| Sumario: | Sentiment analysis has come a long way since it was only about classifying text. Now it uses visuals, audio, and contextual signals to provide a better and more accurate picture of how people feel. This study shows a sentiment analysis approach that takes into account the context and uses multimodal learning to improve emotion recognition. The suggested method uses powerful deep learning methods to extract features and classify them across different types of data. Deep learning-based word embeddings like BERT and FastText are used for textual sentiment analysis to get rich linguistic representations and contextual dependencies. We also employ Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorisation (NMF) to find underlying topic structures and pull out useful latent characteristics from text data. For sentiment analysis based on images, we use EfficientNet-B3 because it is better at extracting features and is faster at doing so. Also, hybrid convolutional neural networks (CNNs) are used for both text and image-based emotion recognition, which makes the sentiment analysis technique more complete. The system is thoroughly tested on benchmark datasets, and the results show that it is far better than unimodal sentiment analysis methods. The Hybrid CNN model was able to tell how someone felt about a piece of text with 92.3% accuracy just looking at the text itself. The EfficientNet-B3 model, which is based on images, has an accuracy of 96.7%. This shows that it can pick up on emotional cues from visual data and that deep learning is useful for image-based sentiment analysis. Multimodal context improves the accuracy of categorisation, which makes it beneficial for social media analysis, AI that can understand emotions, and initiating research on sentiment and emotion detection.
|
|---|