SENTIMENT-GUIDED MULTIMODAL CONTENT RECOMMENDATION USING GRU AND CNN-LSTM ARCHITECTURES

In the era of social commerce, recommendation systems must transcend traditional paradigms by incorporating multimodal data to personalize content more effectively. This study proposes a sentiment-guided, multimodal recommendation framework that integrates text and image-based features to enhance product recommendation strategies, particularly in influencer marketing and fashion retail. The proposed architecture utilizes Gated Recurrent Units (GRUs) for analyzing user-generated textual comments to capture emotional tone and sentiment directionality. Concurrently, a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) pipeline processes product images, extracting semantic features and generating descriptive captions that encapsulate visual attributes such as color, category, and brand. By fusing textual sentiment with visual embeddings, the system produces contextually relevant and emotionally resonant recommendations. The GRU-based sentiment classifier achieves a classification accuracy of 89%, while the CNN-LSTM image captioning module attains a BLEU-1 score of 0.68 and 85% accuracy in visual attribute prediction. The real-time filtering mechanism dynamically modulates content delivery based on sentiment polarity—suppressing recommendations during negative user sentiment to reduce cognitive load and enhance user trust. Designed with scalability in mind, the system supports real-time interaction through streaming APIs and can be deployed on cloud platforms or edge devices. It enables automated visual tagging, reduces manual annotation overhead, and improves content discovery by aligning recommendations with both emotional and aesthetic user preferences. This architecture thus transforms passive browsing into an adaptive, emotionally intelligent interaction—offering a significant leap forward in personalized marketing and user engagement in visually rich environments.