Super Deep Learning Ensemble Model for Sentiment Analysis

• Published in: 2023 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS) pp. 341 - 346
• Main Authors: Garg, Sarita Bansal, Subrahmanyam, V. V.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.11.2023
• Subjects: Analytical models; Classification; Deep Learnin; Deep learning; Ensemble Methods; Predictive models; Recurrent neural networks; Sentiment analysis; Task analysis; Testing
• DOI: 10.1109/ICCCIS60361.2023.10425309

Sentiment analysis, the automated task of discerning and categorizing sentiments conveyed in text, has seen remarkable progress recently, primarily owing to the emergence of deep learning techniques. However, conventional deep learning models continue to grapple with accuracy and resilience limitations, stemming from their inherent deficiencies and biases. To address these constraints, we propose an innovative Super Deep Learning Ensemble Model (SDL-EM) tailored for sentiment analysis. The SDL-EM is crafted to exploit the complementary strengths of various deep learning architectures and harness ensemble learning's potency to enhance sentiment classification accuracy. This model comprises an assortment of base learners, encompassing Convolutional Neural Networks (CNNs) and varieties of Recurrent Neural Networks (RNNs), each trained on distinct feature representations of the input text. To construct the ensemble, we introduce a fusion mechanism that dynamically amalgamates predictions from individual base learners using a Multi-Layer Perceptron Model. Subsequently, we subject the resulting model to testing using the evaluation dataset, yielding discernible outcomes. To substantiate the efficacy of our SDL-EM proposal, we conduct thorough experiments on a widely recognized sentiment analysis dataset. The results unequivocally substantiate our model's superiority over state-of-the-art deep learning models, manifesting in elevated accuracy, recall, precision and Fl-score metrics. This endeavor contributes to the advancement of sentiment analysis, providing pragmatic solutions for scrutinizing sentiments within extensive textual datasets, characterized by enhanced performance and generalization capabilities.