ICASSP2023 MUG Challenge Track4: Keyphrase Extraction (KPE)

Who this competition fits

This suits researchers or engineers who can build keyphrase extraction systems for Mandarin meeting transcripts, especially those familiar with long-document NLP and fine-tuning pretrained models. It welcomes academia and industry participants, but Alibaba employees and Zhejiang University faculty/students cannot participate.

From registration to a first submission

1. 01

   Mandarin text processing

2. 02

   keyphrase extraction

3. 03

   long-document modeling

4. 04

   Transformer fine-tuning

5. 05

   spoken-language transcript handling

6. 06

   ModelScope pre-trained models

Before you commit: The main challenge is extracting salient phrases from long, conversational meeting transcripts rather than clean written text. Disfluencies, redundancies, grammar errors, coreference, and fragmented utterances compound the computational difficulty of processing documents with several thousand words or more.