Knowledge-Graph-Driven Mind Mapping for Immersive Collaborative Learning: A Pilot Study in Edu-Metaverse
Jia, Ye, Wang, Xiangzhi Eric, Sin, Zackary PT, Li, Chen, Ng, Peter HF, Huang, Xiao, Baciu, George, Cao, Jiannong, Li, Qing
IEEE Transactions on Learning Technologies (2024)
Overview of our KG-driven MM editor in an edu-metaverse setting. (a) Players collaborating on an MM. (b) Root node and interaction ray between the controller and a target node. (c) Semitransparent dummy nodes for recommendations. (d)–(g) UI features: the text editor, color picker, voice-to-text, and shape selector. (h) Buttons and triggers bindings on controllers to operate in the system.
Abstract
One of the promises of edu-metaverse is its ability to provide a virtual environment that enables us to engage in learning activities that are similar to or on par with reality. The digital enhancements introduced in a virtual environment contribute to our increased expectations of novel learning experiences. However, despite its promising outcomes, there appears to be limited adoption of the edu-metaverse for practical learning at this time. We believe this can be attributed to the fact that there is a lack of investigation into learners' behavior given a social learning environment. This lack of investigation is critical, as without behavioral insight, it hinders the development of education material and the direction of an edu-metaverse. Upon completing our work with the pilot user studies, we provide the following insights: 1) compared to Zoom, a typical video conferencing and remote collaboration platform, learners in the edu-metaverse demonstrate heightened involvement in learning activities, particularly when drawing mind mapping aided by the embedded knowledge graph, and this copresence significantly boosts learner engagement and collaborative contribution to the learning tasks; and 2) the interaction and learning activity design within the edu-metaverse, especially concerning the use of MM.
The Adoption Gap
Despite years of "edu-metaverse" vision papers, actual adoption of VR for practical learning has been slow. This paper argues that the bottleneck is not hardware cost or content availability — it's the near-total absence of behavioral data on how learners actually interact in social VR learning environments. Without knowing what learners do in these spaces, platform designers can't optimize for it, instructors can't design activities for it, and the field can't move beyond demo-level prototypes. This paper provides that behavioral data for a specific, well-defined activity: collaborative mind mapping augmented by a knowledge graph.
The KGMM System
The Knowledge-Graph-driven Mind Mapping (KGMM) system places a collaborative mind map editor inside a multi-user VR environment. Learners, represented as avatars, stand around a shared mind map and collaboratively build it by adding, connecting, and rearranging concept nodes. The system provides standard editing tools — text editor, color picker, shape selector — plus voice-to-text input to reduce the friction of typing in VR.
The knowledge graph component is what separates KGMM from a straightforward VR whiteboard. As learners work, the KG analyzes the current mind map structure and suggests related concepts as semi-transparent "dummy nodes" positioned near relevant existing nodes. A learner sees a ghosted node labeled "regularization" appear near their "overfitting" node and can choose to accept, reject, or modify it. The KG is not an oracle — it's a cognitive prompt. It doesn't complete the mind map for the learners; it suggests plausible extensions drawn from a pre-built domain knowledge graph, which the learners must then evaluate, integrate, and elaborate.
This design choice is theoretically motivated. Mind mapping is a constructivist learning activity par excellence — learners externalize their mental model of a domain, and the process of organizing concepts spatially forces them to confront gaps, inconsistencies, and assumptions. By embedding KG recommendations into this process rather than presenting them as a pre-built structure, KGMM preserves the constructive element while adding an expert scaffold. The KG doesn't tell you what to think; it nudges you toward connections you might have missed.
The Study Design
The pilot study compared KGMM in the edu-metaverse against an equivalent 2D collaborative mind-mapping task conducted over Zoom. Both conditions used the same KG backend and the same task. The key manipulated variable was the environment: immersive VR with avatar copresence vs. video conferencing with screen sharing.
The behavioral metrics focused on involvement — frequency of contributions, types of interactions (adding nodes vs. connecting nodes vs. editing existing nodes), turn-taking patterns, and the distribution of contributions across group members. The Zoom condition provides a strong baseline because it represents the default remote collaboration tool that most learners already use; if KGMM in VR can't beat Zoom on engagement, it has no practical future.
What the Data Shows
Two headline findings:
-
Heightened involvement in the edu-metaverse. Learners in the VR condition contributed more frequently, made more node connections (as opposed to just adding isolated nodes), and showed more balanced participation across group members. In Zoom, contributions tended to concentrate in one or two dominant participants; in VR, the contribution distribution was flatter. The authors attribute this to copresence — when you see your teammates' avatars standing next to you, manipulating the same mind map, the social norm to contribute is stronger than when you're a rectangle in a grid.
-
KG recommendations were used differently in VR. In the Zoom condition, KG-suggested nodes were often accepted without modification. In VR, learners were more likely to modify suggested nodes — rename them, reconnect them, use them as starting points for further elaboration. The interpretation is that the immersive environment promotes deeper processing: when you're "inside" the knowledge space, you're more invested in getting the map right than in just completing the task. This is speculative — the study wasn't designed to isolate the mechanism — but it's consistent with the broader finding that VR enhances task absorption.
The Copresence Argument
The paper's most theoretically interesting claim is that copresence acts as an engagement multiplier for collaborative knowledge work. This is distinct from saying "VR is more engaging." The argument is structural: copresence changes the social dynamics of collaboration (more balanced participation, more spontaneous verbal contributions), which in turn changes the cognitive dynamics (deeper processing of KG recommendations, more elaboration on peer contributions). The KG provides the cognitive scaffold; copresence provides the social motivation to actually use it.
This framing has a practical implication for edu-metaverse design: don't invest in better AI recommendations until you've invested in better social presence. A sophisticated KG with low copresence will underperform a simple KG with high copresence, because recommendations that nobody discusses or elaborates are wasted.
Boundaries
This is a pilot study. The sample size is small, the task is a single session, and the outcome measures are behavioral engagement rather than learning gains. We don't know whether the heightened involvement translates to better conceptual understanding, better retention, or better transfer — and we don't know whether the effects persist beyond the novelty window. The KG itself is domain-specific and manually constructed; the paper does not address how to build or adapt KGs for new subjects at scale.
The comparison against Zoom is also a comparison of two different modalities (VR vs. 2D video) that differ on many dimensions simultaneously — copresence, input modality, field of view, environmental immersion. Attributing the effects specifically to copresence rather than to, say, the novelty of being in VR or the reduced distraction of a full-field display is not possible from this design. A follow-up that varies copresence within VR (e.g., high-fidelity avatars vs. floating name tags) would isolate the mechanism more cleanly.