Hybrid deep learning: a comparative study on ai algorithms in natural language processing for text classification

Abstract

The objective of this research project is to assess the effectiveness of various machine learning algorithms, including deep learning and combination approaches, in performing tasks such as categorizing products into specific categories using data from an e-commerce platform named "OTHOBA." In this study, a dataset consisting of 19,087 data samples is used to evaluate the effectiveness of seven supervised machine learning models. Among these models are three based on deep learning: long short-term memory (LSTM), bidirectional long short-term memory (Bi-LSTM), and 1D convolutional (Conv1D), as well as a multi-layer model that combines Conv1D and LSTM approaches. The task at hand is the classification of product categories. The LSTM-based model demonstrates the highest accuracy rate of 96.23% among the deep learning models, while the logistic regression (LR) models achieve the highest accuracy scores of 97.00% for product category classification. Overall, the proposed models and techniques show significant progress in natural language processing (NLP) research for text classification, specifically in English, and have practical applications for e-commerce sites.