PubMed Semantic Search

Apply for access Private documentation

Intended use

Problem types:

The PubMed Semantic Search experiment is targeted at scientists and researchers working in biomedical research. It is meant to be a tool to facilitate exploration of knowledge and cut back the amount of time that users spend reading through papers and trying to find the right information.

This experiment offers access to both a state-of-the-art search engine (based on neural-retrieval models) and a focused question-answering engine. The current experiment offers access to the index of papers contained in PubMed. The full-text of papers available through PubMed Central is searched in addition to all abstracts provided by the National Library of Medicine. The retrieval models being used have been shown to be state-of-the-art for biomedical retrieval.

The response to the query API includes both the relevant documents, each containing a set of snippets which are intended to support the relevance of the document. These snippets are generated based on a neural language reader which predicts the parts of the document which are most relevant to the query.

An additional API call allows the user to provide a set of document IDs (PubMed IDs) and a query. We apply the same snippeting approach to return relevant information from the provided set of documents.

Inputs and outputs:

  • Users provide: Biomedical queries in the form of natural questions with sufficient context
  • Users receive: Relevant articles and snippets that highlight passages which support the relevance of the document to the query.

Industries and functions:

Currently, this experiment provides access to an up-to-date PubMed index. Users and systems which benefit from extracting information from PubMed (e.g., biomedical research) are the primary targets of the current experiment.

Technical challenges:

This experiment is most useful for users which:

  • Are seeking scientific information from the biomedical literature in an exploratory fashion (i.e., more than retrieving known papers).
  • For any given query, have multiple additional information needs from each of the retrieved papers.

What data do I need?

Data and label types:

No data is needed to utilize this experiment. The user should supply a set of queries and or queries and PubMed IDs.


What skills do I need?

As with all AI Workshop experiments, successful users are likely to be savvy with core AI concepts and skills in order to both deploy the experiment technology and interact with our AI researchers and engineers.

In particular, users of this experiment should:

  • Be familiar with accessing Google APIs