SLQ: Lightweight-Detection & LLM-Generation SQL SelfOptimization Framework
Tianyou Zhu, Qi Yaru, Jiang Kongchen, Sang Yanting, Yang Chao · DOAJ (DOAJ: Directory of Open Access Journals) · 2026
In real-world database applications, SQL statements written by users often create performance bottlenecks because they violate best-practice rules. Traditional rule-based detectors have limited ability to recognize diverse and increasingly irregular statements and are costly to maintain. To address this, we propose SLQ, a two-stage intelligent SQL optimization framework. First, a lightweight stacked-LSTM module pinpoints problematic statements; then a pre-trained large language model, Qwen3, automatically generates explanations for each flaw and offers targeted rewrite suggestions, helping users quickly improve query quality. Evaluated on a standard dataset, SLQ achieves accuracy, precision, recall and F1 of 0.9841, 0.9974, 0.9702 and 0.9836 respectively, demonstrating superior detection and optimization capability and markedly enhancing SQL compliance and execution efficiency.