The differential impacts of primary mathematics teachers’ multimodal behaviors on student engagement: A deep learning-driven relational matrix network analysis

This study aims to explore the multimodal behaviour characteristics and differences between expert and novice secondary school chemistry teachers during classroom questioning and feedback. Using epistemic network analysis, this study offers deeper insights into the interconnections among different communication modes, beyond traditional methods like frequency counts. By analysing secondary school chemistry teachers in China, this study reveals significant differences in the multimodal questioning and feedback behaviour networks of expert and novice teachers. Expert teachers exhibit a denser and more interconnected network of multimodal behaviours, ask more high-cognitive-level authentic questions, and flexibly use various types of feedback. They effectively combine multiple modalities, such as language, gestures, facial expressions, and eye contact, enhancing student understanding and engagement. Additionally, expert teachers demonstrate greater flexibility in managing interpersonal distance, creating a supportive and inclusive learning environment. In contrast, novice teachers’ multimodal behaviour networks show fewer connections and lower integration, predominantly using low-cognitive-level focused questions with limited feedback types. They are less flexible in using language, gestures, facial expressions, and eye contact. Novice teachers are also more conservative in managing interpersonal distance, resulting in more distant communication with students. These findings provide valuable insights for teacher training and instructional strategy improvement.

Problem‐oriented teaching (POT) activities are important in classroom instruction. The level of student response can be influenced by different teacher behaviours in POT. However, the characteristics of teachers' multimodal behaviours at different levels of response are unclear. This study applied epistemic network analysis to explore the characteristics of teachers' multimodal behaviours in POT with different response levels. Using 256 POT segments from 12 classroom teaching videos in primary and secondary schools, this study found that authentic questions and iconic gestures occupied a central position in multimodal behaviours. The concomitant use of multimodal behaviours and the flexibility in eye contact could improve the response levels of students. These findings help understand how teachers' multimodal behaviours can promote students' responses. Practitioner notesWhat is already known about this topic Various behaviours have been shown to impact learners' response, such as teachers' responses to students and teachers' non‐verbal behaviours. Epistemic network analysis can be used to analyse the structure of the connections between cognitive, social and interaction data. What this paper adds This study analyses the characteristics of teachers' multimodal behaviours during problem‐oriented activities. It also compares the salient properties of non‐verbal, verbal and multimodal behavioural networks generated by interactive fragments of students' responses of different cognitive levels. Implications for practice and policy Teachers should ask more authentic questions in the classroom to promote in‐depth thinking processes in students. The appropriate coordination of multimodal behaviours by teachers is crucial to promote the cognitive level of students' responses. Upon making eye contact with students, teachers should also use iconic gestures and be flexible regarding the use of eye contact.

Traditional teaching quality assessment relies on manual observation, suffering from subjectivity, inconsistency, and inability to provide real-time feedback. This paper proposes a Transformer-based real-time assessment system integrating multimodal fusion and behavior analysis for smart classrooms. We develop a hierarchical multimodal Transformer that processes video, audio, interaction logs, and environmental sensors through intra-modal temporal encoding followed by cross-modal attention, capturing complex teacher-student relationships. The system identifies 32 pedagogical actions across five dimensions (instructional clarity, student engagement, interaction effectiveness, content delivery, classroom management) through interpretable behavior analysis, generating overall and dimension-specific assessments. Experiments on 180 lesson recordings demonstrate superior performance with MAE of 3.24 and Pearson correlation of 0.912, substantially outperforming single-modality approaches (Video-only: 6.43 MAE), simple fusion baselines (Early Fusion: 5.38 MAE), and flat attention architectures (4.56 MAE). Ablation studies reveal video contributes most critically (+2.63 MAE when removed), hierarchical architecture provides substantial benefits (+1.32 MAE), and behavior analysis significantly improves accuracy (+1.65 MAE). The system enables real-time monitoring with 30-second updates, supporting immediate instructional adaptation and teacher development through actionable feedback.

This study examines the expression of students’ motivation and engagement in primary mathematics education through the use of interactive digital tools. The relevance of the topic is highlighted by the increasing integration of digital technologies into the educational process, aiming to enhance students’ active participation, deepen understanding of mathematical concepts, and develop essential competencies that form the foundation for further learning. Mathematics in primary education plays a crucial role in developing conceptual understanding, procedural fluency, strategic problem-solving skills, logical and spatial reasoning, the ability to communicate mathematical ideas, and collaboration with peers. Additionally, fostering a positive attitude toward mathematics is particularly important, allowing students to perceive the subject's relevance and applicability in everyday life. Digital tools such as GeoGebra, Matific, ClassWise, augmented reality (AR), virtual reality (VR), and the EDUKA platform support these objectives by promoting student motivation, engagement, and academic achievement (Bertrand et al., 2024; Radović et al., 2018; Gulbinas & Arkušauskaitė, 2015). The research problem is based on the observation that although interactive digital tools are increasingly used in mathematics lessons, empirical data on their impact on student motivation and engagement are limited. Student behavior, activity levels, and collaboration depend on lesson structure, pedagogical methods, classroom microclimate, and the nature of the tools used, necessitating analysis in authentic educational contexts (Mula et al., 2025; Radišić & Baucal, 2024). The study aimed to analyze students’ motivation and engagement during mathematics lessons employing interactive digital tools. Research tasks included reviewing scientific literature, outlining the advantages and challenges of integrating digital tools, and analyzing observational data to identify forms of behavioral and emotional motivation. The study employed a qualitative research methodology, using structured observation to systematically record students’ engagement and motivation in lessons incorporating GeoGebra, Matific, ClassWise, AR/VR, and EDUKA. Data were analyzed using qualitative content analysis to identify key themes and interpret findings (Kardelis, 2016). The observation protocol focused on three domains: student behavior, emotional expression, and teacher activity. Approximately 40 students and four teachers from two Lithuanian primary schools participated in eight mathematics lessons, allowing for a diverse dataset reflecting different educational contexts. Results revealed that student engagement and motivation are closely linked to the type of interactive tool, the incorporation of gamified elements, and the structure of lesson activities. Lessons using Matific, GeoGebra, or ClassWise promoted higher attention, active participation, collaboration, and emotional involvement. Gamification features, such as point collection and visual performance indicators, enhanced extrinsic motivation, while Matific’s star ratings supported intrinsic motivation, encouraging students to achieve understanding beyond task completion (Noverianto & Munahefi, 2023). Less interactive tools, such as EDUKA modules without gamified features, elicited lower engagement and mechanical task completion. AR and VR applications stimulated curiosity, spatial reasoning, and visual understanding, although technical difficulties occasionally reduced motivation temporarily. Nevertheless, students’ overall responses to AR/VR experiences were positive, demonstrating interest and willingness to repeat activities (Bulut & Borromeo Ferri, 2023; Cao, 2023). The findings also indicated that long-term motivation depends on pedagogical context—consistent teacher support, structured lessons, clear objectives, and high interactivity levels reinforce not only short-term engagement but also sustained interest in mathematics. The study highlighted that technical challenges, including equipment availability, system failures, and platform licensing, can temporarily hinder motivation. Pedagogical preparedness is critical, as even advanced technologies do not yield results without structured lesson planning, gamification, and teacher guidance. Digital tools are most effective when combined with a positive classroom climate, social context, and diverse pedagogical strategies. In conclusion, interactive digital tools significantly enhance students’ motivation, engagement, collaboration, and ability to visualize complex mathematical phenomena. Their effectiveness, however, depends on teacher preparation, technological competence, classroom microclimate, and proper structuring of activities. Integrating interactive, gamified digital resources with social and pedagogical support provides conditions for meaningful, effective, and sustained learning in primary mathematics. The study emphasizes that successful digital tool implementation can maintain intrinsic motivation, foster ongoing learning enthusiasm, and improve academic outcomes, while also providing teachers with flexibility to differentiate tasks according to individual student abilities.

Realizing accurate recognition of Chinese and English information is a major difficulty in English feature recognition. Based on this difficulty, this paper studies the English feature recognition model based on deep belief network classification algorithm and Big Data analysis. First, the basic framework based on deep belief network classification algorithm and Big Data analysis is proposed. Combined with the Big Data analysis training model, the English feature information is processed. Through the recognition of different English text features, the recognition and matching of English features are realized. Then the errors of deep belief network classification algorithm and Big Data analysis are evaluated. Second, this paper describes the quantitative evaluation of deep belief network classification algorithm and Big Data analysis in this system. In the evaluation, the language feature evaluation method is used to improve the evaluation function. At the same time, the deep belief network classification algorithm and Big Data analysis are used to self-study the model, and the English feature recognition method with strong applicability is established. Finally, the effectiveness of the recognition system is verified by the experiment.

Understanding cultural systems in online learning with voluminous data, such as the engagement of students and teachers, has proven to be challenging. This predicament catalyzed the need for an emerging methodology to assess learning and complex thinking in online environments using statistical tools to analyze qualitative data. Quantitative Ethnography (QE) blends qualitative and quantitative approaches and uses statistical tools in analyzing qualitative data. This study advanced Epistemic Network Analysis (ENA), a QE tool, to analyze the engagement of teachers and students in online research courses of six groups of students. Observation of synchronous classes and asynchronous activities was carried out, 4,137 utterances were coded, and network models of engagements were generated. The findings show that the ENA engagement models among teachers and students in synchronous and asynchronous sessions differed significantly, with the teachers giving more frequent and varied feedback during synchronous classes. The network models of engagement in synchronous classes reveal the co-occurrence of students’ cognitive engagement and teachers’ feedback. When students try to connect with the lesson, they are given corrective feedback, and when they extend their understanding of the content, they are given affective feedback. On the other hand, when students are given prompts during asynchronous sessions, they try to connect. When they simply agree, they are given informative-procedural feedback, and when they emancipate their understanding, they receive affective feedback. This study illuminates the methodological advantages of ENA as an ethnographic tool that mixes qualitative and quantitative approaches in understanding cultural systems and processes in online education.

Biosensing technologies, which monitor physiological responses such as Heart Rate Variability (HRL), Skin Conductance Level (SCL), and Electroencephalogram (EEG) activity, offer a novel approach to enhancing the dissemination of health information in ideological and political education (IPE). In this context, health information encompasses topics such as mental health, stress management, and healthy lifestyle practices, all crucial to students’ overall well-being. Traditional health education methods cannot often capture real-time physiological and emotional responses, which can improve engagement and learning outcomes. This research explores the effectiveness of biosensing technology in enhancing the dissemination of health information within IPE. It examines how physiological data can be utilized to assess student engagement, emotional responses, and learning outcomes related to health. A mixed-methods approach was adopted, combining quantitative data from wearable biosensors (heart rate monitors, Galvanic Skin Response (GSR) sensors, EEG headsets) with qualitative feedback from students. Physiological data were preprocessed using signal filtering techniques, such as the Savitzky-Golay Filter, and features such as heart rate variability, skin conductance, and EEG alpha waves were extracted using the Kalman Filter (KF). A Modified Runge-Kutta Optimizer Integrated with Deep Belief Networks (MRKO-DBN) classifier was employed to predict student engagement based on these features. The research revealed that physiological responses, particularly heart rate variability and skin conductance, were strongly correlated with student engagement. The MRKO-DBN model achieved accuracy in predicting engagement. Qualitative feedback further confirmed that Biosensing technology significantly improved students’ engagement. Integrating Biosensing technology into health education within ideological and political contexts offers significant potential for enhancing student engagement and learning outcomes. By providing real-time, personalized feedback, it fosters a more interactive and responsive learning environment.

This article explores whether a learning community can affect students' learning achievement and engagement. Besides, this study also analyzed whether degree centralities of peer interaction affect learning achievement and learning engagement based on social network analysis. While the experimental group combined the English learning system with the online learning community, the control group was simply using the English learning system. The results indicated that the students' engagement from the online learning community were higher than the ones who used the English learning system only, although the learning achievement is not significant difference between these two groups. Moreover, higher interaction learners from the online learning community revealed better performance in learning achievement and student engagement. Other than that, the learners who played the “Center” emerged with a higher learning achievement as well as the students' engagement than the “Periphery” ones. The research provides suggestions for online learning with learning communications as well.

Student engagement is a significant predictor of students’ academic performance that has shown essential benefits for collaborative learning in higher education. Activities of social networking are common practices for college students to pursue higher academic achievement by taking advantage of collaborative learning. Nevertheless, there is a gap in understanding the relationship between student engagement and academic networking patterns. Using social network analysis, this study quantitatively explored the relationships between student engagement and academic networking at the individual level in the contexts of construction education. The self-reported collaboration and engagement data were collected from only dozens of undergraduate construction students at two universities in the US. The regression analysis revealed a positive relationship between student engagement and their academic network degree centrality. The findings suggest that students are more engaged in the coursework when directly collaborating with more peer classmates. Especially, students who seek collaboration with peer classmates can build great social capital and thus high engagement in class. Implications are finally discussed for construction educators to focus on student collaboration and advance the understanding of student needs.

Online learning plays a key role in current education system. Engagement detection in online learning is crucial as the student's success in online courses heavily depends on his/her state of mind. In our previous work, we used facial expressions labeled as engaged and not-engaged for student's engagement detection. In this paper, we use student's behavioral (on-task and off-task) and emotional (satisfied, bored, and confused) information for engagement detection. Five different convolutional neural network models have been tested for the behavioral and the emotional dimensions detection to detect student's engagement in online learning. The models are All Convolutional Network, Network in Network, Very Deep Convolutional Network, Conv-Pool Convolutional Network, and a proposed model combing some special features from the above models. We used the dataset ─ Dataset for the Affective States in E-Environments — for the performance evaluation. Experimental results show that the behavioral and emotional dimensions─based engagement detection provides a high accuracy.

The increasing integration of learning analytics (LA) in education creates new opportunities to improve student engagement and academic achievement, particularly in English language teaching (ELT). However, several critical limitations affect the deployment of these technologies, including privacy concerns and the need for advanced predictions about student behavior alongside adaptations to standard educational frameworks. This investigation examines the role of LA in ELT, focusing on how data-driven strategies powered by artificial intelligence (AI), such as deep learning (DL), can enhance student engagement and academic success. The dataset, sourced from a publicly available Kaggle repository, comprises anonymized real-world student interaction metrics, including demographics, engagement patterns, interactions with the learning system, and academic performance. To efficiently preprocess the dataset, numerical features are normalized using Z-score normalization, and features are extracted using term frequency-inverse document frequency (TF-IDF), which assigns weights to each term based on its frequency in a document and rarity across the dataset. DL models, including the weighted white shark optimized deep residual network (WWSO-DResNet), were employed to examine student engagement patterns, predict learning outcomes, and provide personalized learning paths. The findings indicate that students who received personalized learning interventions based on WWSO-DResNet insights performed significantly better in terms of engagement and academic achievement. The performance of the proposed WWSO-DResNet approach was evaluated using metrics such as MAE (0.09), RMSE (0.21), MSE (0.18), and MAPE (0.25). The WWSO-DResNet method proved effective in identifying at-risk students early, enabling preventive interventions. In conclusion, AI-powered LA holds promise for transforming ELT by fostering personalized learning experiences.

Online anonymity allows individuals to safeguard their personal information. It provides the freedom for anyone to express themselves without being concerned about censorship, discrimination, or retaliation. This pseudonym also provides opportunities for individuals to promote open discourse and diverse viewpoints. However, in today's digital age, this online anonymity has become a growing concern. Although the safeguarding of people’s rights, the advancement of free speech, and the development of a more diverse and democratic online community all depend heavily on the anonymity of online users, there are risks as well, such as cyberbullying, harassment, and the propagation of false information. Because of their anonymity, predators, groomers, and other unscrupulous individuals may be able to take advantage of vulnerable individuals, especially children and adolescents. In order to trick victims into hazardous or abusive situations, adversaries can hide their identities. Deep learning and network analysis can be used to reveal the true identities of anonymous social media users in order to combat this. Deep learning algorithms are capable of analyzing a wide range of social network data, including user behavior, relationships, and content, in order to find patterns and correlations that can lead to the true identities of users that go anonymous. The proposed system includes examining network topology and group dynamics to identify potential anomalies and connections that could lead to de-anonymization. This paper proposes a novel module for de-anonymization integrating deep learning and network analysis in social networks.

When the societies are worried about their educative process and they consider get it better, they are planning the progress in all their dimensions. There is the importance to set up politics that tend to have a high quality education. Nevertheless, the efforts in the Soledad township are not enough. In the development plan SOLEDAD CONFIABLE 2016-2019, the community indicated as a main problem the low quality education in the township. That is reflected in the performance levels measured by the ISCE. Hence, the investigation ́s objective is to analyze the continuous improvement processes of the educative quality in the successful schools of Soledad township. In other matters, this investigation used the paradigm quali- quantitative with a descriptive design to explain the academic process and the description of the factors that have influenced on this continuous process. With the help of four tools: documentary review rubric, semi-structured interview script and two questionnaires; it was achieved determine the specific practices that are using the principals and teachers to support the improvement of learnings and the integral development of the students.

Though higher education scholars have long known that undergraduate student engagement is associated with student persistence, they have yet to fully unravel its character and how it evolves over time. We argue that this is in part due to the individualistic, static ways engagement is measured even though scholars recognize it as a fundamentally relational concept. In our theoretical exploration, we draw on existing thinking about engagement in higher education and fit it to a social network paradigm that is well suited to conceptualizing and measuring relational, multidimensional, and dynamic phenomena such as student engagement. We conclude with suggestions for research and practice to further explore the longitudinal character of student engagement and how network analysis may be employed to explore the roles of engagement in persistence.

We study the eigenvalue distribution of random matrices pertinent to the analysis of deep neural networks. The matrices resemble the product of the sample covariance matrices, however, an important difference is that the analog of the population covariance matrix is now a function of random data matrices (synaptic weight matrices in the deep neural network terminology). The problem has been treated in recent work [J. Pennington, S. Schoenholz and S. Ganguli, The emergence of spectral universality in deep networks, Proc. Mach. Learn. Res. 84 (2018) 1924–1932, arXiv:1802.09979] by using the techniques of free probability theory. Since, however, free probability theory deals with population covariance matrices which are independent of the data matrices, its applicability in this case has to be justified. The justification has been given in [L. Pastur, On random matrices arising in deep neural networks: Gaussian case, Pure Appl. Funct. Anal. (2020), in press, arXiv:2001.06188] for Gaussian data matrices with independent entries, a standard analytical model of free probability, by using a version of the techniques of random matrix theory. In this paper, we use another version of the techniques to extend the results of [L. Pastur, On random matrices arising in deep neural networks: Gaussian case, Pure Appl. Funct. Anal. (2020), in press, arXiv:2001.06188] to the case where the entries of the data matrices are just independent identically distributed random variables with zero mean and finite fourth moment. This, in particular, justifies the mean field approximation in the infinite width limit for the deep untrained neural networks and the property of the macroscopic universality of random matrix theory in this case.