Optimising Text Classification in Social Networks via Deep Learning-Based Dimensionality Reduction

Abstract

Text classification is essential for handling the large volume of user-generated textual content in social networks. Nowadays, dense word representation techniques, especially those yielded by large language models, capture rich semantic and contextual information from text that is useful for classification tasks, but generates high-dimensional vectors that hinder the efficiency and scalability of the classification algorithms. Despite this, limited research has explored effective dimensionality reduction techniques to balance representation quality with computational demands. This study presents a deep learning-based framework for enhancing text classification in social networks, focusing on computational performance, by compressing high-dimensional text representations into a low-dimensional space while retaining essential features for text classification. To demonstrate the feasibility of the proposal, we conduct a benchmarking study using traditional dimensionality reduction techniques on two widely used benchmark datasets. The findings reveal that our approach can substantially improve the efficiency of text classification in social networks without compromising—and, in some cases, enhancing—the predictive performance.