Scholarlink

<p>The increasing complexity of user communication needs, coupled with advancements in Internet technology, has led to the integration of images, emoticons, and memes into web-based chats. This study addresses the challenge of automating meme selection for online communication by investigating the semantic and emotional dimensions of memes within open-domain dialogues. We propose MeCho, a multi-modal fusion framework built on an enhanced multi-choice approach and prompt tuning. MeCho incorporates two primary tasks–Context-Meme Selection and Meme Semantic Prediction–alongside two auxiliary tasks–Masked Context Prediction and Meme Sentiment Classification. Together, these tasks facilitate the extraction of dialogue features, the understanding of dialogue history, emotion analysis, and meme semantics. Utilizing the authoritative MOD dataset (Meme incorporated Open-domain Dialogue), which encompasses a wide range of emotional and semantic categories as the evaluation benchmark, our results demonstrate that MeCho outperforms previous state-of-the-art models, such as CLIP and MMBERT, by 2% in recall, while achieving nearly 75 times faster inference speed. In particular, the multi-choice formulation accelerates inference by jointly evaluating multiple candidates in a single forward pass, thereby avoiding one-by-one matching. Ablation experiments further validate the effectiveness of the individual components of the MeCho model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Meme selection via multiple-choice with masked language models in multimodal dialogue

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Bo Liu,
  • Xiaojun Wu,
  • Jianhao Wang,
  • Bowen Zhang,
  • Li Zhu,
  • Weiping Ding

摘要

The increasing complexity of user communication needs, coupled with advancements in Internet technology, has led to the integration of images, emoticons, and memes into web-based chats. This study addresses the challenge of automating meme selection for online communication by investigating the semantic and emotional dimensions of memes within open-domain dialogues. We propose MeCho, a multi-modal fusion framework built on an enhanced multi-choice approach and prompt tuning. MeCho incorporates two primary tasks–Context-Meme Selection and Meme Semantic Prediction–alongside two auxiliary tasks–Masked Context Prediction and Meme Sentiment Classification. Together, these tasks facilitate the extraction of dialogue features, the understanding of dialogue history, emotion analysis, and meme semantics. Utilizing the authoritative MOD dataset (Meme incorporated Open-domain Dialogue), which encompasses a wide range of emotional and semantic categories as the evaluation benchmark, our results demonstrate that MeCho outperforms previous state-of-the-art models, such as CLIP and MMBERT, by 2% in recall, while achieving nearly 75 times faster inference speed. In particular, the multi-choice formulation accelerates inference by jointly evaluating multiple candidates in a single forward pass, thereby avoiding one-by-one matching. Ablation experiments further validate the effectiveness of the individual components of the MeCho model.