Breadcrumb
- Home
- AI Hub For Education
- Research Study Repository
- Evidence Explorer
- AI Tools
- Custom-built Research Tool
The Repository
Search and Filter
Custom-built research tool
Studies tagged with AI from their full text or abstract; see each study for details.
Showing 391 - 420 of 531 studies
From EduVisBench to EduVisAgent: A Benchmark and Multi-Agent Framework for Reasoning-Driven Pedagogical Visualization
Haonian Ji, Shi Qiu, Siyang Xin, Siwei Han, Zhaorun Chen, Dake Zhang, Hongyi Wang, Huaxiu Yao. (05/2025). arXiv.
EduVisAgent achieved an overall average score of 81.6%, outperforming the best-performing baseline tool, v0, by 23.4 percentage points (a 40.2% relative improvement).
Investigating Pedagogical Teacher and Student LLM Agents: Genetic Adaptation Meets Retrieval-Augmented Generation Across Learning Styles
Debdeep Sanyal, Agniva Maiti, Umakanta Maharana, Dhruv Kumar, Ankur Mali, Clyde Lee Giles, Murari Mandal. (05/2025). arXiv.
The simulation framework enables distinct and interpretable teaching patterns to emerge when interacting with varied student populations.
From Recall to Reasoning: Automated Question Generation for Deeper Math Learning through Large Language Models
Yongan Yu, Alexandre Krantz, Nikki G. Lobczowski. (05/2025). arXiv.
Combining retrieval-augmented generation with Depth of Knowledge levels consistently outperformed prompt-only setups across all tested language models, especially for higher-order thinking skills.
Leveraging Graph Retrieval-Augmented Generation to Support Learners' Understanding of Knowledge Concepts in MOOCs
Mohamed Abdelmagied, Mohamed Amine Chatti, Shoeb Joarder, Qurat Ul Ain, Rawaa Alatrash. (05/2025). arXiv.
Instructors rated generated personalized questions highly for fluency (2.98 out of 3), clarity (2.84), and relevance to slides (2.85) and misunderstood concepts (2.67).
LECTOR: Summarizing E-book Reading Content for Personalized Student Support
Erwin López, Cheng Tang, Valdemar Svábensky, Fumiya Okubo, Atsushi Shimada. (05/2025). arXiv.
LECTOR outperformed existing NLP models in extracting key concepts from lecture slides, reaching an average F1-score of 20.14% compared to 15.06% for AttentionRank.
VTutor for High-Impact Tutoring at Scale: Managing Engagement and Real-Time Multi-Screen Monitoring with P2P Connections
Eason Chen, Xinyi Tang, Aprille Xi, Chenyu Lin, Conrad Borchers, Jionghao Lin, Shivang Gupta, Kenneth R. Koedinger. (05/2025). arXiv.
VTutor combines peer-to-peer screen streaming with an animated panda avatar, enabling a single remote tutor to simultaneously monitor multiple learners and deliver spoken guidance.
VTutor: An Animated Pedagogical Agent SDK that Provide Real Time Multi-Model Feedback
Eason Chen, Chenyu Lin, Yu-Kai Huang, Xinyi Tang, Aprille Xi, Jionghao Lin, Kenneth R. Koedinger. (05/2025). arXiv.
Among 50 participants, 36 preferred VTutor over SadTalker, demonstrating a statistically significant preference.
LLMs to Support K-12 Teachers in Culturally Relevant Pedagogy: An AI Literacy Example
Jiayi Wang, Ruiwei Xiao, Xinying Hou, Hanqi Li, Ying-Jui Tseng, John Stamper, Kenneth R. Koedinger. (05/2025). arXiv.
All participating teachers reported increased confidence in making culturally responsive modifications, with confidence in identifying opportunities rising from 3-5 initially to 5 out of 5.
SlideItRight: Using AI to Find Relevant Slides and Provide Feedback for Open-Ended Questions
Chloe Qianhui Zhao, Jie Cao, Eason Chen, Kenneth R. Koedinger, Jionghao Lin. (05/2025). arXiv.
Participants showed significant pre-to-post learning gains across all feedback conditions, but learning gains did not differ significantly between AI, human, or slide-based feedback.
MSA at BEA 2025 Shared Task: Disagreement-Aware Instruction Tuning for Multi-Dimensional Evaluation of LLMs as Math Tutors
Baraa Hikal, Mohamed Basem, Islam Oshallah, Ali Hamdi. (05/2025). arXiv.
The system achieved top-tier competition ranks across all pedagogical tracks, including first place in Providing Guidance and third in Actionability.
KCluster: An LLM-based Clustering Approach to Knowledge Component Discovery
Yumou Wei, Paulo F. Carvalho, John Stamper. (05/2025). arXiv.
KCluster achieved the closest alignment with expert-designed knowledge component models across all three evaluation datasets.
CAFES: A Collaborative Multi-Agent Framework for Multi-Granular Multimodal Essay Scoring
Jiamin Su, Yibo Yan, Zhuoran Gao, Han Zhang, Xiang Liu, Xuming Hu. (05/2025). arXiv.
The collaborative multi-agent framework improved human-grader agreement across student models by an average of 21% relative to baseline scores, raising Quadratic Weighted Kappa from 0.29 to 0.36.
A Participatory Strategy for AI Ethics in Education and Rehabilitation grounded in the Capability Approach
Valeria Cesaroni, Eleonora Pasqua, Piercosma Bisconti, Martina Galletti. (05/2025). arXiv.
Clinicians emphasized that AI must mediate between system automation and clinician control to avoid dominating the therapeutic interaction or eroding adult agency.
From EduVisBench to EduVisAgent: A Benchmark and Multi-Agent Framework for Pedagogical Visualization
Haonian Ji, Shi Qiu, Siyang Xin, Siwei Han, Zhaorun Chen, Dake Zhang, Hongyi Wang, Huaxiu Yao. (05/2025). arXiv.
The multi-agent system EduVisAgent achieved an overall average score of 81.6%, outperforming the best baseline model by 23.4 percentage points, a 40.2% improvement.
LLM Access Shield: Domain-Specific LLM Framework for Privacy Policy Compliance
Yu Wang, Cailing Cai, Zhihua Xiao, Peifung E. Lam. (05/2025). arXiv.
A domain-specific supervised model identified sensitive data entities with a 0.839 privacy hiding rate, substantially outperforming the base model rate of 0.064.
Personalizing Student-Agent Interactions Using Log-Contextualized Retrieval Augmented Generation (RAG)
Clayton Cohn, Surya Rayala, Caitlin Snyder, Joyce Horn Fonteles, Shruti Jain, Naveeduddin Mohammed, Umesh Timalsina, Sarah K. Burriss, Ashwin T, Namrata Srivastava, Menton Deweese, Angela Eeds, Gautam Biswas. (05/2025). arXiv.
Log-contextualized retrieval outperformed the discourse-only baseline in 4 out of 5 embedding models tested.
Pedagogy-R1: Pedagogically-Aligned Reasoning Model with Balanced Educational Benchmark
Unggi Lee, Jaeyong Lee, Jiyeong Bae, Yeil Jeong, Junbo Koh, Gyeonggeon Lee, Gunho Lee, Taekyung Ahn, Hyeoncheol Kim. (05/2025). arXiv.
The 7-billion-parameter Pedagogy-R1 model achieved the highest scores on teacher decision-making (54.76%), pedagogical knowledge (31.67%), and essay scoring (15.83%), outperforming general base models.
Beyond Retrieval: Joint Supervision and Multimodal Document Ranking for Textbook Question Answering
Hessa Abdulrahman Alawwad, Usman Naseem, Areej Alhothali, Ali Alkhathlan, Amani Jamal. (05/2025). arXiv.
JETRTQA outperformed previous state-of-the-art models, raising accuracy by 2.4% on the validation set and 11.1% on the test set.
AI for Accessible Education: Personalized Audio-Based Learning for Blind Students
Crystal Yang, Paul Taele. (04/2025). arXiv.
Analysis of educator interviews revealed five core design considerations for accessible learning tools: maximizing engagement, keeping interfaces simple, ensuring assistive technology compatibility, accommodating diverse needs, and providing positive reinforcement.
EduPlanner: LLM-Based Multi-Agent Systems for Customized and Intelligent Instructional Design
Xueqiao Zhang, Chao Zhang, Jianwen Sun, Jun Xiao, Yi Yang, Yawei Luo. (04/2025). arXiv.
EduPlanner achieved an overall quality score of 88 out of 100, outperforming standalone GPT-4 (49), Llama-3-70B-Instruct (20), and GPT-3.5-turbo (17).
STRIVE: A Think & Improve Approach with Iterative Refinement for Enhancing Question Quality Estimation
Aniket Deroy, Subhankar Maity. (04/2025). arXiv.
The STRIVE approach increased GPT-4's correlation with human expert ratings across all metrics, reaching correlations up to 0.63 for novelty and 0.62 for appropriateness.
Glocalizing Generative AI in Education for the Global South: The Design Case of 21st Century Teacher Educator AI for Ghana
Matthew Nyaaba. (04/2025). arXiv.
Pilot testing showed the tool enabled language adaptation and bilingual instruction in English and indigenous Ghanaian languages like Twi, Dagbani, Mampruli, and Dagaare.
Synthesizing High-Quality Programming Tasks with LLM-based Expert and Student Agents
Manh Hung Nguyen, Victor-Alexandru Padurean, Alkis Gotovos, Sebastian Tschiatschek, Adish Singla. (04/2025). arXiv.
The multi-agent validation pipeline achieved 87.3% precision and 84.0% coverage when attempting up to 10 trials, significantly outperforming baseline generation methods.
AI-University: An LLM-based platform for instructional alignment to scientific classrooms
Mostafa Faghih Shojaei, Rahul Gulati, Benjamin A. Jasperson, Shangshang Wang, Simone Cimolato, Dangli Cao, Willie Neiswanger, Krishna Garikipati. (04/2025). arXiv.
The fine-tuned expert model scored higher in semantic similarity on 86.02% of test cases and was consistently preferred by an automated language model judge.
Exploring Student Behaviors and Motivations when using AI Teaching Assistants with Optional Guardrails
Amanpreet Kapoor, Marc Diaz, Stephen MacNeil, Leo Porter, Paul Denny. (04/2025). arXiv.
Half of the 885 students disabled guardrails using the "See Solution" feature for at least one problem, and 14% used it for all three.
Assessing AI-Generated Questions' Alignment with Cognitive Frameworks in Educational Assessment
Antoun Yaacoub, Jerome Da-Rugna, Zainab Assaghir. (04/2025). arXiv.
DistilBERT achieved the highest performance across all cognitive levels, reaching an overall validation accuracy of 91%.
Towards responsible AI for education: Hybrid human-AI to confront the Elephant in the room
Danial Hooshyar, Gustav Sir, Yeongwook Yang, Eve Kikas, Raija Hamalainen, Tommi Karkkainen, Dragan Gašević, Roger Azevedo. (04/2025). arXiv.
Educational AI discourse has increasingly equated AI with large language models, with 3,009 of 4,551 generative AI articles published in 2024 alone.
INSIGHT: Bridging the Student-Teacher Gap in Times of Large Language Models
Jarne Thys, Sebe Vanbrabant, Davy Vanacken, Gustavo Rovelo Ruiz. (04/2025). arXiv.
Instructors expressed concern that students turning to tools like ChatGPT would leave teachers with little feedback on student comprehension and course material.
From Prompts to Propositions: A Logic-Based Lens on Student-LLM Interactions
Ali Alfageeh, Sadegh Almahdi Kazemi Zarkouei, Daye Nam, Daniel Prol, Matin Amoozadeh, Souti Chattopadhyay, James Prather, Paul Denny, Juho Leinonen, Michael Hilton, Sruti Srinivasa Ragavan, Amin Alipour. (04/2025). arXiv.
The Prompt2Constraints method correctly translated 225 of 234 sampled student prompts into propositional logic constraints.
Evolution of AI in Education: Agentic Workflows
Firuz Kamalov, David Santandreu Calonge, Linda Smail, Dilshod Azizov, Dimple R. Thadani, Theresa Kwong, Amara Atif. (04/2025). arXiv.
The multi-agent scoring system achieved a lower mean absolute error of 0.561 compared to stand-alone GPT-4o at 0.613 on 17,000 argumentative essays.
