Emotion recognition is an important research area within the field of affective computing, which aims to build systems that can recognize, interpret, and respond to human emotions. The feasibility and costliness of continuous emotion recognition have led to numerous studies. Rather than labeling emotional states continuously, the focus is on modeling time intervals that better represent the presence of affective states. Instead of tracking ongoing affective changes within a specific timeframe, this technique aims to identify the most prominent emotional incidents. This approach offers a practical and natural way of learning in a weakly supervised environment, considering the uncertainty of emotional responses.