NL2SQL-BERT

Content Enhanced BERT-based Text-to-SQL Generation https://arxiv.org/abs/1910.07179

Incorporating database design rule into text-to-sql generation:

1. We use the matching information of the table cells and question string to construct a vector where its length is the same to the question length. This question vector mainly improves the performance of WHERE-VALUE inference result. Because it injects the knowledge that the answer cell and its corresponding table header are bound together. If we locate the answer cell then we locate the answer column which contains the answer cell.
2. We use the matching information of all the table headers and the question string to construct a vector where its length is the same to the table headers’ length. This header vector mainly improves the performance of WHERE-COLUMN inference result.

Requirements

python 3.6

records 0.5.3

torch 1.1.0

Run

1, Data prepare: Download all origin data( https://drive.google.com/file/d/1iJvsf38f16el58H4NPINQ7uzal5-V4v4 or https://download.csdn.net/download/guotong1988/13008037) and put them at data_and_model directory.

Then run data_and_model/output_entity.py

2, Train and eval:

train.py

Results on BERT-Base-Uncased without Execution-Guided-Decoding

Data

One data look:

{
	"table_id": "1-1000181-1",
	"phase": 1,
	"question": "Tell me what the notes are for South Australia ",
	"question_tok": ["Tell", "me", "what", "the", "notes", "are", "for", "South", "Australia"],
	"sql": {
		"sel": 5,
		"conds": [
			[3, 0, "SOUTH AUSTRALIA"]
		],
		"agg": 0
	},
	"query": {
		"sel": 5,
		"conds": [
			[3, 0, "SOUTH AUSTRALIA"]
		],
		"agg": 0
	},
	"wvi_corenlp": [
		[7, 8]
	],
	"bertindex_knowledge": [0, 0, 0, 0, 4, 0, 0, 1, 3],
	"header_knowledge": [2, 0, 0, 2, 0, 1]
}

The Table:

{
	"id": "1-1000181-1",
	"header": ["State/territory", "Text/background colour", "Format", "Current slogan", "Current series", "Notes"],
	"rows": [
		["Australian Capital Territory", "blue/white", "Yaa·nna", "ACT · CELEBRATION OF A CENTURY 2013", "YIL·00A", "Slogan screenprinted on plate"],
		["New South Wales", "black/yellow", "aa·nn·aa", "NEW SOUTH WALES", "BX·99·HI", "No slogan on current series"],
		["New South Wales", "black/white", "aaa·nna", "NSW", "CPX·12A", "Optional white slimline series"],
		["Northern Territory", "ochre/white", "Ca·nn·aa", "NT · OUTBACK AUSTRALIA", "CB·06·ZZ", "New series began in June 2011"],
		["Queensland", "maroon/white", "nnn·aaa", "QUEENSLAND · SUNSHINE STATE", "999·TLG", "Slogan embossed on plate"],
		["South Australia", "black/white", "Snnn·aaa", "SOUTH AUSTRALIA", "S000·AZD", "No slogan on current series"],
		["Victoria", "blue/white", "aaa·nnn", "VICTORIA - THE PLACE TO BE", "ZZZ·562", "Current series will be exhausted this year"]
	]
}

Trained model

https://drive.google.com/open?id=18MBm9qzobTBgWPZlpA2EErCQtsMhlTN2

Reference

https://github.com/naver/sqlova