---

🎓 GenAI-EdSent: Sentiment-Driven Evaluation of AI Educational Apps

Unveiling User Perceptions in the Generative AI Era Through Large-Scale Sentiment Analysis

---

📖 Overview

The integration of Generative AI into education has sparked a digital transformation in e-teaching, yet user perceptions of AI educational apps remain critically underexplored. This research project bridges that gap through a comprehensive sentiment-driven evaluation of user reviews from top AI educational apps on the Google Play Store.

This dataset accompanies the paper Unveiling User Perceptions in the Generative AI Era: A Sentiment-Driven Evaluation of AI Educational Apps' Role in Digital Transformation of E-Teaching.

Code: https://github.com/erfan-nourbakhsh/GenAI-EdSent

🎯 Research Objectives

1. Quantify sentiment distributions and distill key positive/negative themes across app categories
2. Compare performance trends among different app types (homework helpers, math solvers, LMS, etc.)
3. Propose future directions for AI educational ecosystems with hybrid AI-human models

📄 Associated Paper

Title: Unveiling User Perceptions in the Generative AI Era: A Sentiment-Driven Evaluation of AI Educational Apps' Role in Digital Transformation of E-Teaching

Authors: Adeleh Mazaheriyan (Islamic Azad University) & Erfan Nourbakhsh (University of Isfahan)

Abstract: This study performs a sentiment-driven evaluation of user reviews from 22 top AI ed-apps on the Google Play Store to assess efficacy, challenges, and pedagogical implications. Our pipeline leverages RoBERTa for binary sentiment classification, GPT-4o for key point extraction, and GPT-5 for synthesizing top positive/negative themes. Results reveal predominantly positive sentiments, with homework apps leading (e.g., Edu AI: 95.9% positive) while specialized LMS/language apps lag due to stability issues.

---

📊 Research at a Glance

📱 Apps Analyzed	💬 Reviews Processed	🤖 AI Models Used	📚 Categories
22	481,000+	3 (RoBERTa, GPT-4o, GPT-5)	7

🔥 Quick Stats

🎯 96.0% positive sentiment for top performer (Edu AI) ⚡ 481K training reviews • 481K validation samples 📈 4.43M total ratings across all apps 🏆 85% average positive sentiment for homework helpers ⚠️ 21.8% positive sentiment for lowest performer (Teacher AI)

---

🔍 Key Findings

📊 Sentiment Distribution Highlights

• Homework Helpers dominate with ~85% positive sentiment

  • 🥇 Edu AI: 96.0% positive (accuracy, speed, personalization)
  • 🥈 Answer.AI: 92.7% positive (24/7 tutoring, step-by-step explanations)

• Math-Focused Solvers show strong performance (~80% positive)

  • Users praise problem-solving efficiency and photo recognition

• Language/LMS Apps lag behind (20-40% positive)

  • ⚠️ Teacher AI: 21.8% positive (instability, limited features)
  • Issues: crashes, paywalls, feature gaps

🌟 Top Positive Themes

1. ⚡ Efficiency & Speed - Quick solutions for homework and brainstorming
2. 🎯 Personalized Learning - Step-by-step explanations tailored to student needs
3. 🎮 Engagement - Gamification and community features boost motivation
4. 🌍 Accessibility - Democratizing education for under-resourced areas
5. 🤝 Multi-Subject Support - Versatile tools across STEM and humanities

⚠️ Top Negative Themes

1. 💰 Aggressive Monetization - Restrictive paywalls limiting free features
2. ❌ Inaccuracies - Wrong answers eroding trust, especially in specialized domains
3. 📺 Excessive Ads - Disrupting learning flow
4. 🐛 Technical Glitches - Crashes, slow loading, notation failures
5. ⚖️ Equity Concerns - Digital divide and over-reliance risks

---

🛠️ Methodology

Our systematic pipeline combines web scraping, transformer-based NLP, and large language models to analyze authentic user feedback at scale.

Figure 1: Multi-stage workflow from data collection to theme synthesis

🔄 Pipeline Stages

1️⃣ App Selection & Categorization

• Selected 22 prominent AI education apps based on ratings, downloads, and GenAI integration
• Categorized into 7 functional types with overlaps for multifunctional designs:
  • 📝 AI Quiz & Question Generators
  • 🎒 All-in-One Study Companions
  • ✏️ Homework Helpers
  • 🔢 Math-Focused Solvers & Specialized Tools
  • 📄 Document/Content Tools
  • 🏫 Learning Management Systems (LMS)
  • 🌐 Language Learning Apps

2️⃣ Data Collection

• Web scraping from Google Play Store using Python
• Collected app metadata + verbatim user reviews (up to November 2025)
• Dataset: Tens of thousands to millions of reviews per app

3️⃣ Sentiment Analysis & Theme Extraction

Stage	Model	Purpose
Binary Classification	RoBERTa	Transformer-based sentiment labeling (positive/negative)
Key Point Extraction	GPT-4o	Distill recurring themes and pain points
Theme Synthesis	GPT-5	Generate top 5 positive/negative summaries

4️⃣ Aggregation & Trend Analysis

• Computed sentiment percentages per app and category
• Enabled cross-app performance comparisons

---

📈 Results

Figure 2: Positive vs. negative sentiment percentages across 22 AI educational apps

🏆 Top Performers

Rank	App	Category	Positive %
1	Edu AI	Homework Helper	96.0%
2	Answer.AI	Multi-Tool	92.7%
3	Question.AI	Chatbot/Math	87.9%
4	Homework AI	Math/Essay	85.6%
5	Help AI	Homework Helper	85.2%

⚠️ Areas for Improvement

Rank	App	Category	Positive %	Key Issues
22	Teacher AI	Language	21.8%	Instability, limited features
21	Tutor AI	Learning Assistant	41.7%	Narrow functionality
20	Blackboard	LMS	53.3%	Technical glitches

📊 Category-Level Insights

• Homework Helpers: Average 85% positive → Strong personalization and speed
• Math Solvers: Average 80% positive → Photo recognition praised, notation failures criticized
• LMS/Language Apps: Average 35% positive → Urgent need for stability improvements

---

💡 Future Directions

🔮 Proposed Innovations

1. Hybrid AI-Human Models 🤝

   • Combine app strengths (real-time assistance) with teacher oversight
   • Mitigate risks like inaccuracies and over-dependency

2. VR/AR Integration 🥽

   • Immersive learning experiences (virtual labs, interactive simulations)
   • Address gaps in engagement and multimodal inputs

3. Ethical AI Roadmap ⚖️

   • For Developers: Adaptive personalization, plagiarism detection, equitable monetization
   • For Policymakers: Mandated free tiers, accuracy standards, data privacy protections

---

📊 Dataset

The full dataset containing 481,000+ app reviews from 22 AI educational apps is publicly available on Hugging Face:

🔗 GenAI-EdSent Dataset on Hugging Face

The dataset includes:

• App Information: Metadata for all 22 applications (ratings, descriptions, install counts)
• User Reviews: Complete review texts with scores, timestamps, and sentiment labels
• Sentiment Analysis Results: Binary classification outputs from RoBERTa
• Extracted Insights: Key positive/negative themes per application

📁 Repository Data Files

This repository includes curated analysis outputs:

1. App_Ratings.json - Application Metadata

Contains Google Play Store ratings and summaries for all 22 apps:

📋 View Sample Data

{
  "name": "Edu AI - AI Homework Helper",
  "summary": "AI Homework Helper - Math, Physics, Chemical, etc.",
  "app_score": 4.55
},
{
  "name": "Answer.AI - Your AI tutor",
  "summary": "Scan and Get Instant Answers on Your Phone",
  "app_score": 4.73
}

Top Rated Apps:

• 🥇 Studocu: 4.86/5 ⭐
• 🥈 Gauth: 4.77/5 ⭐
• 🥉 Kahoot!: 4.74/5 ⭐

Lowest Rated Apps:

• ⚠️ Teacher AI: 2.29/5 ⭐
• ⚠️ Blackboard: 3.39/5 ⭐
• ⚠️ Tutor AI: 3.40/5 ⭐

2. Top_5_Points.json - Sentiment Themes Per App

Contains the top 5 positive and negative points extracted for each application using GPT-4o and GPT-5:

📋 View Sample Analysis (Edu AI)

Positive Points:

1. ✅ The app enables users to complete homework tasks effectively
2. ✅ The app delivers perfect-quality assignment outputs that meet high standards

Negative Points:

1. ❌ The app returns answers in Spanish instead of matching the user's preferred language
2. ❌ Text is displayed in a font size that is too small to read comfortably

📋 View Sample Analysis (Answer.AI)

Positive Points:

1. ✅ Reliably helps students complete homework across subjects and grade levels
2. ✅ Consistently delivers accurate, correct answers
3. ✅ Explains solutions step by step with reasoning
4. ✅ Provides answers very quickly with fast response times
5. ✅ Easy to use with intuitive interface and camera/scan input

Negative Points:

1. ❌ Frequently provides incorrect or incomplete answers (especially math/graphs)
2. ❌ Core features locked behind expensive paywall
3. ❌ Restrictive energy/points system limits free usage
4. ❌ App instability with crashes and broken features
5. ❌ Unreliable camera/scan and image recognition

Key Insights from Theme Analysis:

• Most Praised: Homework efficiency, step-by-step explanations, speed
• Most Criticized: Paywalls, accuracy issues, technical instability
• Category Trends: Homework helpers excel; LMS/language apps struggle

---

🚀 Installation

Prerequisites

Python 3.8+
pip

Setup

# Clone the repository
git clone https://github.com/erfan-nourbakhsh/GenAI-EdSent.git
cd GenAI-EdSent

# Install dependencies
pip install -r requirements.txt

Required Libraries

• transformers (RoBERTa model)
• openai (GPT API access)
• beautifulsoup4 / selenium (web scraping)
• pandas, numpy (data processing)
• matplotlib, seaborn (visualization)

---

🔬 Sample Usage

1. Sentiment Classification

# Run RoBERTa-based sentiment analysis
python Positive_Negative_Points_Classifier.py

2. Extract Key Points

# Use GPT-4o to extract themes from reviews
python Positive_Negative_Points_Generator.py

3. Generate Top Themes

# Synthesize top 5 positive/negative points with GPT-5
python Top_5_Points_Generator.py

---

📚 Citation

If you use this work in your research, please cite:

@article{mazaherian2025unveiling,
  title={Unveiling User Perceptions in the Generative AI Era: A Sentiment-Driven Evaluation of AI Educational Apps' Role in Digital Transformation of e-Teaching},
  author={Mazaherian, Adeleh and Nourbakhsh, Erfan},
  journal={arXiv preprint arXiv:2512.11934},
  year={2025}
}

---

👥 Meet the Researchers

🌟 Behind the Research

A collaborative effort bridging educational theory and AI technology to understand how generative AI is reshaping learning experiences worldwide.

🎓 Adeleh Mazaheriyan Education Researcher Department of Education Islamic Azad University, Isfahan, Iran Specializes in pedagogical evaluation and digital transformation in e-teaching, bringing educational theory perspective to AI assessment. 📧 adelehmazaheriyan@gmail.com

🤖 Erfan Nourbakhsh AI Researcher & Developer Artificial Intelligence Department University of Isfahan, Iran Expert in NLP, sentiment analysis, and LLM applications, developing the technical pipeline for large-scale review analysis. 📧 erfannourbakhsh2001@gmail.com

🤝 Interdisciplinary Collaboration

This research exemplifies the power of interdisciplinary collaboration, combining:

• 📚 Educational Theory → Understanding pedagogical implications
• 🧠 AI/ML Technology → Analyzing 960K+ reviews at scale
• 📊 Data Science → Extracting actionable insights for stakeholders

Together, we're working to ensure AI in education serves all learners equitably.

---

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss proposed changes.

---

📄 License

This dataset is licensed under the CC-BY-4.0 License. The code is licensed under the MIT License.

---

🌟 Acknowledgments

We extend our gratitude to:

• Microsoft Education for the 2025 AI in Education Report
• RAND Corporation for K-12 AI adoption insights and teacher engagement studies
• OpenAI & Hugging Face for providing access to state-of-the-art language models (GPT-4o, GPT-5, RoBERTa)
• Google Play Store developers and the educational AI community
• Thousands of users whose reviews made this research possible

---

📞 Contact & Collaboration

We welcome questions, feedback, and collaboration opportunities!

For Research Inquiries:

• 📧 Email: erfannourbakhsh2001@gmail.com or adelehmazaheriyan@gmail.com
• 🤗 Dataset: Hugging Face

Interested in:

• Collaborative research on AI in education?
• Extending this analysis to other platforms (Apple App Store, web apps)?
• Developing predictive models for app success?
• Contributing to the codebase?

We'd love to hear from you! 💬

---

⭐ If you find this research useful, please star the repository! ⭐

Advancing equitable, innovative e-teaching through user-driven AI insights

---

Made with ❤️ for the future of education