Classification Model of Traditional Chinese Medicine Prescription Decocting Duration Combining Text Convolutional Neural Network with Attention Mechanism

Abstract

The decocting time of Traditional Chinese Medicine (TCM) formulas is crucial for therapeutic effects. To better capture the important characteristics related to the decocting time of TCM formulas, a formula decocting time classification model TextCNN-attention that integrates the attention mechanism and Text Convolutional Neural Network is proposed. This model predicts the decocting time of formulas and divides them into long-term, short-term, and medium-term decocting. First, the texture information of medicinal herbs is used to expand the TCM prescriptions. The attention mechanism is then used to learn the importance of each word or subsequence in the prescription text, and the medicinal materials and medicinal texture text in the prescription are weighed. Finally, TextCNN was used to classify the prescription extension text with decocting time labels. The experimental results show that compared with the baseline model, the proposed TextCNN-attention model can better understand the expanded text information of the prescription. The experimental prediction accuracy and F1 value are improved by 1.78 and 1.87%, respectively, indicating that the TextCNN-attention model has better performance in classifying text decoction duration after the formula was extended.