Sentiment analysis using osprey-optimized hybrid GoogleNet-ResNet with text and emoji data

Abstract

Sentimental analysis (SA) plays a crucial role in understanding consumer opinions from mobile product reviews. While various Deep learning (DL) approaches, such as deep neural networks (DNN) and convolutional neural networks (CNN), have been developed for SA, they often suffer from limitations like low accuracy, high processing time and reliance on single data modality (text data). Upon considering these issues, a novel hybrid DL model is introduced in this research by integrating both text and emoji data for enhanced sentiment classification. Unlike the existing models that primarily rely on text-based features, the proposed approach utilizes both text and emoji representations to capture richer sentiment context. Initially, data pre-processing and sentiment annotation are performed. Text data is transformed into numerical feature vectors using advanced feature extraction techniques, including Word2Vec and Emoji2Vec models, as well as Continuous Bag of Words (CBOW) and Skip-gram (SG). To improve classification accuracy, a depthwise separable hybrid GoogleNet with ResNet convolutional model is employed, leveraging the strengths of both architectures in feature extraction and representation learning. Furthermore, the classifier’s hyperparameters are fine-tuned using the osprey optimization mechanism, an advanced metaheuristic approach that enhances convergence speed and model performances. The proposed model is evaluated under the Flipkart product reviews sentiment dataset, demonstrating its efficacy through performance analysis. These findings confirm that incorporating multi-modal data (text and emoji) and leveraging optimized hybrid architectures significantly enhance sentiment analysis accuracy. From this analysis, the proposed model outperforms other classifiers. The experimental results prove that the proposed model achieves a high accuracy, precision, recall and f1-score of 98.3%, 96%, 97.6% and 96.82%, respectively.