Bachelor's Thesis Nicolas Klein

Improving Conversational Analytics of a Voice-Based Digital Assistant for Geriatric Care

Abstract

Conversational agents have become more relevant in recent years. Especially with theadvancements in the fields of artificial intelligence and natural language processing, newtechnologies and applications have emerged and gained widespread attention. Conversationalinterfaces present users with the ease of controlling modern technologies through naturallanguage alone. This does not only drastically lower the technical expertise required to engagewith and use these systems, but also motivates their application in new contexts. The medicalcare sector, which suffers from a global staff shortage, presents chances for a conversationalagent to support daily tasks surrounding the care of the elderly and sick.The development process of digital agents is often happening in a non-linear way. Agentsare rolled out multiple times to gather conversational data and user feedback to analyzewhen, where, and why the agent is struggling or failing. Analyzing this data helps tounderstand the agent’s and the user’s behavior. In this approach, known as Conversation-Driven Development, the agent is gradually improved over multiple cycles.This thesis aims to fulfill a part of this development process by analyzing and suggestingimprovements for an agent which is already being used in practice, namely the digital healthassistant ALPHA. Through ALPHA, we get insights into agent analytics and the applicationof conversational agents in specific domains, such as the elderly care sector. While theresearch corpus on voice-based agents and associated technologies is growing, scientific workon voice-based assistants in the elderly-care domain and its unique challenges is still rare. Tothoroughly dissect and understand ALPHA and agents alike, a number of research methodsare carried out. Firstly, a literature analysis of conversational agents, their development cycles,as well as their respective analytics, is conducted. Secondly, collected log data from theALPHA agent is analyzed and restructured for further usage. Next, expert interviews withcontributors to the ALPHA project are held while a data visualization prototype is developedsimultaneously. This prototype aims to make essential metrics and analytics more accessiblethrough concrete visualizations and intuitive interaction possibilities. Finally, the differentfindings are used to infer feasible implications and improvements for the future developmentof ALPHA.

Research Questions

1. What are current insights from literature into “Conversational Analytics” and  “Conversation-Driven Development” of digital assistants?
2. How can insights from literature and expert interviews help to build a conversation analytics tool, and how can this be supportive in uncovering strengths and weaknesses of the digital assistant for geriatric care?
3. What are feasible improvements of the agent and how could they work?