Big Data Categorization for Arabic Text Using Latent Semantic Indexing and Clustering

Documents categorization is an important field in the area of natural language processing. In this paper, we propose using Latent Semantic Indexing (LSI), singular value decomposing (SVD) method, and clustering techniques to group similar unlabeled document into pre-specified number of topics. The generated groups are then categorized using a suitable label. For clustering, we used Expectation–Maximization (EM), Self-Organizing Map (SOM), and K-Means algorithms. In our experiments, we use a corpus that contains 1000 documents of ten topics (100 document for each topic. The results show that using LSI and clustering techniques for Arabic text categorization achieves good performance. The results show that EM clustering method outperforms other investigated clustering methods with an average categorization accuracy of 89%. Keywords—Arabic Text, Categorization, Classification, Clustering, Unsupervised Learning.