基于CPSO的多目标文本分类投影寻踪  

Multi-objective Projection Pursuit for Text Categorization Based on CPSO

作　　者：石松[1] 陈云[1] 

机构地区：[1]上海财经大学上海市金融信息技术研究重点实验室,上海200433

出　　处：《计算机工程》2014年第2期171-174,共4页Computer Engineering

基　　金：上海市科学技术委员会基金资助项目(10dz1123500;10dz1123200;11ZR1411800);上海市自然科学基金资助项目(11ZR 1411800);上海财经大学研究生创新基金资助项目(CXJJ-2012-322)

摘　　要：投影寻踪可有效解决文本分类中的维数灾难问题,而投影方向优化是投影寻踪需要解决的关键问题。传统的投影寻踪方法将投影指标优化看作单目标优化问题,会使解的质量受到影响。为此,提出一种基于多目标优化的投影寻踪方法。将类别之间的距离和类别内数据的聚类紧密程度作为2个优化目标,并将投影扩展到多维,利用混沌粒子群优化算法寻找最优的投影方向。在常用文本数据集上进行实验,确定最优投影指标及维度,并比较不同分类模型的分类结果,结果表明,使用该方法能有效提高文本分类性能。Projection pursuit method is increasingly used in text categorization to solve the curse of dimensionality. Traditional projection pursuit method considers the projection index optimization as a single-objective problem rather than a multi-objective one, which will reduce the quality of the solution. To solve this problem, this paper proposes a projection pursuit mehod based on multi-objective optimization. Measures are taken like class difference and difference between the classes as two objectives of pursuit index, the projection pursuit method is extended to multi-dimensional projections, and a Chaotic Particle Swarm Optimization（CPSO） is suggested to find the optimal projection direction. Experiment on commonly used text datasets determines the optimal projection direction and dimensions, and then compares the results of different classification models. The results demonstrate that the proposed method can improve the text categorization performance effectively.

关 键 词：投影寻踪 文本分类 维数灾难 投影指标 多目标优化 混沌粒子群优化算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]