New Claim2Vec model uses multilingual embeddings to automatically cluster similar fact-check claims and combat recurring misinformation across languages.

Claim2Vec is the first multilingual embedding model designed specifically to represent fact-check claims as vectors for improved semantic understanding

The model uses contrastive learning with similar multilingual claim pairs to fine-tune a multilingual encoder

Testing on three datasets with 14 embedding models and 7 clustering algorithms showed Claim2Vec significantly improves claim clustering performance

Addresses the challenge of recurring claims in automated fact-checking systems, especially in multilingual settings where similar claims can be resolved with the same fact-check