摘要：

介绍一种树状朴素贝叶斯(TAN)文本分类模型，对该模型存在的阈值选取问题进行实验分析，提出不需要进行阈值选取的TAN文本自动分类框架(ATAN)。在中英文非均匀类分布测试集上对基于ATAN的2种算法与手动选取阈值达到最优性能的BL-TAN进行对比，结果表明基于ATAN的算法具有更高性能。

关键词: 文本分类, 树状朴素贝叶斯模型, 贝叶斯网络

Abstract:

This paper introduces a Tree-Augmented Na?ve Bayes(TAN) text categorization model, analyzes its problem of threshold selection, and proposes an Automatic TAN(ATAN) text categorization framework. Two algorithms based on ATAN are compared to the BL-TAN with the best classification performance at a specified threshold both on Chinese and English imbalanced datasets. Results show that algorithms based on ATAN have higher performance than BL-TAN.

Key words: text categorization, Tree-Augmented Na?ve Bayes(TAN) model, Bayesian network

中图分类号: 

• TP391