vec2sparql.pdf

Recent developments in machine learning have led to a rise of largenumber of methods for extracting features from structured data. Thefeatures are represented as vectors and may encode for some semanticaspects of data. They can be used in a machine learning models fordifferent tasks or to compute similarities between the entities of thedata.SPARQL is a query language for structured data originally developedfor querying Resource Description Framework (RDF) data. It has been inuse for over a decade as a standardized NoSQL query language. Manydifferent tools have been developed to enable data sharing withSPARQL. For example, SPARQL endpoints make your data interoperableand available to the world. SPARQL queries can be executed acrossmultiple endpoints.We have developed a Vec2SPARQL, which is a general framework forintegrating structured data and their vector space representations.Vec2SPARQL allows jointly querying vector functions such as computingsimilarities (cosine, correlations) or classifications with machinelearning models within a single SPARQL query. We demonstrateapplications of our approach for biomedical and clinical use cases.