基于测试分类精度的多分类器融合方法

Combing Multi-classifier Based on Their Accuracy on Test Corpus

本文分析了影响分类器精度的因素,并提出了三种基于在测试例集上分类表现效果的多分类器融合方法.这三种方法的基本思想是:当使用多个分类器对未标注文本进行分类时,最终输出在测试例集上表现最好的那个分类器的结果.实验结果表明,这三种融合方法从一定程度上提高了分类器精度.

This paper analyzes the factors affecting the accuracy of classitier, and designs three methods combining multi-classifier based on their performance on test corpus. The main idea of our designation is that we output the result of the classifier which have higher F1, precision, or recall when we use multi-classifier to classify unlabeled text. Our experimental results show that our combining method can improve the effect of classifier in some degrid.