Computational modeling of empathetic response types in textual communication: integrating nonviolent communication methodology and machine learning for explainable AI

Artificial Technology has improved the range and scale of dialogue agents in many spaces; however, these systems tend to fall short when it comes to recognizing types of empathetic responses in written text due to a lackluster understanding of what empathy is. This work introduces a novel empathetic computational model that integrates nonviolent communication techniques with machine learning algorithms to classify empathetic responses in text-based conversations. Our model works in a seeker-listener scenario, merging blame detection (in accordance with Rosenberg’s guidelines) with emotional intensity analysis (utilizing valence, arousal, and dominance scores) to effectively predict an empathetic response as Explorative or Parallel. Based on the Empathetic Dialogs dataset that contains more than 650 annotated conversational instances, experimental results show that our model achieves an accuracy of 84% in predicting the kind of empathetic response, outperforming baseline models in terms of precision and recall. The presented framework enhances Explainable AI’s ability to facilitate empathetic communication in applications like chatbots, virtual assistants and therapeutic settings. This work also pushes the boundary of human–computer interaction, specifically in building more capable empathetic conversational agents.