LOLGORITHM: Funny Comment Generation Agent For Short Videos

ACL ARR 2026 January Submission4552 Authors

Keywords: short video comment generation, multimodal language model, style-controlled generation, multi-agent system, bilingual dataset, humor modeling

Abstract: This paper presents LOLGORITHM, a modular multi-agent system for stylized comment generation in short-form videos. The system employs video segmentation, emotion extraction, and stylized prompt construction to support six distinct comment styles: puns, rhymes, memes, sarcasm, humor, and content extraction. We construct a bilingual dataset covering both TikTok and YouTube platforms, encompassing five categories of popular videos. Experimental results demonstrate that LOLGORITHM outperforms existing methods in both automated metrics and human preference evaluations, achieving user preference rates exceeding 90% on TikTok and 87.55% on YouTube. This research provides a scalable framework for enhancing interactivity and cultural dissemination on short-video platforms.

Paper Type: Long

Research Area: AI/LLM Agents

Research Area Keywords: multimodal generation, comment generation, short video, social media NLP, style-controlled generation, multi-agent systems, cross-platform NLP, bilingual dataset

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Publicly available software and/or pre-trained models, Data resources, Data analysis

Languages Studied: Chinese, English

Submission Number: 4552