Wikidata

Wikidata:WikidataCon 2017/Notes/Data completeness: How to know what Wikidata knows?

< Wikidata:WikidataCon 2017‎ | Notes

Title: Data completeness: How to know what Wikidata knows?

Note-taker(s): Sannita, Bene*

Speaker(s) edit

  • Name or username: Ls1g
  • Contact (email, Twitter, etc.): razniewski@inf.unibz.it

Abstract edit

Wikidata is a great project towards mapping structured information about the world, and exhibits a high degree of correctness. Its degree of completeness, in turn, is much less understood. Anecdotal evidence suggests that it covers many popular topics quite well, but there are few standard means that help in this assessment: At present, editors and consumers have to analyze largely on a case-by-case basis whether given information might be complete or not. This session concerns the automated assessment of the completeness of Wikidata, and consists of two parts:

In the first 35 minutes I will survey techniques to assess the completeness of parts of Wikidata. I will talk about three aspects of completeness, values, properties and entities:

  • For values, I will discuss no-value statements and predicates that talk about object counts, and the COOL-WD tool for asserting metadata.
  • For properties, I will review mandatory properties (like P1963) and the completeness status indicator icon via Recoin and tabular views like discussed here and exemplified here.
  • For entities, I will look at what is currently possible with the Class Browser, SQID, and what faceted browsing should hopefully make possible in the future.

The second part of the session (15 minutes) shall be an open discussion, guided by the questions

  • What kind of (anecdotal) knowledge about completeness of parts of Wikidata do participants have?
  • What kind of structured knowledge about completeness would participants like to obtain?
  • What tools could help towards this?

Collaborative notes of the session edit

What does Wikidata really knows? There are some areas who are "complete" (Physics Nobel Prizes, children of Obama...), other who are not (how many stops of Berlin S1 there are, ...).

Knowledge base engineers only tried to enrich the KBs, but now the point is to understand what they are trying to approximate or "the unknown unknowns" (cit. Donald Rumsfeld).

  • Wikidata quite complete on some topics, but missing data in other domains
  • We are constantly adding more data to Wikidata

Where is Wikidata going? Nobody really knows. Is easier to say that for specific topics.

  • on specific topics, completeness is well defined, but in general hard to tell (in respect to everything)

Bookkeeping: 1) values; 2) properties; 3) entities

"Solutions" and tools
  • value completeness: no-values + "number of ..." statements
  • challenges: possibilty of contradiction + complex querying
  • should completeness only be considered for "important" properties (where important is defined how?)
  • COOL-WD: http://cool-wd.inf.unibz.it/ (mark the statements for one property as complete)
  • property completeness: pre-defined schemata (P1963, manually defined tables)
  • Recoin: https://www.wikidata.org/wiki/User:Ls1g/Recoin (completeness indicator for a whole item, depends on statements on similar items)
  • ProWD (mockup)
  • entity completeness: ???
  • Is a specific query complete?
  • Class browser for a specific class
  • SQID browser
Summary
  1. Without a goal, it's difficult to understand how Wikidata is faring
  2. bookkeeping is possible on a small scale
  3. there is possibility to try to approximate what's missing and what is not by the tools and solutions explained above
  • Related tickets on Phabricator: T150938, T150116

Questions / Answers edit

Retrieved from "https://www.wikidata.org/w/index.php?title=Wikidata:WikidataCon_2017/Notes/Data_completeness:_How_to_know_what_Wikidata_knows%3F&oldid=599510973"