一种高性能的两类中文文本分类方法

提出了一种高性能的两类中文文本分类方法．该方法采用两步分类策略：第1步以词性为动词、名词、形容词或副词的词语作为特征，以改进的互信息公式来选择特征，以朴素贝叶斯分类器进行分类．利用文本特征估算文本属于两种类型的测度X和Y，构造二维文本空间，将文本映射为二维空间中的一个点，将分类器看作是在二维空间中寻求一条分割直线．根据文本点到分割直线的距离将二维空间分为可靠和不可靠两部分，以此评估第1步分类结果，若第1步分类可靠，做出分类决策；否则进行第2步．第2步将文本看作由词性为动词或名词的词语构成的序列，以该序列中相邻两个词语构成的二元词语串作为特征，以改进互信息公式来选择特征，以朴素贝叶斯分类器进行分类．在由12600篇文本构成的数据集上运行的实验表明，两步文本分类方法达到了较高的分类性能，精确率、召回率和F1值分别为97．19％，93．94％和95．54％．

A High Performance Two-Class Chinese Text Categorization Method

Text filtering for topic-sensitive information is one of the important applications in text categorization. To effectively filter out the topic-sensitive information from Chinese text collections is a technical challenge. This paper presents a high performance method employing a twostep strategy to classify texts. In the first step, authors regard the words with parts of speech verb, noun, adjective and adverb as candidate features, perform feature selection on them in terms of the improved mutual information formula, and classify the input texts with a naive Bayes classifier. A portion of texts which are currently thought of being unreliable in categorization are identified, forming a fuzzy area between categories. In the second step, authors regard the bigrams of words with parts of speech verb and noun as candidate features, use the same feature se- lection and classifier to deal with the texts in the fuzzy area. The experiments on a test set consisting of 12600 Chinese texts show that this method achieves a high performance. The precision, recall and F1 is 97.19%, 93.94% and 95. 54% respectively.