-Wikidata is a world readable and writable knowledge base maintained by the Wikimedia Foundation. It offers the opportunity to collaboratively construct a fully open access knowledge graph spanning biology, medicine, and all other domains of knowledge. To meet this potential, social and technical challenges must be overcome - many of which are familiar to the biocuration community. These include community ontology building, high precision information extraction, provenance, and license management. By working together with Wikidata now, we can help shape it into a trustworthy, unencumbered central node in the Semantic Web of biomedical data.
INTRODUCTION
Wikidata is a world readable and writable knowledge base
currently maintained by the Wikimedia Foundation [
In addition to its function as a structured datastore for the
Wikimedia projects, Wikidata is being used to integrate and
distribute biomedical knowledge [
II.
OPPORTUNITIES
Wikidata’s CC0 license, Semantic Web compatible implementation and active community provide a unique opportunity to assemble and disseminate knowledge. Wikidata
Apart from its use as a knowledge graph, Wikidata could
provide great value to the text-mining community as a
multilingual collection of concept labels, descriptions, and links to
encyclopedic text. So-called ‘items’ in Wikidata are roughly
analogous to the concepts in the Unified Medical Language
System (UMLS) [
When creating a knowledge base that spans all domains of knowledge, what are the most effective patterns for representation? How can the community work most effectively together to move iteratively closer to the most useful forms? These questions are currently being tackled by a distributed, mostly-volunteer community of ontologists, technologists, domain experts, and interested citizens in discussions held in forums such as the Wikidata property proposal page6. Before a property (e.g. ‘part of’, ‘MeSH id’, or ‘used to treat’) can be used in Wikidata it must be proposed and approved by community consensus. Once consensus is achieved, an elected community member with administrative powers creates the property and it can then be used to add claims to any item. This property collection, and the guidelines associated with their use, forms a major part of the active ‘ontology’ of Wikidata.
In comparison to other efforts to build large knowledge
graphs, the Wikidata approach is on the chaotic side. There is
no rigid application of an upper ontology, no automated
reasoning to support class inference or quality control, and no
over-arching plan to govern the system’s evolution. Instead,
there is a large, motivated, highly heterogeneous community
doing their best to assemble useful structures one step at a
time. So far, good progress has been made as evidenced by
the early applications of Wikidata content such as the new
infobox for human genes in Wikipedia [
A key enabling feature of the Wikidata infrastructure is the capacity to provide provenance for its claims (the triples that compose the knowledge graph) through references. Each claim can be supported by any number of references to supporting sources of information. Unfortunately, many of the claims that are currently in Wikidata were not assigned references. These unsourced claims are of uncertain quality and may weaken the chances of community uptake. Many long-time Wikipedians are hesitant to embrace Wikidata and use the lack of references as an argument against broadly deploying its contents to support infoboxes. This situation poses a challenge to the information extraction community. Given an unsourced claim (e.g. that a drug treats a particular disease) can we develop automated or semi-automated processes for finding sources to validate or invalidate these claims? Could we apply similar processes to automatically verify references that do exist to ensure high quality? If successful, such automation could greatly help drive Wikidata and other similarly open initiatives forward by allaying concerns about the trustworthiness of content.
The majority of the world’s biomedical knowledge remains locked up in unstructured text. As text mining matures, it is increasingly possible to extract this knowledge automatically; however (1) most people, even within the bioinformatics community, do not have the skills and resources to perform this work themselves and (2) despite many advances, workflows for generating highly reliable content still require human review. If extracted knowledge could be shared through Wikidata, it would reach the broadest possible audience, eliminating the need for consumers to build and run their own extraction pipelines. However, to achieve this, the quality of such workflows would need to be at the same level as institutional biocuration processes – likely with human verification as the final step. A challenge for the textmining research community is to identify ways to engage the thousands of Wikidata community members to define truly scalable, high quality biocuration workflows by effectively integrating machine intelligence with community intelligence.
With diligence, persistence and patience, Wikidata could become the central hub of the Web of data, uniting all domains of knowledge. The biocuration community has an opportunity to help lead this process and, in doing so, benefit all aspects of biomedical research. The time is now.
This work was supported by the US National Institute of Health (grants GM089820 and U54GM114833 to AIS) and by the Scripps Translational Science Institute with an NIHNCATS Clinical and Translational Science Award (CTSA; 5 UL1 TR001114).