Detecting empathy on textual communication

Abstract

Empathy is a necessary component of human communication. The ability to understand and relate to others provides depth to any conversation between people, and is the basis for any exchange that deals with highly emotional topics. Current technological developments have raised interest in human-like behavior from computer systems regarding communication. This has led to the development of the area known as Affective computing, which is based on the study and processing of concepts related to emotions through artificial intelligence. However, in this area, empathy has been largely ignored in favor of other concepts such as emotion and feeling. This can be attributed to the complexity inherent of the concept. Nevertheless, there are now several methods that can be used to finally study and take advantage of empathy in computer applications. We provide a comprehensive study on the nature of empathy and a method for detecting it in textual communication. Thanks to this research, we present a database of conversations with their respective measurement of empathy. This metric, the Empathy score, is the first method for measuring empathy on texts based on psychological research. In order to detect the value of empathy on conversations, we apply machine learning classification. A pattern-based classification approach was taken in order to predict the Empathy score of utterances in our database, which allowed us to explore the advantages presented by these algorithms in psychologically-adjacent computing research. We were able to use methods found in computer science for the study and detection of empathy, and prove the viability of contrast pattern-based classification for measuring empathy levels on textual conversations.