Attention-based Summarization Approach of Clinical Notes

Deployment of Machine Learning for understanding clinical notes in the healthcare sector is crucial to extract meaningful phrases based on disease types. It is tough for human beings to summarize large documents of text manually. Summarization and evaluation of text are considered challenging tasks in the NLP community. We developed CUI Machine learning models to summarize clinical notes based on multi-head attention mechanisms and evaluate the summaries by applying evaluation metrics. In this thesis work, we propose a multi-head attention-based mechanism to perform extractive summarization of meaningful phrases in clinical summaries from the MIMIC-III dataset. This research helps highlight and perceive helpful information from clinical notes to a physician, and this step will increase treatment quality and support doctor's tasks. We conclude with optimal results compared to statistical-based models, proposing certain limitations and employing new evaluation metrics from a different perspective for future work. We achieved a provable score, which indicates that the BERT (Bidirectional Encoder Representations from Transformers) based model output accuracy is better than the statistical-based solution model and can be employed to summarize extractive summarization methods.