Skip to content

Latest commit

 

History

History
55 lines (41 loc) · 2.64 KB

DSTC9_SIMMC_RESULTS.md

File metadata and controls

55 lines (41 loc) · 2.64 KB

SIMMC Track Results (DSTC9), 2020

TL;DR

The first edition of the Situated and Interactive Multimodal Conversations (SIMMC) Track at the Dialog State Tracking Challenge (DSTC) 9 came to a successful end! The challenge saw a total of 11 model entries from 5 teams across the world, setting a new state-of-the-art in all three subtasks:

  1. Subtask 1 (Assistant API Prediction)
    • Action accuracy increased by 3 points (79.3% to 82.5%)
    • Action Attribute accuracy increased by ~10 points (63.7% to 73.9%)
  2. Subtask 2 (Assistant Response Generation)
    • BLEU score increased by 0.067 points (0.061 to 0.128)
    • Recall@1 increased by 45 points (7.2% to 52.6%)
  3. Subtask 3 (Dialog State Tracking)
    • Slot F-1 increased by 16.7 points (62.4% to 79.1%)
    • Intent F-1 increased by 16 points (62.1% to 78.1%)

Congratulations to the winners and runners-up, and a big thanks to all the participants!

Details

We launched the SIMMC challenge in June, 2020. Please checkout the main page, paper, and the Facebook AI blog for more details.

Please checkout the challenge page, paper, and the Facebook AI blog for further details about the challenge, dataset, tasks, and the evaluation metrics.

Results Summary

DSTC9 SIMMC Results Summary

Summary on Test-Std split, average of Furniture and Fashion.
Team Affiliation Github
1 National Taiwan University, Taiwan Link
2 Sogang University, South Korea Link
3 Sogang University, South Korea Link
4 Institute for Infocomm Research, A-STAR, Singapore Link
5 LINKS Foundation and Politecnico di Torino, Italy Link

Detailed Results

All the detailed results are here.