Two years after the NYT’s ‘Year of the MOOC’: how much do we actually know about them?

Timeline of the development of MOOCs and open education
Timeline of the development of MOOCs and open education, from: Yuan, Li, and Stephen Powell. MOOCs and Open Education: Implications for Higher Education White Paper. University of Bolton: CETIS, 2013.

Ed: Does research on MOOCs differ in any way from existing research on online learning?

Rebecca: Despite the hype around MOOCs to date, there are many similarities between MOOC research and the breadth of previous investigations into (online) learning. Many of the trends we’ve observed (the prevalence of forum lurking; community formation; etc.) have been studied previously and are supported by earlier findings. That said, the combination of scale, global-reach, duration, and “semi-synchronicity” of MOOCs have made them different enough to inspire this work. In particular, the optional nature of participation among a global-body of lifelong learners for a short burst of time (e.g. a few weeks) is a relatively new learning environment that, despite theoretical ties to existing educational research, poses a new set of challenges and opportunities.

Ed: The MOOC forum networks you modelled seemed to be less efficient at spreading information than randomly generated networks. Do you think this inefficiency is due to structural constraints of the system (or just because inefficiency is not selected against); or is there something deeper happening here, maybe saying something about the nature of learning, and networked interaction?

Rebecca: First off, it’s important to not confuse the structural “inefficiency” of communication with some inherent learning “inefficiency”. The inefficiency in the sub-forums is a matter of information diffusion—i.e., because there are communities that form in the discussion spaces, these communities tend to “trap” knowledge and information instead of promoting the spread of these ideas to a vast array of learners. This information diffusion inefficiency is not necessarily a bad thing, however. It’s a natural human tendency to form communities, and there is much education research that says learning in small groups can be much more beneficial / effective than large-scale learning. The important point that our work hopes to make is that the existence and nature of these communities seems to be influenced by the types of topics that are being discussed (and vice versa)—and that educators may be able to cultivate more isolated or inclusive network dynamics in these course settings by carefully selecting and presenting these different discussion topics to learners.

Ed: Drawing on surveys and learning outcomes you could categorise four ‘learner types’, who tend to behave differently in the network. Could the network be made more efficient by streaming groups by learning objective, or by type of interaction (eg learning / feedback / social)?

Rebecca: Given our network vulnerability analysis, it appears that discussions that focus on problems or issues that are based in real life examples –e.g., those that relate to case studies of real companies and analyses posted by learners of these companies—tend to promote more inclusive engagement and efficient information diffusion. Given that certain types of learners participate in these discussions, one could argue that forming groups around learning preferences and objectives could promote more efficient communications. Still, it’s important to be aware of the potential drawbacks to this, namely, that promoting like-minded / similar people to interact with those they are similar to could further prevent “learning through diverse exposures” that these massive-scale settings can be well-suited to promote.

Ed: In the classroom, the teacher can encourage participation and discussion if it flags: are there mechanisms to trigger or seed interaction if the levels of network activity fall below a certain threshold? How much real-time monitoring tends to occur in these systems?

Rebecca: Yes, it appears that educators may be able to influence or achieve certain types of network patterns. While each MOOC is different (some course staff members tend to be much more engaged than others, learners may have different motivations, etc.), on the whole, there isn’t much real-time monitoring in MOOCs, and MOOC platforms are still in early days where there is little to no automated monitoring or feedback (beyond static analytics dashboards for instructors).

Ed: Does learner participation in these forums improve outcomes? Do the most central users in the interaction network perform better? And do they tend to interact with other very central people?

Rebecca: While we can’t infer causation, we found that when compared to the entire course, a significantly higher percentage of high achievers were also forum participants. The more likely explanation for this is that those who are committed to completing the course and performing well also tend to use the forums—but the plurality of forum participants (44% in one of the courses we analyzed) are actually those that “fail” by traditional marks (receive below 50% in the course). Indeed, many central users tend to be those that are simply auditing the course or who are interested in communicating with others without any intention of completing course assignments. These central users tend to communicate with other central users, but also, with those whose participation is much sparser / “on the fringes”.

Ed: Slightly facetiously: you can identify ‘central’ individuals in the network who spark and sustain interaction. Can you also find people who basically cause interaction to die? Who will cause the network to fall apart? And could you start to predict the strength of a network based on the profiles and proportions of the individuals who make it up?

Rebecca: It is certainly possible to further explore how different people seem. One way this can be achieved is by exploring the temporal dynamics at play—e.g., by visualizing the communication network at any point in time and creating network “snapshots” at every hour or day, or perhaps, with every new participant, to observe how the trends and structures evolve. While this method still doesn’t allow us to identify the exact influence of any given individual’s participation (since there are so many other confounding factors, for example, how far into the course it is, peoples’ schedules / lives outside of the MOOC, etc.), it may provide some insight into their roles. We could of course define some quantitative measure(s) to measure “network strength” based on learner profiles, but caution against overarching or broad claims in doing so due to confounding forces would be essential.

Ed: The majority of my own interactions are mediated by a keyboard: which is actually a pretty inefficient way of communicating, and certainly a terrible way of arguing through a complex point. Is there any sense from MOOCs that text-based communication might be a barrier to some forms of interaction, or learning?

Rebecca: This is an excellent observation. Given the global student body, varying levels of comfort in English (and written language more broadly), differing preferences for communication, etc., there is much reason to believe that a lack of participation could result from a lack of comfort with the keyboard (or written communication more generally). Indeed, in the MOOCs we’ve studied, many learners have attempted to meet up on Google Hangouts or other non-text based media to form and sustain study groups, suggesting that many learners seek to use alternative technologies to interact with others and achieve their learning objectives.

Ed: Based on this data and analysis, are there any obvious design points that might improve interaction efficiency and learning outcomes in these platforms?

Rebecca: As I have mentioned already, open-ended questions that focus on real-life case studies tend to promote the least vulnerable and most “efficient” discussions, which may be of interest to practitioners looking to cultivate these sorts of environments. More broadly, the lack of sustained participation in the forums suggests that there are a number of “forces of disengagement” at play, one of them being that the sheer amount of content being generated in the discussion spaces (one course had over 2,700 threads and 15,600 posts) could be contributing to a sense of “content overload” and helplessness for learners. Designing platforms that help mitigate this problem will be fundamental to the vitality and effectiveness of these learning spaces in the future.

Ed: I suppose there is an inherent tension between making the online environment very smooth and seductive, and the process of learning; which is often difficult and frustrating: the very opposite experience aimed for (eg) by games designers. How do MOOCs deal with this tension? (And how much gamification is common to these systems, if any?)

Rebecca: To date, gamification seems to have been sparse in most MOOCs, although there are some interesting experiments in the works. Indeed, one study (Anderson et al., 2014) used a randomized control trial to add badges (that indicate student engagement levels) next to the names of learners in MOOC discussion spaces in order to determine if and how this affects further engagement. Coursera has also started to publicly display badges next to the names of learners that have signed up for the paid Signature Track of a specific course (presumably, to signal which learners are “more serious” about completing the course than others). As these platforms become more social (and perhaps career advancement-oriented), it’s quite possible that gamification will become more popular. This gamification may not ease the process of learning or make it more comfortable, but rather, offer additional opportunities to mitigate the challenges massive-scale anonymity and lack of information about peers to facilitate more social learning.

Ed: How much of this work is applicable to other online environments that involve thousands of people exploring and interacting together: for example deliberation, crowd production and interactive gaming, which certainly involve quantifiable interactions and a degree of negotiation and learning?

Rebecca: Since MOOCs are so loosely structured and could largely be considered “informal” learning spaces, we believe the engagement dynamics we’ve found could apply to a number of other large-scale informal learning/interactive spaces online. Similar crowd-like structures can be found in a variety of policy and practice settings.

Ed: This project has adopted a mixed methods approach: what have you gained by this, and how common is it in the field?

Rebecca: Combining computational network analysis and machine learning with qualitative content analysis and in-depth interviews has been one of the greatest strengths of this work, and a great learning opportunity for the research team. Often in empirical research, it is important to validate findings across a variety of methods to ensure that they’re robust. Given the complexity of human subjects, we knew computational methods could only go so far in revealing underlying trends; and given the scale of the dataset, we knew there were patterns that qualitative analysis alone would not enable us to detect. A mixed-methods approach enabled us to simultaneously and robustly address these dimensions. MOOC research to date has been quite interdisciplinary, bringing together computer scientists, educationists, psychologists, statisticians, and a number of other areas of expertise into a single domain. The interdisciplinarity of research in this field is arguably one of the most exciting indicators of what the future might hold.

Ed: As well as the network analysis, you also carried out interviews with MOOC participants. What did you learn from them that wasn’t obvious from the digital trace data?

Rebecca: The interviews were essential to this investigation. In addition to confirming the trends revealed by our computational explorations (which revealed the what of the underlying dynamics at play), the interviews, revealed much of the why. In particular, we learned people’s motivations for participating in (or disengaging from) the discussion forums, which provided an important backdrop for subsequent quantitative (and qualitative) investigations. We have also learned a lot more about people’s experiences of learning, the strategies they employ to their support their learning and issues around power and inequality in MOOCs.

Ed: You handcoded more than 6000 forum posts in one of the MOOCs you investigated. What findings did this yield? How would you characterise the learning and interaction you observed through this content analysis?

Rebecca: The qualitative content analysis of over 6,500 posts revealed several key insights. For one, we confirmed (as the network analysis suggested), that most discussion is insignificant “noise”—people looking to introduce themselves or have short-lived discussions about topics that are beyond the scope of the course. In a few instances, however, we discovered the different patterns (and sometimes, cycles) of knowledge construction that can occur within a specific discussion thread. In some cases, we found that discussion threads grew to be so long (with over hundreds of posts), that topics were repeated or earlier posts disregarded because new participants didn’t read and/or consider them before adding their own replies.

Ed: How are you planning to extend this work?

Rebecca: As mentioned already, feelings of helplessness resulting from sheer “content overload” in the discussion forums appear to be a key force of disengagement. To that end, as we now have a preliminary understanding of communication dynamics and learner tendencies within these sorts of learning environments, we now hope to leverage this background knowledge to develop new methods for promoting engagement and the fulfilment of individual learning objectives in these settings—in particular, by trying to mitigate the “content overload” issues in some way. Stay tuned for updates 🙂


Anderson, A., Huttenlocher, D., Kleinberg, J. & Leskovec, J., Engaging with Massive Open Online Courses.  In: WWW ’14 Proceedings of the 23rd International World Wide Web Conference, Seoul, Korea. New York: ACM (2014).

Read the full paper: Gillani, N., Yasseri, T., Eynon, R., and Hjorth, I. (2014) Structural limitations of learning in a crowd – communication vulnerability and information diffusion in MOOCs. Scientific Reports 4.

Rebecca Eynon was talking to blog editor David Sutcliffe.

Rebecca Eynon holds a joint academic post between the Oxford Internet Institute (OII) and the Department of Education at the University of Oxford. Her research focuses on education, learning and inequalities, and she has carried out projects in a range of settings (higher education, schools and the home) and life stages (childhood, adolescence and late adulthood).

Investigating virtual production networks in Sub-Saharan Africa and Southeast Asia

Ed: You are looking at the structures of ‘virtual production networks’ to understand the economic and social implications of online work. How are you doing this?

Mark: We are studying online freelancing. In other words this is digital or digitised work for which professional certification or formal training is usually not required. The work is monetised or monetisable, and can be mediated through an online marketplace.

Freelancing is a very old format of work. What is new is the fact that we have almost three billion people connected to a global network: many of those people are potential workers in virtual production networks. This mass connectivity has been one crucial ingredient for some significant changes in how work is organised, divided, outsourced, and rewarded. What we plan to do in this project is better map the contours of some of those changes and understand who wins and who doesn’t in this new world of work.

Ed: Are you able to define what comprises an individual contribution to a ‘virtual production network’ — or to find data on it? How do you define and measure value within these global flows and exchanges?

Mark: It is very far from easy. Much of what we are studying is immaterial and digitally-mediated work. We can find workers and we can find clients, but the links between them are often opaque and black-boxed. Some of the workers that we have spoken to operate under non-disclosure agreements, and many actually haven’t been told what their work is being used for.

But that is precisely why we felt the need to embark on this project. With a combination of quantitative transaction data from key platforms and qualitative interviews in which we attempt to piece together parts of the network, we want to understand who is (and isn’t) able to capture and create value within these networks.

Ed: You note that “within virtual production networks, are we seeing a shift in the boundaries of firms” — to what extend to you think we seeing the emergence of new forms of organisation?

Mark: There has always been a certain spatial stickiness to some activities carried out by firms (or within firms). Some activities required the complex exchanges of knowledge that were difficult to digitally mediate. But digitisation and better connectivity in low-wage countries has now allowed many formerly ‘in-house’ business processes to be outsourced to third-parties. In an age of cloud computing, cheap connectivity, and easily accessible collaboration tools, geography has become less sticky. One task that we are engaged in is looking at the ways that some kinds of tacit knowledge that are difficult to transmit digitally offer some people and firms (in different places) competitive advantages and disadvantages.

This proliferation of digitally mediated work could also be seen as a new form of organisation. The organisations that control key work marketplaces (like oDesk) make decisions that shape both who buyers and sellers are able to connect with, and the ways in which they are able to transact.

Ed: Does ‘virtual work’ add social or economic value to individuals in low-income countries? ie are we really dealing with a disintermediated, level surface on a global playing field, or just a different form of old exploitation (ie a virtual rather than physical extraction industry)?

Mark: That is what we aim to find out. Many have pointed to the potentials of online freelancing to create jobs and bring income to workers in low-income countries. But many others have argued that such practices are creating ‘digital sweatshops’ and facilitating a race to the bottom.

We undoubtedly are not seeing a purely disintermediated market, or a global playing field. But what we want to understand is who exactly benefits from these new networks of work, and how.

Ed: Will you be doing any network analysis of the data you collect, ie of actual value-flows? And will they be geolocated networks?

Mark: Yes! I am actually preparing a post that contains a geographic network of all work conducted over the course of a month via oDesk (see the website of the OII’s Connectivity, Inclusion, and Inequality Group for more..).

Mark Graham was talking to blog editor David Sutcliffe.

Mark Graham is a Senior Research Fellow at the OII. His research focuses on Internet and information geographies, and the overlaps between ICTs and economic development.

What are the limitations of learning at scale? Investigating information diffusion and network vulnerability in MOOCs

Millions of people worldwide are currently enrolled in courses provided on large-scale learning platforms (aka ‘MOOCs’), typically collaborating in online discussion forums with thousands of peers. Current learning theory emphasizes the importance of this group interaction for cognition. However, while a lot is known about the mechanics of group learning in smaller and traditionally organized online classrooms, fewer studies have examined participant interactions when learning “at scale”. Some studies have used clickstream data to trace participant behaviour; even predicting dropouts based on their engagement patterns. However, many questions remain about the characteristics of group interactions in these courses, highlighting the need to understand whether — and how — MOOCs allow for deep and meaningful learning by facilitating significant interactions.

But what constitutes a “significant” learning interaction? In large-scale MOOC forums, with socio-culturally diverse learners with different motivations for participating, this is a non-trivial problem. MOOCs are best defined as “non-formal” learning spaces, where learners pick and choose how (and if) they interact. This kind of group membership, together with the short-term nature of these courses, means that relatively weak inter-personal relationships are likely. Many of the tens of thousands of interactions in the forum may have little relevance to the learning process. So can we actually define the underlying network of significant interactions? Only once we have done this can we explore firstly how information flows through the forums, and secondly the robustness of those interaction networks: in short, the effectiveness of the platform design for supporting group learning at scale.

To explore these questions, we analysed data from 167,000 students registered on two business MOOCs offered on the Coursera platform. Almost 8000 students contributed around 30,000 discussion posts over the six weeks of the courses; almost 30,000 students viewed at least one discussion thread, totalling 321,769 discussion thread views. We first modelled these communications as a social network, with nodes representing students who posted in the discussion forums, and edges (ie links) indicating co-participation in at least one discussion thread. Of course, not all links will be equally important: many exchanges will be trivial (‘hello’, ‘thanks’ etc.). Our task, then, was to derive a “true” network of meaningful student interactions (ie iterative, consistent dialogue) by filtering out those links generated by random encounters (Figure 1; see also full paper for methodology).

Figure 1. Comparison of observed (a; ‘all interactions’) and filtered (b; ‘significant interactions’) communication networks for a MOOC forum. Filtering affects network properties such as modularity score (ie degree of clustering). Colours correspond to the automatically detected interest communities.
One feature of networks that has been studied in many disciplines is their vulnerability to fragmentation when nodes are removed (the Internet, for example, emerged from US Army research aiming to develop a disruption-resistant network for critical communications). While we aren’t interested in the effect of missile strike on MOOC exchanges, from an educational perspective it is still useful to ask which “critical set” of learners is mostly responsible for information flow in a communication network — and what would happen to online discussions if these learners were removed. To our knowledge, this is the first time vulnerability of communication networks has been explored in an educational setting.

Network vulnerability is interesting because it indicates how integrated and inclusive the communication flow is. Discussion forums with fleeting participation will have only a very few vocal participants: removing these people from the network will markedly reduce the information flow between the other participants — as the network falls apart, it simply becomes more difficult for information to travel across it via linked nodes. Conversely, forums that encourage repeated engagement and in-depth discussion among participants will have a larger ‘critical set’, with discussion distributed across a wide range of learners.

To understand the structure of group communication in the two courses, we looked at how quickly our modelled communication network fell apart when: (a) the most central nodes were iteratively disconnected (Figure 2; blue), compared with when (b) nodes were removed at random (ie the ‘neutral’ case; green). In the random case, the network degrades evenly, as expected. When we selectively remove the most central nodes, however, we see rapid disintegration: indicating the presence of individuals who are acting as important ‘bridges’ across the network. In other words, the network of student interactions is not random: it has structure.

Figure 2. Rapid network degradation results from removal of central nodes (blue). This indicates the presence of individuals acting as ‘bridges’ between sub-groups. Removing these bridges results in rapid degradation of the overall network. Removal of random nodes (green) results in a more gradual degradation.
Figure 2. Rapid network degradation results from removal of central nodes (blue). This indicates the presence of individuals acting as ‘bridges’ between sub-groups. Removing these bridges results in rapid degradation of the overall network. Removal of random nodes (green) results in a more gradual degradation.

Of course, the structure of participant interactions will reflect the purpose and design of the particular forum. We can see from Figure 3 that different forums in the courses have different vulnerability thresholds. Forums with high levels of iterative dialogue and knowledge construction — with learners sharing ideas and insights about weekly questions, strategic analyses, or course outcomes — are the least vulnerable to degradation. A relatively high proportion of nodes have to be removed before the network falls apart (rightmost-blue line). Forums where most individuals post once to introduce themselves and then move their discussions to other platforms (such as Facebook) or cease engagement altogether tend to be more vulnerable to degradation (left-most blue line). The different vulnerability thresholds suggest that different topics (and forum functions) promote different levels of forum engagement. Certainly, asking students open-ended questions tended to encourage significant discussions, leading to greater engagement and knowledge construction as they read analyses posted by their peers and commented with additional insights or critiques.

Figure 3 – Network vulnerabilities of different course forums.
Figure 3 – Network vulnerabilities of different course forums.

Understanding something about the vulnerability of a communication or interaction network is important, because it will tend to affect how information spreads across it. To investigate this, we simulated an information diffusion model similar to that used to model social contagion. Although simplistic, the SI model (‘susceptible-infected’) is very useful in analyzing topological and temporal effects on networked communication systems. While the model doesn’t account for things like decaying interest over time or peer influence, it allows us to compare the efficiency of different network topologies.

We compared our (real-data) network model with a randomized network in order to see how well information would flow if the community structures we observed in Figure 2 did not exist. Figure 4 shows the number of ‘infected’ (or ‘reached’) nodes over time for both the real (solid lines) and randomized networks (dashed lines). In all the forums, we can see that information actually spreads faster in the randomised networks. This is explained by the existence of local community structures in the real-world networks: networks with dense clusters of nodes (i.e. a clumpy network) will result in slower diffusion than a network with a more even distribution of communication, where participants do not tend to favor discussions with a limited cohort of their peers.

Figure 4 (a) shows the percentage of infected nodes vs. simulation time for different networks. The solid lines show the results for the original network and the dashed lines for the random networks. (b) shows the time it took for a simulated “information packet” to come into contact with half the network’s nodes.
Figure 4 (a) shows the percentage of infected nodes vs. simulation time for different networks. The solid lines show the results for the original network and the dashed lines for the random networks. (b) shows the time it took for a simulated “information packet” to come into contact with half the network’s nodes.

Overall, these results reveal an important characteristic of student discussion in MOOCs: when it comes to significant communication between learners, there are simply too many discussion topics and too much heterogeneity (ie clumpiness) to result in truly global-scale discussion. Instead, most information exchange, and by extension, any knowledge construction in the discussion forums occurs in small, short-lived groups: with information “trapped” in small learner groups. This finding is important as it highlights structural limitations that may impact the ability of MOOCs to facilitate communication amongst learners that look to learn “in the crowd”.

These insights into the communication dynamics motivate a number of important questions about how social learning can be better supported, and facilitated, in MOOCs. They certainly suggest the need to leverage intelligent machine learning algorithms to support the needs of crowd-based learners; for example, in detecting different types of discussion and patterns of engagement during the runtime of a course to help students identify and engage in conversations that promote individualized learning. Without such interventions the current structural limitations of social learning in MOOCs may prevent the realization of a truly global classroom.

The next post addresses qualitative content analysis and how machine-learning community detection schemes can be used to infer latent learner communities from the content of forum posts.

Read the full paper: Gillani, N., Yasseri, T., Eynon, R., and Hjorth, I. (2014) Structural limitations of learning in a crowd – communication vulnerability and information diffusion in MOOCs. Scientific Reports 4.

Rebecca Eynon holds a joint academic post between the Oxford Internet Institute (OII) and the Department of Education at the University of Oxford. Her research focuses on education, learning and inequalities, and she has carried out projects in a range of settings (higher education, schools and the home) and life stages (childhood, adolescence and late adulthood).

Verification of crowd-sourced information: is this ‘crowd wisdom’ or machine wisdom?

Crisis mapping platform
‘Code’ or ‘law’? Image from an Ushahidi development meetup by afropicmusing.

In ‘Code and Other Laws of Cyberspace’, Lawrence Lessig (2006) writes that computer code (or what he calls ‘West Coast code’) can have the same regulatory effect as the laws and legal code developed in Washington D.C., so-called ‘East Coast code’. Computer code impacts on a person’s behaviour by virtue of its essentially restrictive architecture: on some websites you must enter a password before you gain access, in other places you can enter unidentified. The problem with computer code, Lessig argues, is that it is invisible, and that it makes it easy to regulate people’s behaviour directly and often without recourse.

For example, fair use provisions in US copyright law enable certain uses of copyrighted works, such as copying for research or teaching purposes. However the architecture of many online publishing systems heavily regulates what one can do with an e-book: how many times it can be transferred to another device, how many times it can be printed, whether it can be moved to a different format – activities that have been unregulated until now, or that are enabled by the law but effectively ‘closed off’ by code. In this case code works to reshape behaviour, upsetting the balance between the rights of copyright holders and the rights of the public to access works to support values like education and innovation.

Working as an ethnographic researcher for Ushahidi, the non-profit technology company that makes tools for people to crowdsource crisis information, has made me acutely aware of the many ways in which ‘code’ can become ‘law’. During my time at Ushahidi, I studied the practices that people were using to verify reports by people affected by a variety of events – from earthquakes to elections, from floods to bomb blasts. I then compared these processes with those followed by Wikipedians when editing articles about breaking news events. In order to understand how to best design architecture to enable particular behaviour, it becomes important to understand how such behaviour actually occurs in practice.

In addition to the impact of code on the behaviour of users, norms, the market and laws also play a role. By interviewing both the users and designers of crowdsourcing tools I soon realized that ‘human’ verification, a process of checking whether a particular report meets a group’s truth standards, is an acutely social process. It involves negotiation between different narratives of what happened and why; identifying the sources of information and assessing their reputation among groups who are considered important users of that information; and identifying gatekeeping and fact checking processes where the source is a group or institution, amongst other factors.

One disjuncture between verification ‘practice’ and the architecture of the verification code developed by Ushahidi for users was that verification categories were set as a default feature, whereas some users of the platform wanted the verification process to be invisible to external users. Items would show up as being ‘unverified’ unless they had been explicitly marked as ‘verified’, thus confusing users about whether the item was unverified because the team hadn’t yet verified it, or whether it was unverified because it had been found to be inaccurate. Some user groups wanted to be able to turn off such features when they could not take responsibility for data verification. In the case of the Christchurch Recovery Map in the aftermath of the 2011 New Zealand earthquake, the government officials with whom volunteers who set up the Ushahidi instance were working wanted to be able to turn off such features because they were concerned that they could not ensure that reports were indeed verified and having the category show up (as ‘unverified’ until ‘verified’) implied that they were engaged in some kind of verification process.

The existence of a default verification category impacted on the Christchurch Recovery Map group’s ability to gain support from multiple stakeholders, including the government, but this feature of the platform’s architecture did not have the same effect in other places and at other times. For other users like the original Ushahidi Kenya team who worked to collate instances of violence after the Kenyan elections in 2007/08, this detailed verification workflow was essential to counter the misinformation and rumour that dogged those events. As Ushahidi’s use cases have diversified – from reporting death and damage during natural disasters to political events including elections, civil war and revolutions, the architecture of Ushahidi’s code base has needed to expand. Ushahidi has recognised that code plays a defining role in the experience of verification practices, but also that code’s impact will not be the same at all times, and in all circumstances. This is why it invested in research about user diversity in a bid to understand the contexts in which code runs, and how these contexts result in a variety of different impacts.

A key question being asked in the design of future verification mechanisms is the extent to which verification work should be done by humans or non-humans (machines). Here, verification is not a binary categorisation, but rather there is a spectrum between human and non-human verification work, and indeed, projects like Ushahidi, Wikipedia and Galaxy Zoo have all developed different verification mechanisms. Wikipedia uses a set of policies and practices about how content should be added and reviewed, such as the use of ‘citation needed’ tags for information that sounds controversial and that should be backed up by a reliable source. Galaxy Zoo uses an algorithm to detect whether certain contributions are accurate by comparing them to the same work by other volunteers.

Ushahidi leaves it up to individual deployers of their tools and platform to make decisions about verification policies and practices, and is going to be designing new defaults to accommodate this variety of use. In parallel,, a project by ex-Ushahidi Patrick Meier with organisations Masdar and QCRI is responding to the large amounts of unverified and often contradictory information that appears on social media following natural disasters by enabling social media users to collectively evaluate the credibility of rapidly crowdsourced evidence. The project was inspired by MIT’s winning entry to DARPA’s ‘Red Balloon Challenge’ which was intended to highlight social networking’s potential to solve widely distributed, time-sensitive problems, in this case by correctly identifying the GPS coordinates of 10 balloons suspended at fixed, undisclosed locations across the US. The winning MIT team crowdsourced the problem by using a monetary incentive structure, promising $2,000 to the first person who submitted the correct coordinates for a single balloon, $1,000 to the person who invited that person to the challenge; $500 to the person who invited the inviter, and so on. The system quickly took root, spawning geographically broad, dense branches of connections. After eight hours and 52 minutes, the MIT team identified the correct coordinates for all 10 balloons. aims to apply MIT’s approach to the process of rapidly collecting and evaluating critical evidence during disasters: “Instead of looking for weather balloons across an entire country in less than 9 hours, we hope will facilitate the crowdsourced collection of multimedia evidence for individual disasters in under 9 minutes.” It is still unclear how (or whether) Verily will be able to reproduce the same incentive structure, but a bigger question lies around the scale and spread of social media in the majority of countries where humanitarian assistance is needed. The majority of Ushahidi or Crowdmap installations are, for example, still “small data” projects, with many focused on areas that still require offline verification procedures (such as calling volunteers or paid staff who are stationed across a country, as was the case in Sudan [3]). In these cases – where the social media presence may be insignificant — a team’s ability to achieve a strong local presence will define the quality of verification practices, and consequently the level of trust accorded to their project.

If code is law and if other aspects in addition to code determine how we can act in the world, it is important to understand the context in which code is deployed. Verification is a practice that determines how we can trust information coming from a variety of sources. Only by illuminating such practices and the variety of impacts that code can have in different environments can we begin to understand how code regulates our actions in crowdsourcing environments.

For more on Ushahidi verification practices and the management of sources on Wikipedia during breaking news events, see:

[1] Ford, H. (2012) Wikipedia Sources: Managing Sources in Rapidly Evolving Global News Articles on the English Wikipedia. SSRN Electronic Journal. doi:10.2139/ssrn.2127204

[2] Ford, H. (2012) Crowd Wisdom. Index on Censorship 41(4), 33–39. doi:10.1177/0306422012465800

[3] Ford, H. (2011) Verifying information from the crowd. Ushahidi.

Heather Ford has worked as a researcher, activist, journalist, educator and strategist in the fields of online collaboration, intellectual property reform, information privacy and open source software in South Africa, the United Kingdom and the United States. She is currently a DPhil student at the OII, where she is studying how Wikipedia editors write history as it happens in a format that is unprecedented in the history of encyclopedias. Before this, she worked as an ethnographer for Ushahidi. Read Heather’s blog.

For more on the ChristChurch Earthquake, and the role of digital humanities in preserving the digital record of its impact see: Preserving the digital record of major natural disasters: the CEISMIC Canterbury Earthquakes Digital Archive project on this blog.

Ethical privacy guidelines for mobile connectivity measurements

Four of the 6.8 billion mobile phones worldwide. Measuring the mobile Internet can expose information about an individual’s location, contact details, and communications metadata. Image by Cocoarmani.

Ed: GCHQ / the NSA aside … Who collects mobile data and for what purpose? How can you tell if your data are being collected and passed on?

Ben: Data collected from mobile phones is used for a wide range of (divergent) purposes. First and foremost, mobile operators need information about mobile phones in real-time to be able to communicate with individual mobile handsets. Apps can also collect all sorts of information, which may be necessary to provide entertainment, location specific services, to conduct network research and many other reasons.

Mobile phone users usually consent to the collection of their data by clicking “I agree” or other legally relevant buttons, but this is not always the case. Sometimes data is collected lawfully without consent, for example for the provision of a mobile connectivity service. Other times it is harder to substantiate a relevant legal basis. Many applications keep track of the information that is generated by a mobile phone and it is often not possible to find out how the receiver processes this data.

Ed: How are data subjects typically recruited for a mobile research project? And how many subjects might a typical research data set contain?

Ben: This depends on the research design; some research projects provide data subjects with a specific app, which they can use to conduct measurements (so called ‘active measurements’). Other apps collect data in the background and, in effect, conduct local surveillance of the mobile phone use (so called passive measurements). Other research uses existing datasets, for example provided by telecom operators, which will generally be de-identified in some way. We purposely do not use the term anonymisation in the report, because much research and several case studies have shown that real anonymisation is very difficult to achieve if the original raw data is collected about individuals. Datasets can be re-identified by techniques such as fingerprinting or by linking them with existing, auxiliary datasets.

The size of datasets differs per release. Telecom operators can provide data about millions of users, while it will be more challenging to reach such a number with a research specific app. However, depending on the information collected and provided, a specific app may provide richer information about a user’s behaviour.

Ed: What sort of research can be done with this sort of data?

Ben: Data collected from mobile phones can reveal much interesting and useful information. For example, such data can show exact geographic locations and thus the movements of the owner, which can be relevant for the social sciences. On a larger scale, mass movements of persons can be monitored via mobile phones. This information is useful for public policy objectives such as crowd control, traffic management, identifying migration patterns, emergency aid, etc. Such data can also be very useful for commercial purposes, such as location specific advertising, studying the movement of consumers, or generally studying the use of mobile phones.

Mobile phone data is also necessary to understand the complex dynamics of the underlying Internet architecture. The mobile Internet is has different requirements than the fixed line Internet, so targeted investments in future Internet architecture will need to be assessed by detailed network research. Also, network research can study issues such as censorship or other forms of blocking information and transactions, which are increasingly carried out through mobile phones. This can serve as early warning systems for policy makers, activists and humanitarian aid workers, to name only a few stakeholders.

Ed: Some of these research datasets are later published as ‘open data’. What sorts of uses might researchers (or companies) put these data to? Does it tend to be mostly technical research, or there also social science applications?

Ben: The intriguing characteristic of the open data concept is that secondary uses can be unpredictable. A re-use is not necessarily technical, even if the raw data has been collected for a purely technical network research. New social science research could be based on existing technical data, or existing research analyses may be falsified or validated by other researchers. Artists, developers, entrepreneurs or public authorities can also use existing data to create new applications or to enrich existing information systems. There have been many instances when open data has been re-used for beneficial or profitable means.

However, there is also a flipside to open data, especially when the dataset contains personal information, or information that can be linked to individuals. A working definition of open data is that one makes entire databases available, in standardized, machine readable and electronic format, to any secondary user, free of charge and free of restrictions or obligations, for any purpose. If a dataset contains information about your Internet browsing habits, your movements throughout the day or the phone numbers you have called over a specific period of time, it could be quite troubling if you have no control over who re-uses this information.

The risks and harms of such re-use are very context dependent, of course. In the Western world, such data could be used as means for blackmail, stalking, identity theft, unsolicited commercial communications, etc. Further, if there is a chance our telecom operators just share data on how we use our mobile phones, we may refrain from activities, such as taking part in demonstrations, attending political gatherings, or accessing certain socially unacceptable information. Such self-censorship will damage the free society we expect. In the developing world, or in authoritarian regimes, risks and harms can be a matter of life and death for data subjects, or at least involve the risk of physical harm. This is true for all citizens, but also diplomats, aid workers and journalists or social media users.

Finally, we cannot envisage how political contexts will change in the future. Future malevolent governments, even in Europe or the US, could easily use datasets containing sensitive information to harm or control specific groups of society. One only need look at the changing political landscape in Hungary to see how specific groups are suddenly targeted in what we thought was becoming a country that adheres to Western values.

Ed: The ethical privacy guidelines note the basic relation between the level of detail in information collected and the resulting usefulness of the dataset (datasets becoming less powerful as subjects are increasingly de-identified). This seems a fairly intuitive and fundamentally unavoidable problem; is there anything in particular to say about it?

Ben: Research often requires rich datasets for worthwhile analyses to be conducted. These will inevitably sometimes contain personal information, as it can be important to relate specific data to data subjects, whether anonymised, pseudonymised or otherwise. Far reaching deletion, aggregation or randomisation of data can make the dataset useless for the research purposes.

Sophisticated methods of re-identifying datasets, and unforeseen methods which will be developed in future, mean that much information must be deleted or aggregated in order for a dataset containing personal information to be truly anonymous. It has become very difficult to determine when a dataset is sufficiently anonymised to the extent that it can enjoy the legal exception offered by data protection laws around the world and therefore be distributed as open data, without legal restrictions.

As a result, many research datasets cannot simply be released. The guidelines do not force the researcher to a zero-risk situation, where only useless or meaningless datasets can be released. The guidelines force the researcher to think very carefully about the type of data that will be collected, about data processing techniques and different disclosure methods. Although open data is an attractive method of disseminating research data, sometimes managed access systems may be more appropriate. The guidelines constantly trigger the researcher to consider the risks to data subjects in their specific context during each stage of the research design. They serve as a guide, but also a normative framework for research that is potentially privacy invasive.

Ed: Presumably mobile companies have a duty to delete their data after a certain period; does this conflict with open datasets, whose aim is to be available indefinitely?

Ben: It is not a requirement for open data to be available indefinitely. However, once information is published freely on the Internet, it is very hard – if not impossible – to delete it. The researcher loses all control over a dataset once it is published online. So, if a dataset is sufficiently de-identified for the re-identification techniques that are known today, this does not mean that future techniques cannot re-identify the dataset. We can’t expect researchers to take into account all science-fiction type future developments, but the guidelines to force the researcher to consider what successful re-identification would reveal about data subjects.

European mobile phone companies do have a duty to keep logs of communications for 6 months to 2 years, depending on the implication of the misguided data retention directive. We have recently learned that intelligence services worldwide have more or less unrestricted access to such information. We have no idea how long this information is stored in practice. Recently it has been frequently been stated that deleting data has become more expensive than just keeping it. This means that mobile phone operators and intelligence agencies may keep data on our mobile phone use forever. This must be taken into account when assessing which auxiliary datasets could be used to re-identify a research dataset. An IP-address could be sufficient to link much information to an individual.

Ed: Presumably it’s impossible for a subject to later decide they want to be taken out of an open dataset; firstly due to cost, but also because (by definition) it ought to be impossible to find them in an anonymised dataset. Does this present any practical or legal problems?

Ben: In some countries, especially in Europe, data subjects have a legal right to object to their data being processed, by withdrawing consent or engaging in a legal procedure with the data processor. Although this is an important right, exercising it may lead to undesirable consequences for research. For example, the underlying dataset will be incomplete for secondary researchers who want to validate findings.

Our guidelines encourage researchers to be transparent about their research design, data processing and foreseeable secondary uses of the data. On the one hand, this builds trust in the network research discipline. On the other, it gives data subjects the necessary information to feel confident to share their data. Still, data subjects should be able to retract their consent via electronic means, instead of sending letters, if they can substantiate an appreciable harm to them.

Ed: How aware are funding bodies and ethics boards of the particular problems presented by mobile research; and are they categorically different from other human-subject research data? (eg interviews / social network data / genetic studies etc.)

Ben: University ethical boards or funding bodies are be staffed by experts in a wide range of disciplines. However, this does not mean they understand the intricate details of complex Internet measurements, de-identification techniques or the state of affairs with regards to re-identification techniques, nor the harms a research programme can inflict given a specific context. For example, not everyone’s intuitive moral privacy compass will be activated when they read in a research proposal that the research systems will “monitor routing dynamics, by analysing packet traces collected from cell towers and internet exchanges”, or similar sentences.

Our guidelines encourage the researcher to write up the choices made with regards to personal information in a manner that is clear and understandable for the layperson. Such a level of transparency is useful for data subjects —  as well as ethical boards and funding bodies — to understand exactly what the research entails and how risks have been accommodated.

Ed: Linnet Taylor has already discussed mobile data mining from regions of the world with weak privacy laws: what is the general status of mobile privacy legislation worldwide?

Ben: Privacy legislation itself is about as fragmented and disputed as it gets. The US generally treats personal information as a commodity that can be traded, which enables Internet companies in Silicon Valley to use data as the new raw material in the information age. Europe considers privacy and data protection as a fundamental right, which is currently regulated in detail, albeit based on a law from 1995. The review of European data protection regulation has been postponed to 2015, possibly as a result of the intense lobbying effort in Brussels to either weaken or strengthen the proposed law. Some countries have not regulated privacy or data protection at all. Other countries have a fundamental right to privacy, which is not further developed in a specific data protection law and thus hardly enforced. Another group of countries have transplanted the European approach, but do not have the legal expertise to apply the 1995 law to the digital environment. The future of data protection is very much up in the air and requires much careful study.

The guidelines we have publishing take the international human rights framework as a base, while drawing inspiration from several existing legal concepts such as data minimisation, purpose limitation, privacy by design and informed consent. The guidelines give a solid base for privacy aware research design. We do encourage researchers to discuss their projects with colleagues and legal experts as much as possible, though, because best practices and legal subtleties can vary per country, state or region.

Read the guidelines: Zevenbergen, B., Brown,I., Wright, J., and Erdos, D. (2013) Ethical Privacy Guidelines for Mobile Connectivity Measurements. Oxford Internet Institute, University of Oxford.

Ben Zevenbergen was talking to blog editor David Sutcliffe.

Investigating the structure and connectivity of online global protest networks

How have online technologies reconfigured collective action? It is often assumed that the rise of social networking tools, accompanied by the mass adoption of mobile devices, have strengthened the impact and broadened the reach of today’s political protests. Enabling massive self-communication allows protesters to write their own interpretation of events – free from a mass media often seen as adversarial – and emerging protests may also benefit from the cheaper, faster transmission of information and more effective mobilization made possible by online tools such as Twitter.

The new networks of political protest, which harness these new online technologies are often described in theoretical terms as being ‘fluid’ and ‘horizontal’, in contrast to the rigid and hierarchical structure of earlier protest organization. Yet such theoretical assumptions have seldom been tested empirically. This new language of networks may be useful as a shorthand to describe protest dynamics, but does it accurately reflect how protest networks mediate communication and coordinate support?

The global protests against austerity and inequality which took place on May 12, 2012 provide an interesting case study to test the structure and strength of a transnational online protest movement. The ‘indignados’ movement emerged as a response to the Spanish government’s politics of austerity in the aftermath of the global financial crisis. The movement flared in May 2011, when hundreds of thousands of protesters marched in Spanish cities, and many set up camps ahead of municipal elections a week later.

These protests contributed to the emergence of the worldwide Occupy movement. After the original plan to occupy New York City’s financial district mobilised thousands of protesters in September 2011, the movement spread to other cities in the US and worldwide, including London and Frankfurt, before winding down as the camp sites were dismantled weeks later. Interest in these movements was revived, however, as the first anniversary of the ‘indignados’ protests approached in May 2012.

To test whether the fluidity, horizontality and connectivity often claimed for online protest networks holds true in reality, tweets referencing these protest movements during May 2012 were collected. These tweets were then classified as relating either to the ‘indignados’ or Occupy movement, using hashtags as a proxy for content. Many tweets, however, contained hashtags relevant for the two movements, creating bridges across the two streams of information. The users behind those bridges acted as  information ‘brokers’, and are fundamentally important to the global connectivity of the two movements: they joined the two streams of information and their audiences on Twitter. Once all the tweets were classified by content and author, it emerged that around 6.5% of all users posted at least one message relevant for the two movements by using hashtags from both sides jointly.

Analysis of the Twitter data shows that this small minority of ‘brokers’ play an important role connecting users to a network that would otherwise be disconnected. Brokers are significantly more active in the contribution of messages and more visible in the stream of information, being re-tweeted and mentioned more often than other users. The analysis also shows that these brokers play an important role in the global network, by helping to keep the network together and improving global connectivity. In a simulation, the removal of brokers fragmented the network faster than the removal of random users at the same rate.

What does this tell us about global networks of protest? Firstly, it is clear that global networks are more vulnerable and fragile than is often assumed. Only a small percentage of users disseminate information across transnational divides, and if any of these users cease to perform this role, they are difficult to immediately replace, thus limiting the assumed fluidity of such networks. The decentralized nature of online networks, with no central authority imposing order or even suggesting a common strategy, make the role of ‘brokers’ all the more vital to the survival of networks which cross national borders.

Secondly, the central role performed by brokers suggests that global networks of online protest lack the ‘horizontal’ structure that is often described in the literature. Talking about horizontal structures can be useful as shorthand to refer to decentralised organisations, but not to analyse the process by which these organisations materialise in communication networks. The distribution of users in those networks reveals a strong hierarchy in terms of connections and the ability to communicate effectively.

Future research into online networks, then, should keep in mind that the language of protest networks in the digital age, particularly terms like horizontality and fluidity, do not necessarily stand up to empirical scrutiny. The study of contentious politics in the digital age should be evaluated, first and foremost, through the lens of what protesters actually reveal through their actions.

Read the paper: Sandra Gonzalez-Bailon and Ning Wang (2013) The Bridges and Brokers of Global Campaigns in the Context of Social Media.

Uncovering the structure of online child exploitation networks

The Internet has provided the social, individual, and technological circumstances needed for child pornography to flourish. Sex offenders have been able to utilize the Internet for dissemination of child pornographic content, for social networking with other pedophiles through chatrooms and newsgroups, and for sexual communication with children. A 2009 estimate by the United Nations estimates that there are more than four million websites containing child pornography, with 35 percent of them depicting serious sexual assault [1]. Even if this report or others exaggerate the true prevalence of those websites by a wide margin, the fact of the matter is that those websites are pervasive on the world wide web.

Despite large investments of law enforcement resources, online child exploitation is nowhere near under control, and while there are numerous technological products to aid in finding child pornography online, they still require substantial human intervention. Despite this, steps can be taken to increase the automation process of these searches, to reduce the amount of content police officers have to examine, and increase the time they can spend on investigating individuals.

While law enforcement agencies will aim for maximum disruption of online child exploitation networks by targeting the most connected players, there is a general lack of research on the structural nature of these networks; something we aimed to address in our study, by developing a method to extract child exploitation networks, map their structure, and analyze their content. Our custom-written Child Exploitation Network Extractor (CENE) automatically crawls the Web from a user-specified seed page, collecting information about the pages it visits by recursively following the links out of the page; the result of the crawl is a network structure containing information about the content of the websites, and the linkages between them [2].

We chose ten websites as starting points for the crawls; four were selected from a list of known child pornography websites while the other six were selected and verified through Google searches using child pornography search terms. To guide the network extraction process we defined a set of 63 keywords, which included words commonly used by the Royal Canadian Mounted Police to find illegal content; most of them code words used by pedophiles. Websites included in the analysis had to contain at least seven of the 63 unique keywords, on a given web page; manual verification showed us that seven keywords distinguished well between child exploitation web pages and regular web pages. Ten sports networks were analyzed as a control.

The web crawler was found to be able to properly identify child exploitation websites, with a clear difference found in the hardcore content hosted by child exploitation and non-child exploitation websites. Our results further suggest that a ‘network capital’ measure — which takes into account network connectivity, as well as severity of content — could aid in identifying the key players within online child exploitation networks. These websites are the main concern of law enforcement agencies, making the web crawler a time saving tool in target prioritization exercises. Interestingly, while one might assume that website owners would find ways to avoid detection by a web crawler of the type we have used, these websites — despite the fact that much of the content is illegal — turned out to be easy to find. This fits with previous research that has found that only 20-25 percent of online child pornography arrestees used sophisticated tools for hiding illegal content [3,4].

As mentioned earlier, the huge amount of content found on the Internet means that the likelihood of eradicating the problem of online child exploitation is nil. As the decentralized nature of the Internet makes combating child exploitation difficult, it becomes more important to introduce new methods to address this. Social network analysis measurements, in general, can be of great assistance to law enforcement investigating all forms of online crime—including online child exploitation. By creating a web crawler that reduces the amount of hours officers need to spend examining possible child pornography websites, and determining whom to target, we believe that we have touched on a method to maximize the current efforts by law enforcement. An automated process has the added benefit of aiding to keep officers in the department longer, as they would not be subjugated to as much traumatic content.

There are still areas for further research; the first step being to further refine the web crawler. Despite being a considerable improvement over a manual analysis of 300,000 web pages, it could be improved to allow for efficient analysis of larger networks, bringing us closer to the true size of the full online child exploitation network, but also, we expect, to some of the more hidden (e.g., password/membership protected) websites. This does not negate the value of researching publicly accessible websites, given that they may be used as starting locations for most individuals.

Much of the law enforcement to date has focused on investigating images, with the primary reason being that databases of hash values (used to authenticate the content) exists for images, and not for videos. Our web crawler did not distinguish between the image content, but utilizing known hash values would help improve the validity of our severity measurement. Although it would be naïve to suggest that online child exploitation can be completely eradicated, the sorts of social network analysis methods described in our study provide a means of understanding the structure (and therefore key vulnerabilities) of online networks; in turn, greatly improving the effectiveness of law enforcement.

[1] Engeler, E. 2009. September 16. UN Expert: Child Porn on Internet Increases. The Associated Press.

[2] Westlake, B.G., Bouchard, M., and Frank, R. 2012. Finding the Key Players in Online Child Exploitation Networks. Policy and Internet 3 (2).

[3] Carr, J. 2004. Child Abuse, Child Pornography and the Internet. London: NCH.

[4] Wolak, J., D. Finkelhor, and K.J. Mitchell. 2005. “Child Pornography Possessors Arrested in Internet-Related Crimes: Findings from the National Juvenile Online Victimization Study (NCMEC 06–05–023).” Alexandria, VA: National Center for Missing and Exploited Children.

Read the full paper: Westlake, B.G., Bouchard, M., and Frank, R. 2012. Finding the Key Players in Online Child Exploitation Networks. Policy and Internet 3 (2).