基于互信息和众数的类间关系研究

Mining inter-class relationship based on mutual information and mode pattern

簇间关系的评估对于确定许多现实中的关键未知信息具有重要作用,可被广泛应用在犯罪侦查、进化树、冶金工业和生物嫁接等领域。提出了一种称为"众数模式+互信息"的方法对簇间关系进行排序。该方法使用众数模式从每个簇中寻找具有代表性的对象,使用互信息衡量簇间的关联程度。由于该方法利用簇间关系判断簇与簇的联系程度,所以它不同于传统的分类和聚类。在图形和癌症诊断数据上的实验表明了该算法的有效性。

The evaluation of the relationships between conceptual clusters is important to identify vi- tal unknown information in many real-life applications, such as crime detection, evolution trees, metal- lurgical industry, biology engraftment and so forth. A method called ＂mode pattern plus mutual infor- mation＂ is proposed to rank the inter-relationship between clusters. Mode pattern is used to find the out- standing objects from each cluster, while mutual information criterion measures the close proximity of cluster pairs. The proposed method is different from conventional algorithms of classifying and cluste- ring due to focusing on ranking the inter-relationship between clusters. Experiments are carried out on a wide range of real-life datasets, including image data and cancer diagnosis data. The experimental results demonstrate the effectiveness of the proposed algorithm.