As computing becomes involved in every sphere of life, so too is persuasion
a target for applying computer-based solutions. Conversational agents, also
known as chatbots, are versatile tools that have the potential of being used
as agents in dialogical argumentation systems where the chatbot acts as the
persuader and the human agent as the persuadee and thereby offer a costeffective and scalable alternative to in-person consultations
To allow the user to type his or her argument in free-text input (as opposed
to selecting arguments from a menu) the chatbot needs to be able to (1)
“understand” the user’s concern he or she is raising in their argument and (2)
give an appropriate counterargument that addresses the user’s concern.
In this thesis I describe how to (1) acquire arguments for the construction
of the chatbot’s knowledge base with the help of crowdsourcing, (2) how to
automatically identify the concerns that arguments address, and (3) how to
construct the chatbot’s knowledge base in the form of an argument graph that
can be used during persuasive dialogues with users.
I evaluated my methods in four case studies that covered several domains
(physical activity, meat consumption, UK University Fees and COVID-19
vaccination). In each case study I implemented a chatbot that engaged in argumentative dialogues with participants and measured the participants’ change of
stance before and after engaging in a chat with the bot. In all four case studies
the chatbot showed statistically significant success persuading people to either
consider changing their behaviour or to change their stance.