Beyond right or wrong: towards redefining adaptive learning indicators in virtual learning environments

Student learning development must involve more than just correcting or incorrect questions. However, most adaptive learning methods in Virtual Learning Environments are based on whether the student’s response is incorrect or correct. This perspective is limited in assessing the student’s learning level, as it does not consider other elements that can be crucial in this process. The objective of this work is to conduct a Systematic Literature Review (SLR) to elucidate which learning indicators influence student learning and which can be implemented in a VLE to assist in adaptive learning. The works selected and filtered by qualitative assessment reveal a comprehensive approach to assessing different aspects of the learning in virtual environments, such as motivation, emotions, physiological responses, brain imaging, and the students’ prior knowledge. The discussion of these new indicators allows adaptive technology developers to implement more appropriate solutions to students’ realities, resulting in more complete training.

💡 Research Summary

The paper critically examines the prevailing paradigm in adaptive learning within Virtual Learning Environments (VLEs), which largely hinges on a binary assessment of student responses—correct versus incorrect. Recognizing that such a narrow focus fails to capture the multifaceted nature of learning, the authors conduct a systematic literature review (SLR) to identify a broader set of learning indicators that can be leveraged to create more nuanced adaptive systems.

Methodology
Following PRISMA guidelines, the authors defined a search protocol that targeted English‑language peer‑reviewed articles published from 2018 onward. Four major databases—ACM Digital Library, IEEE Xplore, ScienceDirect, and Scopus—were queried using a composite string that combined terms for metrics/indicators (metric, index, criterion, indicator, parameter, factor), learning processes (learn, acquire, obtain, comprehend, gather), and the domains of neuroscience and engagement. The initial retrieval yielded 1,243 records. After automated duplicate removal and exclusion of any record containing the word “review” in its title, 1,018 records remained. A two‑stage screening (title/abstract, then full‑text) applied predefined inclusion criteria (English language, primary research, relevance to learning indicators) and exclusion criteria (non‑relevant topics, review papers, duplicate versions).

A quality assessment form comprising five questions—(1) discussion of learning indicators, (2) presence of an experiment or application, (3) applicability to a VLE, (4) methodological rigor, and (5) clarity of results—was scored on a weighted scale (+2, +1, 0, –0.5). Articles scoring 3.9 out of a possible 6 were retained, resulting in a final set of 16 high‑quality studies for synthesis.

Findings
The selected literature clusters around five major categories of learning indicators:

1. Motivation and Engagement Metrics – Measures such as task completion rates, time spent on content, and self‑report questionnaires. Cheng et al. (2021) demonstrated that an adaptive platform (RealizeIT) that tailors task difficulty based on prior knowledge boosts both engagement and exam performance, especially for highly motivated learners.

2. Affective and Emotional Signals – Facial expression analysis, heart‑rate variability (HRV), skin conductance (GSR), and other autonomic markers. The authors cite Savolainen (2019) and Shao et al. (2021) to argue that positive affect correlates with stronger memory encoding and retrieval, suggesting that real‑time affect detection can inform adaptive interventions.

3. Physiological and Neuro‑Imaging Indicators – Electroencephalography (EEG) metrics such as Frontal Alpha Asymmetry (FAA) for valence and Frontal Midline Theta (FMT) for attentional engagement. Mercier et al. (2020) employed a system‑dynamics model linking affect, cognition, knowledge, and external agents, and collected high‑frequency EEG data from 72 participants engaged in a physics‑learning game. Their analysis revealed slow cyclic variations in EEG markers, punctuated by brief bursts that aligned with performance fluctuations, underscoring the value of neuro‑physiological data for fine‑grained adaptation.

4. Prior Knowledge and Cognitive Load Measures – Pre‑test scores, working‑memory capacity, and Cognitive Load Index (CLI). Studies referenced (e.g., Parija & Singh, 2023) highlight that learners with richer prior knowledge exhibit more pronounced neuroplastic responses, suggesting that initial diagnostic assessments can guide personalized content sequencing.

5. Multimodal and Socio‑Motor Channels – Tchoubar (2019) introduced a four‑channel e‑learning model that adds kinesthetic and socio‑emotional streams to the traditional auditory‑visual framework. Experimental results indicated that students with stronger spatial skills and higher social interaction (e.g., profile photo sharing, virtual competition) performed better on circuit‑design tasks, demonstrating that motor and social dimensions are potent contributors to learning outcomes.

Collectively, these works illustrate that a holistic view—integrating behavioral logs, affective cues, physiological signals, brain activity, and prior knowledge—offers a richer, more accurate portrait of a learner’s state than correctness alone.

Discussion and Implications
The authors argue that adaptive learning systems should evolve from simple rule‑based engines that trigger content changes based on right/wrong answers to sophisticated pipelines that ingest multimodal data streams in real time. Implementing such pipelines would require:

• Sensor Infrastructure: Wearable devices (e.g., heart‑rate monitors, EEG headsets), webcam‑based facial expression analysis, and interaction logging tools.
• Data Fusion Algorithms: Machine‑learning models capable of weighting disparate signals (e.g., combining EEG‑derived attention scores with engagement metrics) to produce a unified learner state estimate.
• Privacy and Ethics Frameworks: Clear consent mechanisms, data minimization strategies, and transparent use policies, given the sensitivity of physiological and neural data.

The paper also acknowledges limitations: the review excluded non‑English literature, relied on a fixed set of databases, and omitted snowballing, potentially overlooking relevant studies from education sociology or anthropology. Moreover, the quality rubric emphasized VLE applicability, which may have filtered out valuable theoretical contributions.

Conclusion
By systematically cataloguing and analyzing recent neuroscience‑informed studies, the authors demonstrate that adaptive learning can be substantially enhanced by incorporating a suite of indicators beyond binary correctness—namely motivation, affect, physiological arousal, brain‑wave patterns, and prior knowledge. They call for future research to develop real‑time multimodal adaptive algorithms, validate them in authentic classroom settings, and address the attendant ethical considerations. Such a shift promises to deliver learning experiences that are truly personalized, responsive to the learner’s holistic state, and ultimately more effective.