文档相似矩阵在提高KNN分类效率中的应用

Application of Document Similarity Matrix in the Improvement of Classification Efficiency of KNN Classification Algorithm

针对传统KNN分类算法在样本数量大、维度高的情况下相似度计算量大的问题,提出了基于相似矩阵的改进KNN分类算法。该算法通过计算样本两两之间的相似度,建立相似矩阵加速KNN算法中寻找K近邻;以搜狗自然语言实验室的文本分类语料库中的新闻文档作为实验对象,采用宏平均F测度值作为分类效果评价标准,用改进KNN方法和传统KNN方法进行对比实验。实验结果表明:通过调节参数,本方法能在不损失精度的情况下减少寻找K近邻时相似度计算的次数。

In view of the fact that the traditional KNN classification algorithm has to calculate a large number of similarity problems when there is a large number of samples with high dimensions, this paper proposes an improved KNN classification algo- rithm based on the similarity matrix. The algorithm uses the similarity matrix established by calculating the similarity between 2 sam- pies to accelerate the search of K nearest neighbors in KNN algorithm; taking the news documents in the text classification corpus in Sogou Natural Language Laboratory as the experimental object, and using the macro-averaging F measuring value as the classifica- tion effect evaluation standard, the paper performs a comparative experiment with the improved KNN method and the traditional KNN method. The experimental results show that by adjusting the parameters, the proposed method can reduce the similarity calcu- lating times without the lass of accuracy when searching K nearest neighbors.