<p>Nowadays, emotion recognition through body gestures has gained attention as a promising approach for psychological assessment and human–computer interaction (HCI). Despite this progress, limited research has addressed the recognition of negative emotions, which significantly impact daily life and mental well-being. Accurately classifying such emotions remains challenging due to their subjective nature and context-dependent variations. This study proposes a novel hybrid deep learning framework combination of a Bidirectional Long Short-Term Memory (BiLSTM) network model with an Attention Domain Discriminator (ADD) to recognize specific negative emotion states from human body gestures. The novelty lies in combining Kinect-based gesture features with a hybrid BiLSTM–ADD architecture for selective negative emotion recognition, which is a less explored and more complex domain than general emotion recognition. Traditional algorithms suffer from limited feature dimensionality, resulting in suboptimal recognition performance. The proposed BiLSTM-ADD model addresses this issue by efficiently learning temporal dependencies and selecting discriminative features, thereby enhancing classification accuracy. Extensive performance evaluations were conducted to validate the effectiveness of the proposed model. Experimental results demonstrate that the BiLSTM-ADD framework significantly outperforms conventional deep learning models in recognizing negative emotions, achieving a recognition accuracy of 97.73% on our custom dataset. These findings highlight the potential of the proposed approach for reliable negative emotion recognition, with applications in psychological assessment, healthcare, and HCI.</p>