计算机工程
計算機工程
계산궤공정
COMPUTER ENGINEERING
2014年
11期
172-177
,共6页
粒子群优化%自调节惯性权重机制%进化程度%云变异算子%k-means算法%文本聚类
粒子群優化%自調節慣性權重機製%進化程度%雲變異算子%k-means算法%文本聚類
입자군우화%자조절관성권중궤제%진화정도%운변이산자%k-means산법%문본취류
Particle Swarm Optimization ( PSO )%self-regulating mechanism of inertia weight%degree of evolution%cloud mutation operator%k-means algorithm%text clustering
针对k-means算法的聚类结果高度依赖初始聚类中心选取的问题，提出一种基于改进粒子群优化的文本聚类算法。分析粒子群算法和k-means算法的特点，针对粒子群算法搜索精度不高、易陷入局部最优且早熟收敛的缺点，设计自调节惯性权重机制及云变异算子以改进粒子群算法。自调节惯性权重机制根据种群进化程度，动态地调节惯性权重，云变异算子基于云模型的随机性和稳定性，采用全局最优值实现粒子的变异。该算法结合了粒子群算法较强的全局搜索能力与k-means算法较强的局部搜索能力。每个粒子是一组聚类中心，类内离散度之和的倒数是适应度函数。实验结果表明，该算法是一种精确而又稳定的文本聚类算法。
針對k-means算法的聚類結果高度依賴初始聚類中心選取的問題，提齣一種基于改進粒子群優化的文本聚類算法。分析粒子群算法和k-means算法的特點，針對粒子群算法搜索精度不高、易陷入跼部最優且早熟收斂的缺點，設計自調節慣性權重機製及雲變異算子以改進粒子群算法。自調節慣性權重機製根據種群進化程度，動態地調節慣性權重，雲變異算子基于雲模型的隨機性和穩定性，採用全跼最優值實現粒子的變異。該算法結閤瞭粒子群算法較彊的全跼搜索能力與k-means算法較彊的跼部搜索能力。每箇粒子是一組聚類中心，類內離散度之和的倒數是適應度函數。實驗結果錶明，該算法是一種精確而又穩定的文本聚類算法。
침대k-means산법적취류결과고도의뢰초시취류중심선취적문제，제출일충기우개진입자군우화적문본취류산법。분석입자군산법화k-means산법적특점，침대입자군산법수색정도불고、역함입국부최우차조숙수렴적결점，설계자조절관성권중궤제급운변이산자이개진입자군산법。자조절관성권중궤제근거충군진화정도，동태지조절관성권중，운변이산자기우운모형적수궤성화은정성，채용전국최우치실현입자적변이。해산법결합료입자군산법교강적전국수색능력여k-means산법교강적국부수색능력。매개입자시일조취류중심，류내리산도지화적도수시괄응도함수。실험결과표명，해산법시일충정학이우은정적문본취류산법。
Clustering result of k-means clustering algorithm is highly dependent on the choice of the initial cluster center. With regards to this, a text clustering algorithm based on improved Particle Swarm Optimization ( PSO ) is presented. Features of particle swarm algorithm and k-means algorithm are analysed. Considering the disadvantages of PSO including low solving precisions, high possibilities of being trapped in local optimization and premature convergence,self-regulating mechanism of inertia weight and cloud mutation operator are designed to improve PSO. Self-regulating mechanism of inertia weight adjusts the inertia weight dynamically according to the degree of the population evolution. Cloud mutation operator is based on stable tendency and randomness property of cloud model. The global best individual is used to complete mutation on particles. Those two algorithms are combined by taking advantages of power global search ability of PSO and strong capacity of local search of k-means. A particle is a group of clustering centers,and a sum of scatter within class is fitness function. Experimental results show that this algorithm is an accurate,efficient and stable text clustering algorithm.
