Research map / Large language models / Agents, code and math
BIRD-INTERACT: Re-imagining Text-to-SQL Evaluation via Lens of Dynamic Interactions
Nan Huo, Xiaohan Xu, Jinyang Li, Per Jacobsson, Shipei Lin, Bowen Qin, Binyuan Hui, Xiaolong Li, Ge Qu, Shuzheng Si, Linheng Han, Edward Alexander and 12 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add function-driven two-stage mapping to the user simulator | GPT-4o user simulator · Alignment with human users | Pearson correlation: 0.61 -> 0.84 | Improved |
| add function-driven two-stage mapping to the user simulator | Gemini-2.0-Flash user simulator · Alignment with human users | Pearson correlation: 0.54 -> 0.79 | Improved |
| add function-driven mapping to the user simulator | user simulator · UserSim-Guard unanswerable questions | failure rate: 67.4% -> 2.7% | Improved |
| add debugging phase in c-Interact | Claude-Sonnet-4 · BIRD-Interact-Full c-Interact priority | success rate (%): +6.67 | Improved |
| add debugging phase in c-Interact | GPT-5 · BIRD-Interact-Full c-Interact priority | success rate (%): +0.67 | No clear effect |
| give GPT-5 the ambiguity-resolution histories of Qwen-3-Coder or O3-Mini (memory grafting) | GPT-5 · BIRD-Interact c-Interact | success rate | Improved |
| switch from c-Interact to a-Interact mode | GPT-5 · BIRD-Interact-Full priority questions | success rate (%): 14.50 -> 29.17 | Improved |
| switch from c-Interact to a-Interact mode | Qwen-3-Coder-480B · BIRD-Interact-Full priority questions | success rate (%): 22.00 -> 13.33 | Hurt |
| switch from c-Interact to a-Interact mode | Claude-Sonnet-4 · BIRD-Interact-Full priority questions | success rate (%): 22.33 -> 27.83 | Improved |
| increase user patience (more interaction turns) | Claude-3.7-Sonnet · BIRD-Interact-Lite | performance | Improved |
| follow-up sub-tasks versus priority sub-tasks | GPT-5 a-Interact · BIRD-Interact-Full | success rate (%): 29.17 -> 17.00 | Hurt |
Closest papers by meaning
- Spider 2.0: Evaluating Language Models on Real-World Enterprise Text-to-SQL Workflows (ICLR 2025 · similarity 0.89)
- CogSQL: A Cognitive Framework for Enhancing Large Language Models in Text-to-SQL Translation (AAAI 2025 · similarity 0.88)
- Structure-Guided Large Language Models for Text-to-SQL Generation (ICML 2025 · similarity 0.87)
- PARROT: A Benchmark for Evaluating LLMs in Cross-System SQL Translation (NeurIPS 2025 · similarity 0.87)
- MARS-SQL: A Multi-Agent Reinforcement Learning Framework For Text-To-SQL (ICML 2026 · similarity 0.87)
- Alpha-SQL: Zero-Shot Text-to-SQL using Monte Carlo Tree Search (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
