Unsupervised feature selection technique based on genetic algorithm for improving the Text Clustering
Unsupervised feature selection technique based on genetic algorithm for improving the Text Clustering
复制标题
DOI：
10.1109/csit.2016.7549453
复制
发表时间：
2016-07
期刊：
影响因子：
--
通讯作者：
L. Abualigah;A. Khader;M. Al-Betar
中科院分区：
文献类型：
--
作者：
L. Abualigah;A. Khader;M. Al-Betar
数字形式的文本文档数量的增加影响了文本分析技术。文本聚类（TC）是用于通过聚类显示大量文本文档的重要技术之一。因此，影响文本聚类技术的主要问题是文本文档上存在稀疏和无信息的特征。特征选择（FS）是一种重要的无监督学习技术。该技术用于选择信息特征，以提高文本聚类算法的性能。近年来，元启发式算法被成功地应用于解决一些困难的优化问题。在本文中，我们提出了遗传算法（GA）来解决无监督特征选择问题，即（FSGATC）。该方法用于创建新的信息特征子集，以获得更准确的聚类。实验采用四个具有不同特征的基准文本数据集。结果表明，本文提出的FSGATC提高了文本聚类算法的性能，与k-均值单独聚类相比，得到了更好的结果。最后，利用文本聚类领域常用的度量f值和精度对本文提出的“FSGATC”方法进行了评价。
The increasing amount of text documents in digital forms affect the text analysis techniques. Text clustering (TC) is one of the important techniques used for showing a massive amount of text documents by clusters. Hence, the main problem that affects the text clustering technique is the presence sparse and uninformative features on the text documents. The feature selection (FS) is an essential unsupervised learning technique. This technique is used to select informative features to improve the performance of text clustering algorithm. Recently, the meta-heuristic algorithms are successfully applied to solve several hard optimization problems. In this paper, we proposed the genetic algorithm (GA) to solve the unsupervised feature selection problem, namely, (FSGATC). This method is used to create a new subset of informative features in order to obtain more accurate clusters. Experiments were conducted using four benchmark text datasets with variant characteristics. The results showed that the proposed FSGATC is improved the performance of the text clustering algorithm and got better results compared with k-mean clustering standalone. Finally, the proposed method “FSGATC” evaluated by F-measure and Accuracy, which are common measures used in the domain of text clustering.