Reproduce “Meeting Action Item Detection with Regularized Context Modeling” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiaqing Liu, Chong Deng, Qinglin Zhang, Qian Chen, Wen Wang
Researchers at Alibaba Group's DAMO Academy developed the AliMeeting-Action Corpus (AMC-A), the first large-scale Chinese dataset for action item detection (AID). They also proposed Context-Drop, a regularization technique that improved AID by +2.98 F1 on AMC-A by enhancing context modeling, and a Lightweight Model Ensemble method for efficient knowledge integration from multiple pre-trained language models.
Key insights
- Action item detection benefits significantly from modeling both local and global context, as relevant information is often distributed non-contiguously within a conversation.
- Regularizing models to ensure consistency between context-rich and context-free predictions (Context-Drop) effectively helps exploit relevant context and reduces distraction from irrelevant information, improving robustness.
- Complementary "knowledge" from different pre-trained models can be efficiently combined at the architectural level (e.g., encoder from one, pooler from another) to improve performance without traditional ensemble's latency cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.