摘要： 分类是知识图谱构建中的一个重要问题,但是目前多数中文百科都采用人工编辑的方式为词条添加分类,耗费人力并且存在漏标和标错等问题。为此,提出一种自动识别百度百科人物领域下全部实体并添加分类的方法。对百度百科词条已有的分类、属性和副标题进行实体集拓展,使用马尔科夫逻辑网络方法联合推断词条的分类。实验结果表明,与支持向量机和逻辑回归算法相比,该方法在实体识别的精确度和召回率方面性能均有所提升。

关键词: 实体分类, 实体集拓展, 马尔科夫逻辑网络, 知识图谱, 机器学习, 联合推断

Abstract: Classification is an important issue in constructing knowledge graph.However,the existing categories are edited by human beings for Chinese Baike websites,leading to increasing manpower cost and lost or error of categories.Aiming at this problem,this paper proposes a method to recognize all entities of Baidu Baike.under people domain automatically and add categories for these entities.Entity Set Expansion(ESE) is used for existing raw data of like categories,attributes and subtitles,and Markov Logic Network(MLN) is used to jointly inference the category of entities.Experimental result shows that,compared with Support Vector Machine(SVM) and Logistic Regression(LR) algorithm,the proposed method performs much better both in precision and recall aspects.

Key words: entity classification, Entity Set Expansion(ESE), Markov Logic Network(MLN), knowledge graph, machine learning, joint inference

中图分类号: 

• TP18