Abstract:

The features of text were selected by pattern clustering and GA(Genetic Algorithm) and were classified by Kohonen network. The dimensions of text could be reduced greatly by using pattern clustering from thousands to hundreds, then reduced to tens by using GA. The experiment results indicate that combining these two methods can greatly reduce the dimension of text and improve the precision of text classification.

Key words: feature selection, pattern clustering, genetic algorithm, Kohonen network

摘要：

采用模式聚类和遗传算法进行文本特征提取，并用Kohonen网络进行分类。模式聚类可以有效降低文本特征的维数，使得特征从几千维降为几百维。但几百维的维数对Kohonen网络来说仍然太高，因此采用遗传算法在此基础上继续降维。实验结果表明，这两种方法结合可以极大地降低文本的维数，并能提高分类准确率。

关键词: 特征提取, 模式聚类, 遗传算法, Kohonen网络