The researchers initially developed an 11-construct framework combining Universal Design for Learning principles with multimodal learning analytics infrastructure, then converted it into a 44-item survey. The infrastructure side comprised data quality, technology platforms and real-time analytics, while the practice outcomes comprised adaptive intervention, formative feedback, inclusive assessment and self-regulated learning support. Experts reviewed the instrument for clarity, relevance and construct alignment, and 25 educators and instructional designers completed a pilot test.

Participants were recruited through 12 universities and 15 K-12 school networks across Kerala, Karnataka, Tamil Nadu, Odisha and Delhi, with additional snowball referrals. Of 687 people contacted, 476 completed the survey, a response rate of 69.3%. Teachers and educators made up 59.9% of the final sample, instructional designers 26.7% and administrators 13.4%. K-12 participants represented 41.6% of respondents and higher-education participants 58.4%. Data were collected from October 2025 to February 2026.

All measures used five-point Likert responses, and the researchers chose PLS-SEM because the analysis prioritised prediction and explained variance, included reflective and formative measurement structures and used practitioner survey data that did not require multivariate normality. Statistical significance was evaluated with 5,000 bootstrap resamples, with path coefficients considered significant at p<.05. The final instrument contained 24 items measuring seven constructs, and all constructs had composite reliability above 0.70.

Nine of the 12 directional hypotheses were statistically significant. Real-time analytics had the largest standardised path coefficient, linking it to adaptive intervention at β=0.546 with p<.001 and a large effect size of f²=0.421. The three infrastructure constructs together explained 46.0% of the variance in adaptive intervention, while explained variance for the other outcomes ranged from 19.4% to 35.4%. All four outcomes had positive predictive relevance, although the model’s explanatory power was bounded by factors that were not included.

Technology platforms showed significant positive associations with all four outcomes, with their strongest relationship being with self-regulated learning support at β=0.323. By contrast, data quality had a limited direct association, reaching statistical significance only with adaptive intervention at β=0.110 and p=.016. The authors interpreted this pattern as consistent with data functioning as an enabling resource whose practical value depends on technology and analytics that can make information accessible and actionable.

The Universal Design for Learning measures did not survive validation and were removed from the final model. Universal Design for Learning Action and Expression and Engagement were empirically indistinguishable, with an HTMT ratio of 1.009, and merging them produced an average variance extracted of 0.313. Personalised Feedback and Formative Assessment were also merged because their HTMT was 0.962, while 20 items with loadings below 0.60 were removed. The authors said these results indicate measurement problems in practitioner self-reports, not that Universal Design for Learning lacks value; possible explanations include holistic practitioner understanding, inadequate survey measures, contextual variation and insufficient professional development.

Several limitations constrain the findings. The cross-sectional, self-report design does not support causal inference and may be affected by common-method and social-desirability bias. Two constructs were measured with only two items each and had low Cronbach’s alpha, while two others met the primary convergent-validity threshold only through a secondary exception. Discriminant validity was not formally established between real-time analytics and adaptive intervention: their HTMT was 0.901, and the 95% bootstrap confidence interval of 0.808 to 0.987 included both commonly used thresholds. The model fit was also mixed, with SRMR at 0.094 and NFI at 0.528.

Purposive recruitment in one country may have overrepresented technologically proficient or enthusiastic practitioners, limiting generalisability beyond the sampled Indian contexts. The model did not test mediation or moderation, distinguish platform- or vendor-specific differences, or include potential influences such as institutional support, teacher artificial-intelligence literacy, assessment culture, professional development, leadership and student characteristics. The validated 24-item instrument can nevertheless support benchmarking of reported multimodal learning analytics implementation, while the observed results indicate that real-time analytics deserve consideration alongside—not instead of—data quality, technology platforms, governance, staff training and technical support.

Subject of Research: Cross-sectional validation of multimodal learning analytics infrastructure and adaptive digital assessment practices among 476 Indian education practitioners.

Article Title: Multimodal learning analytics infrastructure and adaptive digital assessment

Article References: Asha, S., & Joshith, V. P. (2026). Multimodal learning analytics infrastructure and adaptive digital assessment. Discover Education, 5(1), Article 1193. https://doi.org/10.1007/s44217-026-02301-3

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02301-3

Keywords: multimodal learning analytics, adaptive assessment, real-time analytics, learning analytics, PLS-SEM, digital assessment, formative feedback, self-regulated learning, India, survey research, Multimodal, learning

The review, led by Khurram Shahzad of Government College University in Lahore with collaborators from Alfaisal University, Prince Sultan University, and the University of Salford, screened 3,280 records pulled from eleven databases including Scopus, Web of Science, PubMed, and MEDLINE. After removing duplicates, ineligible studies, and non-English publications, only thirty peer-reviewed articles and conference proceedings survived the full-text gauntlet. The team followed PRISMA 2020 guidelines, registered the protocol on the Open Science Framework, and applied a ten-item CASP-informed quality appraisal to each surviving study, with inter-rater agreement checked using Cohen’s Kappa. The result is less a triumphal catalog of AI successes than a field guide to the gap between laboratory performance and real-world deployment.

The central technical finding concerns sentiment analysis itself. Sentiment analysis classifies the emotional polarity, attitudes, and opinions expressed in text, whereas misinformation detection asks whether a claim is factually accurate or misleading. These are fundamentally different tasks, and the review is emphatic that conflating them is a methodological trap. A false claim about a vaccine can be written in cheerful, confident language; a genuine warning about a drug interaction can sound alarming and negative. Both factual and misleading health information may carry positive or negative sentiment, which means emotional polarity alone cannot establish truthfulness. The authors position sentiment-derived signals as an auxiliary feature that must be combined with contextual semantics, source credibility, behavioral patterns, temporal information, and evidence-based verification before any judgment about accuracy can be made.

What does move the needle, according to the synthesis, is a cluster of five enabling factors. Advanced deep learning architectures, including transformer models and large language models, provide contextual reasoning that older lexical and statistical approaches cannot match. Feature engineering, the craft of extracting sentiment scores, named entities, health-specific lexicons, and semantic representations from raw text, remains foundational; high-quality data improved classification performance across transformer-based and machine learning frameworks alike. Real-time analytics, powered by cloud infrastructure and streaming pipelines, allow systems to flag emerging misinformation narratives during fast-moving public health events rather than cataloging them after the fact. And human-centered verification, where domain experts review automated outputs before they inform decisions, emerged as a critical safeguard against the unsupported inferences that even state-of-the-art models confidently produce.

The challenges side of the ledger is equally detailed. Annotation limitations top the list: high-quality labeled health misinformation datasets are scarce, manual annotation struggles to keep pace with rapidly evolving narratives, and inconsistencies between annotators degrade model training. Sampling and algorithmic biases can skew sentiment estimates derived from social media data, reducing the representativeness of any downstream detection system. Informal language, slang, emojis, and code-switching complicate processing, particularly in non-English settings, and mixed or neutral sentiments increase misclassification risk. These are not peripheral annoyances; they strike at the training data on which every detection model depends.

Scalability constraints form the second major barrier. Large language models demand substantial memory and processing power, real-time analytics introduce latency trade-offs between speed and accuracy, and cloud-based deployments carry infrastructure costs and external dependencies that many health systems cannot sustain. During health emergencies, when misinformation volumes spike, resource requirements can overwhelm the very systems meant to respond. Domain adaptation compounds the problem: models trained on one misinformation context lose accuracy when applied to different health topics, populations, or linguistic environments, and distinguishing misinformation from legitimate expressions of psychological distress remains contextually fraught.

Ethical and sustainability concerns round out the picture. Analyzing sensitive patient-generated and mental health data raises privacy questions that current governance frameworks do not consistently answer. Unreliable automated reasoning, hallucinated outputs from large language models, and inadequate external knowledge sources erode trust in AI-based detection. Cyber-security vulnerabilities, algorithmic bias in public health decision-making, and the absence of comprehensive infodemic surveillance frameworks all constrain long-term adoption. The review notes that operational integration into existing healthcare systems, coordination among healthcare decision-makers, policymakers, and technical teams, and cross-platform consistency remain unsolved organizational problems rather than purely technical ones.

From this synthesis the authors derive a five-component conceptual framework: a multi-source data acquisition layer to capture heterogeneous health information across social media, news portals, healthcare repositories, and IoT-enabled systems; an intelligent data management system handling cleaning, normalization, multilingual processing, and feature engineering; an AI-driven analytical engine integrating deep learning, sentiment signals, knowledge graphs, and real-time analytics; a human validation layer incorporating explainable AI, evidence retrieval, and expert oversight; and adaptive governance covering ethics, privacy, bias mitigation, regulatory compliance, and continuous model updating. The components are interdependent rather than sequential, and the authors are explicit that the framework is evidence-informed but not empirically validated; its effectiveness awaits prototype implementation and field testing.

One of the review’s more consequential observations is that data quality may matter as much as, or more than, algorithmic sophistication. Recent studies reporting strong performance from deep learning and LLM-based architectures still depend on representative, balanced, accurately annotated, domain-specific datasets. Increasing model complexity cannot compensate for weaknesses in annotation, contextual representation, or domain adaptation. This challenges a purely technology-centered reading of the misinformation problem and suggests that future research should prioritize computationally efficient, context-adaptive approaches over raw predictive performance, especially in healthcare settings with limited resources.

The review also documents a clear historical shift in the field. Earlier studies emphasized linguistic, statistical, and behavioral features for classification, while recent work increasingly employs BERT, transformer models, and large language models with retrieval-augmented generation for evidence-grounded analysis. Detection itself has moved from static classification of already-disseminated content toward streaming analytics, continuous surveillance, anomaly detection, and predictive assessment, a shift particularly relevant during health emergencies when narratives mutate rapidly. The authors caution that real-time capability is an operational requirement to be balanced against scalability and resource constraints, not an inherent marker of system superiority. For policymakers and health officials, the practical message is blunt: investing in advanced analytics without matching investment in datasets, human expertise, verification workflows, and governance will produce systems that look impressive in benchmarks and fail when it matters.

Subject of Research: Big data analytics and sentiment analysis for detecting health misinformation on social media

Article Title: Big data analytics and sentiment analysis for health misinformation detection: a systematic literature review of influencing factors, implementation challenges and a data-driven adoption framework

Article References: Shahzad, K., Javeed, A. M. D., Ahmad, S., Latif, M., & Saleem, K. (2026). Big data analytics and sentiment analysis for health misinformation detection: a systematic literature review of influencing factors, implementation challenges and a data-driven adoption framework. Frontiers in Big Data, 9, Article 1948398. https://doi.org/10.3389/fdata.2026.1948398

Image Credits: AI Generated

DOI: 10.3389/fdata.2026.1948398

Keywords: health misinformation, big data analytics, sentiment analysis, systematic literature review, deep learning, large language models, real-time analytics, human verification, data quality, AI ethics, public health surveillance, Frontiers in Big Data

The work, published in the journal Knowledge and Information Systems, tackles a setting the authors call a Dynamic and Unknown Domain, or DUD. It is dynamic because the schemas of incoming messages, meaning the set of attributes each message carries, evolve constantly in number, similarity, and distribution. It is unknown because there is no fixed dictionary of attributes; new fields can appear at any moment with names nobody has seen before. In such an environment, the classical assumption underlying most clustering algorithms, that data points live in a fixed-dimensional space, simply breaks down.

The goal of schema profiling is to give practitioners a summarized view of the message structures within a sliding window of the stream. That view is built by clustering similar schemas together and representing each group by its centroid, in the spirit of the k-means algorithm. The resulting profile helps analysts understand which kinds of messages dominate the traffic, and it supports downstream tasks such as formulating queries over schemaless data. But recomputing a full clustering from scratch every time the window slides is both computationally prohibitive and analytically jarring, since profiles computed independently in successive windows would be hard to compare over time.

DSC+ addresses these constraints with a two-phase design executed at every slide of the window. In the first phase, the schemas of incoming messages are pre-aggregated into compact data structures called schema features, or SFs. Rather than storing every individual schema, an SF records, for each attribute, the empirical probability that it appears within the group of schemas it summarizes. These structures are additive and subtractive, which means they can be merged, split, and trimmed as old messages expire from the window and new ones arrive, all without ever reconstructing the original messages.

To decide which schemas should be grouped together in the first place, the algorithm borrows a trick from document comparison: Min-Hash. Each schema is passed through a set of hash functions that produce a short identifier, and the mathematics of the technique guarantees that two schemas with high Jaccard similarity, meaning they share many attributes, are likely to receive the same identifier. The researchers adapted the method so that the hash indexes remain stable even as the attribute domain grows without bound, a property the original Min-Hash formulation lacks. Schemas sharing an identifier are folded into a single schema feature, producing a pane-level coreset, and these are in turn combined into a fixed-size window-level coreset that feeds the clustering stage.

The second phase is where DSC+ departs most sharply from prior work. Instead of rerunning k-means on each window, the algorithm incrementally updates the previous clustering result through a set of formalized rules that capture five distinct evolution phenomena. Sliding occurs when the schemas within a cluster gradually change, shifting the centroid. Fading-in happens when a new source starts emitting recognizably different schemas, and fading-out when an existing source goes silent. Splitting arises when one cluster grows so internally diverse that it should become two, and merging when two clusters drift so close together that they should become one.

Each rule is triggered by statistical signals rather than guesswork. The algorithm monitors the scattering of each cluster, the average distance between its members and its centroid, and the separation between pairs of clusters, the distance between their centroids. When a robust z-score, computed with the median absolute deviation, flags a cluster as an outlier in scattering, a split is executed by locally running k-means with two centers. When a pair of clusters overlaps enough and their separation becomes anomalously low, they are merged. New schema features are simply assigned to the nearest centroid, and empty clusters are deleted. If the quality of the updated profile, measured by the Simplified Silhouette, drops by more than what the worst possible sudden change in the data could plausibly explain, a full reclustering is triggered as a safety net.

The experimental evaluation, run on both synthetic datasets engineered to simulate each evolution phenomenon and on four real-world datasets, including application logs from a consulting company, sensor data from a precision agriculture project, and network traffic logs from the Suricata monitoring tool, shows the approach outperforming the closest state-of-the-art competitors. Against the baseline of rerunning the full clustering algorithm, OMRk++, on every window, DSC+ produced profiles that were actually slightly better in terms of silhouette, an effect the authors attribute to the pre-aggregation step filtering out noise from individual schemas. Compared with the streaming competitors CSCS and FEAC-S, DSC+ tracked changes in the number of clusters far more faithfully while executing faster in most configurations.

The efficiency gains stem from the fact that the expensive clustering operates on a coreset capped at a fixed maximum size, so execution time scales sublinearly with the volume of incoming data. In tests where the window length grew from ten thousand to one million schemas, DSC+ sustained increasing stream rates while a competitor’s throughput actually declined, because that competitor struggled to compress large windows into its fixed-size summary. The researchers also tuned the key parameters, including the coreset size and the number of hash functions, showing that accuracy plateaus quickly as the coreset grows, which means substantial data reduction comes at almost no cost in profile quality.

The authors are candid about limitations. The current implementation runs on a single server, and a distributed version exploiting the additivity of schema features is left for future work. The system also lacks explicit mechanisms for handling outlier schemas that resemble nothing seen before, although such cases did not arise in the real datasets tested. Still, the contribution marks a clear step forward: a clustering method that embraces a variable and unknown attribute domain, tracks the birth, death, splitting, and merging of clusters in real time, and does so within a bounded memory footprint. For the growing ranks of organizations drowning in heterogeneous IoT and log data, DSC+ offers a way to finally see, in real time, what their streams are actually made of.

Subject of Research: Real-time schema profiling of heterogeneous data streams using dynamic stream clustering with variable-k k-means

Article Title: Dynamic stream clustering for real-time schema profiling with DSC+

Article References: Forresi, C., Francia, M., Gallinucci, E., & Golfarelli, M. (2026). Dynamic stream clustering for real-time schema profiling with DSC+. Knowledge and Information Systems, 68(1), Article 276. https://doi.org/10.1007/s10115-026-02889-w

Image Credits: AI Generated

DOI: 10.1007/s10115-026-02889-w

Keywords: schema profiling, data streams, stream clustering, k-means, coresets, Min-Hash, IoT, big data, sliding window, cluster evolution, schemaless data, real-time analytics

The new framework, described in the journal Multimedia Tools and Applications, tackles a problem that researchers call affective state recognition in e-learning. Rather than trying to detect the seven universal emotions that dominate much of facial expression research, the system focuses on four learning-centered states that education scientists have identified as especially meaningful during study: boredom, engagement, confusion, and frustration. These states matter because decades of cognitive science research show they shape how people learn. Positive emotional states are associated with gains in creative thinking, while negative states push learners toward prioritizing performance over mastery. Boredom in particular has been repeatedly linked to suboptimal learning outcomes, whereas mild confusion, somewhat counterintuitively, can signal that a student is grappling productively with difficult material.

At the heart of the system is a multioutput classification model built on a parallel-branch architecture. The researchers trained two versions: one based on a lightweight convolutional neural network with roughly 470,000 parameters, and another built on EfficientNetB2, a more capable backbone with about 31.7 million parameters that remains efficient enough for cloud deployment. Each branch of the network is dedicated to a single affective state and outputs a probability vector across four intensity levels, from very low to very high. The final output is a combined vector capturing the predicted intensity of boredom, confusion, engagement, and frustration simultaneously. This design distinguishes the work from most prior studies, which typically classified only the single dimension of engagement and ignored the richer emotional context surrounding it.

Training relied on DAiSEE, the first multilabel video dataset created for engagement recognition in the wild. It contains 9,068 ten-second video clips of 112 individuals, recorded at full high-definition resolution with ordinary webcams, and annotated with crowd-sourced labels that were correlated against a gold standard produced by expert psychologists. The dataset is notoriously imbalanced: high-intensity engagement labels dominate, while lower-intensity labels are comparatively rare, and the opposite pattern holds for boredom, confusion, and frustration. To counteract the bias this imbalance would otherwise introduce, the team employed a categorical focal loss, a function that applies a modulating term to standard cross-entropy so that learning concentrates on hard, easily misclassified examples rather than being swamped by the majority class.

Preprocessing was deliberately aggressive. Each ten-second clip originally contains 300 frames, but because facial expressions in instructional settings evolve slowly, on the order of one to two seconds, the pipeline subsamples to just ten frames per video, a 96.7 percent reduction in data volume. Each retained frame passes through a Viola-Jones Haar Cascade detector to isolate the face, and frames where facial landmarks or pupils cannot be reliably detected are discarded. Valid crops are resized to 64-by-64 pixels and converted to normalized grayscale. The result is a compact, standardized input that minimizes computational cost while preserving the affective signals that matter, an essential property for a system intended to run continuously during live lectures.

The most novel contribution, however, is not the classifier itself but what the researchers call the attentiveness index. A single classification of four emotional states does not directly tell an instructor whether a student is attentive, so the team needed a way to compress those predictions into one interpretable number. They had multiple instructors score a subset of dataset videos for attentiveness on a scale of one to ten, then applied multiple linear regression to learn weights for each affective state. The resulting formula weights engagement most heavily and positively, gives confusion a smaller positive weight, and assigns negative weights to boredom and frustration, with boredom penalized far more severely. Crucially, these learned coefficients align with the existing cognitive science literature: boredom is strongly associated with poor learning, frustration more weakly so, and moderate confusion often accompanies productive learning.

The index held up under statistical scrutiny. On an unseen test set, the computed index correlated strongly and significantly with the instructor-annotated ground truth, yielding a Pearson correlation coefficient of 0.700 and a Spearman rank coefficient of 0.585, both statistically significant. Three-fold cross-validation produced a mean R-squared of 0.495, and a t-test confirmed the model’s predictive power at p equals 0.0246. Meanwhile, the underlying classifiers achieved state-of-the-art performance on DAiSEE. The CNN-based model reached accuracies of 73.04 percent on boredom, 79.98 percent on engagement, 80.17 percent on confusion, and 85.37 percent on frustration, while the EfficientNet variant performed comparably, peaking at 80.32 percent on engagement. The researchers argue these results outperform prior end-to-end approaches on the same dataset, including methods built on C3D, I3D, and hybrid architectures combining EfficientNet with temporal convolutional and recurrent networks.

What elevates the work from a modeling exercise to a practical tool is the end-to-end pipeline built around it. The system is deployed on a cloud server and supports simultaneous multi-user logins through a web client built with Flask, JavaScript, and HTML5 Canvas. During a live session, the learner’s webcam feed is analyzed in real time, with detected affective states displayed alongside the learning content. Instructors receive a separate dashboard offering both graphical and tabular analytics: class-level trends in average attentiveness over time, post-session distributions of affective states, and timestamped logs of individual learners’ dominant states and index scores. When cumulative class engagement drops below a threshold, the instructor receives an alert, and the analytics can pinpoint exactly when attention sagged during a lecture, for instance, or flag moments when confusion spiked while a difficult concept was being explained. The system can even aggregate data across multiple lectures to suggest how to structure future sessions for maximum engagement.

The researchers are explicit that the tool is meant as a pedagogical aid, not a surveillance instrument. They caution that affective computing models can be sensitive to lighting, camera quality, and demographic variation, and they strongly advise against punitive or comparative uses of the system. Real-world deployments, they stress, must be strictly opt-in, with informed consent from all participants, and individual analytics should be disabled by default in classroom settings in favor of aggregated, class-level feedback. They also acknowledge a limitation long noted in the field: basic universal emotions are rare during genuine learning sessions, which is precisely why the focus on learning-centered states matters.

Future directions outlined by the team include multimodal fusion, combining facial affect with head movement, gaze tracking, and noninvasive EEG signals, along with explainable AI techniques that would visually highlight which facial features drive each prediction. They also plan consent-based periodic self-reports from learners to calibrate the model against individual differences, and the construction of a new, demographically balanced dataset to validate generalizability across racial and cultural backgrounds. If those efforts succeed, the humble webcam could become one of the most informative instruments in the virtual classroom, giving online instructors something they have lacked since lectures moved to screens: a live read on the minds in the room.

Subject of Research: Real-time computer vision analysis of learner affective states and attentiveness in online education

Article Title: Learner attentiveness and engagement analysis in online education using computer vision

Article References: Gogawale, S., Deshpande, M., Kumar, P., & Ben-Gal, I. (2026). Learner attentiveness and engagement analysis in online education using computer vision. Multimedia Tools and Applications, 85(9), Article 743. https://doi.org/10.1007/s11042-026-21884-5

Image Credits: AI Generated

DOI: 10.1007/s11042-026-21884-5

Keywords: computer vision, online education, affective computing, engagement detection, deep learning, convolutional neural networks, DAiSEE dataset, attentiveness index, e-learning, facial expression recognition, focal loss, real-time analytics