基于马尔科夫逻辑网络的实体解析改进算法  

Improvement of Entity Resolution Based on Markov Logic Networks

作　　者：楼俊杰[1] 徐从富[1] 郝春亮[1] 

机构地区：[1]浙江大学人工智能研究所,杭州310027

出　　处：《计算机科学》2010年第8期243-247,共5页Computer Science

基　　金：国家自然科学基金(No.60970081);国家863计划专题课题(No.2007AA01Z197);"十一五"国防预研项目资助

摘　　要：实体解析(Entity Resolution,ER)是数据挖掘过程中关键而又费时的一个步骤。华盛顿大学的Domingos和Singla提出了基于马尔科夫逻辑网络(Markov Logic Networks,MLNs)的ER算法。基于此算法,在原有的MLNs体系中,引入了一个可变权重的规则,试图解决原有系统无法处理的实体二义性问题。实验证明,新算法能够有效缓解数据记录的二义性问题,并且在一定程度上提高了原始算法的精度。Entity Resolution is a crucial and expensive step in the data mining process. Domingos and Singla of University of Washington proposed of well-founded, integrated solution to the entity resolution problem based on Markov Logic. This paper tried to improve Domingos and Singla＇s solution by adding a formula with a changeable weight to it, to handle the problem of ambiguity of entities that the original system cannot distinguish. The new algorithm can effectively handle ambiguity of entities, and improve accuracy compared with the original algorithm, which is proved by experiments.

关 键 词：ER MLNs 可变权重 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]