JMIR Publications

ABSTRACT

Background:

Tobacco addiction remains the top cause of preventable mortality in adults. At any given time, most smokers in a population are ambivalent, with no motivation to quit. Motivational interviewing (MI) is an evidence-based technique that uses open ended questions, reflective statements, affirmations and summary statements to elicit change in ambivalent smokers. It has been successfully applied when sufficient human clinicians are available, but they are scarce and expensive, and smokers are difficult to reach. Smokers are potentially reachable online and if an automated computer chatbot could emulate an MI conversation, it could form the basis of a low-cost and scalable intervention to help motivate smokers to quit.

Objective:

To design, train, and test an automated MI-based chatbot that could engage ambivalent smokers in a conversation designed to stimulate a quit attempt. This paper describes the process to improve the accuracy of automatic MI-oriented responses, particularly reflections and summary statements.

Methods:

An interdisciplinary collaboration between an expert in MI and experts in computer engineering and natural language processing co-designed the text, decision trees, and algorithms underlying a chatbot. A sample of 121 adult smokers were recruited from an online platform for a single arm prospective iterative design study. Participants interacted with a text-based web-accessed platform after they provided online consent. The chatbot was designed to stimulate reflections on smoking that were given in free-form text. Specifically, we wanted subjects to articulate the positive and negative attributes of smoking. The subjects were also asked to confirm the chatbot’s classification of their free-form responses to enable the measurement of the classification accuracy. These responses after every 10-11 participants were then used to further train the chatbot. The University of Toronto research ethics board approved this study.

Results:

An interdisciplinary collaboration between an expert in MI and experts in computer engineering and natural language processing co-designed the text, decision trees, and algorithms underlying a chatbot. A sample of 121 adult smokers were recruited from an online platform for a single arm prospective iterative design study. Participants interacted with a text-based web-accessed platform after they provided online consent. The chatbot was designed to stimulate reflections on smoking that were given in free-form text. Specifically, we wanted subjects to articulate the positive and negative attributes of smoking. The subjects were also asked to confirm the chatbot’s classification of their free-form responses to enable the measurement of the classification accuracy. These responses after every 10-11 participants were then used to further train the chatbot. The University of Toronto research ethics board approved this study.

Conclusions:

Ambivalent smokers online recruitment is a viable method to train a chatbot to increase accuracy in reflection and summary statements, the building blocks of motivational interviewing. The next step is a feasibility study on the use of the automated conversation as an intervention.