Who Shapes Digital Learning in Southeast Asia? A NodeXL Study of X (Twitter)

Across four Southeast Asian countries, 745 accounts spent three years posting about digital learning. Governments, United Nations agencies, global technology companies, local NGOs, and grassroots teacher groups all had something to say about how children should learn online.

They almost never spoke to each other.

That is the central finding of a new open access study in the Journal of International Cooperation in Education, and it is a textbook demonstration of why network structure tells you things that content alone cannot.

The study

“Social network analysis of digital learning in Southeast Asia: trends and insights from X” was written by Richelle Elisan Gerong of the University of Science and Technology of Southern Philippines and Chia Chun Tiew of Universiti Tunku Abdul Rahman in Malaysia. It was published in April 2026.

The researchers used NodeXL to collect and map posts on X from the Philippines, Malaysia, Indonesia, and Thailand, searching on “Digital Learning” alongside each country name. The collection window ran from February 2022 to February 2025. They kept active accounts with at least 100 followers, which produced a network of 745 users connected by 953 interactions.

To find who mattered, they used betweenness centrality to locate the bridges between separate discussion clusters, in-degree and out-degree to measure incoming and outgoing engagement, and follower counts to gauge potential reach. Modularity scores grouped the network into communities.

Finding 1: the network barely holds together

Start with the structure, because it frames everything else.

• The 745 users fell into 505 separate connected components. A connected component is an island of users who can reach each other but not the rest of the graph. Five hundred and five islands among 745 people is close to total fragmentation.
• 408 of those components contained exactly one user. That is 55 percent of everyone in the study posting into the void, connected to no one at all. Network analysts call these isolates.
• The largest connected component held just 31 users.
• Graph density came to 0.00045. Density is the share of all possible connections that actually exist, so this is about as sparse as a network gets.
• The reciprocated edge ratio was 0.03, meaning roughly three percent of connections went both ways. Almost nothing was a two-way exchange.
• Of 953 total edges, 663 were self-loops. In a NodeXL X network, a self-loop is a post that replies to, mentions, or retweets no one. Seven in ten interactions in this dataset were someone talking to themselves.

Modularity landed at 0.445, and the clustering produced 98 distinct groups, of which eight were substantial. So there are real communities here. They just do not touch.

Finding 2: it is a brand cluster, and that has a specific meaning

The authors classify the resulting shape as a brand cluster. In their words, users “are relaying the messages of certain institutions or persons over the topic of digital learning without exchanging extra ideas or information,” and participants “mention brands when they discuss digital learning, but they likely do not connect to each other.”

This is worth pausing on, because the brand cluster is one of the recognised shapes that topic networks take on social media. It is the signature of a subject that is widely broadcast and rarely debated. Lots of voices, pointed at the same handful of institutional names, with no conversation running between them.

A hashtag count would have told you digital learning was a popular topic in Southeast Asia. Only the network map tells you the popularity is an illusion of scale, built from hundreds of disconnected monologues.

Finding 3: the bridges were global institutions, not local ones

Ranked by betweenness centrality, the accounts holding the network together were:

1. Huawei Tech4All, score 592, with 1.59 million followers
2. Malaysia Digital Economy Corporation (MDEC), score 350
3. UNESCO Digital Learning, score 344
4. UNESCO’s SDG4 Global Education Cooperation Mechanism, score 344
5. Indonesia’s Ministry of Primary and Secondary Education, score 216
6. Google Malaysia, score 66
7. Resilience Development Initiative, Indonesia, score 58
8. GM Learning Club, Thailand, score 30
9. UNESCO Bangkok, score 16
10. Philippine Information Technology Organization, score 14

Two things stand out. First, the drop is severe: the top account scores more than forty times the tenth. Influence here is extraordinarily concentrated. Second, the top of the list is dominated by global technology firms and United Nations bodies, while the local and grassroots actors sit at the bottom.

The authors also note that these bridging accounts show low reciprocal engagement. They occupy structurally powerful positions without much back-and-forth, which is a familiar pattern: the accounts that connect a network are not necessarily the accounts that converse in it.

One detail deserves emphasis. Indonesia’s Ministry of Primary and Secondary Education was the only account in the top ten explicitly representing the K-12 sector, in a study specifically about K-12 digital education. The schools were barely present in their own conversation.

Finding 4: what they were actually talking about

The most used hashtags were #digital (44 uses), #learning (37), #indonesia (24), #tech4all (17), then #malaysia, #pgdukungpeningkatansdm, and #petrokimiagresik (16 each), and finally #telkomindonesia, #elevatingyourfuture, and #education (15 each).

Notice how many are corporate or campaign tags rather than topical ones. That is the brand cluster showing up again in the vocabulary.

Shared links pointed mostly to news outlets, universities, and institutional domains. The authors read these as “discursive anchors” that lend credibility, noting the prevalence of official domains such as .id, .my, .org, and .edu. Thematically, the discourse circled equity, infrastructure, sustainability, and digital skills.

Why this matters

The authors’ conclusion is a coordination argument rather than a criticism of any single actor. Ministries focus on reform and training. Companies promote digital skills and employability. Community groups share classroom practice. Each is doing sensible work, but the network shows those efforts running in parallel rather than together, which risks duplicated effort and narratives that never reach school-level practice.

Their recommendation is stronger regional coordination and deliberate partnership across sectors, so that policy, technology, and community engagement point the same direction.

For anyone doing communications or advocacy in this space, the practical read is blunter: if you are trying to influence digital learning policy in Southeast Asia and you are not connected to one of the handful of bridging accounts, your message is very likely going nowhere.

What the study does not claim

The authors are commendably direct about limits, and the most important one is worth repeating because it is easy to over-read this kind of research.

No sentiment or content analysis was performed. The study maps who is connected to whom, not what tone anyone took. As the authors put it, without content-level coding “it is difficult to assess whether the conversations across clusters were collaborative, critical, promotional or passive.”

They also caution against reading network fragmentation as institutional dysfunction. The disconnection visible on X “may be more reflective of the communication behaviors typical of social media ecosystems rather than actual policy or practice gaps.” Organisations that never interact on X may collaborate closely offline. The data covers one platform, publicly, and excludes accounts with fewer than 100 followers, which likely removes exactly the small classroom-level voices the study wishes it could hear more from.

Structure first, then content

What makes this study useful is the discipline of its sequence. Before asking what people said about digital learning, the researchers asked how the conversation was built. The answer, 505 islands and 408 people talking to nobody, reframes everything that follows.

That is the argument for network analysis in a sentence. Volume tells you a topic is popular. Structure tells you whether anything is actually being exchanged.

Read the study

Gerong, R. E., and Tiew, C. C. (2026). Social network analysis of digital learning in Southeast Asia: trends and insights from X. Journal of International Cooperation in Education. https://doi.org/10.1108/JICE-04-2025-0021

Published open access under a Creative Commons Attribution (CC BY 4.0) licence. The study cites Hansen, Shneiderman and Smith (2011), Analyzing Social Media Networks with NodeXL, as the basis for its method.

Map a network like this yourself

Every measure in this study, betweenness centrality, in-degree and out-degree, modularity clustering, connected components, graph density, and top hashtag and URL analysis, is built into NodeXL and runs inside the Excel spreadsheet you already use. No programming required.