The task of emotion recognition from video data has become increasingly important in realms such as human–computer interaction, healthcare services and entertainment. The introduction of Convolutional Neural Networks (CNNs) have shown its promise in this field, however additional techniques are necessary to improve the performance and accuracy of video emotion recognition systems. In laying out our three new algorithms we articulate some of the most immediate research needs in this space, including temporal relation extraction methods and availability of multimodal integration methodology as well as more advanced CNN architectures. In the first part of our algorithm, we build a Temporal Relation Attention Network (TRAN) to capture fine-grained temporal dependencies between frames in reason for better behavioral emotional cue recognition. Dynamic Temporal Convolutional Network (DTCN) attempts to address this problem by introducing dynamic operations into convolution and effectively retaining temporal features so that the model can be more sensitive for subtle changes in emotions related facial expressions. Moreover, MFT algorithm integrates the audio and visual modalities with transformer networks to better fuse information from different sources for more extension emotion recognition. The Temporal Relation Graph Convolutional Network (TR-GCN) introduces graph convolution operations to dynamically represent temporal relations among video sequences, which is a more expressive way while elaborating complex spatio-temporal dynamics with receiver side. Experimental results indicate that our algorithms improve the efficiency significantly and up to state-of-the-art performances are obtained in terms of emotion prediction accuracy for video data. Experiments using benchmark datasets verify the effectiveness of our new techniques in capturing temporal dynamics, integrating multimodal information and taking advantage of finestate-of-the-art CNN architectures to enhance emotion recognition performance. The work helps video emotion recognition systems to achieve better performance, paving the way for more precise detection in real-world usage.