基于SVM的Web文本快速增量分类算法

Fast incremental learning SVM for Web text classification

针对基于支持向量机的Web文本分类效率低的问题,提出了一种基于支持向量机Web文本的快速增量分类FVI-SVM算法。算法保留增量训练集中违反KKT条件的Web文本特征向量,克服了Web文本训练集规模巨大,造成支持向量机训练效率低的缺点。算法通过计算支持向量的共享最近邻相似度,去除冗余支持向量,克服了在增量学习过程中不断加入相似文本特征向量而导致增量学习的训练时间消耗加大、分类效率下降的问题。实验结果表明,该方法在保证分类精度的前提下,有效提高了支持向量机的训练效率和分类效率。

In Web text classification,with extremely large scale of the training set and the characteristic of changing rapidly,this paper proposed an algorithm named FVI-SVM based on incremental SVM for fast Web text classification.In order to conquer the problem of low efficiency of SVM which was aroused by a large scale of training set,datas in incremental training set which violate conditions of KKT would be exterminated.In order to conquer the problem of redundant support vectors which lead to the increasing of taining time consumption and decreasing of classification efficiency in incremental learning,exterminated the redundant support vectors by calculating shared nearest neighbors similarity.Experimental results show that the proposed method enhances the training and classification efficiency on a premise ensure the accuracy of classification.