VLDB 2021 Crowd Science Challenge Workshop

Introduction

Source code of Noisy text sequences aggregation as a summarization subtask

This approach is ranked in the 2nd place in the official run.

Blind Test Result

Ranking	F1-Score
1	95.729869(1)
2-Our Approach	95.659318 (2)
3	95.476175 (3)

Quick Start

Simple demo for our training and testing procedures.

demo.ipynb

We provide all training and testing data in data folder.

Details

preprocess

run

submission

Evaluation

Reproduction

Our best model can be downloaded here https://drive.google.com/file/d/1mh69ZTA2hzXVjRN5JYdeAy72x9xJ_exB/view?usp=sharing

Reference

1. VLDB 2021 Workshop
2. Stacked DeBert
3. DSTC 2/3
4. example code from huggingface

How do I cite this work?

soon