基于和声搜索的近邻传播算法

An affinity propagation algorithm based on harmony search

针对近邻传播算法中偏向参数调优难的问题,提出了一种基于和声搜索的近邻传播算法(HS-AP),利用和声搜索自动为数据集匹配最佳偏向参数,进而提高算法聚类精度。HS-AP算法首先把偏向参数编码为和声,利用和声算法自动搜索最佳和声,并将搜索到的和声解码为偏向参数进行运算。在UCI标准数据集上进行实验对比表明HS-AP算法在准确率,兰德指数,正则化互信息三个指标方面均有提升。准确率平均提升了6.36%,兰德指数平均提升了4.677%,正则化互信息平均提升了19.04%。

In order to reduce the difficulty of adjusting preference parameters in affinity propagation algorithms,an affinity propagation algorithm based on harmony search(HS-AP)is proposed.The algorithm uses harmony search to automatically match the best preference parameter to improve the clustering accuracy.The HS-AP algorithm encodes the preference parameters into harmony,uses the harmony algorithm to automatically search for the best harmony,and uses the searched harmony decoding to preference parameters.Control experiments on UCI standard data sets show that the HS-AP algorithm has improved in accuracy,Rand index and normalized mutual information.Among them,the accuracy increased by 6.36%,the Rand index increased by 4.677%,and the normalized mutual information increased by 19.04%.