Classical Vector Space-based Semantic Representation and Dimensionality Reduction for Arabic Text Clustering: a survey

Abstract

The Arabic language poses many challenges for automatic text processing due to its linguistic complexity, including rich morphology, the use of diacritics, and the frequent omission of vowels in written text. In general, two key factors significantly influence performance in text processing: the incorporation of semantic information in document representation and the control of dimensionality. This paper presents a survey of Arabic document clustering employing classical vector space models for text representation (as opposed to dense embedding models), with a particular focus on semantic representation and dimensionality reduction. The survey explores several applications of clustering, such as text summarization and information extraction, and reviews existing comparisons between clustering algorithms. Although classical vector space methods provide a somewhat limited perspective, the body of work found, covering the period from 2001 to 2025, is quite diverse. Key challenges, major findings, and the Arabic document corpora used in the reviewed literature are concisely presented in comprehensive synoptic tables.