SIMMC 2.0
收藏资源简介:
SIMMC 2.0数据集由Facebook Reality Labs与Facebook AI联合创建,专注于沉浸式多模态对话,特别是在购物领域的应用。该数据集包含11,244条面向任务的用户-助手对话(共117,236条语句),这些对话基于沉浸式和照片级真实感的场景。数据集的创建过程采用两阶段方法:首先使用多模态对话模拟器生成对话流程,然后通过人工改写生成多样化的指代表达。SIMMC 2.0旨在解决真实世界中多模态输入处理和多模态动作执行的挑战,为研究任务导向对话提供了新的基准。
The SIMMC 2.0 dataset was jointly created by Facebook Reality Labs and Facebook AI, focusing on immersive multimodal conversations, particularly applications in the shopping domain. This dataset contains 11,244 task-oriented user-assistant dialogues, totaling 117,236 utterances, which are based on immersive and photorealistic scenarios. The dataset was developed using a two-stage approach: first, a multimodal dialogue simulator was used to generate dialogue flows, followed by manual paraphrasing to generate diverse referring expressions. SIMMC 2.0 aims to address the challenges of real-world multimodal input processing and multimodal action execution, providing a new benchmark for research on task-oriented dialogues.




