1,500 Chinese financial items from East Money scenes; OpenCompass scores nine of 19 files (650 items) with letter accuracy or keyword hit.
unassessed
| Category | domain |
|---|---|
| Subcategory | Chinese financial operations: multiple-choice, entity extraction, and rubriced analysis |
| Page status | active |
| Metric | accuracy |
| Direction | higher_is_better |
| Unit | % |
| Dataset size | 1500 |
| Dataset licence | Apache-2.0 |
| Publisher | East Money (东方财富) and Shanghai Artificial Intelligence Laboratory |
OpenFinData tests whether a model can handle Chinese text from East Money (东方财富) production scenes: market tables, announcements, customer intents, named institutions, and simple indicator arithmetic. Most scored items are a short instruction plus a passage or table, then a lettered choice. One scored file asks the model to list entities. Other files in the zip ask for stock, fund, sector, or announcement write-ups against weighted criteria, or pose compliance questions whose gold field is the string "nan". The suite is Chinese text, not English filing QA and not a licensing exam.
OpenCompass runs nine files as zero-shot generation (ZeroRetriever). Eight are 3-, 4-, or 5-option multiple choice scored by AccEvaluator after last_capital_postprocess. entity_recognition is free-text scored by OpenFinDataKWEvaluator (every gold entity, split on the enumeration comma, must appear in the output). Analysis and interpretation files store criteriumN rubrics and are not in the OpenCompass config.
No model card in ModelSpec reports this benchmark yet.