<p>Electroencephalography (EEG) signals offer a promising avenue for detecting emotional responses during video viewing, enabling the automated recognition of video-induced emotions and providing an objective assessment approach. However, current approaches face two main limitations. First, emotion labels often rely on subjective self-reports that introduce personal bias. Second, most systems require high-density electrode arrays that are costly and impractical for portable applications. To address these challenges, this study explores video emotion recognition using a lightweight EEG setup. We introduce three complementary strategies: (i) a dynamic hierarchical label calibration approach that reduces labeling subjectivity through consistency modeling and boundary refinement; (ii) a multi-dimensional energy ratio analysis that compresses channel requirements while preserving discriminative information; and (iii) a saliency-guided feature selection method to improve generalization capability. By reducing 65% of the channels from the original dataset, our approach achieves 45% accuracy in four-class dominant video emotion prediction using only 11 channels, while maintaining meaningful discriminative performance under cross-subject conditions. Beyond technical advancements, these results demonstrate the potential of EEG-based systems to capture collective emotional responses to video content. This capability supports practical applications in audience sentiment analysis, media content evaluation, and emotion-aware recommendation systems.</p>