Empathetic Conversational Artificial Intelligence Systems: Recent Advances and New Frontiers

Abstract

Recently, empathetic conversational artificial intelligence (ECAI) systems exhibiting emotional and social intelligence have attracted increasing attention from the natural language processing (NLP) community. Recognizing users' affective states (emotions, sentiment, etc.) and behavior and responding accordingly is key to successful communication. While it is straightforward for humans, instilling such human-like capabilities into conversational systems is a challenging task. Since ECAI systems encompass a wide spectrum of human attributes, it becomes difficult for beginners to get started. Therefore, this tutorial aims to present a comprehensive review of empathetic conversational AI systems. The scope of the tutorial includes the background, evolution, datasets, techniques including the state-of-the-art approaches, evaluation methods, and recent trends in ECAI systems. Finally, it points out a few limitations and shortcomings of the existing techniques to facilitate future research.

Outline

The tutorial is organized as follows:

• Conceptual models of empathy (15 minutes)

  We will introduce the fundamental conceptual models of empathy, including the different types of empathy and the individual differences in the perception of empathy.

• Need for empathy in conversational AI systems (15 minutes)

  Empathy is regarded as a necessary trait, and studies have been undertaken to enhance empathy in humans in a variety of contexts. Computational modeling of empathy helps in better comprehending human relations. Thus, we will precisely talk about the works which suggest that the incorporation of empathy in conversational artificial intelligence systems could enhance the user experience and contribute to bridging the human-machine gap.

• Empathy-related concepts in ECAI systems (135 minutes)

  An ideal ECAI system is expected to exhibit emotional and social competence. In this part of the tutorial, we will first introduce the various concepts related to empathy and then points out the works that we will cover in detail.

• Emotion/Sentiment
• Emotion Cause
• Intent
• Persona
• Politeness
• External knowledge
• Multimodal information

Specifically, during this session, we will discuss the most prominent works that have implemented the above mentioned empathy-related concepts for empathetic response generation. We will outline the datasets used in these ECAI systems studies followed by the machine learning/deep learning techniques proposed to imbibe empathetic behavior in such systems, state-of-the-art approaches, and evaluation methods.

• ECAI systems for persuasion and psychotherapy (30 minutes)

  Persuasion and psychotherapy are intricate processes that often involve an empathetic connection between two individuals. Research has shown that empathy leads to positive outcomes in persuasive and therapeutic conversations. Following this, we will discuss studies that attempt to build ECAI systems for such social good applications.

• Conclusion and Future Directions (15 minutes)

  This tutorial discusses the most recent and representative works followed by the current trends in ECAI systems. Despite the remarkable advancements in ECAI systems, many research challenges remain in the context of empathetic systems, which gives future directions in this research area: (i) Combining target-dependent emotion with user modeling would be the next step in this line of research, as emotion is a particular dimension affixed to the speaker and other conversational participants. Emotion and personality should be correlated dimensions of the user, and should therefore be modeled jointly; (ii) Utilizing the existing knowledge base containing sentimental or emotional knowledge, e.g., SenticNet, can aid in detecting the emotional states of the user and understanding background information beyond the context, which eventually leads to generating emotionally-coherent responses.

Materials

Slides

Tutorial Proposal