Who vouched for the work, in their own words.

Recognised as arxiv:2402.03216 · M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity…

The name people use is not in the paper’s own title — it lives in other people’s bibliographies. Both lanes run at once; the aliases win here.

16 positive · 88 neutral · 2 negative — the negative ones are returned too

“More recently, powerful cross-encoder rerankers, such as BGE-reranker-v2-m3 (BGE) (Chen et al., 2024), have significantly improved retrieval accuracy.”

James AllanACM Fellow · University of Massachusetts Amherstendorse

“Strong results are achieved with bge-Gemma2 using layer-normalized average embeddings, our best-performing model.”

Ming-Yu LiuIEEE Fellow · NVIDIA Cosmos Labbuild on

“Local Search Engine: We adopt BGE-M3 (Chen et al., 2024) for hybrid retrieval, which produces both dense and learned sparse embeddings in a single forward pass.”

陈为 · Wei ChenCCF Fellow · Zhejiang Universitybuild on

The work has been cited 106 times by 55 academicians and society fellows, 16 of those citations carrying an explicit endorsement. ACM Fellow James Allan describes it as a powerful cross-encoder reranker that “significantly improved retrieval accuracy”; IEEE Fellow Ming-Yu Liu reports it as their best-performing model; CCF Fellow Wei Chen adopts it as the retrieval engine of their own system.

Each sentence keeps its English original and the paper it came from. search 163 ms · read 27 ms · draft 4.3 s.