ICASSP2023 MUG Challenge Track5: Action Item Detection (AID)

Who this competition fits

This suits NLP practitioners who can build sentence-level classifiers for Mandarin meeting transcripts and fine-tune publicly available pre-trained language models. It is also suitable for academic or industry researchers, except challenge organizers, Alibaba employees, and Zhejiang University faculty or students, who are explicitly ineligible.

From registration to a first submission

1. 01

   Mandarin NLP

2. 02

   meeting transcript processing

3. 03

   sentence classification

4. 04

   pre-trained language model fine-tuning

5. 05

   long-document modeling

6. 06

   precision–recall evaluation

Before you commit: The core difficulty is recognizing actionable-task sentences in long, spoken meeting transcripts containing disfluencies, redundancies, grammar errors, coreference, colloquial language, and fragments. Participants must also work within a constrained-track setup: evaluation data cannot be used for training or fine-tuning, and only specified public models and corpora are allowed.