Hybrid Data Fusion for Elegant Emotion Recognition in Cloud-Based Big Data Systems

In the era of intelligent information, video, audio, and text information are growing explosively. How to extract emotional factors from this information has become very important. This study proposes the novel hybrid data fusion framework based on cloud computing big data system for elegant emotion recognition. This model can effectively solve the multimodal data fusion and recognition tasks. Initially, as the cloud system are always encrypted, data verification mechanism based on linked list index structure is integrated and the novel data extraction methodology is designed. In the image emotion recognition (IER) part, a structure based on a lightweight neural network is proposed, and combined with the Transformer model, spatial and temporal features are extracted to enhance the accuracy of the emotion recognition. In the audio emotion recognition (AER) part, an innovative emotion vector space analysis method is proposed, which converts the emotional expression in music into vector representation and uses the Transformer encoder structure to capture temporal relationship in music features. In the hybrid data fusion aspect, this study uses the Lukasiewicz T-norm to achieve the fusion of image and audio data. Through the experiment, recognition accuracy is tested and performance is evaluated.