Abstract: 【Purpose/significance】This research aims to solve the problem of insufficient professional knowledge when obtaining online
medical misinformation data sets, and helps build online medical misinformation detection models in the field of small samples.
【Method/process】We propose an idea of constructing an online medical misinformation dataset based on the transformed extraction of authoritative misinformation refuting,and construct the traditional machine learning model,CNN model and BERT model for detection.【Result/conclusion】The results show that the construction of an online medical misinformation dataset based on misinformation refut-ing can be achieved at a lower cost and without relying on expert labeling.The comparative experiments based on the data set of online medical misinformation related to the COVID-19 show that the accuracy rate of the BERT model pre-trained based on Weibo data is 95.91%,and the F1 value is 94.57%,which is compared with traditional machine learning models and CNN model the increase is close to 6% and 4% respectively.It means that the pre-trained BERT-based model constructed in this paper achieves better results on the detection of online medical misinformation task.【Innovation/limitation】The method proposed in this paper can build misinformation data sets in professional fields at a lower cost,and the constructed BERT medical misinformation detection model is also of practical value in the field of small samples.However,this study needs to be optimized in terms of datasets,deep learning model comparison,and model performance evaluation metrics.