Adaptive Emotion Cause Pair Detection using Capsule Network

Detecting the causes behind emotions in textual data is crucial for enhancing applications such as dialogue systems, customer service, and mental health assessments. Existing methods, often based on deep learning and machine learning, tend to focus on emotion classification but fall short in pinpointing the causes of those emotions within varied contexts. Proposed is a model that identifies the causes of emotions by analyzing textual data, dynamically adapting to varied emotional expressions in different contexts. The model integrates DistilBERT for generating contextual embeddings, along with a Bi-LSTM and Capsule Network to capture both sequential dependencies and hierarchical relationships in conversations. To achieve efficiency and scalability, distillation is applied, resulting in a lightweight model which ensures both efficiency and adaptability, allowing it to be suitable for real-world deployment in resource-constrained environments. This approach improves the precision of emotion cause detection being optimized for performance in dynamic, real-time scenarios.