Proj AutoWriter Paper Reading: SEQ2SQL: GENERATING STRUCTURED QUERIES FROM NATURAL LANGUAGE USING REINFORCEMENT LEARNING

Abstract

1. 介绍Relational DB
   本文Seq2SQL
   方法: policy-based reinforcement learning
   Task:...
2. 使用in-the-loop query执行来获取rewards
3. 利用SQL的结构来剪枝空间
   本文: 数据集: WikiSQL
   规模80654个人工标注的问题，24241个表
   实验:
4. 比最好的semantic parser更好
5. improving execution accuracy from 35.9% to 59.4%
6. logical form accuracy from 23.4% to 48.3%.