AI for Good

Chapter 29NLP-Supported Chatbot for Cigarette Smoking Cessation

—Jonathan B. Bricker, Brie Sullivan, Marci Strong, Anusua Trivedi, Thomas Roca, James Jacoby, Margarita Santiago-Torres, and Juan M. Lavista Ferres

Executive Summary

Globally, cigarette smoking accounts for 8 million premature deaths and 25 percent of all cancer deaths annually. Despite progress made by government policies, anti-smoking campaigns, and shifting norms, existing public health interventions continue to have very modest treatment engagement and cessation rates. Conversational chatbots are becoming ubiquitous in society. They use natural language processing (NLP)—usually in the form of large language models (LLMs)—to understand user questions and provide conversational responses. Yet whether they are potentially more effective than current standard public health interventions for increasing treatment engagement and helping adults quit smoking has not been examined.

Here we describe the three-year process (2020 to 2023) that lead to the development of a chatbot for smoking cessation capable of answering user's specific clinical questions about quitting smoking. We created the core conversational feature of a chatbot for smoking cessation, called QuitBot, by generating a training bank of 11,000 user questions and clinician answers from linguistic analysis of questions from adults who wanted to quit smoking, call and chatlog transcripts from previous interventions, and chatlog transcripts from the U.S. ...