Multi-Modal Dialogue Dataset
收藏资源简介:
本研究介绍了名为‘Multi-Modal Dialogue Dataset’的数据集,由韩国科学技术院创建,包含45,000条多轮对话,涉及图像与文本的交互。数据集通过预处理文本对话数据,采用文本到图像替换技术,并结合上下文相似性过滤,确保数据集的上下文连贯性。该数据集旨在为多模态对话系统提供训练资源,特别是在需要理解图像和文本的上下文感知方式方面。
This study presents a dataset named 'Multi-Modal Dialogue Dataset', developed by the Korea Advanced Institute of Science and Technology (KAIST). The dataset encompasses 45,000 multi-turn dialogues involving interactions between images and text. To ensure its contextual coherence, the dataset applies preprocessing to textual dialogue data, adopts text-to-image replacement techniques, and incorporates context similarity filtering. This dataset aims to provide training resources for multimodal dialogue systems, particularly in scenarios requiring context-aware understanding of both images and text.

- 1Constructing Multi-Modal Dialogue Dataset by Replacing Text with Semantically Relevant Images韩国科学技术院 · 2021年



