Academic Symposium: Short presentations & panel discussion

Author: Xiaoxiao Liu

Title: Can synthetic medical datasets help AI models speak more like doctors?

Abstract: The development of AI chatbots presents promising approaches to address the difficulty of constructing a medical dialogue generation system in the face of a lack of appropriate datasets. One possible approach is to generate synthetic dialogue datasets using AI chatbots. When using this approach, it is essential to understand how to apply synthetic data in dialogue generation tasks and how to validate the synthetic dialogue dataset. This presentation will show our recent work on validation of the efficacy of synthetic data for fine-tuning language models in the context of dialogue generation.