A substantial portion of today’s multimedia data exists in the form of unstructured text. However, the unstructured nature of text poses a significant task in meeting users’ information requirements. Text classification (TC) has been extensively employed in text mining to facilitate multimedia data processing. However, accurately categorizing texts becomes challenging due to the increasing presence of non-informative features within the corpus. Several reviews on TC, encompassing various feature selection (FS) approaches to eliminate non-informative features, have been previously published. However, these reviews do not adequately cover the recently explored approaches to TC problem-solving utilizing FS, such as optimization techniques. This study comprehensively analyzes different FS approaches based on optimization algorithms for TC. We begin by introducing the primary phases involved in implementing TC. Subsequently, we explore a wide range of FS approaches for categorizing text documents and attempt to organize the existing works into four fundamental approaches: filter, wrapper, hybrid, and embedded. Furthermore, we review four optimization algorithms utilized in solving text FS problems: swarm intelligence-based, evolutionary-based, physics-based, and human behavior-related algorithms. We discuss the advantages and disadvantages of state-of-the-art studies that employ optimization algorithms for text FS methods. Additionally, we consider several aspects of each proposed method and thoroughly discuss the challenges associated with datasets, FS approaches, optimization algorithms, machine learning classifiers, and evaluation criteria employed to assess new and existing techniques. Finally, by identifying research gaps and proposing future directions, our review provides valuable guidance to researchers in developing and situating further studies within the current body of literature.