BanglaMemeEvidence
收藏资源简介:
BanglaMemeEvidence是由阿赫萨努拉科技大学团队构建的一个多模态基准数据集,专门针对孟加拉语模因的解释性证据检测任务。该数据集包含2,917个孟加拉语模因实例,每个实例均附带人工标注的自然语言解释,涵盖模因OCR文本、上下文描述和证据句子,并配有相关性评分以反映模因与标注之间的关联强度。数据收集过程广泛覆盖了社交媒体平台和在线论坛,涉及政治、体育、娱乐、教育、技术等多个主题类别,并通过严谨的注释流程确保数据质量。该数据集旨在支持低资源语言环境下的模因分析研究,特别是通过多模态融合技术解决模因内容理解、有害信息检测和语境推断等关键挑战,推动孟加拉语自然语言处理与计算机视觉的交叉应用发展。
BanglaMemeEvidence is a multimodal benchmark dataset developed by the team at Ahsanullah University of Science and Technology, specifically tailored for the explanatory evidence detection task targeting Bengali memes. This dataset contains 2,917 Bengali meme instances, each accompanied by manually annotated natural language explanations, which cover meme OCR text, contextual descriptions, and evidence sentences, alongside relevance scores to reflect the association strength between the meme and its corresponding annotation. The data collection process spans a wide range of social media platforms and online forums, involving multiple thematic categories such as politics, sports, entertainment, education, and technology, while ensuring data quality through a rigorous annotation workflow. This dataset is designed to support meme analysis research in low-resource language environments, particularly addressing core challenges including meme content understanding, harmful information detection, and context inference via multimodal fusion technologies, and advancing the development of cross-disciplinary applications between Bengali natural language processing and computer vision.

- 1BanglaMemeEvidence: A Multimodal Benchmark Dataset for Explanatory Evidence Detection in Bengali Memes阿赫萨努拉科技大学 · 2026年



