This study examines how AI-based teachable agents with distinct personality traits influence middle school students’ emotions and mathematics learning. Grounded in the Big Five personality framework, six agents were developed: five designed to emphasize one of the Big Five traits and one without a personality emphasis. Students engaged in teaching the agents to solve mathematical problems. Guided by the Control-Value Theory of Achievement Emotions, students’ emotions were coded by valence (positive vs. negative) and activation (activating vs. deactivating) based on their conversations with the agents, while mathematics learning was assessed through coded applications of knowledge during interaction and a posttest. Results showed that extraversion-, openness-, and agreeableness-emphasis agents promoted positive activating emotions (e.g., enjoyment), whereas conscientiousness-emphasis agents were particularly effective in reducing both negative activating emotions (e.g., anxiety) and negative deactivating emotions (e.g., boredom). Emotions were further linked to learning outcomes: positive activating emotions positively predicted knowledge application and posttest performance, whereas negative deactivating emotions negatively predicted students’ knowledge application. These findings highlight the nuanced role of teachable agent personality in shaping students’ emotional experiences and provide design implications for developing teachable agents that effectively support affective and academic dimensions of mathematics learning.

Building on growing evidence of the effectiveness of teachable agents for learning, this study investigated their use of instructional, cognitive, and pastoral conversational strategies, three dimensions of support that learning theories (e.g., the Community of Inquiry framework and Self-Determination Theory) identify as interconnected and critical for student learning, to inform the design of pedagogical conversational agents. By analyzing over 8,000 conversations between teachable agents and students, we found that agents’ cognitive and instructional strategies strongly promoted students’ cognitive elaboration, whereas pastoral strategies were associated with surface-level cognitive engagement and higher affective engagement. Moreover, two-strategy combinations (e.g., cognitive + instructional, instructional + pastoral) were generally more effective for fostering cognitive, affective, and metacognitive engagement than any single strategy, while combining all three strategies often weakened effects. Integration of cognitive and pastoral strategies further enhanced procedural knowledge application, whereas instructional strategies supported conceptual knowledge. Overall, these results clarify the complementary yet distinct roles of instructional, cognitive, and pastoral strategies and provide insights for pedagogical agents to dynamically pair functional strategies to support student learning.

Scaling high-quality mathematics assessment requires balancing automated generation with pedagogical standards. While Large Language Models show promise for educational content creation, systematic evaluation of their output and integration with human expertise remain underexplored. We propose a Generate–Filter–Edit human–AI pipeline combining GPT-4's generative capacity with expert teacher judgment to produce middle-school mathematics items for large-scale deployment. Analyzing 5,634 GPT-4-generated questions, we found 68.74% ready for immediate use, 19.42% requiring revision, and 11.84% discarded. Through comprehensive statistical significance testing on 378 textual features, we identified key linguistic predictors distinguishing quality levels. Low-quality items exhibit significantly higher paragraph-level textual overlap (Cohen's d up to 0.403) and greater verbosity, while items requiring revision show higher rare vocabulary ratios. Teacher editing behaviors follow a light-touch approach, focusing on mathematical language formalization while preserving content fidelity. Post-deployment engagement analysis revealed significant correlations between textual complexity features and student–chatbot interaction patterns. Our results validate scalable human–AI collaboration for mathematics item development and provide evidence-based frameworks for automated quality assessment and editing support systems.

Analyzing multimedia data, which often comprises diverse views such as text, images, and videos, presents unique challenges for data processing. Deep matrix factorization (DMF) provides an elegant way to obtain reduced-dimensional representation of the multiview data produced by multimedia. Compared with single-layer matrix factorization, DMF can better discover the hierarchical information in a layerwise technique. However, the existing multiview DMF methods still have several problems: 1) the standard DMF using Frobenius norm fails to process data containing noises and outliers; 2) most DMF methods neglect to exploit the feature diversity to learn a more discriminative representation; and 3) in graph learning methods for DMF, the$k$NN method is utilized to construct data graphs, which results in many incorrect neighbor assignments. To address these issues, a robust multiview deep nonnegative matrix factorization with feature diversity and optimal graph learning (RMvDNMF-FG) is proposed for clustering in this article. Specifically, a noise-insensitive logarithmic loss function is designed to measure the factorization error, inner products of basis vectors are minimized to achieve feature diversity for obtaining discriminative representation, and an optimal graph construction strategy is proposed to maintain the geometric structure of the data. To solve the proposed model, we explore an iterative updating algorithm that makes the objective function decrease consistently as the number of iterations increases. Additionally, the convergence proof of the iterative updating algorithm is provided with detailed mathematical analysis. Furthermore, through numerous comparative experiments with eleven state-of-the-art algorithms on five multiview datasets, the effectiveness of the proposed method is demonstrated.

The importance of discussing support for learning within online learning communities is widely recognized, yet the diverse user behaviors, especially in the realm of online math learning, are not thoroughly investigated. This work in progress research paper explores the mathematical literacies, and success rates of discussion learning among students taking on different participation roles in an asynchronous online math learning setting. This inquiry is pivotal for comprehending the mechanisms that uphold online learning communities and can provide insights for crafting online discussion activities. This paper employs a mixed-methods approach to analyze big educational data and core-periphery structures within the community. Initially, users are classified into core, periphery, and X-periphery (extra) groups using an extended Surprise detection algorithm, which evaluates interaction quality. The Mann-Whitney U test is applied to assess differences in math literacy and discussion success rates among the groups. The findings reveal that each group is more responsive to its own members, with the core group exhibiting a more balanced response pattern. Notably, the X-periphery group shows the highest success rate in discussions, suggesting that lower activity levels do not compromise communication efficiency. Statistical analysis indicates that while the core and periphery groups have similar levels of math literacy, the X-periphery group excels in all three types of mathematical literacy assessed. These results highlight the importance of considering group dynamics and participation roles when designing online math learning activities to foster effective communication and support within the community. The study's insights into social support traffic offer practical implications for practitioners aiming to enhance the sustainability of online learning communities through tailored discussion activities.

Open-ended mathematical problems are a commonly used method for assessing students’ abilities by teachers. In previous automated assessments, natural language processing focusing on students’ textual answers has been the primary approach. However, mathematical questions often involve answers containing images, such as number lines, geometric shapes, and charts. Several existing computer-based learning systems allow students to upload their handwritten answers for grading. Yet, there are limited methods available for automated scoring of these image-based responses, with even fewer multi-modal approaches that can simultaneously handle both texts and images. In addition to scoring, another valuable scaffolding to procedurally and conceptually support students while lacking automation is comments. In this study, we developed a multi-task model to simultaneously output scores and comments using students’ multi-modal artifacts (texts and images) as inputs by extending BLIP, a multi-modal visual reasoning model. Benchmarked with three baselines, we fine-tuned and evaluated our approach on a dataset related to open-ended questions as well as students’ responses. We found that incorporating images with text inputs enhances feedback performance compared to using texts alone. Meanwhile, our model can effectively provide coherent and contextual feedback in mathematical settings.

The significance of nurturing a deep conceptual understanding in math learning cannot be overstated. Grounded in the pedagogical strategies of induction, concretization, and exemplification (ICE), we designed and developed a conversational AI using both rule- and generation-based techniques to facilitate math learning. Serving as a preliminary step, this study employed an experimental design involving 151 U.S.-based college students to reveal students’ attention patterns, technology acceptance model, and qualitative feedback when using the developed ConvAI. Our findings suggest that participants in the ConvAI group generally exhibit higher attention levels than those in the control group, aside from the initial stage where the control group was more attentive. Meanwhile, participants appreciated their experience with the ConvAI, particularly valuing the ICE support features. Finally, qualitative analysis of participants’ feedback was conducted to inform future refinement and to inspire educational researchers and practitioners.

While virtual learning environments (VLEs) are widely used in K-12 education for classroom instruction and self-study, young students’ success in VLEs highly depends on their self-regulated learning (SRL) skills. Therefore, it is important to provide personalized support for SRL. One important precursor of designing personalized SRL support is to understand students’ SRL behavioral patterns. Extensive studies have clustered SRL behaviors and prescribed personalized support for each cluster. However, limited attention has been paid to the algorithm bias and fairness of clustering results. In this study, we “fairly” clustered the behavioral patterns of SRL using fair-capacitated clustering (FCC), an algorithm that incorporates constraints to ensure fairness in the assignment of data points. We used data from 14,251 secondary school learners in a virtual math learning environment. The results of FCC showed that it could capture six clusters of SRL behaviors in a fair way; three clusters belonging to high-performing (i.e., H-1. Help-provider, H-2) Active SRL learner, H-3) Active onlooker), and three clusters in low-performing groups (i.e., L-1) Quiz-taker, L-2) Dormant learner, and L-3) Inactive onlooker). The findings provide a better understanding of SRL patterns in online learning and can potentially guide the design of personalized support for SRL.

Mathematical stories have demonstrated the ability to bolster the motivation and interest of students in learning mathematics, thereby exerting a positive influence on their academic performance. However, due to a lack of adequate resources and the desire to engage students in the creation process, Generative Artificial Intelligence (GAI) is utilized to generate mathematical stories accompanied by images. This study presents a framework for automatic quality assessment to evaluate the coherence of multimodal text and images generated by GAI, as well as the appropriateness of the stories for different grade levels. The dataset comprises mathematical stories generated by GAI for grades 3, 4, and 5, obtained from an American online learning platform, each story consisting of titles, bodies, and image illustrations generated by GP4, images generated by DALL-E3. Initially, a method is devised based on CLIP model and Mini-GPT4 for extracting multimodal semantic features to establish the relationship between text and images in mathematical stories and their generation parameters. Subsequently, mathematical text features are designed, including nine mathematical text attributes and ten traditional text readability indicators, followed by collinearity feature selection and statistical testing. Finally, five machine learning grade regressor for mathematical stories were trained, and the correlation between these 19 features and grades is explored using genetic algorithm-based factor mining and the interpretable artificial intelligence method SHapley Additive exPlanations (SHAP). To further understand advanced text features, the latest natural language processing (NLP) readability indicators are also integrated into the analysis. Through multimodal features, traditional text readability metrics, and NLP readability indicators, this method introduces a novel approach for automatically assessing the quality of GAI-generated multimodal mathematical stories, providing a tool for grade predictor and shedding light on the factors (Image-text relevance and textual features) influencing the grade level of analyzed stories, thereby offering new insights for leveraging GAI in mathematical education.

This research aims to investigate the allocation mechanisms of written positive affective feedback (PAF) in online mathematics assignments provided by teachers, employing multimodal learning analytics. We extract mathematical text features and readability indicators from teacher comments, utilizing collinearity matrices for linear feature selection. Student multimodal response patterns are obtained through clustering. To analyze the teacher comment strategies under different student response patterns, Mann-Whitney U tests were employed to investigate differences in student scores and feedback readability between scenarios with and without PAF. Our findings uncover the linguistic characteristics of teacher-provided PAF and the corresponding strategies they adopt. Teachers are more inclined to offer PAF to K-12 students with higher scores, challenging assignments, and younger ages. The study points out potential imbalances in the allocation of teacher PAF and emphasizes key factors that teachers need to consider when providing PAF. The findings offer new insights for educators to contemplate on designing and implementing more effective PAF strategies.

Students' participation in online discussions often varies over time. Leveraging over three million discussion interactions from an online math learning platform, the current study aims to investigate the sustainability of student participation in an online mathematical discussion board by examining how students evolve from newcomers in the discussion board into old-timers over an academic year. Additionally, it seeks to explore how the timing of students' joining the discussion board and their early technology and social participation behaviors influence their continued participation, based on Communicative Ecology Theory (CET), which conceptualizes the sustainability of an online community as being influenced by factors related to communicating technology, social interactions, and exchanged information. The results revealed that students decreased their participation or even left the discussion board over time, underscoring the need to maintain students' participation in online mathematical discussions. Students' social interactions and their technology use in the online discussion board were found to influence their sustained participation. The timing of students joining and their role as newcomers or old-timers were also found to affect their participation behaviors and sustained participation. Future investigations have been planned to further use machine learning to examine students' discussion content and understand the role of discussion content on students' continued participation.

Participating in online mathematical discussions is considered a beneficial strategy to improve online math learning. However, achieving high levels of interaction in these discussions is uncommon, and maintaining them is often challenging. To shed light on the potential mechanisms to sustain online mathematical discussions, this study draws on Communicative Ecology Theory (CET), which conceptualizes the sustainability of online communities as influenced by technical, social, and discursive factors, to examine how students' technical, social, and content-related participation patterns on an online mathematical discussion board are correlated. The results indicated that students' engagement with the communication tool and the motivation system significantly facilitated their social interactions and enhanced their demonstration of mathematical knowledge, mathematical literacy, and affect control. These findings demonstrate the interplay among the three types of participation patterns and provide insights for educators and designers of educational applications to enhance student participation in online discussions thereby improving the effectiveness of these discussions in online learning. Future investigations are planned to focus on how such interplay varies over time.

Because of the structural parallelisms between music and computing, it has long been suggested that coding music could be a good way for young children to engage in and learn about computer science (CS). Despite these suggestions, coding music has not reached a wider audience of young children, and the approach’s potential to engage them has not been thoroughly demonstrated. To facilitate the adoption of coding music activities, we created M-flow, a flow-based programming platform that allows young children to code music intuitively from the outset. Then, we developed a standards-aligned curriculum that teachers applied in their fourth-grade classrooms. Surveys indicate that children were greatly engaged, the experience successfully exposed them to and increased their self-efficacy toward programming. Our results indicate that with the appropriate coding platform, coding music can be a powerful way to engage children in CS.

Mathematical discussions have become a popular educational strategy to promote math literacy. While some studies have associated math literacy with linguistic factors such as verbal ability and phonological skills, no studies have examined the relationship between linguistic synchrony and math literacy. In this study, we modeled linguistic synchrony and students’ math literacy from 20,776 online mathematical discussion threads between students and facilitators. We conducted Cross-Recurrence Quantification Analysis (CRQA) to calculate linguistic synchrony within each thread. The statistical testing result comparing CRQA indices between high and low math literacy groups shows that students with high math literacy have a significantly higher Recurrence Rate (RR), Number of Recurrence Lines (NRLINE), and the average Length of lines (L), but lower Determinism (DET) and normalized Entropy (rENTR). This result implies that students with high math literacy are more likely to share common words with facilitators, but they would paraphrase them. On the other hand, students with low math literacy tend to repeat the exact same phrases from the facilitators. The findings provide a better understanding of mathematical discussions and can potentially guide teachers in promoting effective mathematical discussions.

Numerous studies have successfully adopted learning analytics techniques such as machine learning (ML) to address educational issues. However, limited research has addressed the problem of algorithmic bias in ML. In the few attempts to develop strategies to concretely mitigate algorithmic bias in education, the focus has been on debiasing ML models with single group membership. This study aimed to propose an algorithmic strategy to mitigate bias in a multi-group context. The results showed that our proposed model could effectively reduce algorithmic bias in a multi-group setting while retaining competitive accuracy. The findings implied that there could be a paradigm shift from focusing on debiasing a single group to multiple groups in educational attempts on ML.

Discussion forums are important for students’ knowledge inquiry in online contexts, with help-seeking being an essential learning strategy in discussion forums. This study aimed to explore innovative methods to build a peer recommender that can provide fair and accurate intelligence to support help-seeking in online learning. Specifically, we have examined existing network embedding models, Node2Vec and FairWalk, to benchmark with the proposed fair network embedding (Fair-NE). A dataset of 187,450 post-reply pairs by 10,182 Algebra I students from 2015 to 2020 was sampled from Algebra Nation, an online algebra learning platform. The dataset was used to train and evaluate the engines of peer recommenders. We evaluated models with representation fairness, predictive accuracy, and predictive fairness. Our findings suggest that constructing fairness-aware models in learning analytics (LA) is crucial to tackling the potential bias in data and to creating trustworthy LA systems.

Educational researchers have examined artificial intelligence (AI) to automatically support students' learning at a large scale. However, it has been broadly identified that AI models can suffer from algorithmic bias. A recent educational research initiative has shifted to evaluate and address algorithmic bias. Nonetheless, few studies have examined students' perceived fairness towards AI-enhanced systems in education. This study aimed to explore and understand factors that shape students' perceived fairness towards an AI system for math learning at the college level. Specifically, we have conducted a between-subjects randomized experiment with 395 participants to reveal factors that influenced participants' perceptions of fairness. The results showed that students' math anxiety levels and majors could influence participants' reported perceived fairness. In contrast, the outcome favorability of the predictive system did not affect students' perceptions of fairness. Future investigations have been planned to further understand the relationships between the designed manipulations and perceived fairness.

To support online learners at a large scale, extensive studies have adopted machine learning (ML) techniques to analyze students’ artifacts and predict their learning outcomes automatically. However, limited attention has been paid to the fairness of prediction with ML in educational settings. This study intends to fill the gap by introducing a generic algorithm that can orchestrate with existing ML algorithms while yielding fairer results. Specifically, we have implemented logistic regression with the Seldonian algorithm and compared the fairness-aware model with fairness-unaware ML models. The results show that the Seldonian algorithm can achieve comparable predictive performance while producing notably higher fairness.

This study examined two machine learning models for de- signing a learning analytics dashboard to assist teachers in facilitating problem-based learning. Specifically, we used BERT to automatically process a large amount of textual data to understand students' scientific argumentation. We then used Hidden Markov Model (HMM) to find students' cognitive state transition with time-series data. Preliminary results showed the models achieved high accuracy and were coherent with related theories, indicating the models can provide teachers with interpretable information to identify in-need students.