Thursday, September 26, 2019
The lowest hanging fruit in #DBpedia
We should import such agreeable data without further ado. An additional manual step to import to Wikidata is not smart because manual operations introduce new errors. Arguably when there is no unanimous support manual intervention may improve the quality but given the quantity of the data involved, it means that a lot of data will not become available. THAT in and of itself has a negative impact on the quality of available data as well.
So what to do.. Harvest all the data that is of an acceptable quality, that is the data DBpedia accepts for its own purposes. Enable an interface where people verify the data where their project is challenged.
When we truly aim to engage people, we enable them to target the data they want to work on. I will happily work on scientists but do not expect me to work on "sucker stars". More than likely there will be people who care about soccer stars but not about "crazy professors".
Wednesday, September 25, 2019
With #DBpedia to the (data) cleaners
Obviously, it takes skill and resources to get the best result and obviously, some of the data gathered does not pass the smell test. The process the data wranglers use includes a verification stage as described in this paper. They have two choices for when data that should be the same is not; they either have a preference or they go with the consensus ie the result that shows most often.
For data wranglers this is a proper choice.. There is an other option for another day, these discrepancies are left for the cleaners.
With the process well described, the data openly advertised as available, the cleaners will come. First people akin to the wranglers, they have the skills to build the queries, the tools to slice and dice the data. When these tools are discovered, particularly by those who care about specific subsets, they will dive in and change things where applicable. They will seek the references, make the judgments necessary to improve what is there.
The DBpedia data wranglers are part of the Wikimedia movement and do more than build something on top of what the Wikis produced; DBpedia and the Wikimedia projects work together improving our movement's qualities. With the processing data generally available this will become even more effective.
Sunday, September 22, 2019
Comparing datasets, bigger or better or it does not matter?
There are many ongoing efforts to extract information from the Wikipedias. The best organised project is DBpedia, it continuously improves it algorithms to get more and higher grade data and it republishes the data in a format that is both flexible and scalable. Information is also extracted from the Wikipedias by the Wikidata community. Plenty of tools like petscan and the awarder and plenty of people working on single items one at a time.
Statistically on the scale of a Wikidata, individual efforts make little or no impression but in the subsets the effects may be massive. It is for instance Siobhan working on New Zealand butterflies and other critters. Siobhan writes Wikipedia articles as well strengthening the ties that bind Wikidata to Wikipedia. Her efforts have been noticed and Wikidata is becoming increasingly relevant to and used by entomologists.
There are many data sets, because of its wiki links every Wikipedia is one as well. The notion that one is bigger or better does not really matter. It is all in the interoperability, it is all in the usability of the data. Wikipedia wiki links are highly functional and not interoperable at all. More and more Wikipedias accept that cooperation will get them better quality information for its readers. Once the biggest accept data as a resource to curate the shared data the act of comparing data sets is improved quality for all.
Saturday, September 07, 2019
Language barriers to @Wikidata
For most other purposes Wikidata serves the "big" languages best, particularly English. This is awkward because particularly people reading other languages stand to gain most from Wikidata. The question is: how do we chip away on this language barrier.
Giving Wikidata data an application is the best way to entice people to give Wikidata a second look.. Here are two:
- Commons is being wikidatified and it now supports a "depicts" statement. As more labels become available in a language, finding pictures in "your" language becomes easy and obvious. It just needs an application
- Many subjects are likely to be of interest in a language. Why not have projects like the Africa project with information about Africa shared and updated by the Listeria bot? Add labels and it becomes easier to use, link to Reasonator for understanding and add articles for a Wikipedia to gain content.
Key is the application of our data. Wikidata includes a lot, the objective is to find the labels and we will when the results are immediately applicable. It will also help when we consider the marketing opportunities that help foster our goals.
@Wikidata - #Quality is in the network
As always, there has to be a point to it. You may want to write Wikipedia articles about chemists, artists, award winners. You may want to write to make the gender gap less in your face but who to write about?
Typically connecting to small subsets is best. However we want to know about the distribution of genders so it is very relevant to add a gender. Statistically it makes no difference in the big picture but for subsets like: the co-authors of a scientist or a profession, an award, additional data helps understand how the gender gap manifests itself.
The inflation of "professions" like "researcher" is such that it is no longer distinctive, at most it helps with the disambiguation from for instance soccer stars. When a more precise profession is known like "chemist" or "astronomer", all subclasses of researcher, it is best to remove researcher as it is implied.
Lists like members of "Young Academy of Scotland", have their value when they link as widely as possible. Considering only Wikidata misses the point, it is particularly the links to the organisations, the authorities (ORCiD, Google Scholar, VIAF) but also Twitter like for this psychologist. We may have links to all of them, the papers, the co-authors. But do we provide quality when people do not go into the rabbit hole?
Subscribe to: Posts (Atom)