As Artificial Intelligence (AI) becomes increasingly integrated into daily life, there is a growing need to equip the next generation with the ability to apply, interact with, evaluate, and collaborate with AI systems responsibly. Prior research highlights the urgent demand from K-12 educators to teach students the ethical and effective use of AI for learning. To address this need, we designed a Large-Language Model (LLM)-based module to teach prompting literacy. This includes scenario-based deliberate practice activities with direct interaction with intelligent LLM agents, aiming to foster secondary school students' responsible engagement with AI chatbots. We conducted two iterations of classroom deployment in 11 authentic secondary education classrooms, and evaluated 1) AI-based auto-grader's capability; 2) students' prompting performance and confidence changes towards using AI for learning; and 3) the quality of learning and assessment materials. Results indicated that the AI-based auto-grader could grade student-written prompts with satisfactory quality. In addition, the instructional materials supported students in improving their prompting skills through practice and led to positive shifts in their perceptions of using AI for learning. Furthermore, data from Study 1 informed assessment revisions in Study 2. Analyses of item difficulty and discrimination in Study 2 showed that True/False and open-ended questions could measure prompting literacy more effectively than multiple-choice questions for our target learners. These promising outcomes highlight the potential for broader deployment and highlight the need for broader studies to assess learning effectiveness and assessment design.

LLMs promise to democratize technical work in complex domains like programmatic data analysis, but not everyone benefits equally. We study how students with varied experiences use LLMs to complete Python-based data analysis in computational notebooks in a graduate course. Drawing on homework logs, recordings, and surveys from 36 students, we ask: Which experience matters most, and how does it shape AI use? Our mixed-methods analysis shows that technical experience – not AI familiarity or communication skills – remains a significant predictor of success. Students also vary widely in how they leverage LLMs, struggling at stages of forming intent, expressing inputs, interpreting outputs, and assessing results. We identify success and failure behaviors, such as providing context or decomposing prompts, that distinguish effective use. These findings inform AI literacy interventions, highlighting that lightweight demonstrations improve surface fluency but are insufficient; deeper training and scaffolds are needed to cultivate resilient AI use skills.

Learning analytics can guide human tutors to efficiently address motivational barriers to learning that AI systems struggle to support. Students become more engaged when they receive human attention. However, what occurs during short interventions, and when are they most effective? We align student–tutor dialogue transcripts with MATHia tutoring system log data to study brief human-tutor interactions on Zoom drawn from 2,075 hours of 191 middle school students’ classroom math practice. Mixed-effect models reveal that engagement, measured as successful solution steps per minute, is higher during a human-tutor visit and remains elevated afterward. Visit length exhibits diminishing returns: engagement rises during and shortly after visits, irrespective of visit length. Timing also matters: later visits yield larger immediate lifts than earlier ones, though an early visit remains important to counteract engagement decline. We create analytics that identify which tutor-student dialogues raise engagement the most. Qualitative analysis reveals that interactions with concrete, stepwise scaffolding with explicit work organization elevate engagement most strongly. We discuss implications for resource-constrained tutoring, prioritizing several brief, well-timed check-ins by a human tutor while ensuring at least one early contact. Our analytics can guide the prioritization of students for support and surface effective tutor moves in real-time.

We present an empirical study examining how experienced tutors (experts) and non-tutors (novices) evaluate the correctness of tutor praise responses under different AI-assisted decision-support interfaces and explanation styles. We examine human-AI reliance patterns by decomposing interaction errors into over-reliance (accepting incorrect AI suggestions) and under-reliance (rejecting correct AI suggestions), together with time cost as a process-level indicator. Across conditions, human-AI collaboration improved accuracy compared to humans working alone, but consistently underperformed an AI-only baseline, indicating that human judgment introduced additional errors even when assisted by a highly accurate model. Novices benefited more from AI support since they tend to follow AI suggestions, whereas experts frequently overrode correct AI advice, resulting in lower overall performance, revealing a paradox of expertise in educational decision-making. We further compare two explanation modalities: textual reasoning and inline highlighting. Textual reasoning reduced under-reliance when the AI was correct but increased over-reliance when the AI was wrong, while inline highlighting exerted minimal influence on either behavior. Notably, neither explanation modality improved accuracy, and both increased time costs. As a contribution to learning analytics, we demonstrate how reliance patterns (over-reliance and under-reliance) and time cost function as process-level indicators that reveal how users integrate, or fail to integrate, AI recommendations. Our findings underscore the need for adaptive, trust-calibrated explanation strategies in tutor-facing decision support systems that balance accuracy, efficiency, and accountability in human-AI collaboration.

Providing timely, targeted, and multimodal feedback helps students quickly correct errors, build deep understanding and stay motivated, yet making it at scale remains a challenge. This study introduces a real-time AI-facilitated multimodal feedback system that integrates structured textual explanations with dynamic multimedia resources, including the retrieved most relevant slide page references and streaming AI audio narration. In an online crowdsourcing experiment, we compared this system against fixed business-as-usual feedback by educators across three dimensions: (1) learning effectiveness, (2) learner engagement, (3) perceived feedback quality and value. Results showed that AI multimodal feedback achieved learning gains equivalent to original educator feedback while significantly outperforming it on perceived clarity, specificity, conciseness, motivation, satisfaction, and reducing cognitive load, with comparable correctness, trust, and acceptance. Process logs revealed distinct engagement patterns: for multiple-choice questions, educator feedback encouraged more submissions; for open-ended questions, AI-facilitated targeted suggestions lowered revision barriers and promoted iterative improvement. These findings highlight the potential of AI multimodal feedback to provide scalable, real-time, and context-aware support that both reduces instructor workload and enhances student experience.

Measures of disengagement provide insights into unproductive use of learning opportunities. Although measures of active disengagement, such as gaming the system and mind-wandering, are well studied, loss of practice time due to outright task avoidance remains relatively understudied. The current study addresses this gap by extending existing within-task measures (idle time) with two new session-level measures (delayed start and early stop) to capture loss of practice time due to task avoidance. We characterize the combined lost time as coasted time and the associated behavior as coasting behavior. Using ASSISTments logs (N = 1,425), we find that students dedicate only 40% of available classwork time to math practice and coast through the remaining 60%. Of the coasted time, 36% resulted from delayed starts, 2% from mid-practice idling, and 62% from stopping early. Delayed start and early stop showed moderate temporal stability (G = 0.73 and 0.71, respectively), suggesting that coasting is a consistent behavioral pattern. Even after excluding early stops attributable to assignment completion (i.e., early stop = 0), coasted time remained substantial at 32%. While we observe significant differences in coasting by gender and IEP status, we do not observe them by other demographic factors or school locale. Critically, students who continued working beyond the first assignment completion (''extra effort'') performed significantly better on standardized tests. For research, coasting offers a new lens on opportunity loss by combining session-level disengagement with within-task disengagement. For practitioners, our results highlight the need for platform affordances that support sustained engagement and more productive use of available practice time.

Large Language Models (LLMs) excel at generating content at impeccable speeds. However, they are imperfect and still make various mistakes. In Computer Science education, as LLMs are widely recognized as "AI pair programmers," it becomes increasingly important to train students on evaluating and debugging LLM-generated codes. In this work, we introduce HypoCompass, a novel system to facilitate deliberate practice on debugging, where human novices play the role of Teaching Assistants and help LLM-powered teachable agents debug code. We enable effective task delegation between students and LLMs in this learning-by-teaching environment: students focus on hypothesizing the cause of code errors, while adjacent skills like code completion are offloaded to LLM-agents. Our evaluations demonstrate that HypoCompass generates high-quality training materials (e.g., bugs and fixes), outperforming human counterparts fourfold in efficiency, and significantly improves student performance on debugging by 12% in the pre-to-post test.

The role of multiple-choice questions (MCQs) as effective learning tools has been debated in past research. While MCQs are widely used due to their ease in grading, open response questions are increasingly used for instruction, given advances in large language models (LLMs) for automated grading. This study evaluates MCQs effectiveness relative to open-response questions, both individually and in combination, on learning. These activities are embedded within six tutor lessons on advocacy. Using a posttest-only randomized control design, we compare the performance of 234 tutors (790 lesson completions) across three conditions: MCQ only, open response only, and a combination of both. We find no significant learning differences across conditions at posttest, but tutors in the MCQ condition took significantly less time to complete instruction. These findings suggest that MCQs are as effective, and more efficient, than open response tasks for learning when practice time is limited. To further enhance efficiency, we autograded open responses using GPT-4o and GPT-4-turbo. GPT models demonstrate proficiency for purposes of low-stakes assessment, though further research is needed for broader use. This study contributes a dataset of lesson log data, human annotation rubrics, and LLM prompts to promote transparency and reproducibility.

Equity is a core concern of learning analytics. However, applications that teach and assess equity skills, particularly at scale are lacking, often due to barriers in evaluating language. Advances in generative AI via large language models (LLMs) are being used in a wide range of applications, with this present work assessing its use in the equity domain. We evaluate tutor performance within an online lesson on enhancing tutors' skills when responding to students in potentially inequitable situations. We apply a mixed-method approach to analyze the performance of 81 undergraduate remote tutors. We find marginally significant learning gains with increases in tutors' self-reported confidence in their knowledge in responding to middle school students experiencing possible inequities from pretest to posttest. Both GPT-4o and GPT-4-turbo demonstrate proficiency in assessing tutors ability to predict and explain the best approach. Balancing performance, efficiency, and cost, we determine that few-shot learning using GPT-4o is the preferred model. This work makes available a dataset of lesson log data, tutor responses, rubrics for human annotation, and generative AI prompts. Future work involves leveling the difficulty among scenarios and enhancing LLM prompts for large-scale grading and assessment.

published_or_final_version

L@S ’25, Palermo, Italy

Prompting LLMs for complex tasks (e.g., building a trip advisor chatbot) needs humans to clearly articulate customized requirements (e.g., “start the response with a tl;dr”). However, existing prompt engineering instructions often lack focused training on requirement articulation and instead tend to emphasize increasingly automatable strategies (e.g., tricks like adding role-plays and “think step-by-step”). To address the gap, we introduce Requirement-Oriented Prompt Engineering ( ROPE ), a paradigm that focuses human attention on generating clear, complete requirements during prompting. We implement ROPE through an assessment and training suite that provides deliberate practice with LLM-generated feedback. In a randomized controlled experiment with 30 novices, ROPE significantly outperforms conventional prompt engineering training (20% vs. 1% gains), a gap that automatic prompt optimization cannot close. Furthermore, we demonstrate a direct correlation between the quality of input requirements and LLM outputs. Our work paves the way to empower more end-users to build complex LLM applications.

Artificial intelligence (AI) applications to support human tutoring have potential to significantly improve learning outcomes, but engagement issues persist, especially among students from low-income backgrounds. We introduce an AI-assisted tutoring model that combines human and AI tutoring and hypothesize this synergy will have positive impacts on learning processes. To investigate this hypothesis, we conduct a three-study quasi-experiment across three urban and low-income middle schools: 1) 125 students in a Pennsylvania school; 2) 385 students (50% Latinx) in a California school, and 3) 75 students (100% Black) in a Pennsylvania charter school, all implementing analogous tutoring models. We compare learning analytics of students engaged in human-AI tutoring compared to students using math software only. We find human-AI tutoring has positive effects, particularly in student’s proficiency and usage, with evidence suggesting lower achieving students may benefit more compared to higher achieving students. We illustrate the use of quasi-experimental methods adapted to the particulars of different schools and data-availability contexts so as to achieve the rapid data-driven iteration needed to guide an inspired creation into effective innovation. Future work focuses on improving the tutor dashboard and optimizing tutor-student ratios, while maintaining annual costs per student of approximately $700 annually.

We developed GPTutor, a pioneering web application designed to revolutionize personalized learning by leveraging the capabilities of Generative AI at scale. GPTutor adapts educational content and practice exercises to align with individual students' interests and career goals, enhancing their engagement and understanding of critical academic concepts. The system uses a serverless architecture to deliver personalized and scalable learning experiences. By integrating advanced Chain-of-Thoughts prompting methods, GPTutor provides a personalized educational journey that not only addresses the unique interests of each student but also prepares them for future professional success. This demo paper presents the design, functionality, and potential of GPTutor to foster a more engaging and effective educational environment.

Written feedback has long been a cornerstone in educational and professional settings, essential for enhancing learning outcomes. However, multimodal feedback-integrating textual, auditory, and visual cues-promises a more engaging and effective learning experience. By leveraging multiple sensory channels, multimodal feedback better accommodates diverse learning preferences and aids in deeper information retention. Despite its potential, creating multimodal feedback poses challenges, including the need for increased time and resources. Recent advancements in generative artificial intelligence (GenAI) offer solutions to automate the feedback process, predominantly focusing on textual feedback. Yet, the application of GenAI in generating multimodal feedback remains largely unexplored. Our study investigates the use of GenAI techniques to generate multimodal feedback, aiming to provide this feedback for large cohorts of learners, thereby enhancing learning experience and engagement. By exploring the potential of GenAI for this purpose, we propose a framework for automating the generation of multimodal feedback, which we name MuFIN.

Automated feedback systems are pivotal for scaling personalized learning, especially when dealing with large cohorts of learners.This paper introduces HAROR (Highlighting and Rephrasing Openended Responses), a feedback system that utilizes the advanced capabilities of Generative Pre-trained Transformer (GPT) models, including GPT-4 and GPT-3.5, to provide explanatory feedback on learner responses (trainee tutors as learners in our study) to openended questions.HAROR can identify desirable and undesirable parts of open-ended responses, offer explanatory feedback, and rephrase the undesired responses into desirable forms, aiming to foster learners' understanding and improvement.

Visual thinking with diagrams is a crucial skill for learning and problem-solving in STEM subjects. To improve in this area, students need a variety of visual problems for deliberate practice. However, in our interviews, educators shared that they struggle to create these practice exercises because of limitations of existing tools. We introduce Edgeworth, a tool designed to help educators easily create visual problems. Edgeworth works in two main ways: firstly, it takes a single diagram from the user and systematically alters it to produce many variations, which the educator can then choose from to create multiple problems. Secondly, it automates the layout of diagrams, ensuring consistent high quality without the need for manual adjustments. To assess Edgeworth, we carried out case studies, a technical evaluation, and expert walkthrough demonstrations. We show that Edgeworth can create problems in three domains: geometry, chemistry, and discrete math. These problems were authored in just 15 lines of Edgeworth code on average. Edgeworth generated usable answer options within the first 10 diagram variations in 87% of authored problems. Finally, educators gave positive feedback on Edgeworth's utility and the real-world applicability of its outputs.

Addressing negative self-talk by students, such as responding to a student when saying, "I am dumb"or "I can't do this"can be difficult for even the most experienced tutor. Despite potential tutor learning from scenario-based lessons on this topic, human-graded assessment remains time-consuming. Leveraging generative AI for evaluating textual responses in online training presents a scalable solution. Research suggests a tutor validates student's feelings when they speak negatively of themselves, e.g., by a tutor responding, "I understand how you feel"or "I recognize this is difficult."This ongoing work assesses the performance of 60 undergraduate tutors within an online lesson on enhancing tutors' abilities to respond to students engaging in negative self-talk. We find statistically significant tutor learning gains from pretest to posttest. Additionally, we describe a method of using generative AI for assessing tutors' responses to predict the best approach and subsequently explain the rationale behind it. Using the large language model GPT-4, we find high absolute performance when evaluating tutor responses involving predicting (F1 = 0.85) and explaining (F1 = 0.83) the best approach. Minor improvements are needed to the lesson itself. A future goal of this work is to fully develop automated systems of assessing tutor learning attending to barriers to students' motivation and doing so at scale.

This study examines the effects of question type and feedback on learning outcomes in a hybrid graduate-level course. By analyzing data from 32 students over 30,198 interactions, we assess the efficacy of unique versus repeated questions and the impact of feedback on student learning. The findings reveal students demonstrate significantly better knowledge generalization when encountering unique questions compared to repeated ones, even though they perform better with repeated opportunities. Moreover, we find that the timing of explanatory feedback is a more robust predictor of learning outcomes than the practice opportunities themselves. These insights suggest that educational practices and technological platforms should prioritize a variety of questions to enhance the learning process. The study also highlights the critical role of feedback; opportunities preceding feedback are less effective in enhancing learning.

Many SIGCSE attendees are either developing or using online educational tools, and all will benefit from better interoperability among these tools and better analysis of the clickstream data coming from those tools. New tools for analyzing big data leveraged by AI (e.g., deep learning for assessment) in turn improve both content and pedagogy, thus setting up a virtuous cycle fueling learning discoveries and leveraging innovation in AI: Online technologies → big data analysis → better online technologies. This NSF-supported workshop is the latest in a series of SPLICE workshops, and is a continuation of our event at SIGCSE 2023, where the SPLICE-Portal, a dedicated socio-technical research infrastructure for Computing Education Research, was presented. This year, we continue the work with several new SPLICE community working groups, including those on Dashboards, Large Language Models, Parsons Problems, and Smart Learning Content Protocols. We continue to build upon our existing collaborations developed over the course of the project to engage more members of the community in tasks that will advance the project agenda.

Abstract In numerous studies, intelligent tutoring systems (ITSs) have proven effective in helping students learn mathematics. Prior work posits that their effectiveness derives from efficiently providing eventually-correct practice opportunities. Yet, there is little empirical evidence on how learning processes with ITSs compare to other forms of instruction. The current study compares problem-solving with an ITS versus solving the same problems on paper. We analyze the learning process and pre-post gain data from N = 97 middle school students practicing linear graphs in three curricular units. We find that (i) working with the ITS, students had more than twice the number of eventually-correct practice opportunities than when working on paper and (ii) omission errors on paper were associated with lower learning gains. Yet, contrary to our hypothesis, tutor practice did not yield greater learning gains, with tutor and paper comparing differently across curricular units. These findings align with tutoring allowing students to grapple with challenging steps through tutor assistance but not with eventually-correct opportunities driving learning gains. Gaming-the-system, lack of transfer to an unfamiliar test format, potentially ineffective tutor design, and learning affordances of paper can help explain this gap. This study provides first-of-its-kind quantitative evidence that ITSs yield more learning opportunities than equivalent paper-and-pencil practice and reveals that the relation between opportunities and learning gains emerges only when the instruction is effective.

Tutoring is among the most impactful educational influences on student achievement, with perhaps the greatest promise of combating student learning loss. Due to its high impact, organizations are rapidly developing tutoring programs and discovering a common problem- a shortage of qualified, experienced tutors. This mixed methods investigation focuses on the impact of short (∼15 min.), online lessons in which tutors participate in situational judgment tests based on everyday tutoring scenarios. We developed three lessons on strategies for supporting student self-efficacy and motivation and tested them with 80 tutors from a national, online tutoring organization. Using a mixed-effects logistic regression model, we found a statistically significant learning effect indicating tutors performed about 20% higher post-instruction than pre-instruction (β = 0.811, p < 0.01). Tutors scored ∼30% better on selected compared to constructed responses at posttest with evidence that tutors are learning from selected-response questions alone. Learning analytics and qualitative feedback suggest future design modifications for larger scale deployment, such as creating more authentically challenging selected-response options, capturing common misconceptions using learnersourced data, and varying modalities of scenario delivery with the aim of maintaining learning gains while reducing time and effort for tutor participants and trainers.

Gaming the system, a behavior in which learners exploit a system's properties to make progress while avoiding learning, has frequently been shown to be associated with lower learning. However, when we applied a previously validated gaming detector across conditions in experiments with an algebra tutor, the detected gaming was not associated with reduced learning, challenging its validity in our study context. Our exploratory data analysis suggested that varying contextual factors across and within conditions contributed to this lack of association. We present a new approach, latent variable-based gaming detection (LV-GD), that controls for contextual factors and more robustly estimates student-level latent gaming tendencies. In LV-GD, a student is estimated as having a high gaming tendency if the student is detected to game more than the expected level of the population given the context. LV-GD applies a statistical model on top of an existing action-level gaming detector developed based on a typical human labeling process, without additional labeling effort. Across three datasets, we find that LV-GD consistently outperformed the original detector in validity measured by association between gaming and learning as well as reliability. LV-GD also afforded high practical utility: it more accurately revealed intervention effects on gaming, revealed a correlation between gaming and perceived competence in math and helped understand productive detected gaming behaviors. Our approach is not only useful for others wanting a cost-effective way to adapt a gaming detector to their context but is also generally applicable in creating robust behavioral measures.

Advances in Natural Language Processing offer techniques to detect the empathy level in texts. To test if individual feedback on certain students’ empathy level in their peer review writing process will help them to write more empathic reviews, we developed ELEA, an adaptive writing support system that provides students with feedback on the cognitive and emotional empathy structures. We compared ELEA to a proven empathy support tool in a peer review setting with 119 students. We found students using ELEA wrote more empathic peer reviews with a higher level of emotional empathy compared to the control group. The high perceived skill learning, the technology acceptance, and the level of enjoyment provide promising results to use such an approach as a feedback application in traditional learning settings. Our results indicate that learning applications based on NLP are able to foster empathic writing skills of students in peer review scenarios.

In this demonstration, we showcase the recent advancement of scenario-based tutor training with a focus to scale by applying the learn-by-doing approach to teaching strategies to provide socio-motivational support. These short (~15 min.) self-paced lessons use the predict-observe-explain inquiry method to develop mentor capacity in bolstering student motivation (i.e., fostering growth mindset). These custom training modules are being created to provide supplemental mentor support within the Personalized Learning2 system, an app which combines human tutoring and student math software to improve mentoring efficiency by connecting mentors to personalized resources, such as scenario-based mentor lessons, based on individual needs. Enhancing mentor training will aid in better quality mentoring at low cost. Mentor training is most effective when scenario-based practice provides trainees with response-specific feedback. To achieve feedback at scale, we illustrate an iterative design effort toward creating selected-response tasks that maintain some of the authenticity benefits of constructed-response. These scenario-based mentor lessons will be used by national level mentoring organizations as part of our efforts to scale.

Conversational agents (CAs) provide opportunities for improving the interaction in evaluation surveys. To investigate if and how a user-centered conversational evaluation tool impacts users' response quality and their experience, we build EVA - a novel conversational course evaluation tool for educational scenarios. In a field experiment with 128 students, we compared EVA against a static web survey. Our results confirm prior findings from literature about the positive effect of conversational evaluation tools in the domain of education. Second, we then investigate the differences between a voice-based and text-based conversational human-computer interaction of EVA in the same experimental set-up. Against our prior expectation, the students of the voice-based interaction answered with higher information quality but with lower quantity of information compared to the text-based modality. Our findings indicate that using a conversational CA (voice and text-based) results in a higher response quality and user experience compared to a static web survey interface.

Maximizing system scalability and quality are sometimes at odds. This work provides an example showing scalability and quality can be achieved at the same time in instructional design, contrary to what instructors may believe or expect. We situate our study in the education of HCI methods, and provide suggestions to improve active learning within the HCI education community. While designing learning and assessment activities, many instructors face the choice of using open-ended or close-ended activities. Close-ended activities such as multiple-choice questions (MCQs) enable automated feedback to students. However, a survey with 22 HCI professors revealed a belief that MCQs are less valuable than open-ended questions, and thus, using them entails making a quality sacrifice in order to achieve scalability. A study with 178 students produced no evidence to support the teacher belief. This paper indicates more promise than concern in using MCQs for scalable instruction and assessment in at least some HCI domains.

Adaptive Collaborative Learning Support (ACLS) systems improve collaboration and learning for students over individual work or collaboration with non-adaptive support. However, many ACLS systems are ill-suited for rural contexts where students often need multiple kinds of support to complete tasks, may speak languages unsupported by the system, and require more than pre-assigned tutor-tutee student pairs for more equitable learning. We designed an intervention that fosters more equitable help-seeking by automatically detecting student struggles and prompts them to seek help from specific peers that can help. We conducted a mixed-methods experimental study with 98 K-3 students in a rural village in Tanzania over a one-month period, evaluating how the system affects student interactions, system engagement, and student learning. Our intervention increased student interactions by almost 4 times compared to the control condition, increased domain knowledge interactions, and propelled students to engage in more cognitively challenging activities.

Teachers’ perspectives are critical for understanding classroom culture. They create and enforce rules in classrooms and are responsible for educating students using methods that they perceive to be most effective. Therefore, creating supplementary education technologies without understanding teachers and the culture they promote may lead to interventions that are underutilized or ineffective. Our research specifically investigates how technologies that foster student collaboration fit into teachers’ views of learning in a rural context with limited existing collaboration scaffolds. We interviewed 24 teachers and observed 39 classrooms in a rural Tanzanian village to understand how teachers value peer-peer collaboration in their teaching practice, and the unique challenges they face educating students in rural classroom settings. We uncover insights that inform the design and deployment of supplementary education technologies to support teachers in rural Tanzania and similar demographics.

Analytics of student learning data are increasingly important for continuous redesign and improvement of tutoring systems and courses. There is still a lack of general guidance on converting analytics into better system design, and on combining multiple methods to maximally improve a tutor. We present a multi-method approach to data-driven redesign of tutoring systems and its empirical evaluation. Our approach systematically combines existing and new learning analytics and instructional design methods. In particular, our methods involve identifying difficult skills and creating focused tasks for learning these difficult skills effectively following content redesign strategies derived from analytics. In our past work, we applied this approach to redesigning an algebraic modeling unit and found initial evidence of its effectiveness. In the current work, we extended this approach and applied it to redesigning two other tutor units in addition to a second iteration of redesigning the previously redesigned unit. We conducted a one-month classroom experiment with 129 high school students. Compared to the original tutor, the redesigned tutor led to significantly higher learning outcomes, with time mainly allocated to focused tasks rather than original full tasks. Moreover, it reduced over- and under-practice, yielded a more effective practice experience, and selected skills progressing from easier to harder to a greater degree. Our work provides empirical evidence of the effectiveness and generality of a multi-method approach to data-driven instructional redesign.

Practice is essential for learning. However, for many interpersonal skills, there often are not enough opportunities and venues for novices to repeatedly practice. Role-playing simulations offer a promising framework to advance practice-based professional training for complex communication skills, in fields such as teaching. In this work, we introduce ELK (Eliciting Learner Knowledge), a role-playing simulation system that helps K-12 teachers develop effective questioning strategies to elicit learners' prior knowledge. We evaluate ELK with 75 pre-service teachers through a mixed-method study. We find that teachers demonstrate a modest increase in effective questioning strategies and develop sympathy towards students after using ELK for 3 rounds. We implement a supplementary activity in ELK in which users evaluate transcripts generated from past role-play sessions. We have tentative evidence that a combination of role-play and evaluating conversation moves may be more effective for learning. We contribute design implications of using role-play systems for communication strategy training.

As the media landscape is increasingly populated by less than reputable sources of information, educators have turned to argument evaluation training as a potential solution. Unfortunately, the bias literature suggests that our ability to objectively evaluate an argument is, to a large extent, determined by the relationship between our own beliefs and the beliefs latent in the argument we are evaluating. If the argument supports our worldview, we are much more likely to overlook logical errors. Teachers recognize this need to adapt argument evaluation instruction to the specific beliefs of students. For instance, a teacher might intentionally assign a student an argument that the student disagrees with. Unfortunately, this kind of value-adaptive instruction is infrequent due to its unscalability. We propose a novel method for data-driven value-adaptive instruction in instructional technologies. This method can be used to combat bias in real-world contexts and support human reasoning during media consumption.

Intelligent tutoring systems (ITSs) have consistently been shown to improve the educational outcomes of students when used alone or combined with traditional instruction. However, building an ITS is a time-consuming process which requires specialized knowledge of existing tools. Extant authoring methods, including the Cognitive Tutor Authoring Tools' (CTAT) example-tracing method and SimStudent's Authoring by Tutoring, use programming-by-demonstration to allow authors to build ITSs more quickly than they could by hand programming with model-tracing. Yet these methods still suffer from long authoring times or difficulty creating complete models. In this study, we demonstrate that Simulated Learners built with the Apprentice Learner (AL) Framework can be combined with a novel interaction design that emphasizes model transparency, input flexibility, and problem solving control to enable authors to achieve greater model completeness in less time than existing authoring methods.

The goal of this workshop is to bring together the existing community of researchers working on Infrastructure Design for Data-Intensive Research in Computer Science Education and a community of Learning at Scale researchers focused on Computer Science Education. While both communities share many similar goals and could greatly benefit from each other work, the interaction between the communities is small. We hope that the proposed workshop will be instrumental in bringing together like-minded researchers from different communities, establishing collaboration, and expanding the scope of infrastructure project to address critical scaling issues.

Tablet-based educational technologies provide a supplement to traditional classroom-based early literacy education, especially in regions with limited schooling resources. Prior work has probed how children generally interact with and learn from these technologies, however, there is limited research on student engagement with applications that utilize valuable input techniques such as automatic handwriting and speech recognition. In our study, we designed and field-tested early literacy speech and handwriting recognition applications with the primary aim of maximizing student engagement. We designed the applications based on prior research insights and classroom observations from our target population and field-tested the applications with 283 children living in rural Tanzania. We found that observing a small set of classrooms can produce design insights that increase engagement on tablet-based learning systems on a much larger scale. We also demonstrate the importance of domain familiarity in students' choice to persist through activities while learning with technology.

Massive Open Online Courses (MOOCs) often incorporate lecture-based learning along with lecture notes, textbooks, and videos to students. Moreover, MOOCs also incorporate practice activities and quizzes. Student learning in MOOCs can be tracked and improved using state-of-the-art student modeling. Currently, this means employing conventional student models that are constructed around Intelligent Tutoring Systems (ITS). Traditional ITS systems only utilize students performance interactions (quiz, problem-solving or practice activities). Therefore, text interactions are entirely ignored while modeling students performance in MOOCs using these cognitive models. In this work, we propose a Comprehension Factor Analysis model (CFM) for online courses, which integrates student reading interactions in student models to track and predict learning outcomes. Our model evaluation shows that CFM outperforms state-of-the-art models in predicting students' performance in a MOOC. These models can help better student-wise adaptation in the context of MOOCs.

In schools and colleges around the world, open-ended home-work assignments are commonly used. However, such assignments require substantial instructor effort for grading, and tend not to support opportunities for repeated practice. We propose UpGrade, a novel learnersourcing approach that generates scalable learning opportunities using prior student solutions to open-ended problems. UpGrade creates interactive questions that offer automated and real-time feedback, while enabling repeated practice. In a two-week experiment in a college-level HCI course, students answering UpGrade-created questions instead of traditional open-ended assignments achieved indistinguishable learning outcomes in ~30% less time. Further, no manual grading effort is required. To enhance quality control, UpGrade incorporates a psychometric approach using crowd workers' answers to automatically prune out low quality questions, resulting in a question bank that exceeds reliability standards for classroom use.

Smartphone- and tablet-based learning systems are often posited as solutions for closing early literacy gaps between rural and urban regions in emerging economies. These systems are often developed based on experiences with students in urban contexts, limiting their success rates with children from rural areas who have had little to no prior exposure to technology. To explore how such technologies are used in different learning contexts, we deployed an early literacy learning application in school and home settings in a rural village in Tanzania. We use Rogoff's theory of instructional models to understand and describe the interaction between learners, adults, and peers. We found that in the presence of a school teacher, the instructional model was primarily "adult-run" where information was almost entirely disseminated by the teacher, while in home settings, the instructional model was similar to a "community-of-learners" model where children collaborate with other peers and adults to achieve their learning goals. We use these instructional models to surface six themes of support and scaffolding that were expressed differently across settings, and discuss the benefits and drawbacks of the instructional models observed in providing support across these themes.

CS Education makes heavy use of online educational tools like IDEs, Learning Management Systems, eTextbooks, interactive programming environments, and other smart content. Instructors and students would benefit from greater interoperability between tools. CS Ed researchers increasingly make use of the large collections of data generated by click streams coming from them. However, we all face barriers that slow progress: (1) Educational tools do not integrate well. (2) Information about CS learning process and outcome data generated by one system is not compatible with that from other systems. (3) CS problem solving and learning (e.g., coding solutions) is different from the type of data (discrete answers to questions or verbal responses) that current educational data mining focuses on. This BOF will discuss ways that we might support and better coordinate efforts to build community and capacity among CS Ed researchers, data scientists, and learning scientists toward reducing these barriers. CS Ed infrastructure should support broader re-use of innovative learning content that is instrumented for rich data collection, formats and tools for analysis of learner data, and best practices to make large collections of learner data available to researchers. Achieving these goals requires engaging a large community of researchers to define, develop, and use critical elements of this infrastructure to address specific data-intensive research questions.

Many game designers aim to optimize difficulty to make games that are "not too hard, not too easy." However, recent experiments have shown that even moderate difficulty can reduce player engagement. The present work investigates other design factors that may account for the purported benefits of difficulty, such as choice, novelty and suspense. These factors were manipulated in three design experiments involving over 20,000 play sessions of an online educational game.

This workshop will explore community based repositories for educational data and analytic tools that are used to connect researchers and reduce the barriers to data sharing. Leading innovators in the field, as well as attendees, will identify and report on bottlenecks that remain toward our goal of a unified repository. We will discuss these as well as possible solutions. We will present LearnSphere, an NSF funded system that supports collaborating on and sharing a wide variety of educational data, learning analytics methods, and visualizations while maintaining confidentiality. We will then have hands-on sessions in which attendees have the opportunity to apply existing learning analytics workflows to their choice of educational datasets in the repository (using a simple drag-and-drop interface), add their own learning analytics workflows (requires very basic coding experience), or both. Leaders and attendees will then jointly discuss the unique benefits as well as the limitations of these solutions. Our goal is to create building blocks to allow researchers to integrate their data and analysis methods with others, in order to advance the future of learning science.

The current study introduces a model for measuring student diligence using online behaviors during intelligent tutoring system use. This model is validated using a full academic year dataset to test its predictive validity against long-term academic outcomes including end-of-year grades and total work completed by the end of the year. The model is additionally validated for robustness to time-sample length as well as data sampling frequency. While the model is shown to be predictive and robust to time-sample length, the results are inconclusive for robustness in data sampling frequency. Implications for research on interventions, and understanding the influence of self-control, motivation, metacognition, and cognition are discussed.

Making MOOCs accessible to English Language Learners (ELLs) requires that students understand the language of instruction, and that instructional strategies address their unique learning challenges. Through the analysis of clickstream log data gathered from two MOOC courses deployed on Coursera, Introduction to Psychology and Statistical Thermodynamics, we show that ELL students exhibit distinct struggle behaviors in video portions without visual aids e.g., narrations without slides. Our findings challenge widely accepted multimedia design principles such as the split attention effect, provide insights into designing MOOC videos, and emphasize the need for adaptivity to increase MOOC access for ELLs.

"Multi-armed bandits" offer a new paradigm for the AI-assisted design of user interfaces. To help designers understand the potential, we present the results of two experimental comparisons between bandit algorithms and random assignment. Our studies are intended to show designers how bandits algorithms are able to rapidly explore an experimental design space and automatically select the optimal design configuration. Our present focus is on the optimization of a game design space. The results of our experiments show that bandits can make data-driven design more efficient and accessible to interface designers, but that human participation is essential to ensure that AI systems optimize for the right metric. Based on our results, we introduce several design lessons that help keep human design judgment in the loop. We also consider the future of human-technology teamwork in AI-assisted design and scientific inquiry. Finally, as bandits deploy fewer low-performing conditions than typical experiments, we discuss ethical implications for bandits in large-scale experiments in education.

The recent surge in interest in using educational data mining on student written programs has led to discoveries about which compiler errors students encounter while they are learning how to program. However, less attention has been paid to the actual code that students produce. In this paper, we investigate programming data by using learning curve analysis to determine which programming elements students struggle with the most when learning in Python. Our analysis extends the traditional use of learning curve analysis to include less structured data, and also reveals new possibilities for when to teach students new programming concepts. One particular discovery is that while we find evidence of student learning in some cases (for example, in function definitions and comparisons), there are other programming elements which do not demonstrate typical learning. In those cases, we discuss how further changes to the model could affect both demonstrated learning and our understanding of the different concepts that students learn.

The "doer effect" is an association between the number of online interactive practice activities students' do and their learning outcomes that is not only statistically reliable but has much higher positive effects than other learning resources, such as watching videos or reading text. Such an association suggests a causal interpretation--more doing yields better learning--which requires randomized experimentation to most rigorously confirm. But such experiments are expensive, and any single experiment in a particular course context does not provide rigorous evidence that the causal link will generalize to other course content. We suggest that analytics of increasingly available online learning data sets can complement experimental efforts by facilitating more widespread evaluation of the generalizability of claims about what learning methods produce better student learning outcomes. We illustrate with analytics that narrow in on a causal interpretation of the doer effect by showing that doing within a course unit predicts learning of that unit content more than doing in units before or after. We also provide generalizability evidence across four different courses involving over 12,500 students that the learning effect of doing is about six times greater than that of reading.

Student mistakes are often not random but, rather, reflect thoughtful yet incorrect strategies. In order for educational technologies to make full use of students' performance data to estimate the knowledge of a student, it is important to model not only the conceptions but also the misconceptions that a student's particular pattern of successes and errors may indicate. The student models that drive the "outer loop" of Intelligent Tutoring Systems typically do not represent or track misconceptions. Here, we present a method of representing misconceptions in the Knowledge Component models, or Q-Matrices, that are used by student models to estimate latent knowledge. We show, in a case study on a fraction arithmetic dataset, that incorporating a misconception into the Knowledge Component model dramatically improves the overall model's fit to data. We also derive qualitative insights from comparing predicted learning curves across models that incorporate varying misconception-related parameters. Finally, we show that the inclusion of a misconception in the Knowledge Component model can yield individual student estimates of misconception strength that are significantly correlated with out-of-tutor measures of student errors.

Massive scale education has emerged through online tools such as Wikipedia, Khan Academy, and MOOCs. The number of students being reached is high, but what about the quality of the educational experience? As we scale learning, we need to scale research to address this question. Such learning research should not just determine whether high quality has been achieved, but it should provide a process for how to reliably produce high quality learning. Scaling practical learning research is as much an opportunity as a problem. The opportunity comes from the fact that online courses are not only good for widespread delivery, but are natural vehicles for data collection and experimental instrumentation. I will provide examples of research done in the context of widely used educational technologies that both contribute interesting scientific findings and have practical implications for increasing the quality of learning at scale.

Open access and low cost make Massively Open Online Courses (MOOCs) an attractive learning platform for students all over the world. However, the majority of MOOCs are deployed in English, which can pose an accessibility problem for students with English as a Second Language (ESL). In order to design appropriate interventions for ESL speakers, it is important to correctly identify these students using a method that is scalable to the high number of MOOC enrollees. Our findings suggest that a new metric, browser language preference, may be better than the commonly-used IP address for inferring whether or not a student is ESL.

Can experimenting with three-dimensional (3D) physical objects in mixed-reality environments produce better learning and enjoyment than flat-screen two-dimensional (2D) interaction? We explored this question with EarthShake: a mixed-reality game bridging physical and virtual worlds via depth-camera sensing, designed to help children learn basic physics principles. In this paper, we report on a controlled experiment with 67 children, 4--8 years old, that examines the effect of observing physical phenomena and collaboration (pairs vs. solo). A follow-up experiment with 92 children tests whether adding simple physical control, such as shaking a tablet, improves learning and enjoyment. Our results indicate that observing physical phenomena in the context of a mixed-reality game leads to significantly more learning and enjoyment compared to screen-only versions. However, there were no significant effects of adding simple physical control or having students play in pairs vs. alone. These results and our gesture analysis provide evidence that children's science learning can be enhanced through experiencing physical phenomena in a mixed-reality environment.

The possibility of leveraging technology to support children's learning in the real world is both appealing and technically challenging. We have been exploring factors in tangible games that may contribute to both learning and enjoyment with an eye toward technological feasibility and scalability. Previous research found that young children learned early physics principles better when interactively predicting and observing experimental comparisons on a physical earthquake table than when seeing a video of the same. Immersing children in the real world with computer vision-based feedback appears to evoke embodied cognition that enhances learning. In the current experiment, we replicated this intriguing result of the mere difference between observing the real world versus a flat-screen. Further, we explored whether a simple and scalable addition of physical control (such as shaking a tablet) would yield an increase in learning and enjoyment. Our 2x2 experiment found no evidence that adding simple forms of hands-on control enhances learning, while demonstrating a large impact of physical observation. A general implication for educational game design is that affording physical observation in the real world accompanied by interactive feedback may be more important than affording simple hands-on control on a tablet.

The printing press long ago and the computer today have made widespread access to information possible. Learning theorists have suggested, however, that mere information is a poor way to learn. Instead, more effective learning comes through doing. While the most popularized element of today's MOOCs are the video lectures, many MOOCs also include interactive activities that can afford learning by doing. This paper explores the learning benefits of the use of informational assets (e.g., videos and text) in MOOCs, versus the learning by doing opportunities that interactive activities provide. We find that students doing more activities learn more than students watching more videos or reading more pages. We estimate the learning benefit from extra doing (1 SD increase) to be more than six times that of extra watching or reading. Our data, from a psychology MOOC, is correlational in character, however we employ causal inference mechanisms to lend support for the claim that the associations we find are causal.

Online games can serve as research instruments to explore the effects of game design elements on motivation and learning. In our research, we manipulated the design of an online math game to investigate the effect of challenge on player motivation and learning. To test the \'1cInverted-U Hypothesis\'1d, which predicts that maximum game engagement will occur with moderate challenge, we produced two large-scale (10K and 70K subjects), multi-factor (2x3 and 2x9x8x4x25) online experiments. We found that, in almost all cases, subjects were more engaged and played longer when the game was easier, which seems to contradict the generality of the Inverted-U Hypothesis. Troublingly, we also found that the most engaging design conditions produced the slowest rates of learning. Based on our findings, we describe several design implications that may increase challenge-seeking in games, such as providing feedforward about the anticipated degree of challenge.

One essential goal in education is to improve understanding of how humans acquire knowledge and how students vary in their abilities to learn. Building an intelligent agent that models student learning would be a significant achievement in the learning sciences. SimStudent is a state-of-the-art intelligent agent that simulates a human's learning process. However, SimStudent has only been living in the world of graphical user interfaces. To construct a more human-like learning agent, we integrate SimStudent with a cognitive robot, Calliope5KP, to create a physical agent that is able to learn skill knowledge by interacting with users in the physical world. We demonstrate the integration in a tic-tac-toe game, and show that the SimStudent robot is able to learn reasonably well with 12 games.

Building an intelligent agent, which simulates human-level learning appropriate for learning math, science, or a second language, could potentially benefit both education in understanding human learning, and artificial intelligence in creating human-level intelligence. Recently, we have proposed an efficient approach to acquiring procedural knowledge using transfer learning. However, it operated as a separate module. In this paper, we describe how to integrate this module into a machine-learning agent, SimStudent, that learns procedural knowledge from examples and through problem solving. We illustrate this method in the domain of algebra, after which we consider directions for future research in this area.

Gaming the system, attempting to succeed in an interactive learning environment by exploiting properties of the system rather than by learning the material (for example, by systematically guessing or abusing hints), is prevalent across many types of educational software. Past research on why students choose to game has focused on student individual differences. Many student individual differences, including attitudes towards mathematics, have been shown to be associated with gaming, but generally with low correlation. In this paper, we investigate how individual differences between learning environments can increase or decrease the probability of gaming. We enumerate ways intelligent tutor lessons vary from each other, and use data mining to discover hypotheses about how differences in software design and content influence the choice to game the system. We discover a set of tutor features that explain 56% of the variance in gaming, over five times the degree of variance explained in any prior study of student individual differences and gaming. These results provide an important step towards developing prescriptions for designing intelligent tutor software that students game significantly less.

Knowledge tracing (KT)[1] has been used in various forms for adaptive computerized instruction for more than 40 years. However, despite its long history of application, it is difficult to use in domain model search procedures, has not been used to capture learning where multiple skills are needed to perform a single action, and has not been used to compute latencies of actions. On the other hand, existing models used for educational data mining (e.g. Learning Factors Analysis (LFA)[2]) and model search do not tend to allow the creation of a “model overlay” that traces predictions for individual students with individual skills so as to allow the adaptive instruction to automatically remediate performance. Because these limitations make the transition from model search to model application in adaptive instruction more difficult, this paper describes our work to modify an existing data mining model so that it can also be used to select practice adaptively. We compare this new adaptive data mining model (PFA, Performance Factors Analysis) with two versions of LFA and then compare PFA with standard KT.

Giving effective help is an important collaborative skill that leads to improved learning for both the help-giver and help-receiver. Adding intelligent tutoring to student interaction may be one effective way of assisting students in giving and receiving better help. However, such systems have proven difficult to implement, in part due to the challenges of modeling productive dialogue in a collaborative activity. We present a theoretical model of good helping behavior in a peer tutoring context, and validate the model using student tutoring data, linking optimal and buggy behaviors to learning outcomes. We discuss the implications of the model with respect to providing intelligent tutoring for peer tutoring.

The increasing integration of education and technology has led to the development of a range of note-taking applications. Our project's goal is to provide empirical data to guide the design of such note-taking applications by evaluating the behavioral and learning outcomes of different note-taking functionality. The study reported here compares note-taking using a text editor and four interaction techniques. The two standard techniques are typing and copy-paste. The two novel techniques are restricted copy-paste and menu-selection, intended to increase attention and processing respectively. Hypothesized learning gains from the novel techniques were not observed. As implemented these techniques were less efficient and appeared to be more frustrating to use. However, data regarding differences in both note-taking efficiency and learning suggest several important implications for selection-based note-taking applications, such as pasting and highlighting. Our results also indicate that students have strong opinions regarding their note-taking practices, which may complicate potentially beneficial interventions.

Studies have shown that both the act of note-taking and the use of notes for review can promote learning. Many note-taking applications have been developed for computer-based learning content. In general, they include advanced annotation functionality, and are geared toward supporting collaboration and discussion. Though these devices have been shown to change note-taking behavior, their effect on learning has not been evaluated. The goal of our research is to evaluate the effect of specific features of note-taking applications on behavior and learning, in order to develop guidelines for advanced note-taking applications that promote learning. These applications could be used as the basis for a variety of educational activities, including collaboration. In this paper, we present the results of an experiment evaluating a basic feature of note-taking technology: copy-paste. Our findings indicate that copy-paste functionality can be detrimental to learning. We describe potential implications of these results for the developers of notetaking applications.

We believe handwriting input may be able to provide significant advantages over typing, especially in the mathematics learning domain. The use of handwriting may result in decreased extraneous cognitive load on students, and it may provide better support for the two-dimensional spatial components of mathematics when compared to existing typing-based tools. Here we report progress towards the application of a handwriting interface for mathematics learning. We introduce a prototype system that allows students to use handwriting input to solve algebraic equations in an intelligent tutor. We discuss strategies to improve the existing handwriting system and apply it to math learning. Although the recognition accuracy of current handwriting engines may not be at a level suitable for use by students, we hypothesize that this may be realistically improved via advance training of the engine on a large corpus, as well as via techniques similar to co-training

Secondary teachers across the country are being asked to use formative assessment data to inform their classroom instruction. At the same time, critics of No Child Left Behind are calling the bill "No Child Left Untestedö emphasizing the negative side of assessment, in that every hour spent assessing students is an hour lost from instruction. Or does it have to be? What if we better integrated assessment into the classroom, and we allowed students to learn during the test? Maybe we could even provide tutoring on the steps of solving problems. Our hypothesis is that we can achieve more accurate assessment by not only using data on whether students get test items right or wrong, but by also using data on the effort required for students to learn how to solve a test item. We provide evidence for this hypothesis using data collected with our E-ASSISTment system by more than 600 students over the course of the 2004-2005 school year. We also show that we can track student knowledge over time using modern longitudinal data analysis techniques. In a separate paper [9], we report on the ASSISTment system's architecture and scalability, while this paper is focused on how we can reliably assess student learning.

We investigate the prevalence and learning impact of different types of off-task behavior in classrooms where students are using intelligent tutoring software. We find that within the classrooms studied, no other type of off-task behavior is associated nearly so strongly with reduced learning as "gaming the system": behavior aimed at obtaining correct answers and advancing within the tutoring curriculum by systematically taking advantage of regularities in the software's feedback and help. A student's frequency of gaming the system correlates as strongly to post-test score as the student's prior domain knowledge and general academic achievement. Controlling for prior domain knowledge, students who frequently game the system score substantially lower on a post-test than students who never game the system. Analysis of students who choose to game the system suggests that learned helplessness or performance orientation might be better accounts for why students choose this behavior than lack of interest in the material. This analysis will inform the future re-design of tutors to respond appropriately when students game the system.

Although engineering models of user behavior have enjoyed a rich history in HCI, they have yet to have a widespread impact due to the complexities of the modeling process. In this paper we describe a development system in which designers generate predictive cognitive models of user behavior simply by demonstrating tasks on HTML mock-ups of new interfaces. Keystroke-Level Models are produced automatically using new rules for placing mental operators, then implemented in the ACT-R cognitive architecture. They interact with the mock-up through integrated perceptual and motor modules, generating behavior that is automatically quantified and easily examined. Using a query-entry user interface as an example [19], we demonstrate that this new system enables more rapid development of predictive models, with more accurate results, than previously published models of these tasks.