融入混沌与对立学习机制的二进制粒子群特征选择算法

A FEATURE SELECTION ALGORITHM OF BINARY PARTICLE SWARM OPTIMIZATION BASED ON CHAOS AND OPPOSITION-LEARNING MECHANISM

为了实现特征空间降维,提高文本聚类准确性,提出一种融入混沌与对立学习的二进制粒子群优化特征选择算法。设计了新的词条权重计算方法,将文本数据表达为矢量空间模型;提出改进二进制粒子群算法求解特征选择问题,引入混沌系统和对立学习机制对粒子随机搜索方向和初始种群分布分别进行优化;在评估粒子适应度中引入词条方差和平均中位数两种方法对特征子集评估,并设计特征合并和交叉机制融合两种适应度的优势,生成最优特征子集;利用K均值算法对特征选择的文本进行聚类。结果表明,该算法在特征降维、聚类准确率、F度量值上均优于同类算法,可以有效实现特征空间降维并提升文本聚类性能。

In order to achieve the feature space dimension reduction and improve the accuracy of the text clustering,a feature selection algorithm of binary particle swarm optimization based on chaos and opposite-learning is proposed.We designed a new term weight calculation method and expressed the text datasets as the vector space model.We presented an improved binary particle swarm optimization to solve the problem of feature selection,and chaotic system and opposition-learning mechanism were introduced to optimize the random search direction and initial population distribution of particles respectively.In the evaluation of particle fitness,the term variance and mean median were introduced to evaluate feature subset,and we designed characteristics of the merger and crossover mechanism combined with the advantage of two kinds of fitness to generate the optimal feature subset.The k-means algorithm was used to carried on the text clustering analysis for feature selection.The results show that our algorithm performs better than the similar algorithms on the feature dimension reduction,clustering accuracy and F measurements,which can effectively achieve feature space dimensionality reduction and promote text clustering performance.