MultiMind at SemEval-2025 Task 7: Crosslingual Fact-Checked Claim Retrieval via Multi-Source Alignment

TL;DR

TriAligner tackles multilingual and crosslingual fact-checked claim retrieval by fusing native and English representations through a dual-encoder architecture and contrastive learning. It augments and cleans data, uses hard negative sampling, and combines multiple embedding sources to compute robust cross-language similarity, with GPT-4o-based re-ranking to refine top candidates. Across monolingual and crosslingual benchmarks, the approach improves retrieval metrics and demonstrates the value of multi-source alignment for misinformation verification. The work advances multilingual information retrieval by integrating data augmentation, hard negatives, and LLM-based reranking to enhance cross-language evidence retrieval for fact-checking.

Abstract

This paper presents our system for SemEval-2025 Task 7: Multilingual and Crosslingual Fact-Checked Claim Retrieval. In an era where misinformation spreads rapidly, effective fact-checking is increasingly critical. We introduce TriAligner, a novel approach that leverages a dual-encoder architecture with contrastive learning and incorporates both native and English translations across different modalities. Our method effectively retrieves claims across multiple languages by learning the relative importance of different sources in alignment. To enhance robustness, we employ efficient data preprocessing and augmentation using large language models while incorporating hard negative sampling to improve representation learning. We evaluate our approach on monolingual and crosslingual benchmarks, demonstrating significant improvements in retrieval accuracy and fact-checking performance over baselines.

Figures (1)

• Figure 1: Our proposed system's pipeline includes data augmentation, preprocessing, hard negative sampling, the model's core network, score calculation, and re-ranking using a large language model to generate the final ranking for each social media post.