Published Paper Details:

Visual Question Answering (VQA) is an interdisciplinary domain at the intersection of computer vision and natural language processing, focusing on answering questions about images. This study proposes an enhanced framework that incorporates sentiment analysis into VQA systems, enabling context-aware and sentiment-sensitive responses. Such integration is pivotal in applications like e-commerce, healthcare, and social media, where understanding emotions in visual content significantly improves user interactions. The proposed methodology combines state-of-the-art VQA techniques with sentiment analysis models. Visual feature extraction is achieved using a pre-trained convolutional neural network (CNN) such as ResNet, while language understanding employs transformer-based architectures like BERT. A multimodal fusion mechanism integrates visual and textual data, augmented with sentiment features extracted using a separate sentiment analysis pipeline. The fused embeddings are then fed into a deep neural network to generate contextually relevant answers. Experiments are conducted on benchmark datasets such as VQA 2.0 and Visual Sentiment Ontology (VSO), incorporating synthetic datasets with sentiment annotations. Results demonstrate a significant improvement in performance, achieving an accuracy of 89.85% compared to 76.80% for baseline VQA models on the VQA 2.0 dataset. Additionally, contextual relevance is enhanced with sentiment features contributing to improved emotional understanding in responses. This paper contributes a novel multimodal approach that bridges the gap between VQA and sentiment analysis, addressing the lack of emotional intelligence in traditional VQA systems. The findings indicate promising avenues for future exploration in adaptive AI systems.