Hello everyone. I would like to introduce a paper we read at our recent English seminar.

Paper Title: Self-regulated Learning Processes in Secondary Education: A Network Analysis of Trace-based Measures
Authors: Yixin Cheng, Rui Guan, Tongguang Li, Mladen Raković, Xinyu Li, Yizhou Fan, Flora Jin, Yi-Shan Tsai, Dragan Gašević, Zachari Swiecki
Year of Publication: 2025 (LAK 2025, ACM)
DOI: 10.1145/3706468.3706502

1. Introduction and Theoretical Background
Self-Regulated Learning (SRL) is defined as a “cyclical and active process involving cognitive, metacognitive, motivational, emotional, and behavioral aspects” (Zimmerman & Schunk, 2011). This paper presents the perspective that SRL is crucial for academic achievement and personal development in secondary education (SE). It is also noted that some learners in SE may lack the sufficient knowledge and skills to effectively manage their learning and prepare for the transition to higher education (HE).
As a framework for explaining SRL, macro-level cyclical models including forethought, performance, and reflection phases have been widely used. However, it has been pointed out that traditional measurements often rely on self-report questionnaires or think-aloud protocols, which face challenges such as the potential to induce processes, limitations in conscious awareness, and incomplete recall. To address these issues, this paper presents the potential of utilizing trace data. Based on the framework of Bannert (2007), it is stated that breaking down SRL into cognitive, metacognitive, and motivational processes and gaining insights into SRL behaviors from computer-based environment logs is a useful approach.

2. Previous Research and Research Questions
In this paper, while SRL research using trace data has progressed in terms of theory and technology, a gap is identified in the lack of research systematically measuring SRL processes in the context of SE. It is also noted that much of the research linking SRL to performance using trace data targets HE and may not necessarily be applicable to SE (as cognitive and metacognitive strategies may differ between middle school and university students). Therefore, the purpose of this paper is to measure the SRL processes of learners engaged in writing tasks using a digital writing platform and to compare differences between high and low achievers, as well as differences between SE and HE. The research questions are as follows:
• RQ1: What SRL processes are common in SE, and to what extent do they differ between high and low achievers?
• RQ2: What SRL processes are common in HE, and to what extent do they differ between high and low achievers?
• RQ3: To what extent do SRL processes differ between educational contexts (SE vs. HE)?

3. Research Methods
3.1 Experimental Design and Participants
This paper adopts a design comparing two studies (conducted in 2023) that implemented the same task on a writing platform called “FLoRA.” The SE condition consisted of 66 students (ages 12–15, mean 13.44) from two Australian middle schools, with the task time shortened to 45 minutes and reading materials and rubrics simplified. The HE condition consisted of 59 students (mean age 22.63) from a public university in China, with a task time of 120 minutes.
The trace data was large-scale, with a total of 547,812 events recorded in the SE condition (average 8,300 events/session) and 1,329,882 events in the HE condition (average 22,540 events/session).

3.2 SRL Process Modeling and ENA
It is explained that the coding of SRL processes was labeled through an iterative approach combining theory-driven and data-driven methods based on the SRL framework of Bannert (2007) (seven SRL codes: Monitoring, Evaluation, etc.).
This paper adopts a method called Epistemic Network Analysis (ENA). ENA is an analytical method that measures how specific concepts (codes) co-occur (appear together in the same context) within data and visualizes the patterns of these associations as a network model. Nodes correspond to codes, and edges represent the frequency of code co-occurrence.
Regarding ENA settings, each participant is represented as an individual network, the window size is set to 50 events for all models, and regression analysis (M1, M2, M3) controlling for covariates (such as pre-test levels) was used to compare ENA scores.

4. Results (SRL Processes Based on ENA)
4.1 RQ1: Secondary Education (SE)
As an overall trend in SE, it is reported that strong connections were observed between Orientation, Re.Reading, and Elaboration.Organisation. Notably, it is stated that the Evaluation code never appeared in the data in the SE condition, and no connections existed. Regarding performance, statistically significant differences were found; high achievers showed stronger connections between Re.Reading and Elaboration.Organisation, suggesting a tendency to engage in iterative information processing by revisiting materials and integrating them with new knowledge through elaboration. On the other hand, low achievers showed stronger connections between Orientation and Elaboration.Organisation, suggesting a reliance on task instructions and rubrics as guides for writing.

4.2 RQ2: Higher Education (HE)
In HE, strong connections were observed between First.Reading, Re.Reading, Elaboration.Organisation, and Monitoring, suggesting a richer and more diverse set of SRL strategies compared to SE. There were also statistically significant differences by performance; high achievers had stronger connections between First.Reading and Monitoring, reporting a more strategic approach such as actively checking understanding during the initial reading. Low achievers showed stronger associations between Re.Reading and Evaluation/Monitoring/Elaboration.Organisation, suggesting difficulty in understanding during the first reading and the necessity of checking content multiple times.

4.3 RQ3: Between Educational Levels (SE vs. HE)
In the comparison between educational levels, it is stated that HE learners were more deeply involved in processes including Monitoring and Evaluation, showing stronger metacognitive awareness and self-regulation abilities. In contrast, SE learners showed stronger connections between Orientation and Elaboration.Organisation, suggesting a potential reliance on understanding task instructions and rubrics and using them as guides for composition. Furthermore, it is stated that these results are consistent with findings that metacognitive skills are refined with age and the complexity of the educational environment.

5. Discussion, Conclusion, and Educational Implications
In this paper, the complete absence of the Evaluation process in SE is treated as an important implication. Given that self-evaluation is a key component of SRL and that there is a positive relationship between self-evaluation and learning outcomes, it is argued that there may be a potential gap at the SE stage. Also, since self-evaluation was not promoted as expected despite the inclusion of rubrics, it is stated that future research should consider the design of tools such as scripts (scaffolding questions) that are considered more effective than rubrics.
Implications are also shown in measurement methods. It is stated that results based on trace data revealed significant differences, in contrast to previous studies using think-aloud protocols that reported “no significant difference in general metacognitive skills between high and low achievers.” Trace data is positioned as providing a more objective and continuous measurement that can identify subtle differences in engagement with metacognitive strategies that are difficult to capture through self-reports or think-alouds. Based on these, the necessity of designing scaffolding tools and teacher training programs to promote Monitoring and Evaluation for SE learners is emphasized.

6. Limitations of the Study
It is stated that there are four limitations in this study:
1. The task time differs between SE (45 minutes) and HE (120 minutes), making the comparison of educational level differences (RQ3) complex.
2. Since the focus is on group-level comparisons, individual differences related to the self-regulatory nature of SRL may be overlooked.
3. It is unclear whether the transition from SE to HE is a sudden change or a developmental process.
4. There is a lack of similar previous research in the context of SE, limiting comparison and contextualization.

7. Reflections
Below are my thoughts after reading the paper.
The reason for selecting this paper is that I intend to proceed with research incorporating ENA and wanted to review previous studies that applied ENA to SRL. In this paper, by analyzing large-scale trace logs, subtle differences in metacognitive behaviors in the learning process that cannot be captured by think-aloud protocols were extracted, and I recognized once again the usefulness of a data-driven approach.
On the other hand, I felt that the significant difference in task time between secondary and higher education (45 minutes vs. 120 minutes) is highly likely to have influenced the comparison of SRL processes, and that caution is required in interpreting the results. Also, while ENA is effective for visualizing co-occurrence patterns of SRL processes, it cannot handle qualitative aspects such as which materials the learner read and how, or what information was integrated into the text. Therefore, I thought that an approach combining the following analyses would be effective:
・Content analysis of English compositions to clarify which material information was cited or integrated.
・Analysis of revision history to grasp what kind of corrections learning behaviors such as monitoring and self-evaluation actually led to.
By combining these, I believe that in addition to a structural understanding of co-occurrence networks, the basis for learners’ cognitive and metacognitive judgments can be analyzed in more detail. In the future, I would like to incorporate such multifaceted analytical methods into my own research.

Report by: Sayo Tanaka