Drug Labeling Document Summarization
HDMedi
Problem
- Drug Labeling Documents are lengthy, making it difficult to understand important information, which is why experts manually create summaries. Due to the high labor requirements, approximately 80% of medications still lack summary documents.
- Lack of a model that summarizes drug labeling documents in a way that is easy to understand for patients.
Solution
- Implementation: We used a pre-trained LLM and improved the quality of the summary by prompt engineering. Especially, we found that preprocessing the input in XML format rather than plain text can improve accuracy.
- Validation: We used metrics such as BERT, BLEU, METEOR, and ROGUE-L to evaluate the performance.
Contributions
- Leading a team of 2 undergraduate interns
- Literature research, prompt engineering, and design of summary evaluation methods