Empower End-Users to Improve Chatbots: An Experimental Study to Evaluate Explainable AI Features for Collecting Chatbot Response Improvements
- Type:Master Thesis
- Date:now
- Supervisor:
- Add on:
Status: Open
-
Introduction of Chatbots
Many service providers are currently dealing with the development of chatbots. Chatbots are software-based systems that communicate with the user over text- or speech-based natural language. They can be used to communicate with customers, provide information, or offer any other kind of digital service.
Problem Description
Many interactions with chatbots are not perceived as natural. Most chatbot responses are rather short, constrained by limited vocabulary, and contain incomplete statements. The overall quality of chatbot responses is limited by the amount of effort chatbot engineers invest in their development. Because this is a very difficult task, chatbot developers need a system that supports them in this endeavor.
What is currently missing is a system that enables chatbot engineers to efficiently engage domain experts in the response generation process. To address this need, we already developed a system which domain experts can use to improve the responses of a chatbot. However, it is unclear which design features are needed to engage domain experts to contribute many chatbot response improvements. As a consequence, this Master Thesis should come up with different design ideas on how user feedback to improve chatbot responses can be collected. Particularly, the design features should build on knowledge from the research stream about explainable AI. Subsequently, an experimental evaluation should compare the different design ideas to show their effectiveness.
Required Skills
- Interest in human-chatbot interaction
- High motivation to solve interesting real-world problems in an experimental evaluation
Contact
If you are interesting to conduct this these, please contact me directly via email so we can clarify any questions. See you soon, Jasper.