"FiQA (Financial Opinion Mining and Question Answering)"

"The overall benchmark was created to address the gap where general-purpose embedding models inadequately handle the nuanced requirements of the financial domain, such as specific terminology and complex numerical relationships. The benchmark covers: 7 Task Types: Classification, clustering, retrieval, pair classification, reranking, summarization, and semantic textual similarity (STS). Languages: English and Chinese. Key Findings: Domain-specific models (like Fin-E5) significantly outperform general-purpose models, and traditional Bag-of-Words (BoW) models surprisingly surpass dense embedding models on financial STS tasks."