被标注为幽默及其维度(情感和方向),并根据Martin的幽默风格问卷进行标注。该数据集旨在解决现有幽默检测方法基于脚本数据,不适用于‘现实世界’应用的问题。研究通过使用预训练的Transformer、卷积神经网络和专家设计的特征进行了一系列实验,分析了文本、音频和视频在自发幽默识别中的性能,并探讨了它们的互补性。研究发现,面部表情在自动幽默分析中最为有前景,而幽默方向最好通过基于文本的特征来建模。数据集的应用领域包括人机交互的自然化以及人工智能的人性化。
The Passau-SFCH dataset, created by a research team from the University of Augsburg, focuses on multimodal prediction of spontaneous humor. This dataset contains approximately 11 hours of audio recordings, primarily sourced from unscripted press conferences held by coaches of the German Bundesliga. The dataset is annotated for humor and its dimensions (affect and direction), and additionally labeled using Martin's Humor Style Questionnaire. This dataset aims to address the limitation that existing humor detection methods rely on scripted data and thus are not suitable for 'real-world' applications. A series of experiments were conducted using pre-trained Transformers, Convolutional Neural Networks (CNNs), and expert-designed features, which analyzed the performance of text, audio, and video modalities in spontaneous humor recognition and explored their complementarity. The study found that facial expressions are the most promising modality for automated humor analysis, while humor direction is best modeled using text-based features. The application scenarios of this dataset include the naturalization of human-computer interaction and the humanization of artificial intelligence.

- 1Towards Multimodal Prediction of Spontaneous Humour: A Novel Dataset and First Results奥格斯堡大学 · 2023年



