Improving Few-Shot Text-to-SQL with Meta Self-Training via Column Specificity

Authors: Xinnan Guo, Yongrui Chen, Guilin Qi, Tianxing Wu, Hao Xu

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental results on both open-domain and domain-specific benchmarks reveal that our MST-SQL1 has significant advantages in few-shot scenarios, and is also competitive in standard supervised settings.
Researcher Affiliation Collaboration Xinnan Guo1 , Yongrui Chen1 , Guilin Qi1 , Tianxing Wu1 and Hao Xu2 1School of Computer Science and Engineering, Southeast University, Nanjing, China 2Zhejiang Lab, Zhejiang, China
Pseudocode Yes Algorithm 1 details our meta self-training framework F, used to find the optimal parameters θ by mining the potential knowledge of unlabeled data in the following two stages. Our CSML is detailed in Algorithm 2.
Open Source Code Yes Code available at https://github.com/ygxw0909/MST-SQL Contact Author
Open Datasets Yes We evaluated our MST-SQL on the two benchmarks: 1) Wiki SQL [Zhong et al., 2017] is currently the largest open-domain single-table text-to-SQL dataset, which contains more than 20k tables collected from Wikipedia and 56,355 / 8,421 / 15,878 NLQ-SQL pairs for training / development / test. 2) ESQL [Chen et al., 2021] is a Chinese domain-specific singletable text-to-SQL dataset containing 17 tables and 10,000 / 1,000 / 2,000 NLQ-SQL pairs for training / development / test.
Dataset Splits Yes We evaluated our MST-SQL on the two benchmarks: 1) Wiki SQL [Zhong et al., 2017] is currently the largest open-domain single-table text-to-SQL dataset, which contains more than 20k tables collected from Wikipedia and 56,355 / 8,421 / 15,878 NLQ-SQL pairs for training / development / test. 2) ESQL [Chen et al., 2021] is a Chinese domain-specific singletable text-to-SQL dataset containing 17 tables and 10,000 / 1,000 / 2,000 NLQ-SQL pairs for training / development / test.
Hardware Specification Yes Our method ran on Tesla V100 Super GPUs.
Software Dependencies No The paper mentions using "Ro BERTa as the encoder by default" but does not specify any software dependencies with version numbers (e.g., Python, PyTorch, or specific library versions).
Experiment Setup Yes The hyper-parameters were set as follow: (1) The number of table content was set to 5. (2) The learning rate α, β, γ were set to 3e-5, 1e-5, 3e-5. (3) The weight of the loss calculation in CSML η was set to 0.5. (4) The threshold for confidence scores ζ was set to 0.1. (5) N was set to 4, and K was set to 15 / 5 for support / query set. (6) ML-task number n, sampling ratio σ, and warm-boot threshold λ were set to 80, 0.2, 80 in standard supervised setting, and the setting in few-shot tests will be discussed later.