Mixed Hierarchical Networks for Deep Entity Matching

作     者：Chen-Chen Sun De-Rong Shen Chen-Chen Sun;De-Rong Shen

作者机构：Engineering Research Center of Learning-Based Intelligent System(Ministry of Education)Tianjin University of TechnologyTianjin 300384China School of Computer Science and EngineeringTianjin University of TechnologyTianjin 300384China School of Computer Science and EngineeringNortheastern UniversityShenyang 110819China 

出 版 物：《Journal of Computer Science & Technology》 (计算机科学技术学报（英文版）)

年 卷 期：2021年第36卷第4期

页      面：822-838页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 081202[工学-计算机软件与理论] 

基　　金：the National Natural Science Foundation of China under Grant Nos.62002262,61672142,61602103,62072086 and 62072084 the National Key Research and Development Project of China under Grant No.2018YFB1003404 

主　　题：entity matching attention mechanism mixed hierarchical neural network(MHN) domain adaption data integration 

摘      要：Entity matching is a fundamental problem of data *** groups records according to underlying real-world *** is a growing trend of entity matching via deep learning *** design mixed hierarchical deep neural networks(MHN)for entity matching,exploiting semantics from different abstract levels in the record internal hierarchy.A family of attention mechanisms is utilized in different periods of entity ***-attention focuses on internal dependency,inter-attention targets at alignments,and multi-perspective weight attention is devoted to importance ***,hybrid soft token alignment is proposed to address corrupted *** order is for the first time considered in deep entity ***,to reduce utilization of labeled training data,we propose an adversarial domain adaption approach(DA-MHN)to transfer matching knowledge between different entity matching tasks by maximizing classifier ***,we conduct comprehensive experimental evaluations on 10 datasets(seven for MHN and three for DA-MHN),which illustrate our two proposed approaches1 *** apparently outperforms previous studies in accuracy,and also each component of MHN is ***-MHN greatly surpasses existing studies in transferability.