The AI Doctor Is In: A Survey of Task-Oriented Dialogue Systems for Healthcare Applications

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6638 - 6660 May 22-27, 2022 c⃝2022 Association for Computational Linguistics The AI Doctor Is In: A Survey of Task-Oriented Dialogue Systems for Healthcare Applications Mina Valizadeh and Natalie Parde Natural Language Processing Laboratory Department of Computer Science University of Illinois at Chicago {mvaliz2, parde}@uic.edu Abstract Task-oriented dialogue systems are increas- ingly prevalent in healthcare settings, and have been characterized by a diverse range of ar- chitectures and objectives. Although these systems have been surveyed in the medical community from a non-technical perspective, a systematic review from a rigorous compu- tational perspective has to date remained no- ticeably absent. As a result, many important implementation details of healthcare-oriented dialogue systems remain limited or under- specified, slowing the pace of innovation in this area. To fill this gap, we investigated an initial pool of 4070 papers from well-known computer science, natural language process- ing, and artificial intelligence venues, identi- fying 70 papers discussing the system-level implementation of task-oriented dialogue sys- tems for healthcare applications. We con- ducted a comprehensive technical review of these papers, and present our key findings in- cluding identified gaps and corresponding rec- ommendations. 1 Introduction Dialogue systems1 have a daily presence in many individuals’ lives, acting as virtual assistants (Hoy, 2018), customer service agents (Xu et al., 2017), or even companions (Zhou et al., 2020). While some systems are designed to conduct unstructured conversations in open domains (chatbots), others (task-oriented dialogue systems) help users to com- plete tasks in a specific domain (Jurafsky and Mar- tin, 2009; Qin et al., 2019). Task-oriented dialogue systems can potentially play an important role in health and medical care (Laranjo et al., 2018), and they have been adopted by growing numbers of patients, caregivers, and clinicians (Kearns et al., 2019). Nonetheless, there remains a translational 1We follow an inclusive definition of dialogue systems, encompassing any intelligent systems designed to converse with humans via natural language. gap (Newman-Griffis et al., 2021) between cutting- edge, foundational work in dialogue systems and prototypical or deployed dialogue agents in health- care settings. This limits the proliferation of scien- tific progress to real-world systems, constraining the potential benefits of fundamental research. We move towards closing this gap by conducting a comprehensive, scientifically rigorous analysis of task-oriented healthcare dialogue systems. Our un- derlying objectives are to (a) explore how these sys- tems have been employed to date, and (b) map out their characteristics, shortcomings, and subsequent opportunities for follow-up work. Importantly, we seek to address the limitations of prior systematic reviews by extensively investigating the included systems from a computational perspective. Our primary contributions are as follows: 1. We systematically search through 4070 papers from well-known technical venues and iden- tify 70 papers fitting our inclusion criteria.2 2. We analyze these systems based on many fac- tors, including system objective, language, ar- chitecture, modality, device type, and evalua- tion paradigm, among others. 3. We identify common limitations across sys- tems, …

Content truncated to 500 words