A review on the use of complex networks in science education research

Abstract

Network-based approaches have become increasingly prominent in science education research as tools for analysing relational structures in learning, teaching, and knowledge production. This review presents a PRISMA-informed scoping analysis of 82 articles published in nine leading science education journals, which are organised into four main categories: concept networks, social networks, bibliographic networks, and attitudes or behavioural networks. We observe a sustained exponential growth in the use of network methods, indicating a still-emerging and expanding research area. Concept networks dominate the literature, followed by social network analyses linking interaction structure to learning outcomes and persistence, while bibliographic and abilities-oriented networks provide complementary meta-level and practice-focused perspectives. In addition, analysis of the coauthorship network reveals a highly fragmented field, characterised by many small and weakly connected research groups, typically organised within single application categories. Complementary analysis of a citation network that includes all referenced authors shows that, despite this limited collaboration structure, the field is intellectually organised around several major traditions--network science methodology, learning sciences, and argumentation in science education--linked by a small number of bridging authors. Overall, the literature remains largely descriptive, relying on static, single-layer representations and a narrow set of network metrics. We identify substantial opportunities for advancing science education research through stronger theoretical integration and the adoption of dynamic, multilayer, and coevolutionary network frameworks.

Figures (10)

• Figure 1: Frequency of articles in each category.
• Figure 2: Temporal evolution of the number of articles. The line is an exponential fit to the total number of new articles per year with a doubling time of 3.38 years.
• Figure 3: Examples of the networks extracted from Riihiluoma2024: the upper panel shows an expression–concept network built from the responses of 139 students, where nodes represent survey expressions and edges indicate that two expressions were selected simultaneously for at least one concept (edge weights reflect the number of students). The lower panel shows the same network after community analysis, revealing six stable forms of grouping among the expressions..
• Figure 4: Network extracted from Wheatley2022: Pretest partial correlation networks constructed from student responses to a conceptual survey. Communities indicate groups of answer patterns that tend to co-occur across students; edge width represents the strength of the association between responses. Correct answers are marked with an asterisk.
• Figure 5: Network extracted from Williams2019: Evolution of a student interaction network at two time points during a course. Early ties largely reflect seating proximity, while later the network shows a more complex interaction structure, illustrating its development over time. Node size represents closeness centrality at the later measurement; highlighted nodes correspond to students who did not complete the course.