The primary objective of text classification in the field of Natural Language Processing (NLP) is to methodically categorize a wide range of textual input, including sentences, documents, and queries. The increasing prevalence of large digital document collections, especially in business environments seeking to enhance efficiency and profitability, has emphasized the growing importance of text classification procedures. This study aims to enhance the field of automated text classification methodologies by utilizing advanced Deep Learning (DL) frameworks such as Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Bi-Directional Encoder Representation from Transformers (BERT). The research carefully deals with the pretreatment of the text database by removing unnecessary special characters and non-relevant stop words, using two modern and publicly accessible databases. The subsequent process of dividing the text into tokens and providing it to the deep learning frameworks outlined before enables effective extraction of important characteristics and accurate categorization. A sequence of experimental rounds is performed to determine the most effective architectural configurations and hyperparameter values, resulting in the development of a final design that achieves an impressive validation accuracy of 98%. This architecture is ready to be used in real-time text categorization scenarios, proving its effectiveness as a powerful tool in the current field of Natural Language Processing (NLP).

Index Terms- Text Classification, LSTM, Deep Learning, BERT and Convolutional Neural Network.