Open Source Labels

Sysrev recently enabled ‘shared’ labels.  When you create a sysrev project, you can import labels created in other projects or found on labels.sysrev.com.  We think this is a pretty big deal, let me tell you why.

Open source software projects allow the public to contribute to and/or copy the underlying code base.  Many important software projects are open source.  

Sysrev’s shared labels bring the concept of open source code to document review.

This means that anybody can create a new label and host it on labels.sysrev.com. Labels can be modified and versioned, which means you can start with a good idea for a label and improve it over time, without disrupting your downstream users.

Reducing redundancy
Shared labels enable a lot of super powers.  When users agree that they are using the same label in different projects, then redundant work can be reduced.  

Learnable Labels
Supervised learning is a type of machine learning where models are created to ‘classify’ digital entities.  Creating supervised learning models usually requires a large annotated corpus, often created by humans assigning labels to entities.

Shared labels allows sysrev to build machine learning models that span multiple projects.  If you create a label that identifies genes in text, and a successful sysrev model is created from that

Search Index
The NCBI takes on many large projects to make PubMed a more and more useful tool.  They have their very own special ‘shared’ labels in the form of mesh terms, article attributes, and article types, which can be used to create and filter PubMed searches. For example, PubMed allows you to filter down to publications with the article type ‘clinical trial’.  

I don’t know how PubMed maintains their labels, but popular open source labels that have been assigned to enough documents, or accurately modeled can be used as search indexes for future reviews.  

What’s next?