Multimodal Sentimental Analysis of Text and Image-Based Product Reviews Using Deep Learning Framework
Suresh C, M. Saravanan · 2025
Sentiment analysis of product reviews has become an essential factor for understanding customer feedback. While traditional methods focus only on text-based reviews, the fusion of multiple modalities, such as text and images, can provide a more complete understanding of customer feedback. An extensive deep learning methodology for multimodal sentimental analysis of product reviews is proposed. Text feature extraction and Image Feature Extraction with OpenCV and ResNet is done using the BERT model. The extracted features are fused using a multimodal fusion technique to capture the corresponding information between text and image modalities. We test our framework on the Amazon Product Reviews dataset with text reviews and product images. Tests of the suggested approach show that the multimodal approach outperforms the conventional unimodal approaches, which can significantly increase sentiment classification accuracy. Overall, these findings indicate accuracy, reliability, and classification performance improvements, highlighting the significance of multimodal integration for sentiment analysis and suggest ways to improve customer feedback analyses in e-commerce systems.