The experimental results show that Kernel PCA works almost as well as Latent Semantic Indexing (LSI) over Reuters21578 while the micro-F1 of Kernel PCA is 2% higher than that of LSI over Chinese 863 Evaluating corpus.

• 实验结果表明,核主成分分析在英文语料reuters21578上达到了潜在语义索引分类性能,而在中文863评测语料上微平均F1值比潜在语义索引高2%25。