基于免疫算法的文本分类研究

Research of Text Categorization Based on Immune Algorithm

借鉴免疫的生物学机理,本文提出了一种基于抗体浓度的克隆选择算法,该算法中抗体的选择概率由亲和度与浓度共同决定,具有高亲和度和低浓度的抗体才受到促进。该算法在文本分类领域得到了成功应用。在文本分类的应用中,抗原、B细胞和抗体分别对应训练文本、分类器的一个解和分类器的解与训练文本的亲和度,最后训练完成的分类器含有多个记忆细胞,有效保证了解的多样性。在数据集20_newsgroups上的实验结果显示,该方法的综合性能指标F1可达80.90%,优于Rocchio法与Naive Bayes法。

The clonal selection principle and density control mechanism are used by the natural immune system to define the features of an immune response to an antigenic stimulus. It establishes the ideas that only those ceils that have higher affinity and lower density are selected to proliferate. A new algorithm, called the Clonal Selection Algorithm Based on Antibody Density （CSABAD）, is brought forward and successfully implemented in text categorization. In text categorization, antigen, B cell and antibody are respectively corresponded with training text, an individual of classifier and affinity between the individual and training texts. The final classifter is composed with many memory B cells. The method is applied to the 20_newsgroups dataset and we obtains a F1 score of 80.90%. The result shows that CSABAD significantly outperform Rocchio and Naive Bayes.