United StatesHealthcare and life sciences2017 – ongoingNational Institutes of Health
PubMed Best Match machine-learning search ranking
PubMed's Best Match ranking increased click-through by over 20% in real-world testing
Problem
PubMed's growing biomedical literature collection made relevant papers harder to find. Date-based sorting could place useful older research below newer results.
Approach
National Library of Medicine researchers trained a relevance-ranking model using past searches and document features. Best Match reranked retrieved results with LambdaMART. PubMed deployed it in June 2017 as an alternative to date sorting.
Results
Takeaway
The evaluation measured search engagement. Higher click-through did not establish better clinical decisions or research outcomes.
Categories
- Size
- Large enterprise
- Industry
- Healthcare and life sciences
- Business function
- Production and quality
- Innovation type
- Product and service innovation
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
NLM researchers described their deployed learning-to-rank system. They reported over 20% higher click-through in testing and millions of searches assisted weekly after deployment.
Full details are in the original sources.