 Hi and welcome to my talk are qualifiers enough context compatible information fusion for Wikimedia data. So this is a joint work with Wolf-Tilo Balker. So what is Wikidata? In the first place, Wikidata is a knowledge base, right? It has items and has statements about these items. So if you asked for terms about Wikidata, you might say something like, okay, it's free, it's linked, it's open, structured, collaborative and multilingual, right? Terms associated with Wikidata. So how is the knowledge actually be represented in Wikidata? It's represented by so-called statements. We have a subject, in this case, earth, and then we have predicate object like statements. So the earth has the English label earth. It has alternatives label like world or the world. It is an instant of a planet, right? So subject predicate object based statements in Wikidata. What can we do with Wikidata? On the one hand, we can go to the website, browse through the knowledge and learn things, right? This is how humans could interact with Wikidata. But we have also the option to use it in an automated way. For knowledge basis, it's called Sparkle, and we can ask structured queries. Something like, hey, Wikidata, what is the earth? And then Wikidata could reply, it's an instance, it's a planet. Okay. And then we might learn something. Another option could be, hey, Wikidata, how old is the earth? Then we might find the right property like the inception. And then Wikidata could say something like, it's 4.5 billion years old. Okay. This is how we can use Wikidata in an automated way, right? The question is now, what do we also know about statements, right? In Wikidata, we have so-called qualifiers. Qualifier, a qualifier is a property value pair which is attached to some statement. So in brief, we could store additional information which we know about the statement. For example, how it was determined, which was a point in time. Is it subject of some theory? Is it distributed by some other guys? For example, has it a reference to some source or was it a site from some source? So provenance information or additional context information about the statement, right? This is what we can do with qualifiers. If we now think about a more complicated example in Wikidata, like, hey, Wikidata, what was the population of Leningrad, then we should at best do two things. On the one hand, we must find out that St. Petersburg was previously called Leningrad and that we select the right city, right? And on the other hand, we have to select the right population here. One way could be to select the last population which was counted for the city. But this does not apply to the context of the question, right? It was called what is the population of Leningrad. And the city was only called Leningrad between 1924 and 1991, right? So what should we do then, right? Select a population within the timeframe. So Wikidata states that the time that Leningrad is only valid in this time span but which population should be used. And our argument here is we should not use some population. We should only use populations which were counted in the same time interval, right? So that the information becomes so-called context-compatible. They were both valid in the same timeframe. And this is the argument for our work and why I'm here. So I would like to talk about context-compatible information fusion for knowledge basis. My previous works were based on this thing. And our basic idea is that a statement is only valid in its context, right? Think about a therapy here. A therapy in biomedicine is only valid within a certain dose, treatment group, duration, and so on. Or statements like names or populations are only valid within some timeframe, right? There might be statements which are valid in general, right? Birth dates. And then you can use them in any context, no problem. But the point is if a statement is only valid in a context, then we should consider these contexts in the information fusion step. So if we combine two statements, then we should ensure that they are valid in the same context or compatible context. This is the main argument of our work. So how can you do it? On the one hand, there are explicit context models, yeah? Then you need to model every relevant condition like qualifiers do, for example, right? We model the point in time, we model the location, the determination method, for example. But then on the one hand, we must know every condition and this can be exhausting. On the other hand, we also need automated rules how we should combine these contexts, right? So for time, it might be simple. Just say it must be valid in the same time. That means point in time must be equal or intervals must be overlap or must overlap. This is one way, but you see if we see how many different qualifiers we have in Vicky Data, think about how many rules we would need to combine them. So it's a very challenging approach. In our work, we focused on so-called implicit contexts, right? The idea is if we have a scientific article and we have statements in it, then the author should at best mention every important context condition in its written narrative. And if we know which statements were extracted from which document, we could argue something like, okay, we only combine statements which were observed in the same document, written in the same article, because the context is quite stable within a short article, for example, right? It's cheaper to implement and to use, but also has a lower quality. In the end, I would love to talk or to discuss the following question with you. Are qualifiers in Vicky Data enough for a reliable context compatible information fusion? I don't know the answer to the question. I think maybe if we think about geospatial or temporal context, right? We could build rules, right? Few statements, same time and everything is fine. But if we think about the different plethoras of combination, I think it becomes impossible to do it. And another point to think about is Vicky Data tries to collect as much knowledge as possible, right? And if we have things which are, I say, let's say we have viewpoints, we have assertions, we have belief and negations. Some might say it does not belong to Vicky Data. But if we think about recent conflicts, for example, or the inception of the earth, then we see that we have qualifiers like the statement subject of some theory and is disputed by some other theory. How should we engage it, right? I don't know, but I think it's important to talk about these problems in Vicky Data. So thank you so much for your attention. If you have any questions, just contact me. We are one of the following channels and I'm open for discussions right now. Thank you.