基于后验概率制导的B-KNN文本分类方法

B-KNN Text Categorization Method Based on Posterior Probability Guidance

针对K最近邻(KNN)方法分类准确率高但分类效率较低的特点,提出基于后验概率制导的贝叶斯K最近邻(B-KNN)方法。利用测试文本的后验概率信息对训练集多路静态搜索树进行剪枝,在被压缩的候选类型空间内查找样本的K个最近邻,从而在保证分类准确率的同时提高KNN方法的效率。实验结果表明,与KNN相比,B-KNN的性能有较大提升,更适用于具有较深层次类型空间的文本分类应用。

Considering K Nearest Neighbor（KNN） method has high accuracy but poor efficiency,this paper proposes a text categorization method based on the guidance of posterior probability named B-KNN.By using the posterior probabilities collected from the training text,B-KNN prunes the multi-branch-static-searching tree of the training dataset and reduces the candidate class set where K nearest neighbors can be found so that the efficiency of KNN method can be improved while preserving its classification accuracy.Experimental results show that B-KNN method remarkably outperforms KNN method,and it is more suitable for classification tasks with deep hierarchy categorization space.