 I'm very excited to present our next speaker, who is a research professor for information visualization at the Potsdam University of Applied Sciences in Germany. Motivated by the design opportunities and research challenges arising from growing information spaces, he's particularly interested in the potential of visual interfaces to support exploratory information practices. Please give a warm welcome to Marian Dirk. Alright, thank you. So I think this talk will depart a bit from the last two presentations. I'm, as Arina was already mentioning, I'm really interested in using visualization to support exploratory information practices for not just quantitative data sets, but rich collections of data that may include text, imagery, maybe video. And this talk is going, kind of, promoting, so it's kind of, here and there, there might be a tone of antagonism, so hopefully this will not offend you. But promoting a slightly different mode of using visualization to explore rich data spaces. So before I do that, a little, sort of, premise. So if you go along, I'm going to use the analogy of the city and how we are engaging with the city to talk about how we engage with data. So if this is not too cheesy or too cliche, please let me proceed with that. The reasons are also that, actually, there's lots of commonality between cities and rich data sets. In the past, and still so in certain regions of the world, cities are growing in size, they are culturally significant, they have become the cultural backdrop for everything that we do, and they are also struggled in these places. And we can say the same about data today. We have growing data sets, they are becoming sort of the, especially with social media, they are becoming the backdrop for everything that we do, and there are also conflicts in terms of access to data and data poverty and so forth. Now to think about how visualization works and comparing it to the city, I feel that most visualizations, most data visualizations take an overview approach, a bird's eye perspective on the phenomena that we want to explore. And I think that's fine. I think we have seen wonderful work during this conference starting with a billion dots on maps to wonderful stuff today by the New York Public Library. So I have nothing negative to say about this, but I would suggest, and I would try to argue over the next minutes, that this primacy and that sort of overview first dogma is maybe problematic for certain types of experiences and for certain types of data collections. So for example, well actually, first before I go into maybe the alternative, there are many examples I'm going to go into these now, but for example, the Angram Viewer by Google allows us to compare how certain certain words are used differently or have been used differently in books over the last decades and centuries. And so I've plotted here the varying use of micro versus macro. If you pay close attention, micro has been catching up recently, but these types of tools are very useful, very powerful, and they are sort of promoted and they're promoted to sort of allow us to grapple with big data challenges in the sense that they allow us to provide big picture perspectives on these data sets to find patterns and relationships across time, across space, and so it's great stuff. But when we deal with data visualization in contexts, in more cultural, maybe literary contexts, when we are visualizing text collections, image collections, then I think there are some limitations with these reduced abstract views. So some concepts that have been thrown around like distant reading by Franco Moretti, sort of a literary, sort of a macro approach to literary history, or cultural analytics promoted by Lev Manovich. So this idea that you can apply data analytics to cultural collections, I think these initiatives and these concepts are very useful, but I think especially in the cultural sphere, when we're trying to make sense of our cultural legacy and trying to make sense what it means to be human, I think they make a promise implicitly, and I think that's also the promise that visualization is making here and there, that we can see everything from nowhere. And here I'm citing Donna Haraway, she was using that phrase in the late 80s, sort of criticizing objectivist, positivist science, where she was arguing actually, in many cases, the more situated, more partial knowledges, the more partial perspectives might be the more powerful insights to gain. Now picking up on that idea of partial situated perspectives, I would argue that if we want to understand a phenomenon, maybe we actually have to get closer to the phenomenon. So here a few sort of scenes from Paris, the first image was also from Paris, if we actually go onto the street level of a city, we will probably engage more closely with the phenomenon that we are trying to understand, what maybe in this case city life is about. And this sort of notion of situated views is also implicitly also advocates movement. So you're actually moving between these views and that the changes that we experience as we move through a rich data set is also part of the analytical process. Yeah. So, and to actually make sense of that and inform visualization design, interface design, I found the cultural figure of the flaneur to be quite instructive. So it was sort of promoted by Walter Benjamin and picked up by loads of researchers in cultural studies. The flaneur has been sort of characterized as this urban character who is roaming the streets and squares of Paris in the middle of the 19th century. He so is actually a masculine persona, so that's something to be closely looked at. But his attitude towards the city I think is quite interesting. He is wandering around, not really with the clear destination, but with a very expressed goal to make sense of what city life is about to, and actually, so the flaneur has been sort of framed from different angles, but I want to take three character traits, which I think are very useful when we talk about visualization. The first, the flaneur as a curious explorer who has a great sense of interest in both maybe the shop windows, but also the side streets, the alleyways, and he's being pulled in by his different senses, smells, sights, sounds, and he's open to actually change his path along the way. As he moves in that sort of non-goal directed or this sort of more open-minded way, he also develops his own pace, his own privileged perspective on the city that maybe everyone around him who is just running from A to B might not actually have. And by doing so, he's also portrayed as a critical spectator who outlines or who actually states where the issues are. So when the flaneur was coming up as a sort of urban figure, there was really the alienation in the city, mechanization, and sort of, yeah, negative side effects of capitalism that were sort of happening around him. And he was pointing these things out and as a third character trait, he is sort of working around these issues as a poet, as a painter. He's a creative person who tries to reimagine what city life could be. So in a way, he's quite paradoxical. He's quite excited about the city, but he's also taking issue with quite a few problems that are arising. Now, that's a big detour maybe. I don't know, I'm excited about cities, so I can talk about that for a while. But let's come back to data and information spaces. So to me, how the flaneur relates to the city can be, we can use that to think of our users, the viewers of our visualizations, not around deficits, that they may not have a stats background, that they may not have, or that they have information needs or knowledge gaps and so forth, but rather as people sort of in a sort of optimistic and idealistic sense as people that we treat as information flaneurs that can be curious, they're pursuing diverse paths through our data sets, through our interfaces. They're critical, they question the visual representations, maybe also the stuff that is left out. We had a great talk yesterday by Andy about the sort of visualizing the blank, the missing values, but this notion, this persona, this idealistic notion of the information flaneur would also be a creative person, so she would actually contribute new types of content, new representations, new visualizations. So maybe I'm setting the bar quite high, but if we design our tools for that kind of person, I think we can go much further in sort of the reach and the impact that we can have with our tools. Okay, so let's see, I'm not sure if I have convinced everyone that this is a good idea, but I will try to break it down a bit more to sort of what this could mean for interface design, for visualization design. So basically, what I'm interested in is negotiating and sort of improving how people relate to information spaces. And for that, I find explorability a very useful sort of umbrella term, and it's in a way, I would position that next to usability. While usability is about the sort of fine-grained mechanics of the interface, the sort of user-facing interface, I would say explorability negotiates the relationship between the searcher or the viewer and the information space or the data set. And the three, and this is not an exhaustive list, but the three principles that I find very useful in my design work, and I'm going to get to some of these visualizations in a bit. I think I'm doing quite well with time, maybe I'm rushing a bit. But the three principles that I find really useful are orientation, continuity, and serendipity. So orientation, this, and again, that relates back to how a planero might walk through a city. By knowing where I am and by sort of having a certain confidence that I would find my way back home, more or less, I can actually be quite exploratory. So if we design visualizations in a way that gives our users, our viewers, the confidence in trying things out and actually making invitations to navigate, to look around, we might trigger their curiosity. As we change things in the visualization as viewers, as we change parameters, as we navigate, oftentimes these changes are abrupt. In search interfaces, it's actually still the case as you enter a search query and change it, the list will update very rapidly and you don't have a sense what you have looked at already, unless you click on it so it will get highlighted. The same is true just for browsing. If I go from one link to the next, I go from the start page from New York Times to an article, there might be an overlap. The title might still be the same, but just perceptually, it's very abrupt. So in our visualizations, what I'm promoting is actually have more continuous experiences. So there's this perceptual notion of visual momentum. So if we have display state changes, can we actually connect these display states that we're moving from if there's some kind of overlap? It's actually fairly easy to do. And then the third one, to be honest, it's quite hard to design for serendipity. These fortitious discoveries when we deal with data and information, then when people sort of encounter bits that they haven't looked for, but once they found them, it actually was quite useful or inspiring for their own work or their life. And how can we support that in our visualizations so that people might actually move more laterally and not just target or try to answer a very specific question? OK, so now I'm going to actually move from this sort of conceptual, philosophical preamble that hopefully didn't bore you too much to demo mode, which should be exciting for all of us because demos are always a risk. So let's see how this works. So I'm going to start. Actually, I'm going to start now take that back, cut that from the video. I'm actually going to introduce that project a bit before I show it in action. PivotPass is a visualization project. I've done it in Microsoft Research. And the motivation here was very tightly linked to that notion of the information planner strolling through data. And so we had the goal to help the sort of information planner relate the resources. In this case, the resources will be publications, academic publications and the facets that are related to them and encourage a gradual movement, a gradual pivoting through a very large database of publications. We started out with just trying to understand the basics of that information space, the tripartite information space that has resources at the middle that can reference each other. Then there are people involved that are the authors. And we have concepts such as keywords that actually portray the aboutness or topicality of these resources. We designed sort of very basic elements for these different entity types that have some basic interactive capabilities. We arrange that, we give them different sort of visual states, depending on how you interact with them. We arrange them in a sort of three-band layout and connect. So the resources are in the middle. And now I'm actually going to demo mode so it becomes more obvious. So the resources, in this case, the publications. So you probably can't read the text, but that's OK. But there is text. That's important. So what's happening here is on the left, we have Mary Shavinsky. She's a Microsoft researcher. And at this moment, at this time in the interface, she's the anchor for that view. So her most cited papers are selected. And they're arranged in the middle diagonally in green. And the point here, I mentioned in terms of orientation, actually making invitations to go around. So everything that is on the display, before I tell you how things are arranged, everything can be clicked on and you can navigate around these elements. So the papers are arranged by a number of citations. Something, for some reason, academics care about. But above that, you have the authors of these papers. So her co-authors. And they are arranged horizontally based on where the publications are. And that will become interesting in other layouts in a second. But they are also sort of ordered vertically based on their degree. So as you see, for example, here George Robertson, he's on the top because he's actually connected with quite a few papers. So he's a very active co-author of hers. And the same, we can look at the sort of larger. I also adjusted the font size slightly. The larger keywords on the bottom, user study, information visualization. So these are important research topics for Mary Shavinsky. Now the point of this interface is that it encourages you to stroll around. So I can click, for example, on George. And as I clicked on George, the interface takes me along, makes a state transition. And I see, OK, that's easy. I can move on. OK, who else is doing interesting things? I can check out what George is about, look at his keywords, who is he working with. I see, OK, here's an interesting paper, cone trees. I can actually click on the paper as well. Then we have a sort of bifurcated view that arranges references that are mentioned in the paper on the left. So that's sort of the before part of the display. And on the right, you have publications that are actually citing that one. So you're sort of taking that paper as a pivot point. And you see in a sort of micro fashion how the field has developed. So you see that cone trees has taken sort of inspiration or influence from papers on input devices, inverted indices, case studies. And on the other hand, it has been picked up by researchers working in the information visualization field. What is also interesting is that the main of the three authors that are shown here are citing themselves as well. So they have been influenced for their own work. So that's something that's common, but it's also interesting to actually see that this is exposed here, that there are some self-citation happening. Now, this is a web-based tool. So actually, there's not much graphics happening, to be honest. I mean, it's text. I moved text around with CSS transitions. I have as much graphics that I have as really only the curves that I don't know if you can actually see them. They're really subtle so that it's actually not a very complex. It's not very overloaded. So it was intentional to keep all the curves fairly subdued until you hover over something and you actually see how they're connected. But so this is web-based. So I'm actually using the back button. And I can use it. So I go back to Mary Shavinsky. And I want to show you a third layout that this prototype supports, the comparison layout. So here we now have the most cited papers that Mary and George have written together in the center. So kind of like a Venn diagram. And then a selection of papers on the left and right side that they have authored without the respective other. There's a caveat to this. I come to this in a second. A huge caveat, actually. But what is interesting here, so we now actually see, or we can see how the sort of social network on the top around these papers that they have written together or apart actually unfolds. Who's writing with whom? And are there maybe people that they work only apart from each other? So that's one thing that you can explore with that. But you also see on the bottom that there might be a certain weighting of interest, of research interest between these two people. So user study is positioned slightly towards the left because Mary has also been writing user study papers on her own. But George hasn't, at least based on the sample that we have on display here. On the other hand, we have user interface on the right. So slightly moved more towards George's side. And it suggests that George maybe has been more involved with creating user interfaces. And this actually confirms what we know about these two people. Mary Shavinsky is a cognitive psychologist and George Robertson is a software engineer. So they were actually sort of a dream team for several years at Microsoft Research. She was doing the studies. She was creating really with bank interfaces and they were making stellar research. And that sort of role allocation actually comes out in the arrangement of the keywords of the papers below here. Okay, so taking a breath. So basically that's pivot path is a prototype that sort of illustrates this notion of strolling through a rich collection. We deployed that for a couple of weeks within Microsoft and there was largely positive feedback. So mainly focusing on sort of aesthetics on the movements that are happening. There was some confusion or sort of false expectations because people were quite used to the typical search and refine model where you enter a keyword and you try to just arrive at maybe one paper or something. So that tool didn't really support that that well. Something else that didn't really come up so much in the feedback but that we really grappled with throughout that project was that we're actually cutting a lot. So when I select Mary, I haven't selected a keyword and I did that purposefully. We're only showing the most in this case that based on the screen resolution we're showing the top-sighted papers of hers. 26 of in this database I think it's a little bit under 100 papers. So we're cutting a lot of papers and we don't actually give presence to these papers in that interface. So that was something we were sort of dealing with. We didn't really find a solution within that timeframe but I will come back to that in a different project where we try to address that to some degree. Okay, so that's pivot paths. It requires a fairly well-structured database but I wanna show you another project that takes some of these ideas of sort of moving around, of strolling, taking anchors for one view actually from the data and then sort of piggybacking from one view to the next. I wanna apply that to an unstructured type of data which is text which you could argue whether it's unstructured or not but so the previous project was at Microsoft Research. This now has been done by myself and Corpus Linguist at Newcastle University in the UK and we wanted to sort of bring playfulness and sort of exploratory playfulness to sort of Corpus Linguist visualizations to invite non-experts, non-techies, non-linguists to play with their data and so this is an early prototype to manage your expectation. So this is what happens, what it looks like when we visualize the English text of Hansel and Gretel, one of the famous fairy tales of the Grimm's collection. So it just starts out as a fairly mundane tech cloud but it's an interactive tech cloud that allows you to actually navigate between different words. So we can see how words correlate with each other. So when I just hover over a word like little, I can see that little Hansel is actually often used in that tail because it's highlighted in bright pink. So I can then click on little and it becomes a pivot point just as we had pivot points in the previous project but now I'm actually operating on the words used in that tail in this case. I can see how they are used, whether they are used more before or after that word. I don't do collision detection. I think we had collision detection mentioned by Mike yesterday, not doing anything here. So you see some, not so fancy overlaps but they're positioned based on whether they are used more before or after and you can also see a sample of the words used below the visualization so I can see that Hansel is often used more or less before a little and you can also interact with these words below and sort of move through that text word by word. I wanted to show you an example so there's also a little search included and I've shown you two anchor views in PivotPass and we've done something similar here where we can compare how the text corpus which is a fairly small corpus here just a short fairy tale but how the words that are used in that fairy tale how they associate with two anchor words. So we have now children and forest so they are sort of taking actually data elements as my sort of dimensions or as my poles and I'm saying show me all the words that correlate with that and position them based on how much they actually are attracted to one or the other. The font size now is just the combined sort of correlation but the positioning shows sort of a weighting. So we see for example that the word woman is used more often in relationship to children whereas wood and thickest and Hansel and animals are used more often with forest. So you can get a sort of long sense of how certain words are used together. Okay, so there's a bunch of options so we actually have some NLP, some language processing behind that so we can actually focus on certain word types if you're interested in that. Actually what is maybe interesting when you're looking at Hansel and Gratle that Hansel and Gratle are actually sort of overpowering the visualization so you can throw out these words. If I can spell. So we released this for several months and we have a little feedback mechanism in the tool and we got actually loads of feedback from teachers, from language teachers and they were interested in using such a tool in the context of language teaching where pupils have a hard time understanding how certain words are used in which context and how they could use them themselves and they were thinking that such a tool here could be useful when we want to look at how to use a certain word on the street, in the supermarket, at a conference with their friends. So we want to explore that further how this could be used in the context of language learning. Okay, so that's word wanderer, it's online, it's not yet on GitHub but you can play around with it, encourage you and send us feedback and now I'm getting to the last and I'm running out of time to the last case study to illustrate so this idea of moving between partial situated views, the more bottom up approach to data. So this is a monadic exploration, this is the title of that project, also a project at Newcastle University and we collaborated for that project with the book team, the team of editors and also tech people behind Beautiful Trouble. Beautiful Trouble is sort of a recipe book for creative forms of non-violent action, sort of protest and civic engagement pretty cool book, I encourage you to check it out, maybe through that visualization. So the book is from a data standpoint, it's really interesting because you flip through it, they send me a copy and there are quite a few books like that on the margins you see these sort of see also references. So immediately as a visualization person, oh, I'm holding a network in my hand, it's actually a network data set, we could do something with that. So what we want to explore though in that visualization, to negotiate sort of the individual chapter, so individual element of a collection and the whole collection. So, but in a way that doesn't end up in sort of only abstract shapes that give you sort of the macro view or only the individual pages of an individual chapter. So what we did here is we create that circular arrangement with a search box in the middle, so I can actually start with a search if I want. So considering protest, oftentimes this happens in the streets, so I can actually enter street. You have to trust me that I'm entering this right now because I think it's tiny on the screen. And as I enter that, the relevance, the search relevance actually then used for the displacement around the circle. So I see for example, billionaires for Bush, apparently did some street actions, reclaim the streets, streets into gardens. So I'm selecting one of these because I'm interested in maybe reclaim the streets has been around for a few years now and now selecting a node replaces the search that I've done before and now the other elements of the collection are arranged based on their attraction, based on their sort of affinity within the network. So the direct neighbors in the network are highlighted in with these sort of dark dots and then there are sort of indirect neighbors that have certain neighbors in common and they are also shown with their label but in a more subdued and a brighter color. So we can get a sense, they have some kind of affinity but they're actually not mentioned by the editors as directly related. So this idea of having a more nuanced sense of how one element relates to the whole collection is what we were interested in this project. And now I mentioned movement earlier what was important here as well was that you can actually click on any element now in the collection, even something that hasn't been identified as a neighbor sort of and navigate around. So I can now move to balance art and message. I can look at theater of the oppressed. So I can sort of let my interest in that collection drive my navigation. And if I, for example, actually want to depart from that path so that path right now might look like it's all about similarity. Similar items are popped up. No, I can actually say I'm explicitly moving to the periphery of that visualization that may have nothing to do or very little to do with the selection. So in an attempt to maybe avoid the sort of filter bubble effect. So that type of visualization will allow us to give the viewer or the searcher that ability to actually depart from that path. Okay, we have a legend here so you can actually filter out or add different types. And we had that also online for several months. Actually in collaboration with the team behind the book. And the feedback from the book team was very positive. We actually worked closely with them. We try to also, I can actually bring up one page here if I'm online. So actually try to use some of the aesthetics of that collection, the color choice, the icons in the visualization. So it actually links to the content that is being represented. And so their feedback was, this is like a wonderful new way of showing a table of contents. They called it a living table of contents. And the feedback from sort of passersby, I don't really know sort of what their background was, was an interesting mix of being excited about the content and being excited about the arrangement, the type of visualization that we were sort of offering. And in a way, to me, that was actually quite nice to see that the sort of richness and the sort of interconnectedness that the editors and the authors have put together into this book actually shines through in that visualization. I think that's a good thing to strive for. Okay, I have one and a half minutes left to get to wrap that up. Okay, we have that. Okay, I've showed you three case studies. And I wanna sort of bring that together and argue for that visualization doesn't always need to start with an overview. We can start at the bottom with individual elements. And I think a navigational approach to data can have some merit and allow us to engage more closely with the data and the collection items that we are interested in. I mentioned earlier three principles, orientation. In these tools that I've shown you, I actually used the data elements. I used facet values and words and the titles of the chapters to actually engage with the data. Actually, I didn't show you much GUI. Most of what you've seen were actually data elements and labels and they are used to then pivot through the data. The display state changes were animated and especially in the last project, these animations are meaningful. They're loaded to portray the difference between different perspectives. When I go from one case study to a principle, actually the way things move in and out portray the difference in relationship. So it's not just a sort of a tweening, but it's actually meant to help you understand the differences between the elements. And serendipity in all these views, there are sort of these tangential ways that help you to explore different aspects of a data set and hopefully positively get lost and find your way back if you choose so. So in one sentence, so the plea or the battle cry that I would try to sort of convey here was, let's take data visualization to the streets. Let's look eye to eye with our data elements. Let's not remove ourselves too far all the time with tiny dots and abstract shapes. I think there's lots of insight to be gained on the streets. Thank you.