 Do you hear me? Actually, yeah. Excellent. Well, ladies and gentlemen, hello again. Well, we'll be having a session, a brief panel session actually, that will be occurring for one hour about upgrading medical practice with Wikidata. So this is a part of the project, Wikimedia Research Funded Project. You can scan actually the report for that project on the right. And that project is a collaboration between Wikiproject-made data engineering at Semantec's research unit from Tunisia and the University of Virginia School of Data Science from the United States. Yeah, so that's the name of the project, just for your information. And well, just to reveal how the panel session will be occurring, actually we will have 20 minutes to present with the state of the art of medical Wikidata and then we will have the 35 minutes of discussion between the panelists regarding the state of the art of the usage of Wikidata in healthcare. Yeah, so now we will have a brief tour of introduction of the panelists. So we were supposed to have Daniel Michen, but he was not here due to some circumstances, but we have myself, Laine Raspberry, Thomas Shafi, James Heyman, and Mosa Bennett. So we'll get everyone to have a brief introduction of himself, we'll start with James. Perfect, James Heyman, I'm the chair of Wikiproject Med Foundation and I've been involved with working on Wikipedia's medical content since 2007. I'll hand it over to Mozab. My name is Mozab, I'm a medical doctor. I am also part of Wikimedia Medicine. I'm a membership admin and I have been contributing to Arabic Wikipedia since 2013, mostly in Arabic content, medical. Yeah, so my name is Thomas. I've been contributing, I guess, since about 2014. I don't have a medical background. I'm a biochemist originally and subsequently a bioinformatician and data scientist. But my interest is in, I guess, academic Wikipedia collaboration and ways to bring in Wikimedia contributors from those sorts of demographics that don't normally contribute, but could. My name is Lane Raspberry. I'm a Wikipedia at a university in the United States. I do medical information and content related to LGBT people and LGBT topics. Yeah, so I am a husband in Turkey. I used to be a medical student. I am still a medical student at my home city in Sfax, Tunisia. I'm also involved in Wikimedia research, particularly computer science research and have been involved in creating our Wikimedia research structure at my university, data engineering and semantics research unit. So I have served in several positions, including the Wikimedia CUT of last year, the board of Wikimedia Tunisia as well. So, all right, after this introduction, let's go straight to the point. Yeah, so as you have seen, we are all Wikimedians, but not only Wikimedians. So we should benefit from every single open resource within our movement. And this is not what many Wikimedians think. Well, why we should actually look at Wikidata? Well, it is structured ontological database. Having statements in the form of triples, it has a multilingual support, semantic unlimited to external resources, plenty of tools for a person at a rich data base, and it is linked to Wikipedia. So we can play with it in a variety of shapes. So, it is a rich network of medical knowledge and it can be easily extensible to cover new topics. For example, during the COVID-19 pandemic, there was nothing by the early month of the disease, but just in four months, look at that. We have actually a whole taxonomy about COVID-19 related topics. We have the three pillars of the graph, the COVID-19 pandemic, the disease COVID-19, and the virus, which is SARS-CoV-2. And we have all these three items, actually major items, are related to themselves and actually they are related to variety of medical classes and non-medical classes. We have genes, proteins, taxons, actually humans who contracted the disease, kind of symptoms, vaccinations, drugs, et cetera. And that just in a few months. However, although this is a huge achievement, actually, UkiData, mainly when we talk about medical information, has several limitations. And that's why it did not go to the next stage of using it in developing medical applications. The first limitation is that medical statements and UkiData have limited number of references. Some statements are not imprecise. For example, in this example, you have the sign of appendicitis as abdominal pain. It did not show which part of the abdomen actually is the place where the pain is. And actually it should be the right lower side of the abdomen or at the early stage it can be gastric pain. So that lack of actually precision can actually let people not take seriously the database. Also, we have seen that a lot of external resources are not fully important. And when we talk, for example, about OBU ontologies, the open biomedical ontologies, we find that 40% of these ontologies are mass important to UkiData. Although they are open-licensed and there is no legal constraints to actually download, upload this information to UkiData. Yeah, several types of information are still not supported. For example, where diseases are located. For example, for the Ebola hemorrhagic fever. We have them up on Wikipedia, but actually this is not converted into statements in UkiData. Also, several medical classes do not support actually validations schemes. For example, we have no entity schema, no data model, and no entity schema for many important medical classes in UkiData. For example, we have no entity schema for surgical therapists, for example. And this will lead people to define the thing in a different layout leading to some scattering, some inconsistency, and so on. Also, some kind of relation types are misused. For example, Besheh disease has phenotype uveatus. Probably this is because there is no, the relation type that should be used is not in UkiData. So there are a number of cases that are misused in UkiData. So there are a number of cases that are misused in UkiData. This is not in UkiData. So they are doing some approximative work to actually get the closest relation type and use it instead. So here are the problems. So we need a plan for action. So what we can do, actually we can raise awareness about UkiData as an open medical resource, and you can arrange and validate medical knowledge in UkiData. Finally, we can create tools to reuse medical knowledge in UkiData. How to raise awareness about UkiData? That's pretty simple. Actually, there are several works in this context. So first, we can make UkiData useful for the UkiMedia community. For example, doing like supporting info boxes for medical articles in UkiPedia, kind of for travel guides for UkiVoyage project, if they are traveling to epidemic countries, et cetera. Also, we can share UkiData with the scientific community and show in how UkiData can be a flexible hub for the semantic research and how they can be useful actually for scientists to develop their ideas, et cetera. Now, let's show how to arrange and validate UkiData in a variety of paths. Yeah, so actually one of the main ideas that we propose to do is actually to make a framework of interaction between UkiData, open biomedical ontologies, and PubMedMesh keywords. For those who do not know, mesh keywords or medical subject headings are controlled keywords that are existing in PubMed. So the association of mesh keywords in a single article can mean that the two words that are co-occurring are related. And so using this combination, we can actually generate semantic relations just by leveraging PubMed. Concerning open ontologies, many ontologies are aligned to UkiData. We have now 15 ontologies that have operational UkiData properties. And so they can play on that, actually ameliorate the coverage of these ontologies, and work more of that. As well, we can actually use PubMed to get references to unsupported UkiData statements. We can verify several relations in UkiData by comparing them to external resources, et cetera. So all kinds, actually, of interaction between these three sources is possible. And we can play more and discover more things and actually develop that, et cetera. Yeah. Finally, concerning the creation of tools for reusing medical nodes in UkiData, well, this is mainly not done actually to take the place of the physicians. Many people actually understand biomedical informatics wrong. Actually, it is there to help physicians to do the dirty work, the dull work, the dangerous work, the dear work, et cetera. So what we mean here is that physicians have serious work in conditions, and they need to go through night shifts, et cetera. Lots of information to handle at the same time. Patient data are not like in the book, as we may say. So that thing actually cannot be handled by the doctor without any help. So he should have some resources, some tools to help him around this mission. And even artificial intelligence will not take over the physicians even in 100 years because there is even machine learning models will not deal with rare diseases where data is not available. So we will always need physicians at medical specialist. Yeah. So a few examples about how actually to use UkiData in this context. Actually, we can use UkiData actually, mirrors of UkiData, to actually annotate electronic health records and actually for doing some medical reasoning about the patient's status to do some prognosis or to revere drugs, et cetera. So that's the first application. The second application is that we can actually give UkiData kind of some information, anonymized information about patients. And UkiData can provide us the answer about how to deal with the patient. For example, if we give the list of the drugs for a given patient, it can return to us the drug interactions, the symptoms and you will have the medical diagnosis or the complications, et cetera. So this tool actually is called medicine. It exists. We have developed it. And if you scan the QR code on the right, actually you will get into the URL. You can test it yourself. It is multilingual and that's an advantage about clinical decision support based on UkiData by contrast to the ones that are on the market and that are unproperatory and mainly monolingual. And if you click, if you actually scan the QR code on the left, actually you will get all the source code and it is open source and you can easily figure out how it works. Yeah. There is also the COVID-19 dashboards. I think most of you have learned of that. Actually, what we do is that we enrich UkiData with real-time information about COVID-19. And actually, we have some predefined Sparkle queries. And what the dashboard does is that it gets the information as soon as you access the tool, the dashboard, at this actually information is updated in near-lead time. This means as UkiData is updated with new information. And by this fact, actually you don't modify your tool but actually as it gets information all the time from UkiData, you will have it there and it will be updated automatically to give you all the consistent information and recent information that can be used by physicians or by any person who is interested in medical code topics. Yeah. So these are all the references that I have used for this research. Most of them are developed by us, the panelists. So you can see in both our names in some of the papers. Yeah. And now, actually, let's jump to our panelists actually to discuss more about actually how to move forward with the application and the use of UkiData in medicine. So the question number one, what are the good examples of UkiData benefited in patient care or public health on a local or global level? Anyone to answer? Okay, Hassan. So during our session, we talked about medicine. If you remember the previous slides, medicine is a tool that is used to see the drug interaction. Most of us, like even physicians, there's hundreds of drugs. Not every physician can remember or memorize by heart every single interaction. So, for example, if you go to the tool and you put, for example, reef and basin, a drug used to treat TB, tuberculosis. If you press to find the interaction, you will find that it interacts with paracetamol, for example. So this is one of the fees that UkiData can help physicians in finding out what's the drug interaction. Other thing that we work together with our friends and panelists and also Hassan, we worked in the association between signs and symptoms and disease manifestations using UkiData items. So this is also another example. After making signs and symptoms more specific like the example Hassan gave about appendicitis and lower quadrant pain, this will help, for sure, it will not replace the physician because medicine is not one plus one equal to two. It's more complicated relationship. But for sure, when we give the computer or give the database a bunch of symptoms, it will help us to find at least a smaller list of differential diagnosis that will help the physician to then figure out the correct and exact diagnosis. So this is two points that I can make to this question. Anybody can think of something? So this is not necessarily an example of what's being done exactly right now, but it's an extension of what you were just talking about. So the way that UkiData handles drug interactions is not fundamentally different from the way in which it can handle other interactions. So for example, there are certain drugs that interact poorly with the parasite of humans called pregnancy. But also there are other conditions that can interact with drugs. For example, sleep deprivation can have effects on some drugs or even certain things like talking therapies or talking therapies interacting with other types of talking therapies. So the way in which UkiData is modeling drug interactions is extensible to a wide range of other relevant interactions outside of just the narrow scope of pharmacology. Exactly. And for those who are not from a medical background, I want to explain that why is pregnancy is called a parasite in a medical point of view because the fetus is dependent on the mother. He takes the food and he shares the blood with her. That's why in a medical point of view, we look at the baby as a parasite. Yeah, for sure, in the future. Thanks, Moza. So, you know, one of the other common uses of UkiData within the Wikipedia world is the use of UkiData items within info boxes and that allows a bunch of languages to collaborate on keeping certain topics up to date. You know, one of the unfortunate limitations of that is that, you know, the English Wikipedia hasn't bought into the use of UkiData at this point in time. And some of the concern around that is, you know, we don't quite have good mechanisms to feedback changes that occur within UkiData to the Wikipedia's themselves. And, you know, they're still working on efforts to allow UkiData to be used outside of the Wikis run by the Wikimedia Foundation. So hopefully, you know, from what I understand, the Wikimedia Deutschland, which is running Wikidata, is working on both of those problems and hopefully they'll expand the potential of these info boxes to be powered by Wikidata items. Yeah, I think so. Yeah, the question, member two. So what are the main pros and cons of UkiData as a central medical knowledge hub versus the traditional databases we discussed earlier? The advantage of traditional databases is the authority behind them, the control, the high quality that you get of not letting anyone edit. Of course, the problems with these kinds of things, they come with traditional values like closeness, inability to export and reuse to adapt to do translation or to move into other platforms to remix in different ways if it's a traditional database. Sometimes it's not positioned to remix and combine with other databases when that's necessary. When you have something like Wiki, anything's fair game. You can combine these things. You can translate them. You can remix them. You can export them into other products. Disadvantage with Wiki, of course, is maintaining the quality control. But in every talk that you hear at this conference, people are going to say something about quality. Quality is definitely a big concern for the Wiki community and in my opinion, it gets much better every year. I mean, another thing that a lot of people in this audience can talk to a lot better than I can is the interlingual aspects of it. I'm a native English speaker and as such, I only speak English, so I can't really talk about that. But I am interested in interdisciplinary research and one of the limitations with a lot of existing databases, they're hyper-specialized. You go to a medical database. It's only got information about medical topics. You go to a biochemistry database. It's typically only got biochemical topics. You go to an evolutionary database. It makes it hard to ask interesting interdisciplinary questions like which genera of animals tend to produce peptides that tend to interact with receptors that are associated with pain. Which is an interesting research question from the point of view of, which animals should I be testing to try and find new painkillers? That's the sort of question that is already actually still quite narrow in scope. There are far more interdisciplinary questions and asking questions across multiple topics. The WikiData is, to the best of my knowledge, the only database that even comes close to being able to have these sorts of hyper-diverse questions asked of it. Yes, and also other point is that the accessibility. WikiData for sure is a free and accessible database, but for other specialized databases, you need to pay monthly or yearly a certain amount of payment to get access to this. This is another thing. Also, it makes our work on Wikipedia easy because if I edit on Wikipedia and WikiData, a number, for example, a statistical number or I added some symptoms, it will be reflected on more than 300 languages, which also makes it a broader audience than some sort of special databases. Next question. Indeed. The next question is, what is WikiData's role in enhancing situational awareness for emergency responders and medical teams? That's a question. I will read the question again. What is the WikiData's role in enhancing situational awareness for emergency responders and medical teams? Okay. I will try to answer this question as best I can. As far as, to my knowledge, when we deal with sometimes with emergency situations, one of the things that we want to know sometimes when there is a poison, for example, was the antidote. And when there is some medication taken, what is if there is a medical interaction? For example, like sensitivity reaction, like anaphylactic shock and these things. If I think this is an area that WikiData can help if it is developed in a more simple way, for example, emergency and the first aid sometimes are provided by a layperson, not even a specialized person. So if our tools become more simplified, that would help people because when we are talking about coma or losing consciousness, there is sometimes medication as the cause, sometimes there is blood pressure as the cause, there are differentials. But if it is medication, for example, we can use WikiData to find out what's the antidote or if it's a scorpion bite, WikiData can tell us there is something called antiscorpion. If it is like, for example, drug toxicity, it can tell us what to do. Is it like activated charcoal? Is it like lavage? Is it like a certain antidote? So I think this is an area that WikiData can help. I'll answer the question. And also, I guess there's been an ongoing topic throughout the whole of this conference around the increasing role of AI, and I suspect that this is exactly the sort of situation that there's going to be an increasing reliance on AI in emergency situations, asking some AI assistance, not necessarily for entirely what to do, but maybe for some of the details of what to do, particularly from non-specialists or people coming across a particularly complex situation for the first time. A classic problem of current AI models, they hallucinate, they come up with false information, but they present it very confidently. And there's a lot of interesting work going on around how do you anchor those large language AI models to reality, and this is a more complex topic than just WikiData, but WikiData, I think, is a part of that that allows you to anchor some level of reality into these sorts of models. There's something else that's relevant to this conference, relevant to interdisciplinary multi-linguality. So many things are happening so quickly, it's amazing. I'd like to call out, disasters always happen, crisis always happens, it can happen for weather or epidemics or any other kind of disaster that you can imagine will never be done with these. And you need all kinds of diverse data every time a disaster happens. There's people at this conference from a project called OpenStreetMap. They make free and open maps. And if you have a crisis somewhere, you need high-quality open maps of the region. And this starts with interconnectedness with Wikipedia. So outside of our movement, but very much related, there's something called the Humanitarian OpenStreetMap team, and every time there's a crisis, people map out the region where the crisis is so that medical responders can go in. And often this starts with something seemingly silly and unrelated in the Wiki movement, like what is the public art? Children of natchoos are the artworks in the places when the crisis. But if a crisis happens, that artwork and those Wikipedia articles about the monuments, those become landmarks for other people to use to orient other things. In an OpenStreetMap, in the same way that Wikipedia has collaborative editing, OpenStreetMap will say where are the grocery stores, where are the buildings, that we can house people in case of a big storm, where people suddenly need housing? And this interconnects. OpenStreetMap in so many ways interconnects with Wiggy data. you've heard of this conference this is just a fraction of it because we're working with so many other open data communities and this is getting very complex and developing very rapidly. And actually just to add to that I don't do a huge amount of work with open street maps one of the things that is really impressive is also the speed at which it's updated you come across disaster scenarios well the map that you were relying on a month ago may not be accurate anymore you know this bridge no longer works these roads are no longer traversable this landsliders taken out a particular a particular road a particularly important other piece of infrastructure and the speed at which that information is updated is exactly like you know our wiki movement it's highly relevant and highly timely. Yes and also for wiki data we are talking about wiki now but what about the future we always hear about image processing for example in medical field radiology there is a there is like a saying that nowadays says after maybe five ten years the radiologist will be replaced by you know AI because imaging process by computer to a certain existence can help like for example lay people to know what is what is happening in this picture is it for example ebidural hemorrhage is it is it like you know berforated bowel what what what is the situation so it can give us information yeah next question yeah as James James you know just one one extra addition on the open street maps topic one of the wonderful things about their project is it functions so amazingly well in an offline environment and you know that that is key for a disaster is you know people need you know because during a disaster often people lose their telecommunications infrastructure they lose their electricity power infrastructure so having offline functionality which wikipedia does which open streets maps does but which all the for-profit organizations out there simply don't so we will head to the next question and yeah so the next question is in what ways does the utility of wikidata differ between common and rare diseases okay so for example for the common diseases information are plenty we are talking about rare diseases the information are markedly less than common diseases also the language difference is very big so when we are we are searching about rare diseases in for example in English we will find good information in comparison with like when we search about this the same rare disease in arabic so i think wikidata role in in in the trans lingual translations can help to bridge this gap because even when you talk about signs and symptoms um if we search in arabic for for a rare disease it's it's it's sometimes there the content is zero and that's what we are trying to do in in rich arabic content on wikipedia to to make this data accessible to wikidata and to make it in different languages that will help bridge the gap i mean even though the question is phrased in terms of common and rare diseases i think there's also a similar issue that comes up with diseases of the rich and diseases of the poor and the same sorts of biases exist in terms of the amount of content that is out there how in depth it's documented how accessible it is and you know the multilingual aspect as well so these are kind of very linked topics it's not just a question of the common versus rare aspect there can be extremely common diseases or common conditions that are treated very differently one of the things that wikidata still also doesn't do particularly well is treatments outside of uh pharmaco interventions so so drug interventions when we're talking about other sorts of interventions surgical interventions as an example that's come up but also various different types of talking therapy for psychological interventions right if you look at a wikidata item on a pharmacological intervention for severe depression those items are going to be far more in-depth than the equivalent items on a talking intervention and in part that's maybe because the the structure of the information and the data on a talking intervention is maybe a bit more complex it sometimes requires um more subtlety more nuance um more qualification in the way that that information is put up but it's also a reflection of the existing data that's out there and the current interests of the the community so i don't think it's a fundamental problem it's something that's going to improve over time i'm gonna say this from a perspective of kindness this is a non-medical perspective and i i i'm going to tell you something about the wiki community so if somebody shows up and they have a rare disease and they come to wikipedia or wikidata and they say i'd like to get more information about it they're going to be approached by wiki community members who are not have no background in healthcare who don't know anything about the disease i've never really even care about the disease but we have a community of people with library skills and they want to help people answer any kind of questions and i tell you tell you truthfully and i mean mean this sincerely in all kindness they will make a game of the rare disease these wikipedia editors are very proficient in library research and they will go out into the world and they will try to find every source of information that has ever been published on this rare disease and when they do this they will feel pleasure from this they will think it's fun they will think it's a game to find all available information so if you're talking about a very common disease it's less fun because there's tens of thousands of papers and tens of thousands of sources of information but if you're talking about a rare disease in our wiki community people get a great amount of pleasure to find the 10 sources in the world that talk about this and then deliver those to the person so that's something unusual that you will see in the wiki community and you can't play these kinds of games if you have a closed database or a closed community yeah so the last question actually how can healthcare providers be motivated to contribute to creating wiki data and maintaining data quality all right well I guess I've got a few things to say about that I mean one of the things is as many people have mentioned visibility within wikipedia wikipedia has huge visibility there are very few people who you'll come across who haven't interacted with wikipedia in some way and even though the reputation of wikipedia used to be extremely poor amongst academics researchers physicians clinicians I think that that reputation has been improving pretty rapidly and that's you know due to concerted work of the community in terms of quality referencing timeliness in-depth thoroughness etc etc all topics you know about in terms of wiki data it's not well known I mean people at this conference will typically know about wiki data but I think it's pretty knowledge about it is pretty limited outside of our community and so part of what can be done is I guess a boring answer of awareness raising under you know reaching out to practitioner communities who could find contributing to or drawing from this sort of data useful and interesting but simply don't know about it yet and the other aspect I think is working out systems of reward for it so people who know me know that I do a lot of work with a project called wiki journals now this is mostly focused on the wikipedia pros type content it's a way to try and draw in more academic expert professional etc communities into contributing content that can be put into wiki media projects by effectively effectively rewarding them with peer reviewed citable publications now that model currently works well for pros type content but I don't think it's impossible for it to eventually also work for curating a subsection of wiki data type content so you can imagine a scenario in which some expert on a particular drug class is asked the question hey we'd love you to go through subset of wiki data to do with this drug class is the information up to date are the references the best references to use are there other interactions that you've come across that will be interesting are there are other qualifiers that could be added that we're not including you're an expert in this topic but not an expert in wiki data so we can pair you up with someone who works in wiki data and we can work with you to write up the improvements that you made the observations that you made in a small mini paper few you know a few pages long but the ability to then publish that get it peer reviewed get it get a citable object means that you can put that on your CV and justify the time that you've spent you know that then means you can justify that time to your institution to your promotions committee to the next job you're applying for for the next grant you're applying for so those systems of reward can be really important for people who have extremely limited time and unfortunately the the systems that are set up currently have very strict definitions of reward for different types of activities and if your activities don't fit within those particular categories I'm sorry there's no reward for doing that so I think there's an extent to which we can make that interface between the wiki media world in general and wiki data in particular compatible with the ways of working that some other expert communities of potential contributors work but but don't yet exist and you know to add on that point as as a non-technical expert wiki data is difficult you know I think in pros as a physician I don't think in structured data so you know my work on wiki data always generally requires collaboration with someone who has the technical expertise to run scripts to you know upload the the data that I want uploaded to wiki data to you know when I propose new properties to make sure of you know I've structured it correctly so you know if we're wanting to pull an academia we need to form these partnerships between those who have the technical expertise and those who have the conceptual conceptual and and subject matter expertise and this is going to require pairs of people and I think that keeps a lot of academics away you know we within wiki data within the wiki media movement you know I know who to poke and who I can whose arm I can twist to get me you know get what I want into wiki data but as a new person to the movement it takes a lot of time to build those relationships and find people who are willing to help you on these technical projects and you know having a place within wiki data where people who are willing to volunteer and join experts in in solving their issues I think would be would help our movement grow over the lane so these two have talked about telling organizations during our reach about wiki data I'm gonna have a different perspective on this don't tell organizations about wiki data at all don't get them to engage what we really need to do and where the better fight is is just convincing people to have open content period because if people will use open access licenses for their papers apply open licenses to their data sets they don't need to understand wiki at all there's plenty of wiki people who will migrate their content into the wiki platform and universities and research institutes already understand this concept of openness if you just tell them open source software open access licenses open copyright do not make something private unless there's a reason to be private I'm not saying release personal information but if something ought to be public anyway then just give it an open license and the wiki people will take care of the rest I'm gonna do something I did earlier and answer a different question that wasn't actually asked which is kind of the but I think it's the mirror image to this so this is about you know contributing curating maintaining wiki data the other aspect is also getting it easier to pull information back out of wiki data and there have been some interesting early experiments again to be kind of a techno optimist in this with around AI and the ability to bypass the need to know sparkle the querying language in order to ask questions of wiki data so the example I mentioned earlier from this kind of multidisciplinary point of view about you know which which evolutionary groups of organisms produce peptides that might be that interact with receptors that are related to pain to be able to actually ask that question of wiki data is actually still quite complicated you need to be able to to write spark or code and even though we do have tools to help out with that it's still not trivial there's a significant barrier to entry and so you know you talked about pairing up people with subject domain knowledge and people with wiki knowledge I think that they will increasingly be a third member of that marriage which will be you know the machine tools to be able to facilitate that partnership so that we can for example ask that question just in plain language have the sparkle career written behind the scenes by an AI and maybe it has to be checked over by someone with with more wiki data knowledge but at least it's a first approximation to be able to being able to ask that question without having to have quite such a level of technical knowledge we have five minutes anybody has any question please raise your hands yeah I can why else some people take actually the time to reflect off questions I can make a brief clarification about what Thomas said actually the medicine tool that we showed earlier aims to bypass the sparkle thing actually you just need to choose using the elastic search using the search part that of a school year tool the same search bar we just get it and you search for the items the medical concepts and your reason about that you say you search for drug interaction diagnosis association complications and you click and the tool does that for you it writes the sparkle query and it generates the answer and that's all what it does yeah any question yeah thank you very much it was very interesting to listen to yeah how far we are going with wiki data and and I think it's there's a lot you're talking about this from a medical perspective but if we kind of try to expand this it's really huge what can be done there what I am particularly interested about is if you could tell me about your research journey like you have research questions you start with those questions right then you probably say I need to prepare the data to look if it's healthy if it's enough if it's probably I need to upload some data and some cues and piece probably to define some if they are missing assume you are having that you have also good references and you have probably common image which are linked to the items and how does it go after that like are you your models or your research is focused on the text or on the relationships of the data that you have or you are more and more using probably image classification or AI based tools to do kind of more out of the data that is there so I am not sure if the question is clear but I am interested to see how deep do you go on the data and in particular the model selection you use to come out with your research because I did not see the research yet but probably you could give a brief answer about that thank you yeah I can answer that actually well you are right actually actually when you are collecting data about a given disease actually you should formulate a research question you say that I know I want to know the semantic relations related to that thing to a disease to a drug etc and so what we would do is that you formulate the query at the bibliographic database which is we permit in our situation and so using kind of a prisma method but it is not totally prisma you can you can actually search for the the articles with higher level of evidence that are not retracted they are quite recent etc and then from that after you do that preprocessing actually you get the mesh keywords and in that part a second part actually begins in fact the mesh keywords have have qualifiers and the qualifiers actually informs you about the aspect of of the keyword that exists in the paper for example if you see a hepatitis c drug therapy it means that the paper is about drug therapy of hepatitis c so these qualifiers are predefined so we know them there are around 75 ones and so using kind of matrix of correspondence between the qualifiers of couples actually you can train machine learning model to recognize the relation type based on wiki data so that that's it is as simple as that then you will have the human validation step to actually reveal if this is correct or not yeah just very little extension to that do you use classical AI or you are focusing more on the deep learning models yeah because if I think about the amount of data probably it could be huge not that huge actually the relations between between mesh keywords in wiki data are are around 100 000 that's not that huge actually so and we are using concerning the models we are using actually three models one is svm support vector machine one is the cnn convolution neural networks and one is the dense model so quite state-of-the-art models okay we have the last minute any last question thank you very much for the presentation it was very informative for someone coming from a non medical background I just have a very simple question the tools are excellent we can see their their use and their added value to the field but how can you avoid having like avoid any tiny risk of having vandalism on wiki data because the information is very sensitive so if one thing is changed or misleading maybe it can goes that are some very harmful consequences thank you the best answer that I can give you is that there's 20 years of criticism of wikipedia and how vandalism works there's hundreds of papers written about this everyone in every country in the world has come up with this like every three months for the past 20 years so you read that and then you get a feel for the quality in wikipedia and at this very conference there's multiple talks of people developing artificial intelligence bots that are doing mass surveillance on every aspect of wikipedia to try to detect vandalism I will not tell you that the quality of wikipedia is good I'm not going to argue that I do think it's good but I'm not gonna not gonna try to impress that on you what I what I will tell you is that nobody in the world has identified an accessible source of information that is better than wikipedia I'm absolutely sure of that I guess one small thing to add is also just the number of eyes on on information is important because I actually think the the greater risk is missing information right if I'm standing here with you know an injection ready to give someone and I'm relying on some tool that tells me you know oh you know what are the contra indications when should I not give this injection and there's some information missing and I go fab time to give the injection like that's I would say a more common problem than someone going into wikidata and changing a piece of existing information it's it's actually going to be the the mistaken lack of information rather than a deliberate piece of vandalism yes for example for example that is a study in nature comparing the accuracy of wikipedia comparing to to encyclopedia britannica it founds that wikipedia is one and a half more accurate than encyclopedia britannica although encyclopedia britannica all know it's written by experts so we still have a trust in wikipedia we have we can argue academics in that but it is it's a long journey I think with time with time this problem will be solved and you know there doesn't need to be more checks and balances with respect to wikidata and you know my hope is that those will continue to be developed you know one of the key aspects is you know if if a change occurs in wikidata and that property is used in a wikipedia then then that change should appear in someone's watchlist so that then you have a greater number of human beings reviewing changes that occur to wikidata currently that doesn't work very well I know that is on wikidata's efforts but they have a backlog of things they want to do so it continues to be a work in progress one other aspect is also something that is incredibly baked into wikipedia's culture and wikidata is still catching up on references citation citations citations there's still a disturbing number of wikidata statements that are unreferenced and that's I think also a huge problem because that that also raises the barrier to vandalism and make can make vandalism easier to detect in the long run that if the it makes it easier to spot hey this statement either doesn't have a supporting reference or I can go to that reference and very quickly check to to verify it and so the the way that wikidata has so far evolved has led to a lack of references for some of this information that's changing pretty quickly because there's been a concerted push in that direction but I think that's still a current significant limitation thank you very much actually this is my contact so if you can if you have any further questions and you need to reach out to me these are my contacts you can reach out the phone is actually available on whatsapp and telegram you can reach out to me and there are also on the back there are my business cards as well as the business card of the head of research for my institution so you can grab it and actually contact me thank you