A methodology for building and evaluating a synthetic human-chatbot dialogue dataset for customer-service conversation management. Dialogues are generated with ChatGPT-3.5 under two controlled conditions -- a target user language-proficiency level and a specific user emotion -- then assessed for overall quality and for the linguistic complexity of both the human and AI turns using standard complexity measures. The chatbot's per-turn attitudes and interaction patterns are stored to later detect common conversation patterns within specific emotional contexts, toward systems that can learn from user interactions.
A first-step methodology for building and evaluating a synthetic human-chatbot dialogue dataset intended for customer-service conversation management. Dialogues are generated with ChatGPT-3.5 under t…