ABSTRACT
Understanding how human emotions are expressed in text and images depends heavily on sentiment analysis, a crucial field of natural language processing (NLP). In order to improve multimodal sentiment analysis (MSA), this study suggests an Integrative Deep Learning Framework that combines Convolutional Neural Networks (CNN) with Bidirectional Long Short-Term Memory (Bi-LSTM) networks. The framework leverages CNN to extract spatial features from images, capturing local patterns and pictorial semantics. Simultaneously, Bi-LSTM processes textual data, effectively capturing sequential dependencies and contextual relationships by analyzing information bidirectionally. The integration of these architectures enables a robust multimodal feature fusion strategy, allowing the model to analyze sentiment comprehensively across text and image modalities. The framework addresses key challenges in multimodal sentiment analysis (MSA), such as the alignment of heterogeneous data sources and the preservation of contextual accuracy. To validate the model, extensive experiments are conducted on high-quality datasets comprising text-image pairs, annotated for sentiment classification tasks. Metrics including recall, accuracy, F 1 score and precision, are used to evaluate performance. Results reveal that the hybrid CNN-Bi-LSTM model outperforms traditional unimodal and existing multimodal approaches, demonstrating superior capabilities in capturing complex interactions between modalities. The results highlight the probable of the proposed framework to advance applications in domains such as social media analytics, user behavior prediction, and human-computer interaction. This research contributes to the growing field of multimodal sentiment analysis by presenting a novel, efficient, and scalable approach to understanding sentiment in diverse data environments.
