#Wikidata - ten questions about #Kian

The quality and quantity of Wikidata relies heavily on technology. When the people who develop their tools collaborate, the results increase exponentially. Amir has made his mark in the pywikibot environment and now he spreads his wings with Kian.

I have mentioned Kian before and I am happy that Amir was willing to answer some questions now that has over 82,000 edits.
Enjoy,
      GerardM

What is Kian

Kian right now is a tool that can give probability of having certain statement based on categories but the goal is to become a general AI system to serve Wikidata.

Why did you write Kian

Huge number of items without statement always bothered me and I thought I should write something that can analyse articles and take out some data out of articles.

How is Kian different from other bots

It uses AI to extract data, I have never seen something like this in Wikipedia before. The main advantage of using AI is adaptability. I can now run Kian on languages that I have no idea about them. 

Another advantage of using AI is having probability which can be useful in lots of cases such as generating list of mismatches between Wikipedia and Wikidata that shows possible mistakes in Wikidata or Wikipedia.

Is there a point in using Kian iteratively

With each run of Kian Wikidata becomes better. After a while we would have so much certainty in data that we can assure Wikipedia and other third party users using our data is a good thing

What can Kian do other bots cannot do

First is generating possible mistakes and building a quality assurance workflow. 

Another one is adaptability of adding broad range of statements with high accuracy.

What can Kian do for data from Wikipedias in "other" languages

We can build a system to create these articles in languages such as English since using Kian now we have data about those articles.

Let me give you an example: Maybe there is an article in Hindi Wikipedia, we can't read this article but Kian can extract several statements out of that article. Then using resonator or other tools we can write articles in English Wikipedia or other languages.

What question did I fail to ask

Plans about Kian. What I'm doing to make Kian better. Hopefully we would have a suggesting tool using Kian very soon.

What does it take for me to use Kian

We have a instruction in github you only need an account in Wikimedia Labs

Does Kian use other tools

Yes, right now it uses autolist which makes it up-to-date.

What is your favourite tool that is not a bot

Autolist, Wikidata can't go on without this tool.

#Wikimedia - more #contributors or more #editors?

The foundation of the Wikimedia projects are its people. Whatever effort these people do, the more it generates data. The data may be in the form of text, images, sources, software or statements but it is all about mangling it into information. The question is not so much what has value for us all, as it all has its own value, its own merit. It gets its value from the people who take an interest.

The question is how to generate more merit. How do we get people involved to do their "own" thing. One way of doing this is by not seeking for the perfect solution. Yes, we can do a lot in an automated way. However, this will only get us mostly more of the same and not necessarily more of what is of interest to some.

Consider, there are people who demand better quality. When all they can do is look helplessly from the sidelines, they get frustrated. When you give them something to do, they have a choice; to put up or to shut up. Personally I care about human rights so I enrich content related to human rights. The data is not perfect but I notice improvements. I notice when other people contribute as well. It feels positive.

The problem with many tools is that they are great for what they aim to do but once they get into the grey area of doubt and uncertainty they flounder. Technically the negative results from Kian are perfect. It is just that it does not make it easy for people to work on these results. It is not obvious what result will be enough for Kian.

What we really want is tools that people can use, tools that are as obvious as we can make them, tools that have descriptions and workflows. Tools that do not need nerds or developers to use. Tools that can be used by you and me. Tools that get us more contributors. Contributors that like me work on a subject they care about.
Thanks,
     GerardM

#Wikidata - Tiberius, Modestus and Florentia

Tiberius, Modestus and Flotentia are three Catholic martyrs. There is a Dutch article by that name and Kian has it on its "problems list".

Lists like these "lists of possible mistakes" are necessary. At best they are an evolutionary step between having no awareness and being aware of issues. It would be wonderful when there are workflows for fixing issues in a way that prevents them from reappearing on such a list.

For these three martyrs new items were added and they are linked to the item for a "group of people". Joseph Guislain is in the Dutch article also a museum in Ghent. It is easy enough to add an item for the museum and link him to the person. But will it fix this issue for Wikidata?

Workflows for the issues that we face would be wonderful:

• when done, an item should disappear from a "to do" list
• it should be more obvious what it is that will fix an issue
• when we identify martyrs or whatever, we should involve people who are into the related subjects

Magnus has many tools that have people fix things. They are workflows we could adopt and by doing this make them even easier to use.

Thanks,

      GerardM

#Wikidata - #StrepHit, the package damaged the message

When a good idea is posted, the message of the announcement can completely blow it away.

First the good news. StrepHit has the potential of becoming a valuable tool for new content for Wikidata. It is all about Natural Language Processing and consequently it is all about harvesting facts from text. The idea is to harvest structured facts and provide references for statements and harvest references for existing statements. This is really welcome, it may prove to be important.

For the bad news, the plan is based on a number of awful assumptions that prevent it from being taken seriously at first glance.

The best thing the authors can do is appreciate that what they are building is a tool. A tool that analyses text, a tool that can be trained to do a good job. A tool that can be integrated with other tools. A tool that is not defined by particular use cases or assumptions.

When it runs in an optimal way, it is much like Kian. It runs and makes changes to Wikidata directly. This week it added 21.426 statements with a very high rate of certainty. Problematic data is identified and lists are created and this is where people are invited to make a difference.

Kian works in the Wiki way, it does its thing and it invites people to collaborate. It does not assume that people have to do this that or the other. Contrast this with StrepHit where the author suggests that people should not be allowed to add statements without references. If that is not enough, it will not even add data to Wikidata but considers the data it generates a "gift" and condemns its data to the "Primary sources tool". It is a sad place where valuable data lingers that is not finding its way into Wikidata.

StrepHit and tools like it may become valuable. Its value will be in a direct relation to how it integrates in other tools.  When it does it will be great, otherwise it will sit in its corner gathering dust.
Thanks,
      GerardM

#Wikidata - Kian and #quality

Last time Kian was very much a promise. This time, after the announcement by Amir, Kian is so much more. Kian is a tool that can be trained to identify items for what they are. Training means, that parameters are provided whereby the software can act on its own and based on likelihood will make the identification or list it as a "maybe".

Obviously once it is known what an item, an article is about, so much more can be deduced. That is something Kian will do as well.

The thing that pleases me most, is that Kian for its learning makes use of autolists, it means that Kian became part of the existing ecosystem of tools. Eventhough hard mathematics are the background of Kian, it is relatively easy to train because prior knowledge is of value.

In the announcement mail Amir asks for collaboration. One area where this will be particularly relevant is where people are asked to decide where Kian has its doubt. It currently uses reports in the Wiki but it would be awesome if such questions can be asked in the same environement where Magnus asks for collaboration.

Yes, Kian makes use of hard scientific knowledge but as it is structured in this way, it makes a real difference. It is possible to learn to train Kian and when ambiguous results can be served to people for a result, Kian will be most glorious. Its bus factor will not be Amir.
Thanks,
      GerardM

#Wikidata - Kian dehumanising disambiguation pages

Logic has it that a disambiguation page like the one for Hristo Nikolov is not a human. There are at least two of them. There are more of them but this disambiguation page does not know about the others..

Hristo Nikolov is the first item of a recent list produced by Kian. All of them include Wikipedia disambiguation pages that have been recognised as a human. It takes attention to fix the issues involved. It includes linking Wikidata items to the correct article and, it means removing and adding statements.

Once all items of this list are done, quality has improved. We can turn over a page and go into "maintenance mode". When this report is run weekly or monthly, there will be only a few new cases. They can be fixed quickly and we can be more confident about the Wikidata quality.
Thanks,
      GerardM

#Kian does #Kannada #Wikipedia

Mr A. Surya Prakash is an author and journalist.There is an article about him on the Kannada Wikipedia. The item on Wikidata exists thanks to Kian. It is one of 48 articles created based on articles. These 48 items are identified as human.

Using Google Translate it is easy to find the English name. Finding additional information about Mr Prakash was easy as well. Several awards he received are not known in English. I added one in Wikidata. Many of the things I did I could have left for Kian to do in a later run. It is however so cool to be able to do share in this knowledge. One thing I did was to merge the Kannada category for journalist.

I asked Amir to have Kian run for all the languages from India. It finished with 500 new humans for Hindi, my latest news is that it is running for Tamil. What we know from the Kian run on the German Wikipedia is that most items have been already been identified for what they are. I found that for only 11 people I could add that they were a journalist.
Thanks,
       GerardM

Categoria:Morti a Rottenburg am Neckar

The category on the Italian Wikipedia has been expanded with statements that help Reasonator find all the people who died in Rottenburg am Neckar. This was done by bot and it was done for the categories who identify the place of death.

This is awesome! Not only do we find all the people who are included in the Italian category, we also find all the people who have been  identified based on information from elsewhere. Caspar Adelmann for instance only has an article on the German Wikipedia.

With Kian we have a tool to make the most of information that exist in our projects. The relevance of tools like Kian is in sharing the information of Wikidata. The value is in its use. Reasonator is how we add value to the data we have and it is wonderful to see how things become connected.
Thanks,
      GerardM

#Kian - Adding humans known to #German #Wikipedia

The first feat of Kian was to recognise humans based on articles from the German Wikipedia. Now that is the holy grail because once you know that an item is human, it is trivially easy to make all kinds of assumptions.

It was done by learning from what is already there. Given that for 90% of all Wikidata items it is already known what they are, there is plenty of learning material for Kian.

Once everything was said and done, 3600 articles are now known to be about humans. Given that much of the learning is based on Wikidata data itself, it follows that languages like Malayalam or Chinese can be tackled in the future. The future is bright and we happily welcome the first 3600 humans Kian recognised.
Thanks,
       GerardM

#Kian - the future is bright

Send someone bright to a course on artificial intelligence, someone with a pedigree of great bot-work and you get "Kian". In Farsi it means "glory" and,yes I have seen the light.

Amir introduced Kian in an e-mail: "Kian is a three-layered neural network with flexible number of inputs and outputs. So if we can parametrize a job, we can teach him easily and get the job done." If I understand properly what this means, I will stop doing most of the work I have been doing on Wikidata with AutoList. I will not reach three million edits and I will be ever so happy.

Amir's example for a first job: Add P31:5 (human) to items of Wikidata based on categories of articles in Wikipedia. The only thing we need to is get list of items with P31:5 and list of items of not-humans (P31 exists but not 5 in it). then get list of category links in any wiki we want[2] and at last we feed these files to Kian and let him learn. Afterwards if we give Kian other articles and their categories, he classifies them as human, not human, or failed to determine. As test I gave him categories of ckb wiki (a small wiki) and worked pretty well and now I'm creating the training set from German Wikipedia and the next step will be English Wikipedia. Number of P31:5 will drastically increase this week.

I am ever so happy because once we know what an item is, it becomes easy to make all kinds of inferences.
Thanks,
      GerardM