Question-Matching-on-Qianyan

• 2024.10.1 - 10.18: Machine Learning Practice (2024 Fall), for Baidu"Qianyan Dataset: Question Matching Robustness" Competition.
• 2024.10.30: NLP (2024 Fall) Course Lab, adding some new exploration of this project (including dataset, model, training, etc).

Get Started

Repo Stucture

Github Repo Root
 |---- corrector  # Contains code for text correction
 |---- data      # Contains datasets
 |---- result    # Contains results and logs
 |---- train     # Contains training scripts and configurations
 |---- misc     # New things from NLP course lab

Introduction

This is the demo implementation (which only includes the core code but not an excutable version, since the competition hasn't ended) for the technical report "Qianyan Dataset: Technical Report on the Question Matching Robustness Competition", which ranked 6th (until 2024.10.18) in the "Baidu Qianyan: Question Matching Robustness Competition". This is also the assignment 1 of Machine Learning Practice (2024 fall).

Performance

Untill 2024.10.18 0:00, ranked 6th in the leader board.

Rank	score	OPPO	DuQM_pos	DuQM_named_entity	DuQM_synonym	DuQM_antonym	DuQM_negation	DuQM_temporal	DuQM_symmetry	DuQM_asymmetry	DuQM_neg_asymmetry	DuQM_voice	DuQM_misspelling	DuQM_discourse_particle(simple)	DuQM_discourse_particle(complex)
1	90.845	88.895	76.943	95.441	89.889	99.672	95.726	84.188	94.747	83.3	87.755	85.496	97.65	96.714	95.42
2	90.331	89.155	73.568	94.779	88.376	99.344	94.302	79.487	94.371	82.093	91.837	87.023	98.932	96.714	94.656
3	90.31	89.48	77.966	97.206	90.287	99.672	96.011	87.607	92.308	82.294	65.306	90.84	99.359	99.061	96.947
4	90.208	89.23	78.315	96.912	89.729	99.344	94.587	86.752	93.058	82.495	65.306	92.366	99.573	99.061	96.183
5	90.132	89.09	79.752	96.471	86.943	98.361	94.587	94.444	92.308	84.708	67.347	85.496	98.504	97.653	96.183
6	89.901	89.07	80.318	96.912	87.978	99.672	93.732	91.453	91.932	86.72	65.306	81.679	99.359	99.061	95.42

Reference

Qianyan: Question Matching Robustness Competiton Competition Baseline
2021 CCF BDCI Thousand Words - Problem Matching Robustness Review - 12th Place Scheme
2021 CCF BDCI Thousand Words - Problem Matching Robustness Review - 5th Place Scheme
2021 CCF BDCI Thousand Words - Problem Matching Robustness Review - 1st Place Scheme

Acknowledgements

We would like to thank the organizers of the Qianyan Question Matching Robustness Competition for providing the dataset and the opportunity to participate in this competition. Also, authors of the blogs in the Reference helped me so much (part of the code in this repo is developed based on their code or idea). Thanks a lot!

License

This project is licensed under the MIT License.