This study innovatively developed an intelligentmodel of syndrome differentiation and machine learning trainingmethod with zero-clinical samples which could intuitively infer with diagnostic knowledge of traditional Chinesemedicine，and conducted case analysis on the syndrome differentiation of vertigo.

Based on the deep learningmodel Bert，a new inference algorithm was proposed to establish a new intelligent inferencemodel of TCM diagnosis. According to the key points of six types of dizziness syndrome differentiation（such as liver Yang hyperactivity and Qi-blood deficiency）in Internal Medicine of TCM，a random function was used to generate 400 samples of sentence sequences for each syndrome differentiation class，to let the model learn diagnostic knowledge directly. In order to compare the prediction performance of the newmodeland the random forest model，it used the 5-fold crossovermethod to evaluate themodel for all syndrome differentiation classes with indicators such as Accuracy，F1 score，ROC curve，P-R curve，etc. Furthermore，the reasoning diagrams outputted by the newmodelwere showed to demonstrate the algorithm reasoning process of simulations of TCM syndrome differentiation thinking. Based on the generated combination data of the symptoms and probabilities during the model reasoning process，the probability density distribution map of syndrome types，the correlation clustering heatmap of symptom types and the Pearson correlation analysismap of syndrome types were calculated and drawn to further verify the reliability of the syndrome differentiation results.

The average accuracies and F1 scores of the new model tested with the 5-fold crossmethod were 0.990 and 0.971，respectively. In all the syndrome types，the accuracy（0.975）and F1 score（0.923）of liver-kidney Yin deficiency were the lowest. The ROC curve and P-R curve showed that the newmodel had a little poor ability of prediction for liver-kidney Yin deficiency and liver Yang hyperactivity，while the random forestmodel had a little poor ability of prediction for Qi-blood deficiency syndrome. The data distribution of the probability density distributionmap and the correlation clustering heatmap were consistentwith the inference conclusion of the new model. The Pearson correlation map showed that the syndrome type diagnosed with the new model was negatively correlated with other syndrome types，and the P-value of the correlation coefficientwas less than 0.01，indicating that the newmodel had a good discrimination.

It can intelligently diagnose and demonstrate the reasoning process of dizziness based on the new intelligent diagnostic reasoningmodel，whichmay help to solve the bottleneck problems in the current field of artificial intelligence diagnosis in traditional Chinesemedicine.