My current research work focuses on building novel machine learning models to support information retrieval from a large number of medical notes in the VBOSSA crowdsourcing system.

Specifically, I developed methods to:

• extract clinical similar terms from electronic medical records (EMRs) to refine and recommend query for complex tasks,
• learn from users’ behaviors and rapidly adapt to their needs for search terms. documents or medical events.

I have already published the first work in Journal of Biomedical Informatics (JBI) and AMIA Summits on Translational Science Proceedings, and I am now preparing to submit the second work to JBI.

EMR-based Word Embeddings

Introduction of word embeddings:

Clinical Similar Terms:

EMR-subsets Method

Word embeddings were trained on each clinical note type in an electronic medical record (EMR) system. These embeddings were then combined, weighted, and truncated to select a refined set of similar terms to be used in keyword search and text highlighting. To evaluate their quality, we measured the similar terms’ information retrieval (IR) performance using precision-at-K (P@5, P@10). Additionally, a user study evaluated users’ search term preferences, while a timing study measured the time to answer a question from a clinical chart.

Search Query Recommendation

Medical researchers often use chart reviews, whereby they manually go through a large number of medical notes, searching for evidence for specific questions. Semantically similar terms are proved to be an effective resource to expand query for search evidence in chart reviews. Previous work, such as EMR-based word embeddings, identified clinically similar terms as the ones that frequently coexistence in medical notes, without considering their relationships in other medical contexts (e.g., medical events). We present methods to capture and leverage the medical contexts of clinical terms to improve the quality of query recommendation in chart reviews.