A Novel Hybrid Feature Engineering Approach for Enhanced Sentiment Analysis

Abstract

Sentiment analysis, a critical subfield of Natural Language Processing (NLP), involves identifying and categorizing sentiments expressed in textual data. However, analyzing human language is inherently challenging because of its complexity and ambiguity. In this study, we propose a hybrid approach that integrates Latent Dirichlet Allocation (LDA) for feature extraction with Non-Negative Matrix Factorization (NMF) for dimensionality reduction to enhance the sentiment classification performance. The effectiveness of the proposed method is evaluated using two datasets to assess its robustness and generalizability. experimental results demonstrate that the hybrid LDA–NMF model achieves high accuracy scores, with performance improvements reaching up to 93% on one dataset and 79% on the other, significantly outperforming standalone LDA and NMF-based approaches. These findings highlight the potential of hybrid feature engineering techniques to transform complex textual data into more discriminative representations, thereby substantially improving sentiment analysis outcomes.