摘要： 讨论了用传统性能评估指标(精度和错误率)评估文本分类器时面临的困难,分析了目前几种常用的文本分类器性能评估指标:查 全率-查准率曲线、AUROC、F1值以及BEP值的优点与不足,并提出了两种新的可用于文本分类器的性能评估指标。

关键词: 文本分类, 性能评估, BEP值, ROC曲线

Abstract: Afterdiscussing on the deficiency of accuracy and error rate as performance metrics for text categorization problems, this paper analyzes the strength and shortcoming of recall-precision curve, area under the ROC curve, F1value and break-even point, which are the most prevailing performance metrics for text classifiers, and proposes two kinds of new performance metric for text classifier.

中图分类号: 

• TP391.1