We present Waste Genie+ (WG+), an AI-infused web-based educational technology to improve sustainability awareness and waste management skills. WG+ utilizes Large Language Models (LLMs) to transform complex environmental regulations and waste management information into accessible and digestible content, enabling learning in the complex, legislative-involved sustainability field. WG+ features AI-infused content, interactive quizzes, and various sustainability awareness simulations (i.e., waste classification scanner, virtual carbon credit tracker, carbon dioxide emission converter). We used LLM chain-of-thought reasoning to implement a systematic prompt-based evaluation of the decomposed legislative content quality. A 10-day user study with 10 participants was designed and conducted to evaluate the effectiveness of WG+ in improving users’ understanding of waste management practices and regulations. Results demonstrated that our Environmental Legislative-guided LLM successfully extracted coherent, highly readable content while preserving the original legal information. The results also indicated a significant improvement in the participants’ sustainability awareness and waste sorting abilities. It also revealed that users found the quizzes engaging, and the AI-generated bite-sized content was more digestible and easier to understand compared to the original bills or articles. These findings contributed to our understanding of how AI-enhanced educational technologies can support continuous informal learning in this growing, convoluted field and promote environmental stewardship, facilitate public awareness of sustainability practices and policies.

This work presents a platform that facilitates sustainability learning through waste management principles - reducing and reusing waste from everyday items. The platform features (a) an AI assistant that helps users express and develop their ideas about sustainable practices, and (b) a commentary interface to support asynchronous collaboration. The AI agent provides context-aware suggestions while encouraging users to modify and personalize these recommendations, creating a collaborative approach to sustainability ideation. A user study was conducted, and the effectiveness of the approach was evaluated. Results demonstrated increased sustainability awareness among participants after using the platform, with varying patterns of improvement across different sustainability approaches. In particular, users who actively modified AI suggestions produced higher-quality contributions with more specific actionable recommendations compared to those who directly copied AI's responses. The study also reveals insights into how AI assistance affects content quality. These findings contribute to understanding how AI can be effectively integrated into sustainability education platforms to enhance learning outcomes.

Many pose recommendation applications have been proposed and studied, primarily focusing on providing exercise feedback and fine-tuning postures to enhance precision for skill acquisition. These applications often utilize pose extraction and identification to help users perfect their form, whether in yoga, fitness routines, or other sports activities. However, these specialized applications are not generalizable for leisure use, such as capturing a photogenic or novel pose in front of an iconic tourist attraction. To address this issue, we present InstaPose, a mobile-friendly framework that delivers real-time pose recommendation and correction feedback. The core of our approach leverages Mobile Vision Transformers (MobileViT) to retrieve refined pose images from our dataset by matching the background context of the input photo. In addition, the proposed model outperforms VGG16 on our newly curated, large-scale high-quality dataset. This demonstrates the potential of our approach for real-time mobile pose guidance.

Responsible waste management is important for our sustainable living. However, there are still many challenges ahead, such as properly sort the wastes. In this paper, we conducted a Theory of Planned Behavior (TPB)-based study to understand the impact of waste management literacy on people's intentions. Our results indicate that increasing the literacy level can positively influence people's intentions to actively sort their wastes. Additionally, we also proposed a mobile application that includes an intelligent Waste Detect Engine (WDE) and a Waste management Knowledge Agent (WMKA) to facilitate people learn. In a subsequent user study, we observed a significant improvement in users' waste sorting knowledge with the assistance of the proposed educational technology.

Augmented reality (AR) has been shown its prominence in immersive learning across various fields of education, especially in STEM curricula. Researchers have demonstrated that AR visualization can improve learning and cognitive outcomes by multimodal interaction. However, compared to passive modes of learning like reading and viewing, there is little research focused on the effects of AR on a collaborative problem-solving task like pair programming in computer science education, which is difficult to analyze due to the increased complexity in the task and user-user interaction. We designed an AR application, where two participants can work together to solve a programming task, and conducted a within-subject study to evaluate how the participants' code editing and vocal communication are different in the AR and non-AR visualizations. The result showed that participants using the AR visualization exhibited strategic programming behavior and had more interactive communication than using the non-AR visualization. Overall, this preliminary work contributes to the understanding of how AR can support collaborative problem-solving and engage users in verbal communication by multimodal analytics.

Learning from erroneous examples involves the intentional inclusion of errors as part of the learning process. Prior works, mostly from the field of mathematics, have investigated how this can be used in blended learning environments to help students. Due to the Covid-19 pandemic, most learning activities have shifted to online, motivating us to study and utilize students' use of an existing grading platform. Students were tasked to evaluate various degrees of erroneous answers as their learning opportunities, resembling program debugging. The grading process was engineered to supply feedback to students by revealing the actual marks and remarks to help them address their misconceptions and prepare them for an upcoming exam. This study presents our findings from clickstream data of students taking a synchronous online Computer Informatics class. How different students approached the activity was looked into: the amount of time spent and the difference of their assigned grade to that of a subject expert's. Although it is still inconclusive whether students learned from erroneous computer programs, we found that students who were proactive in seeking feedback had better midterm scores than those who were not. This underscores the importance of feedback in this learning process.

The uniqueness of Augmented Reality (AR) is its affordance to support learning abstract concepts by rich information, visualization and the integration of user-content interaction. Research has shown that abstract idea and invisible phenomena can be learned better with the support of AR. However, high complexity and conceptual Computational Thinking (CT), such as algorithm design and related programming concepts, are rarely studied empirically in the intersection with immersive technology. This study is aimed at addressing this issue and providing a piece of quantitative and empirical evidence to AR-supported Computer Science discipline-based learning. We designed a mobile AR-enabled application based on a CT framework and related AR affordances in the literature. This app can contextualize a programming debugging task and support program editing & CT learning. A controlled laboratory study was designed and conducted. The result of statistical analyses shows that participants with the AR support made better quality of programs with lower errors and less amount of code edits, compared to those without the AR support.

A student's short-term study behavior may not necessary infer his/her long-term behavior. It is very common to see a student changes study strategy throughout a semester and adapts to learning condition. For example, a student may work very hard before the first exam but gradually reducing the effort due to several possible reasons, e.g., being overwhelmed by various course work or discouraged by increasing complexity in the subject. Consistency or differences of one student's behavior is more likely to be discovered by multiple granularity of learning analytics. In this study, we investigate students' study persistence on a self-assessment platform and explore how such a behavioral pattern is related to the performance in exams. A probabilistic mixture model trained by response streams of log data is applied to cluster students' behavior into persistence patterns, which are further categorized into "micro" (short-term) and "macro" (long-term) patterns according to the span of time being modeled. We found four types of micro persistence patterns and several macro patterns in the analysis and analyzed their relations with exam performances. The result suggests that the consistency of persistence patterns can be an important factor driving student's overall performance in the semester, and students achieving higher exam scores show relatively persistent behavior compared to students receiving lower scores.

This Research Full Paper aims to investigate the learning analytics of students' problem solving when working on distributed programming practices. Typical programming practice activities (i.e. assignments) in a lecture-dominant course may violate the principles of distributed retrieval practice. However, there are tradeoffs between managing depth and breadth of the content and classroom disruptions with the modern platforms and technologies. In this work, we investigate students' behavioral analytics in distributed programming practices. A classroom study was conducted in an introductory programming course and the learners' patterns were observed. Results showed that there were three distinct patterns found: affirmative, experimental, and surrendering. Better-performing students demonstrated more affirmative behaviors and fewer surrendering acts; Below-average students showed a lack of persistence in distributed practices. Additionally, the study reconfirmed the value of spacing effects on learning, which is the importance of spending time and to spreading the working sessions to solve diverse quizzes. Ineffective trial-and-error strategy and neglect the power of practices can be two alarming behaviors in distributed programming practices. Finally, predictive models of performance were presented based on the behavioral patterns.

This Research Full Paper investigates potential quantitative self-regulated learning (SRL) analytics of problem-solving behaviors on a multiple-choice question (MCQ) platform. Conventionally, SRL skills are assessed by qualitative and self-report instruments which can provide insightful details of SRL behaviors. Nevertheless, such kind of tools might be under criticism of reproducibility and lack scalability. Additionally, it is difficult to provide real-time feedback based on such measurements since they require active participation and longer data collection processes. These factors thereby deter educational practitioners who wish to utilize SRL-based pedagogy in larger classrooms. To make SRL research more accessible to educational practitioners, we propose to use quantitative analysis based on trace data to measure potential SRL behaviors from two aspects: study persistence and study regularity, which we assume are implicit measurement of SRL. We conducted the experiment in an undergraduate computer science course by a homegrown self-assessment platform, QuizIt and successfully identified four types of study persistence. The study regularity is measured as scores and associated in a linear regression model. The evaluation based on students' formal exam scores in the course shows that high persistence and regular behaviors are positively correlated to exam performance. Although the examination of linkage from persistence and regularity to conventional self-report SRL measures is left for the future study, our approach has implied a potential for scalable educational tools that can avoid subjectivity of self-report SRL measurement.

In this work, we propose a new methodology to profile individual students of computer science based on their programming design using a technique called embeddings. We investigate different approaches to analyze user source code submissions in the Python language. We compare the performances of different source code vectorization techniques to predict the correctness of a code submission. In addition, we propose a new mechanism to represent students based on their code submissions for a given set of laboratory tasks on a particular course. This way, we can make deeper recommendations for programming solutions and pathways to support student learning and progression in computer programming modules effectively at a Higher Education Institution. Recent work using Deep Learning tends to work better when more and more data is provided. However, in Learning Analytics, the number of students in a course is an unavoidable limit. Thus we cannot simply generate more data as is done in other domains such as FinTech or Social Network Analysis. Our findings indicate there is a need to learn and develop better mechanisms to extract and learn effective data features from students so as to analyze the students' progression and performance effectively.

There are numerous studies have reported the effectiveness of example-based programming learning. However, less is explored recommending code examples with advanced Machine Learning-based models. In this work, we propose a new method to explore the semantic analytics between programming codes and the annotations. We hypothesize that these semantics analytics will capture mass amount of valuable information that can be used as features to build predictive models. We evaluated the proposed semantic analytics extraction method with multiple deep learning algorithms. Results showed that deep learning models outperformed other models and baseline in most cases. Further analysis indicated that in special cases, the proposed method outperformed deep learning models by restricting false-positive classifications.

This Research Work in Progress Paper presents our ongoing work to encourage student to reflect while participating in a distributed self-assessment process. We shed light on the case of low reflect participation and the change in process we made that led to a significant increase in both the quantity and quality of students reflects. To evaluate the effect of the intervention, we conducted two classroom study for an introductory Java course with over 500 students. From the latest enhancements in the current study, we observed that the overall reflecting engagement from the students has significantly increased compared to our initial work. To evaluate the reflects quality, we deployed an automated measure that classifies the constructiveness of the comment based on its relevant and subjectivity to the domain. This work describes the details of the interventions along with the resulted observations. The main objective was to encourage learners to provide thoughtful reflects which we were able to increase along with the positive effect on their performance.

This Research Full Paper implements a framework that harness sources of programming learning analytics on three computer programming courses a Higher Education Institution. The platform, called PredictCS, automatically detects lower-performing or “at-risk” students in programming courses and automatically and adaptively sends them feedback. This system has been progressively adopted at the classroom level to improve personalized learning. A visual analytics dashboard is developed and accessible to Faculty. This contains information about the models deployed and insights extracted from student's data. By leveraging historical student data we built predictive models using student characteristics, prior academic history, logged interactions between students and online resources, and students' progress in programming laboratory work. Predictions were generated every week during the semester's classes. In addition, during the second half of the semester, students who opted-in received pseudo real-time personalised feedback. Notifications were personalised based on students' predicted performance on the course and included a programming suggestion from a top-student in the class if any programs submitted had failed to meet the specified criteria. As a result, this helped students who corrected their programs to learn more and reduced the gap between lower and higher-performing students.

This Research Work in Progress paper presents a study on improving student learning performance in a virtual hands-on lab system in cybersecurity education. As the demand for cybersecurity-trained professionals rapidly increasing, virtual hands-on lab systems have been introduced into cybersecurity education as a tool to enhance students' learning. To improve learning in a virtual hands-on lab system, instructors need to understand: what learning activities are associated with students' learning performance in this system? What relationship exists between different learning activities? What instructors can do to improve learning outcomes in this system? However, few of these questions has been studied for using virtual hands-on lab in cybersecurity education. In this research, we present our recent findings by identifying that two learning activities are positively associated with students' learning performance. Notably, the learning activity of reading lab materials (p <; 0:01) plays a more significant role in hands-on learning than the learning activity of working on lab tasks (p <; 0:05) in cybersecurity education.In addition, a student, who spends longer time on reading lab materials, may work longer time on lab tasks (p <; 0:01).

Engaged students are more likely to spend longer time on study, and obtain a better academic performance. Previous studies investigated the role of student engagement in virtual learning environments (e.g., online course, online discussion forum, and intelligent tutoring systems). However, it is still challenging to engage students on a virtual hands-on lab system. Comparing to other virtual learning environment, students have a unique learning model -- learning by doing in virtual hands-on lab. To successfully engage students in a large hands-on lab in cybersecurity education, instructors need to understand how students engage in a lab session, and how their engagement affect lab learning outcome in this specific educational setting. In this paper, we developed a conceptual model, especially for virtual hands-on lab education, to describe student engagement during learning processes in working on virtual hands-on lab tasks. This model adopts two existing educational models on engagement behavior. Preliminary data was collected from 109 students' lab project in a computer network security course at Arizona State University in 2016 Fall semester. Pearson correlation coefficient analysis results reveal two statistically significant preliminary results: the longer time a student spends on reading lab instructional material, the more likely the student works longer time on lab tasks (p < 0.01); the longer time a student works on lab tasks, a better learning performance the student archives (p < 0.01).

Programming novices usually find acquiring the ability to write programs challenging at first, however, they overcome this obstacle as they encounter more opportunities in the learning process. Providing learners with distributed practices and the ability to self-assess their programming knowledge is key to measure their development and guide them towards programming proficiency. In this work, we introduced QuizIT, a programming learning tool designed for novices. We conducted a classroom study and collected a semester long data to measure the effectives of the tool to achieve the design objectives. We analyzed the study data and provided the preliminary results from statistical perspective, as well as evaluating the effectiveness of the tool from learners' outcome. The data showed the positive effect of learners' usage of the tool on their course performance. We reported correlations exists in the data between effort (by actively benefiting and reflecting to the small learning opportunities) and the course outcome.

The usage of online search engine is growing rapidly not only in daily life, but also in education. We are interested in understanding what strategies students apply during search, especially the tactics they use to decompose a programming task. In this paper, we report a lab study to investigate students' programming information seeking behavior via Google search engine. Students were given a programming task with limited time, and they were also required to report their online search process including search query and web pages browsed. We analyze the web pages they browsed, model student's behavior, and cluster them into groups with different search tactics. The results show that the web pages they browsed during the task consisted of either conceptual knowledge or coding technical content. The students who performed better would browsed more about conceptual knowledge. Students who set more and smaller unit of subgoals outperformed the students with fewer and larger subgoals.

In this paper, we study students' learning effectiveness through their use of a homegrown educational technology, Web Programming Grading Assistant (WPGA), which facilitates grading and feedback delivery of paper-based assessments. We designed a classroom study and collected data from a lower-division blended-instruction computer science class. We tracked and modeled students' reviewing and reflecting behaviors from WPGA. The results show that students demonstrated an effort and desire to review assessments regardless of if they were graded for academic performance or for attendance. Hardworking students achieved higher exam scores on average and were found to review their exams and the correct questions frequently. Additionally, student cohorts exhibited similar initial reviewing patterns, but different in-depth reviewing and reflecting strategies. Ultimately, this work contributes to the aggregation of multidimensional learning analytics across the physical and cybersphere.

We design and study ExamParser, an innovative intelligent semantic automatic indexing method, for orchestrating today's programming classes. ExamParser automatically processes paper-based exams by associating sets of concepts to the exam questions, which provide graders semantic grading guidelines and leave personalized semantic feedback. Results showed that the ExamPraser significantly extract more and diverse concepts from exams. It also achieves high coherence within exam, indicating the automatic concept extraction from exams is promising and could be a potential technological solution to provide personalized feedback for large-size programming classes.

We survey across generations of work in authorized cheat sheets, and emphasize on extracting generic features for designing one. We construct generic models to address cognitive aspects of learning and focus on the content of knowledge components in the domain of programming language learning. We evaluate the models with real classroom study data and found that the amount of notes on a cheat sheet is associated with gains in students' declarative knowledge, but not procedural knowledge. We also discovered that working in creating new cheat sheets can help students to achieve better exam grades accumulatively beyond and above the effect of their performance in a prior exam. We discuss how these findings can guide the design of technology-based study guides.

In this work, we studied how programming novices explore and search on online discussion forums. We collected novices' intentions and search logs. The results indicated that novices had significant different behavior patterns compare to experienced learners, and they indeed require specific assistant in generating search queries, and filtering helpful information among the search result. This work also suggests directions to assist future learners in seeking programming information in online environment.

We designed and studied an innovative semantic visual learning analytics for orchestrating today's programming classes. The visual analytics integrates sources of learning activities by their content semantics. It automatically processs paper-based exams by associating sets of concepts to the exam questions. Results indicated the automatic concept extraction from exams were promising and could be a potential technological solution to address a real world issue. We also discovered that indexing effectiveness was especially prevalent for complex content by covering more comprehensive semantics. Subjective evaluation revealed that the dynamic concept indexing provided teachers with immediate feedback on producing more balanced exams.

In this paper, we modelled constructive engagement activities in an online programming discussion. We built a logistic regression model based on the underlined cognitive processes in constructive learning activities. The findings supported that there is passive-proactive behaviour and suggested that detecting constructive content can be a helpful classifier in discerning relevant information to the users and in turn creating opportunities to optimize learning. The results also confirmed the value of discussion forum content, disregarding the crowd approves or not.

Online programming discussion forums are popular trouble-shooting and problem-solving sites for programmers and learners to reach out for help. The massive volumes of forum threads harbor tremendous amounts of information, but at the same time increase the complexity of search and navigation. In this work, we make use of programming discussions' syntactic, semantic and social features to model content associated with learning activities based on the ICAP learning framework. Our main goal is to detect useful content for learning programming in a large scale of questions and answers, while at the same time experiment with an artificial intelligence approach to detect learning-inductive content. We build regression models based on the defined constructive learning activities. Results reveal a passive-proactive learning behavior in an online programming discussion forum. The findings also reconfirm the value of programming discussion content, disregarding the crowds' approval. The automatic detection of constructive learning activities from programming discussions can be a helpful classifier in identifying relevant educational resources. Overall, this project contributes to our understanding on analyzing and utilizing mass programming discussion content for online programming language learning.

In this paper, we investigated constructive engagement activities in an online programming discussion forum. We engineered features according to the forum content and built a logistic regression model based on the underlined cognitive processes in constructive learning activities. We hypothesized that performing any engaging activities after reading constructive content contributes to learning. This work aimed to discover such behavior in a free-formed online discussion setting and assess the impacts. Evaluation results revealed and reconfirmed the value of content constructiveness in the discussion forum. Our findings also suggested that detecting constructive content could be a helpful classifier in discerning relevant information to the users and in turn creating opportunities to optimize learning.

In this paper, we propose a novel Topic Facet Model (TFM), a probabilistic topic model that assumes all words in single sentence are generated from one topic facet. The model is applied to automatically extract forum posts semantics for uncovering the content latent structures. We further prototype a visual analytics interface to present online discussion forum semantics. We hypothesize that the semantic modeling through analytics on open online discussion forums can help users examine the post content by viewing the summarized topic facets. Our preliminary results demonstrated that TFM can be a promising method to extract topic specificity from conversational and relatively short texts in online programming discussion forums.

With the explosion of online learning technologies and increasing demands to make learning more accessible and effective for all, personalized and social learning has been a suggested method of addressing these challenges. We present mSchool, an open, scalable educational learning platform which addresses the challenges of learner & community centered open learning space. We study mSchool's effectiveness on knowledge discovery and management. The preliminary results suggest the straightforward design resulted in effective navigational support for learners.

Quantitative methods in education research have long been limited by the ability to collect detailed learner data in a consistent, scalable way. As education continues to move online we are presented with an unprecedented opportunity to study learner interactions within learning systems. However, doing so requires infrastructure to collect and store massive interaction data from which we can learn. In this paper we present Data Depository, a flexible, pluggable, data hub for tracking interaction data from any browser-based application, aiding the measurement of usage and effectiveness.

Social learning has confirmed its value in enhancing the learning outcomes across a wide spectrum. To support social learning, a visual approach is a common technique to represent and organize multiple students' data in an informative way. This paper presents a design of comparative social visualization for E-learning, which encourages information discovery and social comparisons. Classroom studies confirmed the motivational impact of personalized social guidance provided by the visualization in the target context. The visualization encouraged students to do some work ahead of the course schedule. Moreover, class leaders provided an implicit social guidance for the rest of the class and successfully led the way to discover the most relevant resources creating good trails for the rest of the class. We summarized the evidence of students' engagement and performance through the social visualization interface.

In this paper, we present an asynchronous video discussion tool called 'Vialogues' and empirical evidence of system usage for supporting modern social learning environment. Statistical analysis showed that Vialogues not only enriches online learning experiences but also successfully retains users and creates social learning opportunities. Evidence of learning taking place in a new technology, called Vialogues, is reported.

Over the last 10 years two major research directions explored the benefits of visualizing student learning progress. One stream of research on learning performance visualization attempts to build a visual presentation of students' learning progress, targeting the needs of instructors and academic advisors. The other stream of research on Open Student Modeling (OSM) attempts to visualize the state of individual student's knowledge and present the visualization directly to the student. The results of the studies in that area show that, presenting students with basic representation of their knowledge will result in facilitating their metacognitive activities and promoting self-reflection and awareness. This paper tries to study the impact of a more sophisticated form of performance visualization on students. We believe that our visualization tool can positively influence students by granting them the opportunity to get a view of their performance in the content of the class progress. Moreover, we tried to boost their motivation by building a positive sense of competition using a representation of average class performance. In this paper we present study comparing two groups of students, one using the visualization and another without visualization. The results of the study shows that: 1) the students are likely to use the social visualization tool during the whole semester to monitor their progress in comparison with their peers; 2) the visualization tool encourages students to use the learning materials in a more continuous manner during the whole semester and 3) students will achieve a higher success rate in answering self-assessment quizzes.

This paper presents Progressor, a visualization of open student models intended to increase the student's motivation to progress on educational content. The system visualizes not only the user's own model, but also the peers' models. It allows sorting the peers' models using a number of criteria, including the overall progress and the progress on a specific topic. Also, in this paper we present results of a classroom study confirming our hypothesis that by showing a student the peers' models and ranking them by progress it is possible to increase the student's motivation to compete and progress in e-learning systems.

Problem-tracing questions are popular among teachers of various programming languages. In an assessment mode these questions allows to evaluate student knowledge of language semantics. In a self-assessment mode, they provide an excellent learning tool. A 2004 ITiCSE working group report [4] stressed the importance of this type of questions to build foundation of higher-level knowledge. Yet the use of problem-tracing questions is still limited due to a large authoring overhead. To resolve this bottleneck, we explored the idea of parameterized question generation [2]. We developed QuizPACK [1], a system which can generate parameterized problem-tracing questions for C programming language. We also developed QuizGuide [1], a personalized guidance system for QuizPACK, which models student knowledge and guides students individually to most appropriate questions to try. The results of our studies demonstrated that QuizPACK strongly benefits student knowledge and that QuizGuide personalized guidance technology increased student ability to answer questions correctly and encouraged them to use the system more extensively (which, in turn, positively impacted their knowledge) [1]. However, parameterized questions in area of C programming were not as diverse from the complexity point of view as parameterized questions explored in other areas such as physics [2]. As a result, it was left unclear whether personalized guidance technology can successfully guide students to a broader range of questions from relatively simple to very difficult.The work reported in this poster expands our work on parameterized questions to a more sophisticated domain of object-oriented Java programming, which allowed us to introduce questions of much broader. Capitalizing on our experience with QuizPACK, we developed QuizJET (Java Evaluation Toolkit), which supports authoring, delivery, and evaluation of parameterized questions for Java [3]. We also implemented JavaGuide system (Figure 1), which provides personalized guidance for QuizJET questions. We assessed the impact of adaptive navigation support to student work with questions of different complexity as well as the impact of this technology on weaker and stronger students. The results of two classroom studies indicate that personalized guidance encouraged students to use parameterized questions more extensively and also helped them to access right questions at the right time. Students were 2.5 times more likely to answer a quiz correctly with personalized guidance than without such it. In addition, we found that personalized guidance especially benefited weak students to achieve scores comparable with the scores of strong students on each complexity level of questions.