Supervised Models for Clinical Text: Evaluating SVM and BERT Performance | HackerNoon

The article discusses the use of sentence-level classification for clinical notes using Support Vector Machine (SVM) and Bidirectional Encoder Representations from Transformers (BERT). Initial experimentation using term frequency-inverse document frequency (TF-IDF) for vector conversion was found less effective compared to word2vec, which produced superior results. The research highlights crucial aspects of data sources, annotation methods, and the lexicon expansion process. It ultimately suggests that advanced embedding techniques alongside SVM offer promising capabilities for improving sentence classification in clinical contexts, underscoring the importance of accurate annotations in health data analysis.

Our results indicate that the word2vec-based system outperformed the TF-IDF-based system for sentence-level classification, showcasing the importance of effective embedding techniques in clinical note analysis.

The study highlights the effectiveness of SVM and BERT in classifying sentences within clinical notes, emphasizing the need for precise annotations and robust machine learning methods for clinical data.

Read at Hackernoon