Real-Time Sentiment analysis of Multimodal Data streams – A Deep learning-based framework in Social Media Platforms

The companies of social media have exploded exponentially and generated enough multimodal data, such as text, photos, videos, and audio. Sentiment analysis of such different data streams is highly challenging due to the difficulty of processing the volume, velocity, and variety of such data streams. In the proposed research, a solution to real-time sentiment mining of multimodal messages in interactive social media platforms is suggested, based on deep learning. It stitches multiple structures of deep neural networks into one system and comprises audio processing models to identify a sentiment in the speech (RNNs and transformers to process the text-based input and CNNs to do image analysis). To efficiently integrate the learnt features from several modalities, a multimodal fusion layer is added. To guarantee low-latency sentiment classification, the system uses Apache Kafka and Apache Flink for real-time stream processing. The robustness and scalability of the system are demonstrated through evaluation on real-time Twitter and Instagram data streams as well as benchmark datasets. The suggested approach performs better in terms of accuracy and reactivity than conventional unimodal sentiment analysis models, offering insightful information for use in crisis management, political analysis, brand monitoring, and consumer feedback systems. This study emphasises how crucial real-time processing and multimodal integration are to thorough sentiment analysis in dynamic online settings.

Перевод пока недоступен