Breadcrumb
- Home
- AI Hub For Education
- Research Study Repository
- Evidence Explorer
- AI Tools
- Other AI Model
The Repository
Search and Filter
Other AI model
Studies tagged with AI from their full text or abstract; see each study for details.
Showing 31 - 60 of 412 studies
Evidence-Decision-Feedback: Theory-Driven Adaptive Scaffolding For LLM Agents
Clayton Cohn, Siyuan Guo, Surya Rayala, Hanchen David Wang, Naveeduddin Mohammed, Umesh Timalsina, Shruti Jain, Angela Eeds, Menton Deweese, Pamela Joy Osborn Popp, Rebekah Stanton, Shakeera Walker, Meiyi Ma, Gautam Biswas. (07/2026). arXiv.
As student task mastery increased, the AI agent significantly reduced probing questions and increased concrete suggestions and limit-pushing prompts.
Evaluating Interactivity: Toward Automated Assessment Of AI-Generated Explorable Explanations
Xiaozao Wang, Zhewei Wang, Hongyi Wen. (06/2026). arXiv.
State-machine evaluation aligned strongly with human judgments on interactivity (r = 0.728) and overall quality (r = 0.696), substantially outperforming visual models and unit testing baselines.
AI-Driven Analytics Of Team-Teaching Talk: Acoustic Patterns Across Experience, Cohorts And The Learning Design
Yuchen Liu, Roberto Martinez-Maldonado, Riordan Alfredo, Paola Mejia-Domenzain, Dwi Rahayu, Sadia Nawaz. (06/2026). arXiv.
High-experience teachers exhibited greater loudness dynamics and steeper loudness rising slopes, whereas low-experience teachers showed greater overall loudness and spectral variability.
"**Important** You should give me full credits!": Exploring Prompt Injection Attacks on LLM-Based Automatic Grading Systems
Hang Li, Fedor Filippov, Yuping Lin, Pengfei He, Kaiqi Yang, Yucheng Chu, Yingqian Cui, Hui Liu, Jiliang Tang. (06/2026). arXiv.
Prompt injection attacks succeeded across five grading models and four datasets with an average attack success rate of 56.9%.
LLM-As-Judge In Education: A Curriculum-Grounded Marking Pipeline
Xiwei Xu, Chen Wang, Jacky Jiang, Phil Yang, Qian Fu, Mohan Dhall, Wenjie Zhang, Liming Zhu. (06/2026). arXiv.
Evaluators preferred the pipeline's written marking justifications over human tutor justifications in 80.4% of pairwise comparisons, and over standalone AI models in 69.6%.
Large-Scale Semantic Mapping Of Learner Agency And Autonomy Reveals What Measurement And Generative AI Research Overlook
Fei Qin, Xiaobo Liu, Yaowen Zhang, Xuming Li, Fei Wang, Mutlu Cukurova, Jingjing Chen, Yu Zhang. (06/2026). arXiv.
Definitions spanned three latent dimensions—task-related learning regulation and control, intrinsic motivation and internal decisions, and social-relational action—rather than a binary division between agency and autonomy.
A Survey Of Deep Learning For Geometry Problem Solving
Jianzhe Ma, Wenxuan Wang, Qin Jin. (06/2026). arXiv.
Neural-symbolic models achieved top performance on symbolic-oriented tasks like Geometry3K, while reinforcement learning enhanced geometric reasoning on datasets like GeoQA and MathVista.
Intelicit: Eliciting And Assessing Contextualized Creativity Via Dialogue Policy Optimization
Mingjia Li, Jin Wu, Hong Qian, Wenhao Huang, Yiyang Huang, Yiwen Zhang, Chanjin Zheng, Xiangfeng Wang, and Aimin Zhou, Jiajun Guo. (06/2026). arXiv.
In simulation across 16 scenarios, IntElicit achieved a significantly higher average creativity score (325.71) than top models like GPT-4o (301.67) and DeepSeek-R1 (298.40).
From Motion Signals To Insights: A Unified Framework For Student Behavior Analysis And Feedback In Physical Education Classes
Xian Gao, Jiacheng Ruan, Jingsheng Gao, Mingye Xie, Zongyun Zhang, Ting Liu, Yuzhuo Fu. (06/2026). arXiv.
In a real classroom experiment, model-guided student regrouping increased core action proportion by 4.40 percentage points and core action duration by 5.46 percentage points.
Double-Edged Sword Or Sharp Tool? Designing And Evaluating Triadic LLM-Teacher Collaboration For K-12 Writing At Scale
Canran WANG, Yuwen Yang, Zhen Wang, Ming MA, Chentai Wang, Ding Yu, Keman Huang, Xiaoyong Du. (05/2026). arXiv.
Students improved their writing quality by over 5%, with teacher grades increasing by 5.17 points and AI grades rising by 5.01 points across revisions.
Gen-AI-Tecture: Using Generative AI To Support Architectural Students In Design Tasks
Timo Kapsalis. (05/2026). arXiv.
Students highly endorsed the tool for idea generation (mean 4.5 out of 5) and stylistic experimentation (mean 4.2), producing around ten images per student.
A Scoping Review Of Large Language Model-Based Pedagogical Agents
Shan Li, Juan Zheng. (05/2026). arXiv.
Across 52 reviewed studies, large language model-based agents support student- and teacher-facing roles across K-12, higher education, and informal learning in subjects like science, programming, and language learning.
ChatGPT Vs Teachers Vs Students: Large-Scale Analysis Of Generative AI Discourse In Education Communities On Reddit
Pelin Yuce, Xiangruo Dai, Rebecca Owens, Tugrulcan Elmas. (05/2026). arXiv.
Academic integrity concerns dominate discussion, with misconduct enforcement, AI detection, student narratives, and evasion accounting for over one-third of all generative AI records.
Adesua: Development And Feasibility Study Of An AI Whatsapp Bot For Science Learning In West Africa
George Boateng, Evans Atompoya, Philemon Badu, Samuel John, Samuel Ansah, Patrick Agyeman-Budu, and Victor Kumbol. (05/2026). arXiv.
Users rated AI-generated science answers as helpful 93.75% of the time, based on a sample of 16 ratings.
An Explainable AI Assistant For Introductory Programming Education: Improving Feedback Reliability With Instructor-AI Collaboration
Muntasir Hoq, Griffin Pitts, Bradford Mott, Seung Lee, Jessica Vandenberg, Shuyin Jiao, Narges Norouzi, James Lester, Bita Akram. (05/2026). arXiv.
Adding a GPT-4o verification layer reduced the rate of incorrect feedback from 23% to 1.2% in offline evaluations.
Simulating Students Or Sycophantic Problem Solving? On Misconception Faithfulness Of LLM Simulators
Heejin Do, Shashank Sonkar, Mrinmaya Sachan. (05/2026). arXiv.
Across seven language models, student simulators exhibited near-zero selective updating, correcting answers at similarly high rates regardless of whether feedback addressed their specific misconception.
Generative AI For Visualizing Highway Construction Hazards Through Synthetic Images And Temporal Sequences
Trevor Neece, Mason Smetana, and Lev Khazanovich. (05/2026). arXiv.
Single-pass images achieved 81.1% educational acceptability, with fidelity and alignment scores of 4.14 out of 5 and 4.07 out of 5, respectively.
Dipser: A Dataset For In-Person Student Engagement Recognition In The Wild
Luis Marquez-Carpintero, Sergio Suescun-Ferrandiz, Carolina Lorenzo Álvarez, Jorge Fernandez-Herrero, Diego Viejo, Rosabel Roig-Vila, Miguel Cazorla. (05/2026). arXiv.
The DIPSER dataset compiled 1,311,761 uncompressed images across 3,080.73 minutes of in-person classroom instruction with synchronized smartwatch biometric and motion data.
Temperature And Persona Shape LLM Agent Consensus With Minimal Accuracy Gains In Qualitative Coding
Conrad Borchers, Bahar Shahrokhian, Francesco Balzan, Elham Tajik, Sreecharan Sankaranarayanan, Sebastian Simon. (05/2026). arXiv.
Multi-agent consensus coding did not consistently improve accuracy over a single agent, and in five of six models, it slightly reduced overall agreement with human coders.
First, Do No Harm: AI Supervisor Scaffolds Novice Growth In Counselor Education
Chen Xu, Zhenyu Lyu, Tian Lan, Yang Yi, Yu Ji, Yu Ji, Jian Shen, Zhihua Wang, Leyang Cui, Jieshuo Zhang, Xiaohua Wan, Qunxi Dong, Minqiang Yang, Juan Wang, Xiuling Liu, Bin Hu. (05/2026). arXiv.
Fine-tuning Qwen3-14B on the ETHICSCAFF dataset achieved a 94.37% F1-score in classifying violated principles, outperforming both its base model and GPT-4o with retrieval-augmented generation.
From Surface Learning To Deep Understanding: A Grounded AI Tutoring System For Moodle
Anna Ostrowska, Michał Kukla, Gabriela Majstrak, Jan Opala, Sebastian Pergała, Jan Skwarek, Anna Wróblewska. (05/2026). arXiv.
Technical evaluation showed Gemini Flash 2.0 achieved high faithfulness up to 0.97 in STEM subjects and 0.98 in humanities when properly configured.
Cognitive Twins: Investigating Personalized Thinking Model Building And Its Performance Enhancement With Human-In-The-Loop
Wu-Yuin Hwang, Nur Alif Ilyasa, Muhammad Irfan Luthfi, and Yuniar Indrihapsari. (05/2026). arXiv.
The model achieved an overall accuracy F1 score of 74.57% before human refinement and 75.48% after human refinement, exceeding the predefined 70% threshold.
Lata: A Drop-In, Ferpa-Compliant Local-LLM Autograder For Upper-Division STEM Coursework
Jesse A. Rodríguez. (05/2026). arXiv.
The local autograder processed submissions in 1 to 3 minutes each, with an instructor-confirmed grading-error rate of roughly 0.02% to 0.04% per rubric line item.
Automatic Reflection Level Classification In Hungarian Student Essays
Zsolt Csibi, Kristian Fenech, Mónika Sándor, Mónika Serfőző, Kinga Gyöngy. (05/2026). arXiv.
A classical model using SMOTEBoost and embeddings achieved the highest overall score of 71.76%, slightly outperforming transformer models on aggregate metrics.
Investigating The Effects Of Different Levels Of User Control In An Interactive Educational Recommender System
Qurat Ul Ain, Mohamed Amine Chatti, William Kana Tsoplefack, Rawaa Ala Trash, Shoeb Joarder. (05/2026). arXiv.
Perceived control was the only outcome significantly affected across control levels (p = .033), with input control scoring significantly higher than output control.
Math Education Digital Shadows For Facilitating Learning With LLMs: Math Performance, Anxiety And Confidence In Simulated Students And Ais
Naomi Esposito, Anthony Tricarico, Luisa Porzio, Ali Aghazadeh Ardebili, and Massimo Stella. (04/2026). arXiv.
Several models displayed myopic overconfidence, with confidence scores exceeding 0.90 despite actual problem-solving accuracies remaining near 0.55.
What Don'T You Understand? Using Large Language Models To Identify And Characterize Student Misconceptions About Challenging Topics
Michael J. Parker, Maria Guadalupe Zavala-Cerna. (04/2026). arXiv.
Subject matter experts rated 93.8% of AI-identified student misconceptions as excellent across evaluated immunology and physiology questions, with no observed hallucinations.
From Test-Taking To Cognitive Scaffolding: A Pedagogical Diagnostic Benchmark For LLMs On English Standardized Tests
Luoxi Tang, Tharunya Sundar, Yuqiao Meng, Shuai Yang, Ankita Patra, Lakshmi Manohar Chippada, Jiqian Zhao, Yi Li, Weicheng Ma, Zhaohan Xi. (04/2026). arXiv.
Models frequently succeeded at initial problem formulation (up to 97% accuracy) but suffered sharp performance drops during intermediate constraint integration and execution steps.
Maic-Ui: Making Interactive Courseware With Generative Ui
Shangqing Tu, Yanjia Li, Keyu Chen, Sichen Zhang, Jifan Yu, Daniel Zhang-Li, Lei Hou, Juanzi Li, Yu Zhang, Huiqin Liu. (04/2026). arXiv.
In a high school deployment, the pilot class achieved 9.21-point average gains in STEM subjects compared to -2.32 points across other classes.
Applications Of The Transformer Architecture In AI-Assisted English Reading Comprehension
Ping Li. (04/2026). arXiv.
The interpretable transformer achieved higher accuracy (0.86) and macro F1 (0.84) than baseline BERT and RoBERTa models, alongside a 0.72 attention alignment score.
