Source-linked AI summary
Beyond Problem Solving: Large Language Models for Emotional and Reflective Support in Mathematics Learning
Vera Rief, Mirella Hladký, Minju Yoo, Stephanie Heel, Shintaro Sato, Tomohiro Nagashima
TL;DR
Intelligent tutoring systems rarely provide scalable real-time emotional support alongside cognitive instruction, leaving mindfulness in adaptive math learning underexplored. The paper develops an LLM-enhanced algebra tutor with mindful chat, breathing, and mindful feedback, and evaluates it against cognitive-only support in a classroom study. Overall learning and executive state-math anxiety improved, while mindfulness was associated with fewer solved problems and hints at comparable learning and with stronger perceived socio-emotional support, without significant condition differences.
Problem
Existing intelligent tutoring systems emphasize cognitive support, while scalable mindfulness interventions for students’ emotional states in adaptive math learning remain underexplored.
Method
The study developed an LLM-enhanced algebra ITS combining cognitive tutoring with mindful chat, breathing exercises, and mindful hints and feedback, then compared it with cognitive-only support.
Results
The ITS reduced executive state-math anxiety and improved math learning overall; mindfulness conditions showed comparable learning with fewer problems solved and hints used, plus stronger perceived socio-emotional connection.
Takeaways & Limitations
LLM-driven mindfulness can be integrated into ITSs as scalable socio-emotional support delivered automatically and without specialized instructor training.
Takeaways & Limitations
Limited English proficiency and a smaller-than-expected sample may have masked intervention effects and constrain generalizability.
Abstract
from arXiv · showhide
Intelligent Tutoring Systems (ITSs) traditionally focus their adaptive support on cognitive aspects of learning. Although effective, little is known about how such systems can be enhanced by addressing students' emotional states. In particular, the role of mindful interventions for supporting student learning and experiences in adaptive math learning remains underexplored. We developed "Math with Matt", an ITS that leverages Large Language Models (LLMs) to provide both cognitive and emotional support in algebra learning. The system offers 1) an LLM-based mindful chat that delivers context-sensitive emotional support through a pedagogical agent Matt, and 2) mindful feedback and hint messages (not just evaluative) to enhance learning experiences and reduce math anxiety. We conducted a classroom study with 7th graders, comparing a Mindful version against a version with cognitive support only. Overall, the ITS reduced executive state-math anxiety and improved students' math learning, though no significant differences emerged between the conditions. However, students with the mindfulness interventions showed higher learning efficiency and well-balanced problem-solving behavior, since they achieve a similar level of math learning with less learning time and fewer requested hints compared to the Cognitive version. Additionally, they reported that the pedagogical agent felt more supportive and caring than students in the cognitive condition. Our study demonstrates the feasibility and scalability of integrating mindfulness into ITSs through LLM-based interactions and positions LLMs as an adaptive, socio-emotional layer within cognitive math tutoring.
1 Introduction
Math anxiety can impair memory, performance, and engagement, while existing emotional interventions are difficult to scale within instruction. The study therefore developed and evaluated an LLM-enhanced ITS combining cognitive tutoring with mindfulness support.
- Math anxiety involves fear and tension during numerical or mathematical tasks and can impair working memory, performance, and engagement.
- Existing mindfulness interventions can reduce math anxiety in real time but often depend on trained instructors and multisession programs.
- The system combines LLM-based context-sensitive emotional support, breathing exercises, and mindful cognitive feedback and hints for algebra learning.
- A classroom comparison tested the mindfulness-enhanced ITS against a cognitive-support-only version on anxiety, learning, usability, and learning experiences.
- The study found reduced executive state-math anxiety and improved math learning overall, without significant differences between conditions.
2 Background and Related Work
Math anxiety is a context-sensitive emotional and cognitive challenge, while mindfulness and ITS research suggest complementary support mechanisms. However, scalable digital systems rarely provide real-time socio-emotional support alongside adaptive math instruction.
- Math Anxiety: Math anxiety includes stable trait anxiety and context-dependent state anxiety, with anticipatory and executive forms during math activity.
- Math Anxiety: State-math anxiety is associated with lower performance and can consume cognitive resources by reducing working-memory capacity.
- Mindfulness Interventions: Mindfulness involves non-judgmental present-moment awareness and may support cognitive control and adaptive self-regulation.
- Mindfulness Interventions: Brief guided breathing has been associated with reduced negative emotions and improved calmness and performance during math tasks.
- Intelligent Tutoring Systems: ITSs provide personalized instruction and adaptive feedback, but existing systems rarely embed real-time socio-emotional support such as mindfulness.
- Intelligent Tutoring Systems: LLM-based pedagogical agents offer a potentially scalable way to deliver flexible emotional support within adaptive tutoring.
3 Design of the Mindful Intelligent Tutor for Algebra
Math with Matt combines self-paced algebra practice with LLM-guided mindful interactions. Its design integrates dynamic chat, breathing exercises, and carefully controlled mindful hints and feedback.
- Math Learning Units: The ITS teaches linear systems of equations through three units covering substitution, elimination, and equalization methods.
- Math Learning Units: Students solve nine core problems comprising 57 problem-solving steps, with ten optional bonus exercises adding 65 steps.
- Mindfulness Interventions: Dynamic chat units appear after each math unit, allowing learners to express thoughts and emotions in brief conversations with Matt.
- Mindfulness Interventions: Matt’s responses use a prompt based on seven mindfulness principles, with runtime prompts incorporating dialogue state, chat history, and student input.
- Mindfulness Interventions: Students are offered a guided 60-second breathing exercise after each mindful chat unit.
- Mindfulness Interventions: Mindful hints and feedback were generated by converting cognitively oriented messages, then hard-coded after accuracy checking rather than generated dynamically during tutoring.
4 Classroom Evaluation
The classroom evaluation compared mindfulness-enhanced and cognitive-only versions of the ITS across anxiety, learning, usability, and perceptions of the pedagogical agent. It used research questions, pre/post measures, questionnaires, and interaction logs.
- Research Questions: The evaluation asked whether mindful ITS use changes state-math anxiety, supports math learning, and affects perceived usability and agent support.
- Experimental Conditions: The Mindful condition included chat, breathing, and mindful hints and feedback, whereas the Cognitive condition provided only cognitively oriented tutoring support.
- Participants: Participants were 252 seventh-grade students aged 12–13 across seven classes, randomly assigned within classes to the two conditions.
- Measures: Math learning was measured with pretests and posttests assessing conceptual and procedural knowledge of linear systems of equations.
- Measures: Additional measures covered trait anxiety, system usability, perceptions of Matt’s socio-emotional support, warmth, competence, and logged tutor interactions.
- Measures: State anxiety was measured before testing for anticipatory anxiety and during testing after four of six items for executive anxiety.
- Procedure: The study took place across two sessions in a weekly 50-minute class, with introductory instruction, testing, ITS use, and debriefing.
5 Results
The classroom study found reductions in executive state-math anxiety and algebra learning gains over time, without significant condition differences. Mindful students completed fewer problem-solving steps and requested fewer hints, while reporting stronger support and concern from Matt on specific items.
- 5.1 Math Anxiety: Executive state-math anxiety decreased significantly over time in both conditions, with no significant group or interaction effect.The effect was large: F(1, 15)=5.422, p=.034, partial η2=0.27.
- 5.2 Math Learning: Algebra test scores improved significantly from pretest to posttest, but gains did not depend on condition.The time effect was significant with F(1, 40)=6.418, p=.015, partial η2=.14.
- 5.3 System Usability: System usability ratings were low in both groups, and the higher Mindful rating was not statistically significant.The Cognitive group averaged 43.96 and the Mindful group 55.56 on the 0–100 SUS scale.
- 5.4 Pedagogical Agent’s Characteristics: The Mindful group rated Matt significantly higher on feeling supported and believing he was concerned about their well-being.The item comparisons were significant for “I feel supported by Matt” (p=.028) and “I believe that Matt is very concerned about my well-being” (p=.046).
- 5.5 Additional Insights: Exploratory Log Data Analysis: Students in the Mindful condition completed 12.43 problem-solving steps on average versus 21.28 in the Cognitive condition, a significant difference.The intervention contained 122 total problem-solving steps; the group comparison yielded p < .001.
- 5.5 Additional Insights: Exploratory Log Data Analysis: Mindful students used 0.61 hints per completed step versus 1.06 for Cognitive students, and this difference was significant.Each step offered three available hints; the group comparison yielded p = .002.
6 Discussion
The study found that mindfulness-enhanced tutoring did not significantly outperform cognitive support on anxiety or learning, but it supported socio-emotional connection and more efficient, balanced problem solving. Classroom and language constraints may have obscured differential effects.
- Anticipatory state-math anxiety showed no significant reduction over time or difference between groups, whereas executive state-math anxiety decreased significantly in both groups.
- Peer influence in the classroom may have diluted the affective impact of self-paced digital mindfulness practices, including breathing exercises.Students could feel uncomfortable closing their eyes around peers, and conversations comparing interface versions were distracting.
- Students in both conditions improved math performance and showed reduced executive state-math anxiety over time, without significant between-condition differences.
- Mindfulness-intervention students solved significantly fewer problems overall and used fewer hints per solving step while achieving similar learning outcomes.The authors interpret this pattern as improved learning efficiency and self-regulation.
- The mindful pedagogical agent was perceived as more supportive, empathetic, and caring than the agent in the cognitive condition.
- Limited English proficiency and reliance on translation tools disrupted students’ workflow and may have masked potential effects of the mindful language.
- The smaller-than-expected sample limits how confidently the findings generalize across contexts, while classroom conditions constrained individual mindfulness engagement.
- The study demonstrates that LLM-driven mindfulness can be integrated into ITSs and delivered automatically in real time without specialized instructor training.