HFL at SemEval-2022 Task 8: A Linguistics-inspired Regression Model with Data Augmentation for Multilingual News Similarity

TL;DR

This paper presents a multilingual news similarity system using a linguistics-inspired regression model with data augmentation, achieving top performance in SemEval-2022 Task 8.

Contribution

It introduces a novel combination of data augmentation, multi-label loss, adapted R-Drop, and sample reconstruction techniques for multilingual news similarity.

Findings

Ranked 1st on the leaderboard

Achieved Pearson's Correlation of 0.818

Demonstrated effectiveness of linguistics-inspired methods

Abstract

This paper describes our system designed for SemEval-2022 Task 8: Multilingual News Article Similarity. We proposed a linguistics-inspired model trained with a few task-specific strategies. The main techniques of our system are: 1) data augmentation, 2) multi-label loss, 3) adapted R-Drop, 4) samples reconstruction with the head-tail combination. We also present a brief analysis of some negative methods like two-tower architecture. Our system ranked 1st on the leaderboard while achieving a Pearson's Correlation Coefficient of 0.818 on the official evaluation set.