文本聚类算法的质量评价

文本聚类是建立大规模文本集合的分类体系实例的有效手段之一。本文讨论了利用标准的分类测试集合进行聚类质量的量化评价的手段，选择了k-Means聚类算法、STC（后缀树聚类）算法和基于Ant的聚类算法进行了实验对比。对实验结果的分析表明，STC聚类算法由于在处理文本时充分考虑了文本的短语特性，其聚类效果较好；基于Ant的聚类算法的结果受参数输入的影响较大；在Ant聚类算法中引入文本特性可以提高聚类结果的质量。

Quality Evaluation for Three Textual Document Clustering Algorithms

Textual document clustering is one of the effective approaches to establish a classification instance of a huge textual document set. Clustering Validation or Quality Evaluation techniques can be used to assess the efficiency and effectiveness of a clustering algorithm. This paper presents the quality evaluation criterions. Based on these criterions we take three typical textual document clustering algorithms for assessment with experiments. The comparison results show that STC(Suffix Tree Clustering) algorithm is better than k-Means and Ant-Based clustering algorithms. The better performance of STC algorithm comes from that it takes into account the linguistic property when processing the documents. Ant-Based clustering algorithm's performance variation is affected by the input variables. It is necessary to adopt linguistic properties to improve the Ant-Based text clustering's performance.