基于竞争神经网络的全局K-Medoids聚类算法研究

Research on Global K-Medoids Clustering Algorithm Based on Competed Neural Network

针对全局K-Medoids算法在处理大规模数据聚类分析时搜索效率低的问题,提出了基于竞争神经网络的全局K-Medoids聚类算法。由于神经网络算法对输入模式要求为数值向量,不适合处理文本序列数据的聚类问题,通过定义文本序列数据在聚类分析时的属性描述方式,利用竞争神经网络对数据进行初始分类,在此基础上运行全局K-Medoids算法进行详细的分类,使算法适合于处理文本序列数据聚类问题。文章分别利用UCI数据库中的8组实验数据和机械加工企业工艺数据中的工艺路线数据进行算法验证,结果证明该方法的效率和精度均高于K-Medoids算法和全局K-Medoids算法。

Aiming at the problem of low research efficiency of the global K-Medoids algorithm in clustering analysis dealing with large scale data, the global K-Medoids algorithm based on competed neural network was presented. But the neural network algorithm being inappropriate for dealing with clustering analysis of text sequence data because of numerical vector as the input mode of neural network, by defining the method of attribute description, and this algorithm is appropriate for clustering analysis of text sequence data when facing on text sequence data clustering problem, it used competed neural to cluster initially, then ran the global K-Medoids to cluster further based on the results of competed neural network. The results proved the efficiency and accuracy of this algorithm for clustering was higher than K-Medoids algorithm and global K-Medoids algorithm by testing of utilizing 8 sets of data of UCI dataset and the process data of a mechanical enterprise.