Dynamic Text Augmentation for Robust Sentiment Analysis: Enhancing Model Performance with EDA and Multi-Channel CNN

Abstract

The rapid growth of social media has revolutionized the way individuals share and access opinions, generating vast amounts of textual data containing diverse sentiments. Sentiment analysis enables organizations to derive valuable insights from this data, facilitating improved decision-making. However, the limited availability of labeled datasets and challenges related to data imbalance often hinder the development of robust sentiment analysis models. In this study, we propose a novel framework dynamic Easy Data Augmentation (EDA) technique with a Multi-Channel Convolutional Neural Network (MCNN) to address these issues. The EDA techniques including synonym replacement (SR), random insertion (RI), random deletion (RD), and random swap (RS), enrich the training data by introducing diversity, thereby enhancing model robustness. The MCNN architecture effectively captures local patterns in text using multiple filter sizes, enabling comprehensive feature extraction with low computational overhead. Furthermore, the integration of Word2Vec embeddings ensures meaningful text representation, leveraging context-independent word relationships for improved sentiment analysis. Experimental evaluations demonstrate the effectiveness of the proposed approach, highlighting significant improvements in F1-score compared to baseline models. The findings underscore the potential of dynamic text augmentation and efficient deep learning architectures to overcome the challenges of limited training data, offering a scalable solution for sentiment analysis in real-world applications. Additionally, the proposed model achieves an average performance improvement of 11.34% over previous models.