Frictionless Linked Data

I was reminded of the importance of approachable, low-barrier-to-entry tools for data management by Monica Granados and Lily Zhao in their presentation of the Frictionless Data toolkit.¹

They showcased use of a browser-based interface² for a simple yet valuable task: associating title and description metadata with potentially cryptic column header names in a CSV file, and exporting that metadata together with the raw data to an open-source package format.

First, they directed the app to load an Algae.csv CSV file from an open data archive.³ The app detected tabular data and offered to infer field names and data types:

Next, they showed how a user could fill in field title and description metadata by hand, and select correct data types from a drop-down list:

This kind of interface, where you are guided to fill in metadata and can produce shareable artifacts that are much more likely to be interpretable both by other humans and machines, is crucial for science that is more reproducible/trustable, and thus citable.

When I saw this demo, I immediately thought of the potential for linked open data to get this kind of interface even closer to the Frictionless ideal. The Title and Description fields are analogous to the rdfs:label and rdfs:comment fields ubiquitous in linked open data sets such as Wikidata, and there is a rich vocabulary of data types as URIs under the namespace typically prefixed as xsd:.

If user input to Title and Description were to be fuzzy-matched to corresponding linked-data field values, the app could auto-suggest additional metadata for the tabular data column. Unambiguous URIs from Wikidata or other sources could be associated as metadata, making it even more likely that the recipient of a data package could understand the semantics of the data – by following links.