Mapping Twitter-based social learning spaces

Throughout this semester—and especially over the last few weeks— I’ve been using the digital humanities seminar I’m enrolled in to explore how building a map might help me better represent some of the Twitter-based social learning spaces that I study. In this post, I’d like to weave together some themes that have emerged from this class and my previous work; in doing so, I don’t expect to generate any profound or conclusive answers, but I do hope to lay a foundation for my thinking as I continue to fiddle with this medium and this genre of data.

So, here are the themes that I’m bringing with me into this post:

First, it’s possible to map Twitter data in interesting and compelling ways. This isn’t really new; I’ve been experimenting with mapping data from the #educattentats hashtag for over a year now. So, why am I continuing to come back to it? To be honest, part of this is because I’m learning fancier and flashier methods for mapping these tweets: Of the two maps below, the one on the left is clearly a couple steps above the one on the right, even if there’s still a lot of work to be done.

The second theme I have in mind is that the digital humanities teach us the importance of aligning methods and theory. I wrote about this at the beginning of the semester, and it’s been a theme throughout the whole seminar. In other words, “fancier and flashier” doesn’t by itself justify any particular set of methods—if I’m going to demonstrate that I really learned something from this seminar, I’m going to need to come up with a compelling theoretical reason for mapping this Twitter data.

This leads nicely into my third theme: It’s easy to fall short of that ideal when mapping Twitter data. Earlier this semester, I suggested that Kenya-Tweet (a DH project I was reviewing) didn’t make a clear theoretical contribution when it mapped tweets, and just a few days ago, I found myself trying to use a map to represent something that would have been better represented in some other form.

In short, I know that as I move forward with projects like these that there is a trap waiting for me… and that I’m prone to fall into it despite my awareness of it. I think the most helpful thing that I could do with the rest of this post is to review the phenomenon I’m studying, summarize why it’s interesting from a theoretical perspective, and then explore how building a map could help me represent that.

The Phenomenon

#educattentats is a hashtag that plays on the French words éducation (education) and attentat (terrorist attack). It was created by the Twitter user @padagogie on the morning of Saturday, November 14, 2015 in response to a series of terrorist attacks that had taken place in Paris the night before. As seen in the tweet below, @padagogie introduced the hashtag with a question to his fellow teachers: How are you going to react in your classes? The goal of the hashtag was for teachers to coordinate their efforts and pool possible resources as they prepared to return to classes (and teach their students) on Monday.

For other background information on #educattentats, I’m embedding the slides that I used to give a presentation on this subject at the 2016 AECT Conference. Of particular note in these slides is a summary of the scope of this hashtag. Over the first 28 days of its existence, about 3,600 unique Twitter accounts used (or interacted with) the hashtag, collectively posting about 1,200 original tweets and about 4,300 retweets. In the grand scheme of viral media on the Web, that may not be a lot, but in my mind, it is something to get excited about.

In fact, to speak informally, I personally find this phenomenon to be quite moving. In the face of a national tragedy, people throughout France and all over the world (one conclusion we can already draw from maps) leveraged a social networking site frequently—and jokingly—described as a place for people to post pictures of what they’re eating for breakfast as a resource for supporting and celebrating teachers as they engaged in talking to their students about terrorist attacks that had already begun to leave a deep impression on the national consciousness.

The Theory

From an education researcher’s point of view, #educattentats is interesting because it represents a social learning space that we’ve only recently started to conceive of. Twenty years ago, Greeno, Collins, and Resnick (1996) argued that educational research was headed in a direction that was increasingly taking into consideration the social dimensions of learning. Although scholars of learning like John Dewey and Lev Vygotsky had emphasized the role of social interaction in learning in the early 20th century, their work had either (respectively) fallen out of vogue or never come to prominence, and educational psychology was therefore dominated by a more individual view of teaching and learning.

One of the most influential ways of conceiving of social learning has been Lave and Wenger’s (1991) community of practice. According to this perspective, learning happens in groups that share professional (or other) practices. Those on the periphery of these groups learn thee practices from more experienced members and, in so doing, integrate themselves more closely into the community. The community of practice perspective makes some key contributions to our understanding of social learning: First, the group plays an essential role in the learning process. Second, learning is no longer restricted to formal settings (such as school); rather, it is something that is constantly happening as part of people’s everyday lives.

Although some scholars (e.g., Gao & Li, 2016) use the community of practice to describe learning that happens on Twitter, others (e.g., Carpenter & Krutka, 2014; 2015) prefer a newer perspective—Gee’s (2004) affinity space. Gee suggested that despite the utility of the community of practice, it had two important shortcomings. First, the idea of a community implies close connections and a sense of belongingness that are not always present in social learning spaces. This is true of physical spaces (for example, different students in the same classroom may not feel connected by the same goals or practices) but especially true of virtual spaces, where people may interact around the same subject despite having very little else in common. Second, Gee argued, few people used the community of practice perspective the way it was intended, which diluted its potency and contributions.

In contrast, Gee argued, it is often more instructive to talk about a space than a community. Talking about a space does not presuppose connections (which may not exist) among the people that collectively occupy that space but still allows us to discuss the learning that happens within. It is important to note that the affinity space is technically a subset of a social semiotic space—while the first term is catchier and has gained more traction, Gee argues that a space only qualifies as an affinity space when it meets several criteria that set it apart from other social semiotic spaces. It’s not without irony that researchers (myself included) have used Gee’s term without the proper nuance, given that that was part of the problem Gee was trying to solve.

The Method

To summarize up to this point, my intention is to use a map (the method) to display data related to the #educattentats hashtag (the phenomenon) in such a way that it emphasizes its nature as a social learning space that corresponds with Gee’s (2004) conception of the social semiotic space. In my mind, the most distinctive thing about Gee’s social semiotic space is the way that it breaks with earlier conceptions of learning. It breaks with a traditional classroom-centered, individual view by describing teaching and learning as activities that can occur in a wide range of physical and virtual spaces and through social interaction. It also breaks with the community of practice view in suggesting that people need not have much in common in order to learn from and teach each other within these spaces. This latter part is especially interesting in that it suggests that participants in an affinity space may be wildly different from each other in all ways except for their having engaged with the space. This allows for high levels of diversity and variety within the space, and that’s one thing I think a carefully-constructed map could get across.

So, what forms of diversity and variety could I display on a map? Broadly speaking, I think there are two kinds of picture that I’d like to paint here: a picture about the different participants that have engaged with this learning space, and a picture about the different ways that they have participated in the space.

Diversity of Participants

For all of the diversity that might exist within a classroom or a community of practice, we can still make basic assumptions about what all of those learners have in common. With a few exceptions, we can probably expect everyone in a high-level teacher education class to be a teacher-in-training and everyone in a teacher professional development workshop to be an in-service teacher.

We may be tempted to apply this same logic to the #educattentats affinity space; that is, we might feel comfortable assuming that everyone participating in the space is a teacher working in France—after all, that seems to be the intended audience of @padagogie’s original tweet. However, even a quick survey of #educattentats participants demonstrates that this is far from the case. Even as we identify exceptions to our assumptions about classrooms and communities of practice, I would argue that the scale of those exceptions would be dwarfed by what we see here.

Geographic Diversity

Let us first consider the question of geography, since this kind of data is the most well-suited to being displayed on a map. Geographic data for the #educattentats space isn’t as readily available as I would like, but it’s possible to make some estimates and then plot them on a map. The results are striking, especially if you are considering the assumptions that one might bring into an analysis of the #educattentats space. While activity is concentrated the most highly in France, there are participants from all over the world. This raises all sorts of questions around one main theme: Why are people outside of France interested in this? Obviously the Bataclan attacks got worldwide attention, but attention to this particular hashtag seems more surprising to me, and I think the answer to this question would be very interesting.

As a side note, mapping user profiles in this way has actually helped me have methodological insights, not just the theoretically-relevant ones I was hoping to achieve. I’ve always known that that my methods for estimating user locations were flawed, but this foray into Leaflet mapping has helped me associate points on the map with specific users in a way that I couldn’t with my previous maps. I’ve thus been doing a fair amount of spot checking to see if points in particularly interesting areas actually belonged there… and I’ve been a little bit disappointed with the results. I need to do some more work to figure out just how (in)accurate my estimation methods are; knowing this will help me know if I need to improve/replace my current methods.

I’ve also noticed that there are a few places where dots are superimposed. This usually isn’t a problem, but it does pose the danger of underrepresenting the participants in a very specific area or not finding the participant one expected when exploring a particular location.

Linguistic Diversity

Above, I asked why people outside of France would be interested in the #educattentats hashtag, given that the content of associated tweets is (mostly) directed towards teachers in France. If the content of the tweets is generally directed to teachers in France, we would also expect it to be mostly in French. We might then ask ourselves how many of the participants in this affinity space are native French speakers. Although Twitter metadata doesn’t tell us the birth language of its users, it does mention what language it users have Twitter set to, and that metadata is included in the data that I’ve collected.

In the current version of my Leaflet map (which can be viewed above), I’ve represented this data by assigning each user language a different color and plotting the dots representing these users in those colors. I think the results are instructive: The dominance of French in France, Spanish in Spain, Italian in Italy, and English in the UK are all to be expected, but they all also contrast with the blue (i.e., Francophone) dots throughout the world. This contrast shows us that there are both Francophones outside France (and often outside Francophone countries) and non-Francophones throughout the world who are participating in #educattentats, and I think that the presence of each of these populations is worth knowing about. There are also fun little glimpses into the diversity of the European linguistic landscape, like the cluster of Catalan-speakers in, well, Catalonia, the man with his Twitter profile written in English but his Twitter interface set to German, and the Spaniard living in Germany (who’s taken interest in a French hashtag).

One thing I’d like to mention here before moving on is how effective using different colors is for portraying this kind of diversity: It’s very easy to get a feel for this just by scanning the map. There’s no inherent connection between colors and language, though, so as I discuss the remaining forms of diversity that I’d like to be able to represent, one thing at the back of my mind will be whether something else deserves to be represented in this way instead.

Identity Diversity

While the last two forms of diversity implicitly address the assumption that only people in or from France would be interested in this affinity space, this one addresses the assumption that only educators would find value from participating in this space. In the published versions of this research, I’ve simplified this diversity by analyzing a sample of Twitter profiles and summarizing the different identities represented in them with a concise coding scheme. When translating this into an interactive map rather than a paper (or presentation), that doesn’t work as well. It wouldn’t do to only have this kind of data be present for a small sample of the points on the map, and I’d rather not have to code 1,600 profiles by hand. The compromise I’ve found for this version of the map is just to make it possible to access the Twitter profile corresponding to each participant by clicking on the corresponding dot. On one hand, this makes things hard to assess at a glance; on the other, it allows the person exploring the map to explore the participants on their own and develop their own conclusions.

I really enjoy this addition of the map, since it makes the participants seem a little more real. However, directly quoting participants’ Twitter profiles gets us into the tricky world of Internet research ethics. In most cases, the profiles are anonymized, and all of the data is public data, but traditional paradigms of research ethics don’t hold up very well in the realm of research on the Web (Markham & Buchanan, 2012), and while I feel comfortable doing this for the time being, it’s worth further consideration as I move forward.

There are also several participants whose profiles have some encoding errors in them. I did some very basic data cleaning when setting up my map, but I wasn’t as thorough as I needed to be.

Diversity of Participation

The participants in the #educattentats learning space are diverse not only in terms of who they are, where they’re from, and what they speak but also with regards to how they participated in this space. I would also argue that, like the diversity of participants, the diversity participation in an affinity space such as #educattentats is generally greater than the diversity we might see in a classroom or a community of practice.

Scale of Participation

Not all of the participants in the #educattentats learning space participated on the same scale. One Paris-based education researcher contributed more than 70 tweets and retweets to the space during the 28 days that I’m concerned with, but many of the participants (maybe even most—I’d have to recrunch the number) only posted (or reposted) a single tweet that contained the hashtag. I think it’s important to represent this kind of diversity, and the size of individual dots seems like the most intuitive way to show it.

However, making it a strict 1:1 relationship (i.e., increasing the radius of each dot by 1 for each additional tweet or retweet) creates some visibility problems: Our education researcher friend and similarly enthusiastic participants wind up with dots so large that they blot out surrounding dots, especially when the map is set at larger scales. For the time being, I’ve fixed the sizes so that they only range between 1 and 10, and that seems to be doing the trick, though it’s hard to make nuanced distinctions about the scale of participation. That is, it’s easy to tell those who participated a lot from those who only participated a little, but it’s hard to tell that Participant A only sent one or two more tweets than Participant B. Before I managed to fix the size to a certain range, I first explored the possibility of using color to represent the scale of participation, but the problem was even more pronounced there, so I’m happier with it the way it currently is.

The elephant in this particular room is trying to define what count as participation for these purposes. As mentioned above, it’s currently defined only as the number of tweets and retweets someone has sent out that use the #educattentats hashtag. I also have some data available for how often people “liked” tweets having the #educattentats hashtag, but it doesn’t lend itself quite as well to mapping, in part because I can only (currently) map dots for people who have tweeted or retweeted using the hashtag. In other words, I could use the “likes” data to boost the participation figures for people already on the map, but I’m not in a good position to add people to the map who only liked tweets without composing or reposting any. That feels incomplete, so I’m leaving it be for now.

On the other hand, though, liking-without-tweeting is an important part of this diverse range of participation, so it needs to get in there eventually. Are there other forms of participation, too? What about “replies” to #educattentats tweets that don’t themselves include the hashtag? What about “quote-retweeting” #educattentats tweets? It’s hard not to see these as a way of engaging with the learning space, though I’m not 100% confident I can capture the first and very doubtful that I can capture the second. Finally, Carpenter (2015) has explained that there are at least some teachers who learn from these kinds of spaces without ever leaving any traces of their engagement. It’s downright impossible to capture data on them (at least comprehensive data using the collection methods that I prefer), so it may be that I can only represent part of this range of participation.

Forms of Participation

By asking what counts as participation for the purposes of measuring the scale, I’ve already summarized some of the different ways that people might engage with this learning space. These forms range from the highly-active (composing an original tweet including the hashtag) to the highly-passive (reading #educattentats tweets without “liking” or retweeting them), though for practical purposes, I can concentrate on three different forms: tweeting, retweeting, and liking. I know from my previous work on this dataset that there are some participants who engage in all three of these forms of participation and some who only engage in one. None of that is currently represented on the map, and I wonder if it should be. On one hand, it would certainly be handy to be able to distinguish those who only “liked” tweets from those who actively composed their own tweets, but I’m not sure what the best way to portray that is. Different shapes might do the trick, but if my math is right, I’d need seven different shapes if I wanted to distinguish between all different combinations of tweeting, retweeting, and liking. In the end, it may not be worth it.

Another issue related to forms of participation that I haven’t yet represented is connections between different dots on the map. This is the sort of thing that would be better represented on a sociogram than on a map, so it could be set aside as a non-issue, but I feel like depicting network connections on this map would reinforce some of the attributes of Gee-style social learning spaces that I’d like to get across, namely the fact that a shared virtual space diminishes the importance of geographic distance between participants. However, my efforts to depict this have hit a couple of snags. It takes a while to calculate all of these connections, and trying to display them on an interactive map seems to slow the whole process down… not to mention serve as a visual distraction from the other features of the map. It’s worth exploring this further, but I feel like I’m at somewhat of a dead-end right now.

Other Considerations and Conclusions

There’s one more big issue that needs to be brought up when considering whether a map is the right way to go with these data, and that’s all of the users for whom I do not have geographic data. Does the payoff from seeing the geographic diversity of this learning space (since that’s the only thing that couldn’t be depicted with another visual representation) outweigh the costs of all of the participants that I can’t display?

There are several other questions worth adding to this one: Is a map really the most compelling way to depict the forms of diversity I’ve listed here? Does that answer change when I take into account the flaws I’ve seen in my methods for estimating geographic locations? Is insisting on a geographic element constraining my ability to represent other forms of diversity (e.g., social connections, scale of participation)?

These are tough questions, and I actually find myself rethinking a map more than ever before. Abandoning the map for some other form of representation (maybe a souped-up sociogram?) would probably let me more accurately and more fully depict some of the other forms of diversity that I value in that space. Who knows, I might even be able to find a way to include a geographic element—maybe something along the lines of different colors for different continents.

Despite all of this, I’d like to continue working with a map to see if I can’t overcome the challenges that I’ve identified for myself in this post. There’s something intuitive about a map that I think would help people get a sense for this space without needing too much explanation. Plus, for all our experience with the Internet, there remains a certain power in reminding people that they can use Twitter to learn from and communicate with people from entirely different continents. Finally, there’s something humanizing and personal about being able to see where these participants are from; to re-use a phrase from earlier, knowing where people are participating from makes it seem much more real.

So, onwards and upwards! Plenty of work to be done, plenty of questions to answer, and probably plenty more questions yet to be asked, but this feels like a step in the right direction.

References

Carpenter, J. (2015). Preservice teachers’ microblogging: Professional development via Twitter. Contemporary Issues in Technology and Teacher Education, 15, 209-234.

Carpenter, J. P., & Krutka, D. G. (2014). How and why educators use Twitter: A survey of the field. Journal of Research on Technology in Education, 46, 414-434, doi:10.1080/15391523.2014.925701

Carpenter, J. P., & Krutka, D. G. (2015). Engagement through microblogging: Educator professional development via Twitter. Professional Development in Education, 41, 707-728, doi:10.1080/19415257.2014.939294

Gao, F., & Li, L. (2016). Examining a one-hour synchronous chat in a microblogging-based professional development community. British Journal of Educational Technology. doi:10.1111/bjet.12384

Gee, J. P. (2004). Situated language and learning: A critique of traditional schooling. New York: Routledge.

Greeno, J., Collins, A., & Resnick, L. (1996). Cognition and learning. In D. Berliner & R. Calfee (Eds.), Handbook of educational psychology (pp. 15-46). New York, NY: Macmillan.

Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. New York: Cambridge University Press.

Markham, A., & Buchanan, E. (2012). Ethical decision-making and Internet research: Recommendations from the AoIR Ethics Working Committee (Version 2.0). Chicago: Association of Internet Researchers.

Leave a Reply

Your email address will not be published. Required fields are marked *