Multiple-choice questions (MCQs) are central to instruction and assessment, with distractors revealing student understanding and misconceptions. However, creating high-quality distractors is time-consuming, especially for emerging domains like K–12 AI education. This study explores using generative AI to support distractor creation in a self-paced online module integrating AI and Algebra 1. Five MCQs were selected to compare distractors written by human developers and ChatGPT, using expert reviews and log data from 80 students. Experts rated human distractors higher overall, though AI ones consistently ranked second. Log analysis showed human distractors drew more initial selections, while students who chose AI distractors spent more time engaging without differences in hint use or revisits. Transition patterns across attempts suggest AI-generated distractors can effectively guide students toward correct answers, highlighting their potential for scalable MCQ design.

With the increasing integration of Augmented Reality (AR) in education, learners can investigate scientific phenomena through embodied and interactive experiences, while collaborating in shared perceptual spaces. Although prior research has highlighted the conceptual and motivational benefits of AR, less is known about how students’ behavioral and emotional processes unfold during AR-supported collaboration. This study investigates 80 middle school students’ collaborative inquiry in an AR-based collaborative activity using a multimodal learning analytics approach. We combined video-based coding of verbal and non-verbal behaviors with audio-based emotion detection across three dimensions: arousal, dominance, and valence. Cluster analysis revealed three distinct collaboration patterns, which we further examined in relation to students’ emotional trajectories and learning outcomes. Findings show that groups in the reciprocal–balanced pattern engaged in active and coordinated behaviors, accompanied by more synchronized trajectories, and achieved higher learning gains. Dominant groups displayed high activity but uneven participation and imbalanced emotional dynamics. By contrast, passive–disengaged groups demonstrated limited coordination and unstable affect. This work advances understanding of how collaboration unfolds in AR-supported inquiry by linking behavioral and affective dimensions. Our results provide implications for the design of AR learning environments and analytics-driven supports for productive and emotionally balanced AR learning experiences.

Artificial Intelligence (AI) literacy is increasingly recognized as a critical competency for K–12 students, yet little is known about how learners engage with AI-focused modules in virtual school contexts. To address this gap, we designed an online narrative-driven AI literacy module (AI4VS) that integrates AI learning with Algebra 1. In this pilot study, data from 80 secondary school students who completed the 250-minutes module in three weeks were analyzed, including 117,866 system log records (e.g., submissions, clicks) and pre-/post-surveys on mathematics motivation, AI self-efficacy, and AI literacy. Using K-means clustering, we identified four distinct behavioral patterns: reflective learners, low-revision committers, high-frequency trial-and-error learners, and balanced learners. These groups demonstrated different outcomes: while all clusters showed significant improvement in AI self-efficacy, only some showed notable gains in motivation (i.e., low-revision committers and balanced learners) and AI literacy (i.e., balanced learners). The findings underscore the need for tailored scaffolds to better support varied learning strategies and highlight the potential of the AI literacy module in accommodating diverse learner profiles.

While teachers’ discourse is central to shaping student learning in online video-based contexts, little is understood about how specific linguistic features of discourse affect student behaviors and performance. This study examines teachers’ discourse patterns in large-scale online mathematics learning environments and their relationships with student dropout during videos and post-video accuracy. A total of 100 videos (25 topics) were collected from four teachers with differing levels of teaching experience (two senior, two junior). Drawing on systemic functional linguistics, discourse features were analyzed using natural language processing, focusing on process types, participant features, logical relations, and interactional markers. Results showed that senior teachers employed more logical extensions and imperatives but fewer interrogative and inclusive clauses than junior teachers. While dropout rates did not differ significantly, students of junior teachers achieved higher post-video accuracy. Regression analyses revealed that dropout was significantly lower in videos with greater use of relational processes. In contrast, post-video accuracy was positively associated with the use of material, mental, and relational processes, as well as elaboration, enhancement, interrogative, and inclusive features. Negative predictors of accuracy included technical terms, complex noun phrases, extension, and imperative forms.

Understanding how students reason about code is essential for providing tailored scaffolding in computer science (CS) education. Prior work has used think-aloud protocols with the Structure of the Observed Learning Outcomes (SOLO) taxonomy to examine students' code comprehension and programming levels. However, analyzing such data is labor-intensive and requires expert judgment. Recent advances in large language models (LLMs) offer a promising avenue for scaling this analysis, though their reliability for fine-grained coding remains uncertain. To address this gap, our study investigates the extent to which GPT-5 and 4o can classify SOLO levels and identify code-comprehension strategies from think-aloud transcripts of 27 high-school students working on block-based and text-based tasks. Results show modest alignment with human ratings for SOLO, with one-shot prompting improving agreement over zero-shot, though distinctions between adjacent lower levels (e.g., Prestructural 1 vs. 2) remained difficult. Strategy detection demonstrated stronger performance, achieving accuracies of 75–77% (block) and 62–67% (text), particularly for surface-visible strategies such as 'walkthroughs', 'control-structure identification', and 'pattern recognition', but weaker for less frequent, abstract, meta-cognitive strategies such as 'strategizing' (planning an approach) or 'thoroughness' (systematically checking work). These findings highlight both the potential and the limitations of using GPT-5 and 4o to analyze think-aloud data. While this work represents an initial step, with plans to examine more models, our preliminary results indicate that a human-in-the-loop approach is essential to ensure reliability and interpretive depth. Future work will extend this evaluation to other LLMs to better understand their role in supporting instructional decision-making.

Creativity is a critical learning outcome in K–12 computer science, yet assessing it at scale remains challenging. Human-scored approaches, such as the Consensual Assessment Technique (CAT), are resource-intensive and prone to rater variability. Leveraging advances in large language models (LLMs), this study investigates whether GPT-4o can reliably assess creativity in student-generated code. We collected 383 flow-based music programs from 194 upper-elementary students (ages 10–12) between 2022 and 2024. Each artifact was rated by five human experts across four dimensions: originality, complexity, efficiency, and emotional expressiveness. We evaluated three prompting strategies: zero-shot, few-shot with theory-driven exemplars (ECD), and few-shot with human-selected examples. Among them, the ECD-based few-shot prompting yielded the best performance, achieving the lowest mean absolute error (MAE = 0.582) and highest agreement within ±1.0 of human scores (82.4%). Zero-shot prompting, while slightly less accurate, achieved the highest correlation with human scores (Spearman's p = 0.53), suggesting its potential for lightweight deployment.

Augmented Reality (AR) has demonstrated significant potential in enhancing inquiry-based learning in K-12 classrooms. However, challenges such as communication barriers and unequal participation during collaboration highlight the need for structured support when conducting AR-based collaborative learning activities. This study introduces collaboration scripts in AR-based collaborative inquiry settings to examine their effects on learning outcomes and student experiences. A quasi-experimental study was conducted with 78 sixth-grade students, divided into an experimental group using collaboration scripts and a control group engaging in unscripted AR inquiry. Key findings indicate that collaboration scripts significantly enhance knowledge acquisition and reduce cognitive load. Observation and interviews further explained the effect of collaborative scripts in facilitating a more structured and effective collaborative inquiry process. The results highlight the importance of well-defined scripts and clear guidelines in improving AR-based collaboration.

This research explores how generative AI (GenAI) can enhance online AI education by automating the generation of multiple-choice distractors and personalized feedback.Building on the federally funded project AI Across the Curriculum for Virtual Schools, this study focuses on improving assessment and learning experiences for high-need high school students enrolled in Algebra I. Using a design-based research approach, the project develops a GenAI module integrated into existing AI-in-Math lessons and evaluates its impact on learning outcomes, AI self-efficacy, and student perceptions.Expert reviews and pilot studies will assess the pedagogical quality of GenAI-generated content.The study aims to address equity and scalability challenges in virtual AI education and contribute to the growing field of AI-enhanced learning environments.

Multiple-choice questions (MCQs) serve as fundamental assessment tools in computing education, where high-quality distractors are critical for evaluating conceptual understanding and debugging skills.While large language models (LLMs) show promise in automating distractor generation, their effectiveness in reasoningintensive programming domains remains understudied.Another challenge is that current evaluation metrics often emphasize surfacelevel semantics rather than the logical reasoning required in programming tasks, limiting their practical utility for educators.To compare AI-generated and human-authored distractors, this study collected 925 MCQs from two online high school courses.The collected data include the question stem, correct answer, and three human-authored distractors for each question.For AI-generated distractor generation, we employed the GPT-4 API through a structured prompt containing: (1) question stem, (2) correct answer, (3) Bloom's taxonomy level, and (4) instructional constraints.To determine the Bloom's taxonomy level for each question, two assessment experts independently classified all questions based on Bloom's taxonomy (Remember, Understand, Apply, Analyze, Evaluate, Create), achieving moderate inter-rater reliability (Cohen's 𝐾 = 0.67).Discrepancies, which occurred in 33% of cases, were resolved by a third expert to ensure accurate cognitive-level categorization.The GPT-4 model generated three plausible distractors per question while maintaining cognitive-level alignment.This study proposes a human-AI collaborative framework to evaluate distractor quality in programming education.Our human-AI collaborative evaluation framework combined human expertise with AI analysis (using GPT-4 and DeepSeek-V3) through a three-phase process: First, human-created and AI-generated distractors were anonymized and randomized.Next, human experts and AI models independently selected the three most pedagogically effective distractors per question based on plausibility and challenge potential.Finally, we implemented a ranking system prioritizing distractors with the highest selection frequency across evaluators, with ties resolved by cross-validator agreement.Results demonstrate that AI-generated distractors achieve comparable quality to human-crafted ones for foundational programming concepts (e.g., syntax recall and basic logic).However, significant gaps emerge in higher-order cognitive domains, particularly

Some online learning platforms frequently recommend educational materials to attract student engagement, with visual elements playing a critical role in capturing attention. To optimize the visual design of mathematical stories, this study examines the relationship between visual features and click frequency, based on log data from a U.S. platform featuring AI-generated mathematical stories for elementary students. Our methodology involves a multi-level visual feature extraction framework, categorizing features into low-, mid-, and high-level. Low-level features capture fundamental visual elements like color, texture, shape, and composition, commonly used for their simplicity. Mid-level features, inspired by psychological and artistic theories, more directly link to emotional impact, including attributes like brightness and contrast. High-level features focus on semantic content, using AI models to extract aesthetic scores and identify entities. Based on the correlation analysis between visual features and clicks, our findings indicate that images featuring characters and natural landscapes positively correlate with student interest, aligning with theories of situational interest. In contrast, images with pronounced brightness contrasts negatively impact engagement, likely due to increased cognitive load. The study highlights the limited influence of mid-level aesthetic features on elementary students' engagement, emphasizing the importance of visual clarity and educational relevance over purely aesthetic considerations.

Despite extensive research on code plagiarism detection in higher education and for programming languages like Java and Python, limited work has focused on K-12 settings, particularly for pseudocode. This study aims to address this gap by building explainable machine learning models for pseudocode plagiarism detection in online programming education. To achieve this, we construct a comprehensive dataset comprising 7,838 pseudocode submissions from 2,578 high school students enrolled in an online programming foundations course, along with 6,300 pseudocode samples generated by three versions of generative pre-trained transformer (GPT) models. Utilizing this dataset, we develop an explainable model to detect AI-generated pseudocode across various assessments. The model not only identifies AI-generated content but also provides insights into its predictions at both the student and problem levels, thus enhancing our understanding of AI-generated pseudocode in K-12 education. Furthermore, we analyzed SHAP values and key features of the model to pinpoint student submissions that closely resemble AI-generated pseudocode. This research offers implications for developing robust educational technologies and methodologies to uphold academic integrity in online programming courses.

Music and computer science (CS) have profound historical and structural connections, with programming music offering a promising avenue for engaging children in CS through creative expression. To foster this engagement, our team developed M-Flow, a flow-based music programming platform designed to introduce students to CS via music. Despite extensive existing research in music and CS education, experience reports and empirical studies on K-12 teachers' implementation and its impact on young kids' learning are limited. Therefore, we recruit elementary school teachers and students with no or limited prior programming experience, introducing them to M-Flow and its curriculum through a professional development workshop, a semester's job embedded support, and classroom implementation. We describe the experiences of teachers as they attempt to integrate music and CS, the challenges they face, and the influence on students' attitudes toward learning computing concepts. Specifically, we reflect on our intervention by conducting a sequential mixed-method evaluation. During the qualitative phase, we collected multiple sources of data from three teachers through focus groups and debriefings after a semester of classroom implementation. Thematic analysis of workshop activities, interviews, and debrief videos revealed three themes with seven sub-themes on teachers' integration of flow-based music programming and two themes with five sub-themes on challenges faced by the teachers. In the quantitative phase, we gathered data on attitudes and self-efficacy from 75 students taught by these teachers. Results indicate that the flow-based music programming environment provided an engaging programming experience for students and significantly increased their self-efficacy towards learning programming.

This work-in-progress research paper aims to explore students' dropout behavior during video engagement in online learning platforms. As online learning becomes increasingly popular, analyzing how students engage with video content provides important insights into their learning behaviors. This study explores multiple factors influencing K-12 students' in-video dropout rates in online math education. We examined 34,666,481 log entries from Math Nation, covering 1313 videos and 14,251 students. Using survival analysis, we evaluated how 27 variables, including demographic details, video interaction behaviors, and video characteristics(e.g. length, category), affect in-video dropout. Our findings reveal that video length significantly predicts dropout, with each additional minute increasing the dropout rate by 1.26%. Videos with higher dropout rates often feature more frequent pauses, jumps, and rewatches. The study also highlights that the quality of video content, the creators of the videos, and how students interact with the videos are crucial factors affecting dropout rates. Further research is needed to determine the specific causes of video dropout.

This study examines public sentiment towards AI in education, focusing on the impact of ChatGPT's launch by OpenAI on November 30, 2022. Analyzing around 80,000 Twitter posts from before and after the launch, we conducted a comprehensive sentiment analysis using a fine-tuned BERT, outperforming traditional methods such as VADER and SVM. We applied an RDD to assess the causal impacts of ChatGPT's introduction on public sentiment track sentiment shifts, highlighting how the introduction of AI technologies like ChatGPT has influenced educational discourse. Our findings reveal significant public sentiment changes post-launch, contributing new insights into AI's role in education and public discourse.

Prompt-based methods with Pre-trained Language Models (PLMs) have demonstrated remarkable zero-shot performance in text classification tasks. Among these methods, verbalizers are employed to convert model-predicted vocabulary logits into task-specific labels. However, most existing zero-shot approaches face several challenges: (1) a reliance on unlabeled data or additional knowledge bases, which limits their effectiveness in varied data scenarios; (2) a disregard for semantic ambiguity issues in verbalizer construction. This paper presents a novel fully zero-shot approach named Hierarchical Generative Semantic-Aware Generative Verbalizer (HGSVerb). (1) We propose a prompt-based module, Hierarchical Semantic Verbalizer Generation, that explores the full utilization of PLMs and the semantics of label names. As a result, without requiring any extra data or knowledge, our model iteratively generates a verbalizer with hierarchical semantics and structure. (2) We propose a Semantic-Aware Refinement method based on semantic distance between label words and categories to reduce the semantic ambiguity issue. (3) Additionally, HGSVerb introduces a Hierarchical Weight Aggregation to optimize the utilization of our verbalizer by using a decay coefficient to distinguish the importance of label words generated by different iterations. We evaluate HGSVerb on four topic and sentiment classification datasets, and our average accuracy even outperforms those methods utilizing extra resources. We also examine its transferability on datasets with diverse numbers of classes and topics. HGSVerb achieves the best results compared with existing fully zero-shot methods.

Public cloud providers embrace multi-tenancy as a strategy to enhance the utilization and efficiency of resources. However, co-located virtual machines (VMs) suffer from qualityof-service (QoS) degradation caused by shared resource interference. Existing solutions for predicting QoS degradation often rely on the assumption of online access to application-level information. However, in a production environment, this assumption proves invalid as the VMs are black boxes to the providers. This intrinsic characteristic of the IaaS cloud necessitates the prediction model to generalize to unfamiliar applications and imposes specific criteria on the monitorable metrics.To meet the black-box scenario under Infrastructure as a Service (IaaS) cloud computing, we present a novel framework, FEDGE, that can predict interference-aware QoS (IA-QoS) of co-located VMs using only low-level monitorable metrics before migration. Specifically, FEDGE utilizes a stochastic gates layer to select the most informative features from the high-dimensional resource and hardware metrics, which helps to reduce the monitoring overhead. Furthermore, we design a multi-domain MMD-based adversarial denoising autoencoder to regularize the learned hidden representations and prevent over-fitting on the source domains. Next, we employ a multi-layer perceptron (MLP) to accurately predict complex QoS degradation using the learned representations with domain generalization. Experimental results demonstrate that FEDGE outperforms other state-of-the-art methods in terms of both generalizability and effectiveness.

In light of the emergence of privacy breaches in federated learning, secure aggregation protocols, which mainly adopt either homomorphic encryption or threshold secret sharing techniques, have been extensively developed to preserve the privacy of each client's local gradient. Nevertheless, many existing schemes suffer from either poor capability of privacy protection or expensive computational and communication overheads. Accordingly, in this paper, we propose an efficient and multi-private key secure aggregation scheme for federated learning. Specifically, we skillfully design a multi-private key secure aggregation algorithm that achieves homomorphic addition operation, with two important benefits: 1) both the server and each client can freely select public and private keys without introducing a trusted third party, and 2) the plaintext space is relatively large, making it more suitable for deep models. Besides, for dealing with the high dimensional deep model parameter, we introduce a super-increasing sequence to compress multi-dimensional data into one dimension, which greatly reduces encryption and decryption times as well as communication for ciphertext transmission. Detailed security analyses show that our proposed scheme can achieve semantic security of both individual local gradients and the aggregated result while achieving optimal robustness in tolerating client collusion. Extensive simulations demonstrate that the accuracy of our scheme is almost the same as the non-private approach, while the efficiency of our scheme is much better than the state-of-the-art baselines. More importantly, the efficiency advantages of our scheme will become increasingly prominent as the number of model parameters increases.

Wireless Charging Vehicles (WCVs) have been widely explored as a means of enabling continuous operation of sensors that are powered by batteries. However, the energy consumption of WCVs can be inefficient, leading to insufficient energy supply for sensors that are located in challenging-to-access areas. Consequently, there is a need to design an effective charging and energy sharing scheme for sensors to improve the quality of service in this setup. This paper focuses on the Joint optimization of Mobile charging and Energy sharing of sensors (JOIN-ME) problem, which is known to be NP-hard. To address this challenge, we first transform JOIN-ME into a submodular maximization problem with general constraints. Subsequently, we propose the Routing planning, Mobile charging, and Energy sharing for Sensing devices (RMES) algorithm, which has an approximation ratio of 1/8(1-1/e). Finally, we conduct experiments to showcase the superior performance of RMES compared to existing baselines, under varying scales and constraints. Our work on the design of an efficient charging and energy sharing scheme for sensors can significantly improve the reliability and longevity of wireless sensor networks, enabling the deployment of these networks in critical applications such as environmental monitoring, crowd sensing, and security surveillance.

Preschool children have difficulty learning and comprehending abstract concepts, and the cognition of numbers has always been the key to mathematical enlightenment for young children. Our research aims to help preschool children build their cognition of cardinal and ordinal numbers, comprehend simple logical relationships, and master simple digital addition. We developed an Augmented Reality learning tool based on theories related to number cognition and a theoretical framework of software design for preschool children. We also conducted a teaching experiment in a kindergarten, and interviewed the teachers of the kindergarten to learn about their attitudes towards the application of AR in preschool education. Through data analysis, interviews, and discussions, we conclude that (a) AR application can positively influence children’s cognitive digital skills; (b) children have positive attitudes and positive evaluations toward AR application use, but there are some unavoidable problems in children’s attention allocation; (c) proficiency in operating AR tools has a large impact on children’s learning effects.

Augmented Reality (AR) has great potential in science education, and Collaborative Inquiry-based Learning (CIBL) in the AR environment is of great significance. However, there is a problem of low collaborative performance in technology-based CIBL. This study applied the strategy of role assignment to AR-based CIBL, aiming to explore the effect of role assignment on students’ collaboration. Forty-seven sixth-grade students in elementary school were randomly divided into Group A (without role assignment) and Group B (with role assignment) to participate in AR-based collaborative scientific inquiry activities. Data on students’ scientific knowledge achievement, attitudes toward science learning, cognitive load, and flow experience were collected. In addition, interviews were conducted to investigate students’ opinions on role assignments. It is found that the strategy of role assignment could significantly improve students’ science knowledge achievement. The interview results revealed how role assignments facilitate students’ collaboration from three aspects.

It is a common strategy to predict oncogenes by differential analysis between somatic mutations and background mutations. Most previous methods only utilize mutations in the cancer population to model its background mutation, which have an obvious bias. A recent method, DiffMut, improves this issue by conducting differential mutational analysis with both mutations in the cancer population and the natural population. However, it assumes the impacts of all mutations are equal, neglecting their functional difference. Thus, we developed a method, DGAT-onco that integrated the functional impacts of mutations to the differential mutational analysis framework of DiffMut. We performed DGAT-onco analysis with 33 cancer types from the Cancer Genome Atlas (TCGA) dataset. Its reliability was further evaluated on an independent test set including 22 cancers from other sources (TS22). Using oncogenes from the Cancer Gene Census (CGC) as the gold standard, our method achieves higher classification performance in oncogene discovery than five alternative methods (i.e., DiffMut, WITER, OncodriveCLUSTL, OncodriveFML, and MutSigCV) with an average AUPRC of 0.197 and 0.187 in TCGA and TS22 respectively. The source code and supplementary materials of DGAT-onco are available at https://github.com/zhanghaoyang0/DGAT-onco.

Fast and accurate extraction of vascular structures from medical images is fundamental for many clinical procedures. However, most of the vessel segmentation techniques ignore the existence of the isolated and redundant points in the segmentation results. In this study, we propose a vascular segmentation method based on a prior shape and local statistics. It could efficiently eliminate outliers and accurately segment thick and thin vessels. First, an improved vesselness filter is defined. This quantifies the likelihood of each voxel belonging to a bright tubular-shaped structure. A matching and connection process is then performed to obtain a blood-vessel mask. Finally, the region-growing method based on local statistics is implemented on the vessel mask to obtain the whole vascular tree without outliers. Experiments and comparisons with Frangi’s and Yang’s models on real magnetic-resonance-angiography images demonstrate that the proposed method can remove outliers while preserving the connectivity of vessel branches.