LiViBench: An Omnimodal Benchmark for Interactive Livestream Video Understanding
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (31) · 26517-26525
DOI: 10.1609/aaai.v40i31.39859 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Xiaodong Wang und 6 weiteren Autor:innen.
Abstract (Originalsprache)
The development of multimodal large language models (MLLMs) has advanced general video understanding. However, existing video evaluation benchmarks primarily focus on non-interactive videos, such as movies and recordings. To fill this gap, this paper proposes the first omnimodal benchmark for interactive livestream videos, LiViBench. It features a diverse set of 24 tasks, highlighting the perceptual, reasoning, and livestream-specific challenges. To efficiently construct the dataset, we design a standardized semi-automatic annotation workflow that incorporates the human-in-the-loop at multiple stages. The workflow leverages multiple MLLMs to form a multi-agent system for comprehensive video description and uses a seed-question-driven method to construct high-quality annotations. All interactive videos in the benchmark include audio, speech, and real-time comments modalities. To enhance models' understanding of interactive videos, we design tailored two-stage instruction-tuning and propose a Video-to-Comment Retrieval (VCR) module to improve the model's ability to utilize real-time comments. Based on these advancements, we develop LiVi-LLM-7B, an MLLM with enhanced knowledge of interactive livestreams. Experiments show that our model outperforms larger open-source models with up to 72B parameters, narrows the gap with leading proprietary models on LiViBench, and achieves enhanced performance on general video benchmarks, including VideoMME, LongVideoBench, MLVU, and VideoEval-Pro.
Zitieren
APA
Wang, X., Huang, L., Wu, Z., Zhao, X., Xu, T., Xia, X., & Peng, P. (2026). LiViBench: An Omnimodal Benchmark for Interactive Livestream Video Understanding. Proceedings of the AAAI Conference on Artificial Intelligence, 40(31), 26517-26525. https://doi.org/10.1609/aaai.v40i31.39859
Vancouver
Wang X, Huang L, Wu Z, Zhao X, Xu T, Xia X, et al. LiViBench: An Omnimodal Benchmark for Interactive Livestream Video Understanding. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(31):26517-26525. doi:10.1609/aaai.v40i31.39859
BibTeX
@article{Wang2026,
author = {Wang, Xiaodong and Huang, Langling and Wu, Zhirong and Zhao, Xu and Xu, Teng and Xia, Xuhong and Peng, Peixi},
title = {LiViBench: An Omnimodal Benchmark for Interactive Livestream Video Understanding},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {31},
pages = {26517--26525},
doi = {10.1609/aaai.v40i31.39859}
}
EUVIMED