A Keyword Extraction Method for Traditional Chinese Medicine Constitution Identification Based on Batch Data Hybrid Sampling

Keyword extraction for traditional Chinese medicine (TCM) constitution identification refers to using keyword extraction methods to extract critical symptoms from task corpora to achieve the TCM constitution identification task. In this paper, we constructed a dataset for keyword extraction in TCM constitution identification and implemented keyword extraction using sequence labeling methods. However, we discovered an issue of uneven category distribution during the keyword extraction process, which affected the model's performance. To address this problem, we propose a model training method based on batch data hybrid sampling algorithm. This method divides the training dataset into several batches. It strategically balances the ratio of common and rare category data within each batch, achieving an even distribution of data categories within each training batch and reducing the impact of insufficient learning on rare categories by the model. Experimental results demonstrate that our proposed method improves the performance of keyword extraction for TCM constitution identification, with an average F1 score improvement of 4.29% for rare category keyword extraction.