Optimizing Clinical Text Segmentation with Latent HMM-Greedy Algorithm

In this paper, a novel method for accurate and efficient word segmentation in medical texts is presented: the Latent HMM-Greedy algorithm. In order to refine the process and uncover hidden linguistic patterns, the method combines a latent Hidden Markov Model (HMM) with a greedy corresponding technique for initial swift segmentation. The latent HMM model improves accuracy by segmenting the remaining text and utilizing latent dependencies, while the greedy matching phase rapidly finds and segments well-known terms using a large dictionary. To improve performance over time, the algorithm also dynamically adds newly segmented terms to the dictionary. When used on clinical symptoms texts, experiments demonstrate the Latent HMM-Greedy method outperforms conventional greedy matching algorithms by 70% in processing speed and by achieving a 3% increase in precision. This work compared with the existing CRF, LDA, MRFS, and GMM models and it shows the latent Hidden Markov Model Greedy Algorithm high accuracy with compared with the existing model.