基于矩阵分解和同文正则化的甲骨文本聚类分析  

Clustering analysis of oracle bone inscription based on matrix factorization and identical oracle bone inscriptions regularization

作　　者：马园园 柳利芳[2] 涂克强 刘国英 刘永革[1,3] MA Yuanyuan;LIU Lifang;TU Keqiang;LIU Guoying;LIU Yongge(School of Computer Science and Information Engineering,Anyang Normal University,Anyang 455000,Henan,China;School of Education,Anyang Normal University,Anyang 455000,Henan,China;Key Laboratory of Oracle Bone Inscriptions Information Processing(Anyang Normal University)of Ministry of Education,Anyang 455000,Henan,China)

机构地区：[1]安阳师范学院计算机与信息工程学院,河南安阳455000 [2]安阳师范学院教育学院,河南安阳455000 [3]甲骨文信息处理教育部重点实验室(安阳师范学院),河南安阳455000

出　　处：《山东大学学报（工学版）》2021年第6期69-74,共6页Journal of Shandong University（Engineering Science）

基　　金：教育部人文社科一般项目(20YJC740042);甲骨文信息处理教育部重点实验室开放课题(OIP2019E003);国家自然科学基金资助项目(U1804153)。

摘　　要：针对甲骨卜辞数据,提出一种基于对称非负矩阵分解的无监督文本聚类方法,根据卜辞之间的同文关系构建卜辞邻接网络,并将其作为约束信息引入到目标函数中,使得语义上相关的卜辞能够在分解的低维子空间中也互相靠近。引入潜在的卜辞语义网络与利用相似性计算获得的卜辞相似性网络之间的差异,避免将后者直接用作对称非负矩阵分解输入的做法。该模型能够充分挖掘卜辞中潜在的语义信息。试验结果表明,该方法能够快速、有效地对卜辞文本进行主题聚类和辅助甲骨文考释研究。This paper proposed an unsupervised text clustering method based on symmetric nonnegative matrix factorization with identical oracle bone inscriptions(OBI)regularization,which was called oracle bone inscriptions clustering based on symmetric nonnegative matrix factorization(SNMFobi).It first constructed the affinity network of OBI according to the sematic information between OBIs,and then this network was introduced into the objective of SNMFobi so that the OBI with related semantic relationships was close to each other in the factorized low-dimensional space.This objective could tolerate the difference between the ground truth OBI semantic network and the similarity network computed by kernel method,and it avoided the inflexible practice that the computed similarity matrix was directly used as the input of symmetric non-negative matrix factorization in previous studies.This proposed model could fully excavate the potential semantic information in OBI.The experimental results showed that the proposed SNMFobi could effectively perform topic clustering dynamically in the OBI text analysis tasks and further provided important information for the explanation of OBI.

关 键 词：同文卜辞 对称非负矩阵分解 文本聚类 甲骨文 图正则化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]