Emotionally-Aware Chatbots: A Survey

Abstract

Textual conversational agents, also known as chatbot technology, have garnered significant attention in both academia and industry in recent times. Chatbots have found wide application as agents for human communication in various services, such as booking assistance, customer service, and even personal companionship. The primary challenge in chatbot development lies in creating a machine capable of human-like interaction to enhance user engagement. Several studies have highlighted the importance of incorporating emotion in humanizing machines, including chatbots. However, to the best of our knowledge, there has been no focused work in this specific area. Therefore, this paper aims to provide a systematic review of approaches to building emotionally-aware chatbots (EACs). We propose three key focuses for studying EACs. Firstly, we explore the history and evolution of EACs. Next, we review the approaches adopted by previous studies in building EACs. Lastly, we delve into available resources for constructing EACs, with a particular emphasis on datasets and affective resources. Our investigation reveals that early EAC development predominantly relied on a rule-based approach, while more recent works have embraced neural-based approaches. We observe that most EACs incorporate emotion classifiers within their architecture, leveraging various existing affective resources. We argue that the field of EAC development will continue to attract increasing attention from researchers, as evidenced by recent studies proposing new datasets for building EACs in diverse languages.