Semantic Scholar Releases New Recommendations API
Semantic Scholar’s newest addition to its growing list of open data services will recommend recently published papers and preprints to users
Today, the Semantic Scholar team at the Allen Institute for AI released their Recommendations API, a new public service that can recommend recently published papers or preprints to researchers based on a learned model of their topical interests. The Semantic Scholar Recommendations API will enable third-party applications to suggest new and relevant papers to their users — delivering the latest scientific research to people’s doors.
How the New Recommendations API Works
The Recommendations API can be used in two modes: Single paper recommendations and List-Based Recommendations.
Single Paper Recommendations
Single paper recommendations can be called by appending a Paper ID to the API request URL. For example, the top 10 recommendations for the paper SPECTER: Document-level Representation Learning using Citation-informed Transformers (2020, Cohan et al.) which has the paper ID of ““a3e4ceb42cbcd2c807d53aff90a8cb1f5ee3f031” can be retrieved with the following request:
Alternative paper identifiers (such as arXiV IDs and digital object identifiers, DOIs) can also be used. For example, recommendations for the same paper above can also be fetched by substituting “doi:10.18653/v1/2020.acl-main.207” in place of the paper ID:
List-Based Recommendations
List-based recommendations allow applications to pass in either a list of positive paper examples or negative paper examples. This mode can be utilized by research servers and publishers to serve up more relevant research to users. To illustrate how list-based recommendations can work for these applications, we can take a look at Semantic Scholar’s Research Feeds, which utilizes list-based recommendations to power its website.
In Research Feeds, the papers that a user has saved into a Library folder serve as the set of positive paper examples, helping to inform new recommended papers which will appear in the user’s feeds.
Users can designate any irrelevant papers recommended in their feed as “not relevant,” enabling the recommendation process to improve for each user over time. The more papers that a user saves to a Library folder and rates as “not relevant,” the better the list-based recommendations will be.
For both single-paper and list-based modes, the recommendations are sorted by relevance, and a maximum of 500 recommendations can be returned.
Who benefits from the new Recommendations API
Every day thousands of scientific papers and preprints are published, expanding the number of resources available to researchers, academic institutions, and interested readers. While this means more and more information at our fingertips, navigating these new papers to find the research relevant to your work can sometimes feel like a daunting task.
The Semantic Scholar Recommendations API is helping bridge this knowledge gap for researchers around the world, making it easier for them to find the new resources they need to advance their own work. And as a free and open data service, Semantic Scholar is bringing this tool–and its other open AI services–to both academic research publishers and the broader research community.
One example of such an application comes from the team at Cold Spring Harbor Laboratory which is integrating the Recommendations API into bioRxiv, one of the leading archives for unpublished preprints in the life sciences. The result: researchers can more easily filter preprints to find and review what’s most relevant to their own area of expertise.
“With the deluge of preprints, it’s more and more important that scientists have good tools for filtering the literature,” says Richard Sever, co-founder of bioRxiv. “The new Recommendations API will really help bioRxiv readers to identify papers of interest.”
This newest offering joins the growing list of open data services from Semantic Scholar that are building better ways to search and discover scientific knowledge, including the Semantic Scholar Academic Graph (S2AG), the Semantic Scholar Open Research Corpus (S2ORC), and the Peer Review API.
For more information on using the new Semantic Scholar Recommendations API, please visit our documentation page: https://api.semanticscholar.org/api-docs/recommendations
To request an API key, please contact our Partner team using the following form: https://www.semanticscholar.org/product/api#Partner-Form
Follow @allen_ai and @semanticscholar on Twitter, and subscribe to the AI2 Newsletter to stay current on news and research coming out of AI2.
