深圳大学学报（理工版）2025，Vol.42Issue(2)：242-248,7.DOI:10.3724/SP.J.1249.2025.02242
基于DDQN的否定约束规则迁移
A DDQN framework for denial constraints transfer
摘要
Abstract
Traditional data cleaning methods rely on experts to manually define data quality rules,making the process complex and time-consuming.Moreover,the cleaned data may not be reusable,reducing the efficiency and quality of data cleaning.To address this issue,we propose a double deep Q-network for denial constraints transfer(DDQN-DCT)algorithm.This algorithm designs a similarity measurement method for denial constraints(DCs)and,by considering the simplicity and coverage of DCs,modifies the predicates in DC rules using double deep Q-network(DDQN)to achieve DC transfer.The goal is to make the transferred rules as similar as possible to the original rules,thereby retaining the information of original rules.Building upon DDQN-DCT,we further design an enhanced algorithm,DDQN-DCT+,which separates the action selection strategy of DDQN into two stages:addition and deletion.Comparative experiments demonstrate that DDQN-DCT+performs better in terms of DC simplicity.In comparison with algorithms of brute-force dependency constraint transfer(BFDC),DDQN-DCT+,and structure expansion/reduction(SER),the DDQN-DCT algorithm achieves an improvement of average rule similarity by approximately 10%over BFDC,10.6%over DDQN-DCT+,and 16.4%over SER.The experiments show that DDQN-DCT can effectively transfer rules from the source domain to similar target domain data.关键词
计算机技术/规则迁移/否定约束/相似度度量/强化学习/数据清洗Key words
computer technology/rule transfer/denial constraints/similarity metric/reinforcement learning/data cleaning分类
信息技术与安全科学引用本文复制引用
秦建斌,杜玉琪,林毅斌..基于DDQN的否定约束规则迁移[J].深圳大学学报（理工版）,2025,42(2):242-248,7.基金项目
National Key R&D Program of China(2021YFB3301500) 国家重点研发计划资助项目(2021YFB3301500) （2021YFB3301500）
