Siamese capsule gorilla troops network-based multimodal sentiment analysis for car reviews

Sentiment analysis of online car reviews is a significant process in natural language processing. Numerous car enthusiasts and influencers create video reviews of automobiles on platforms for example YouTube. In realistic settings for sentiment analysis within this context, the task includes evaluating spoken reviews, the video content, and the emotions expressed by both the content creators and the viewers as reflected in the comments section. The vulnerability to facing circumstances in real-world scenarios where critical components are necessary for performing sentiment analysis on multimedia car reviews may unexpectedly be absent. To overcome this issue, siamese capsule gorilla troops network-based multimodal sentiment analysis for car reviews is proposed in this research work. Initially, pre-processing and feature extraction are performed on the audio, video and text information. The feature fusion is performed by enhancing spiking neural networks with geometric attention fusion network. The fused feature is fed as an input to siamese capsule gorilla troops network which classifies the emotion states as valence, arousal, and trustworthiness. Also, the algorithm called gorilla troop optimization (GTO) optimizes the classification error of the proposed siamese capsule network. Finally, this research work assesses high- and low-level features across modalities, introduces a novel enhancing spiking neural networks with geometric attention fusion network, enhances classification accuracy through new audio, video, and text embeddings, and boosts system robustness by integrating incomplete multimodal data. The achieved accuracy, precision, recall, and F-score values are 99%, 98.55%, 98.75%, and 98.02%, respectively, for the active multimodal sentimental analysis in car review (MuSe-CaR) dataset.