基于M^3-DGMF的专利数据聚类方法研究  

STUDY ON PATENT DATA SETS CLUSTERING BASED ON M^3-DGMF

作　　者：田冬阳[1] 

机构地区：[1]中国人民银行乌鲁木齐中心支行,新疆乌鲁木齐830002

出　　处：《计算机应用与软件》2013年第3期297-303,共7页Computer Applications and Software

摘　　要：朴素贝叶斯、KNN、决策树及支持向量机等方法已经应用到文本聚类中,然而专利文献数据有着大规模、不平衡、层次化、多标号的特点,导致传统的方法在解决此类问题中往往性能不佳。鉴于此,通过引进基于双高斯合成函数的最小最大模块化神经网络的监督聚类算法,来有效地修剪训练数据的规模,并将其成功应用到专利数据的聚类中,实验结果证明了并行化及监督聚类的有效性。Methods of Naive Bayesian classification,KNN,decision tree and support vector machine,etc.have been applied to text clustering.However,the data set of patent literatures has the characteristics of big size,unbalanced,hierarchical and multi-label,which lead to traditional classification methods performing poor in resolving the problems like these.In view of this,in the thesis we introduce the supervised clustering algorithm of min-max modularised neural network based on dual-Gaussian synthesis function to effectively prune the size of training data,and successfully apply it to patent data clustering.Experimental results indicate the validity of the parallelism and supervised clustering.

关 键 词：监督聚类 双高斯合成函数 专利分类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]