基于改进近邻传播聚类挖掘算法的竞争情报研究

Improvement of affinity propagation clustering and its application in competitive intelligence

根据竞争情报分析需要,会产生不同竞争情报分析模型,这些分析模型的构造大多建立在竞争情报数据的聚类统计之上。提出采用改进的近邻传播(Affinity propagation,AP)聚类算法实现大规模竞争情报数据聚类统计。根据竞争情报数据样本建立相似矩阵,初始化偏向参数;通过布谷鸟搜索优化偏向参数,将偏向参数作为布谷鸟巢进行训练,设置轮廓指标值作为布谷鸟算法适应度函数;通过鸟巢位置更新优化后的偏向参数进行AP聚类运算,不断更新AP算法的决策和潜力阵;最终获得稳定的聚类结果。试验证明,通过合理设置布谷鸟宿主发现概率、移动步长和AP算法阻尼因子等参数,能够获得较好的聚类效果。相比常用竞争情报聚类算法,所提改进AP聚类算法能够获得更高的轮廓指标值和最短的欧式距离性能,在竞争情报数据分析统计中的适用度高。

According to the needs of competitive intelligence analysis,there are different competitive intelligence analysis models,most of which are based on the clustering statistics of competitive intelligence data.Therefore,an improved affinity propagation(AP)clustering algorithm is proposed to realize large-scale competitive intelligence data clustering statistics.Firstly,the similarity matrix is established according to competitive intelligence data samples,and the bias parameters are initialized.Then,the bias parameter is optimized by cuckoo search,and the bias parameter is used as cuckoo nest for training.The contour index value is set as the fitness function of cuckoo algorithm.The AP clustering operation is carried out by updating the optimized bias parameters of the nest position,and the decision and potential matrix of the AP algorithm are constantly updated.Finally,a stable clustering result is obtained.Experiments show that better clustering results can be obtained by reasonably setting the parameters of cuckoo host discovery probability,moving step size and damping factor of AP algorithm.Compared with the common competitive intelligence clustering algorithms,the improved AP clustering algorithm can obtain higher contour index values and the shortest Euclidean distance performance,and has high applicability in competitive intelligence data analysis and statistics.