ENHANCED SENTIMENT ANALYSIS IN TEXT DATA USING A CNN GRU ATTENTION FRAMEWORK WITH GEPSVM CLASSIFICATION

Abstract

Sentiment analysis, a core natural language processing (NLP) task, is concerning the detection of emotional polarity of text material and is a highly relevant application field in customer experience management, social media tracking, and public opinion excavation. Building on previous work that employed CNN and a twin SVM-based system with an extended pinball loss function, this paper introduces CNN-GRU-GEPSVM, a deep learning-based hybrid approach to improve sentiment classification accuracy on diverse dataset. The novelty of CNN-GRU-GEPSVM is the combination of GloVe word embedding and CNN-GRU architecture, which efficiently learns local phrase-level information and long-distance dependencies of a document. A stronger attention mechanism dynamically emphasizes sentiment-conveying words while encouraging feature importance and interpretability. The model also adopts the Generalized Eigenvalue Proximal Support Vector Machine (GEPSVM) for end classification, facilitating the formation of superior decision boundaries in high-dimensional feature space. CNN-GRU-GEPSVM is tested against benchmark sentiment analysis datasets and demonstrates a remarkable improvement in performance with a 6.8% accuracy improvement, as well as similar improvements in precision, recall, and F1 score compared to traditional deep learning models. The model performs well across various domains and best suits real-world applications like product review extraction, health sentiment analysis, and opinion monitoring during elections. This work highlights the importance of integrating neural network structures, efficient classification, and attention mechanisms to provide scalable, context-aware, and interpretable sentiment analysis solutions.