Predicting Humor in Visual and Language Modalities

• Item type: SIG Technical Reports(1)
• 公開日: 2021-05-13
• タイトル: Predicting Humor in Visual and Language Modalities
• タイトル / 言語: en
• タイトル: Predicting Humor in Visual and Language Modalities
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 画像と言語・音響
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Osaka University
• 著者所属: Osaka University
• 著者所属: Osaka University
• 著者所属(英): en
• 著者所属(英): Osaka University
• 著者所属(英): en
• 著者所属(英): Osaka University
• 著者所属(英): en
• 著者所属(英): Osaka University
• 著者名: Zekun, Yang; Yuta, Nakashima; Haruo, Takemura
• 著者名(英): Zekun, Yang; Yuta, Nakashima; Haruo, Takemura
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Finding humor in videos is an interesting but challenging task because humor can be induced by various signals in the visual, linguistic, and vocal modalities emitted by human. Previous methods mainly predict humor in the sentence level and with single modality, which often ignore humor caused by e.g., actions. In this work, we propose a multi-modal humor prediction method to find temporal segments that involve humor in videos. Our method adopts a sliding window to divide the video and uses the visual modality described by pose and facial features, along with the linguistic modality given as subtitles for humor prediction. We use long short term memory (LSTM) networks to model poses and faces and use pre-trained BERT to model the subtitles. Experimental results with the dataset based on a sitcom TV drama series show that our method helps improve the performance of humor prediction.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Finding humor in videos is an interesting but challenging task because humor can be induced by various signals in the visual, linguistic, and vocal modalities emitted by human. Previous methods mainly predict humor in the sentence level and with single modality, which often ignore humor caused by e.g., actions. In this work, we propose a multi-modal humor prediction method to find temporal segments that involve humor in videos. Our method adopts a sliding window to divide the video and uses the visual modality described by pose and facial features, along with the linguistic modality given as subtitles for humor prediction. We use long short term memory (LSTM) networks to model poses and faces and use pre-trained BERT to model the subtitles. Experimental results with the dataset based on a sitcom TV drama series show that our method helps improve the performance of humor prediction.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11131797
• 書誌情報: 研究報告コンピュータビジョンとイメージメディア（CVIM）; 巻 2021-CVIM-226, 号 7, p. 1-6, 発行日 2021-05-13
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8701
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会