一种改进的免疫算法及其在文本分类中的应用

Improved Immune Algorithm and Its Application in Text Categorization

借鉴了免疫系统的分类本质以及免疫系统的克隆选择和抗体浓度控制原理,提出了基于抗体浓度的克隆选择算法。该算法基于抗体的浓度和亲和度选择免疫反应细胞,具有高亲和度和低抗体浓度的细胞其选择概率相对较高。通过对多个免疫反应细胞经过多次克隆变异后选取最优解作为记忆细胞,由最终保留的记忆细胞群生成分类器。整个过程既保证了解的正确性,又保证了解的多样性。在数据集20_newsgroups上的实验结果显示:该算法的分类性能优于Rocchio和Nave Bayes,与SVM性能相当。

The classification of immune algorithm is divided into self-ones and non-self ones. On the basis of immune algorithm, the authors propose a new method of text categorization called clonal selection algorithm based on antibody density. According to the clonal selection principle and density control mechanism, only those cells that have higher affinity and lower density are selected to proliferate. The ultimate classifier is composed of many memory cells. On the whole, the approach considers the accuracy of individuals as well as variety and is successfully implemented in text categorization. The experiment results show that it significantly outperforms Rocchio and Na?ve Bayes and has similar performance with SVM on 20_newsgroups data set.