SNS-Bench-VL
收藏资源简介:
SNS-Bench-VL是一个全面的多模态基准,旨在评估视觉语言大型语言模型(LLM)在社交网络服务(SNS)场景中的表现。该数据集包含4001个精心策划的多模态问答对,涵盖单选、多选和开放式任务,跨越8个多模态任务,包括笔记理解、用户参与分析、信息检索和个性化推荐。数据集来自一个拥有超过30亿用户的社交平台,确保了真实反映现实世界社交媒体互动。数据收集过程包括数据收集和处理、不同SNS相关任务的仔细注释以及跨模态质量控制和人机验证。SNS-Bench-VL旨在推动下一代社交网络服务中的鲁棒、情境感知和人性化的多模态智能研究。
SNS-Bench-VL is a comprehensive multimodal benchmark aimed at evaluating the performance of vision-language large language models (LLMs) in social network service (SNS) scenarios. This dataset contains 4001 carefully curated multimodal question-answer pairs, covering single-choice, multiple-choice, and open-ended tasks, spanning 8 multimodal tasks including note comprehension, user engagement analysis, information retrieval, and personalized recommendation. The dataset is sourced from a social platform with over 3 billion users, ensuring it authentically reflects real-world social media interactions. The data collection process includes data collection and processing, careful annotation of various SNS-related tasks, cross-modal quality control, and human-machine verification. SNS-Bench-VL seeks to advance research on robust, context-aware, and human-like multimodal intelligence in next-generation social network services.

- 1SNS-Bench-VL: Benchmarking Multimodal Large Language Models in Social Networking Services北京航空航天大学, 小红书有限公司, 复旦大学, 牛津大学 · 2025年



