jjjycaptain/BIRD-Spider-Ent

Citation text from the card

@inproceedings{ji-etal-2026-text,
    title = "Text-to-{SQL} Benchmarks Toward Enterprise Realities: Under Massive Scopes, Complex Schemas and Scattered Knowledge",
    author = "Ji, Yuchen  and
      Yu, Chengxi  and
      Shi, Jie  and
      Zhou, Weixin  and
      Cao, Xi  and
      Liang, Jiaqing  and
      Chen, Hai  and
      Mao, Yu  and
      Yang, Deqing  and
      Xiao, Yanghua",
    editor = "Lee, Hung-Yi  and
      Dev, Sunipa  and
      Li, Jessy  and
      Elliott, Desmond",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
    month = oct,
    year = "2026",
    address = "Budapest, Hungary",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-emnlp.1971/",
    pages = "42028--42060",
    ISBN = "979-8-89176-462-0",
    abstract = "Existing Text-to-SQL benchmarks remain idealized compared with enterprise scenarios, which require massive query scopes, complex schemas, and scattered external knowledge. We present BIRD-Ent and Spider-Ent, two enterprise-oriented benchmarks constructed via a refinement framework, together with Dual-Retrieval-Augmented-Generation (DRAG) Text-to-SQL, which formalizes table-schema and knowledge retrieval before SQL generation. The benchmarks feature over 4,000 columns per query and, for BIRD-Ent, documents totaling 1.5M tokens. Evaluation on state-of-the-art models shows sharp drops, with only 39.1{\%} EX on BIRD-Ent and 60.5{\%} 

All datasets marked Permissive or share-alike | JSON