Source-linked AI summary
Unveiling User Perceptions in the Generative AI Era: A Sentiment-Driven Evaluation of AI Educational Apps' Role in Digital Transformation of e-Teaching
Adeleh Mazaherian, Erfan Nourbakhsh
TL;DR
Evidence on real-world perceptions of AI education apps remains limited, despite their growing role in e-teaching. This study analyzes Google Play reviews using NLP and large language models, finding predominantly positive sentiment, especially for homework apps, alongside recurring user complaints.
Problem
Large-scale sentiment-driven evidence about real-world perceptions of AI education apps remains limited, while surveys and interviews offer less scalable insight.
Source: §II · p. 2Method
The study analyzes Google Play app reviews with an automated NLP and large-language-model pipeline and categorizes apps by function.
Source: §III · p. 3, A. App Selection and Categorization · p. 3Results
Most apps received predominantly positive reviews, with Edu AI at 95.9% positive and Teacher AI at 21.8% positive; homework and math tools averaged 85% positive sentiment.
Source: §IV · p. 4, §V · p. 5Takeaways & Limitations
The findings identify homework and math tools as a strong area for personalized learning while highlighting monetization and equity concerns that call for refinement.
Source: §V · p. 5
Abstract
from arXiv · showhide
The rapid integration of generative artificial intelligence into education has driven digital transformation in e-teaching, yet user perceptions of AI educational apps remain underexplored. This study performs a sentiment-driven evaluation of user reviews from top AI ed-apps on the Google Play Store to assess efficacy, challenges, and pedagogical implications. Our pipeline involved scraping app data and reviews, RoBERTa for binary sentiment classification, GPT-4o for key point extraction, and GPT-5 for synthesizing top positive/negative themes. Apps were categorized into seven types (e.g., homework helpers, math solvers, language tools), with overlaps reflecting multifunctional designs. Results indicate predominantly positive sentiments, with homework apps like Edu AI (95.9% positive) and Answer.AI (92.7%) leading in accuracy, speed, and personalization, while language/LMS apps (e.g., Teacher AI at 21.8% positive) lag due to instability and limited features. Positives emphasize efficiency in brainstorming, problem-solving, and engagement; negatives center on paywalls, inaccuracies, ads, and glitches. Trends show that homework helpers outperform specialized tools, highlighting AI's democratizing potential amid risks of dependency and inequity. The discussion proposes future ecosystems with hybrid AI-human models, VR/AR for immersive learning, and a roadmap for developers (adaptive personalization) and policymakers (monetization regulation for inclusivity). This underscores generative AI's role in advancing e-teaching by enabling ethical refinements that foster equitable, innovative environments. The full dataset is available here(https://github.com/erfan-nourbakhsh/GenAI-EdSent).
II. RELATED WORK
Prior work describes GenAI’s educational uses and adoption patterns while noting concerns about efficacy, integrity, and equity. This paper addresses a gap in scalable, sentiment-driven analysis of real-world AI education app reviews.
II. RELATED WORK
- Prior reviews describe GenAI applications including adaptive tutoring and automated assessment, while noting risks such as bias and limited long-term efficacy evidence.The cited K-12 review attributes the evidence gap to early adoption.Source: §II · p. 2
- Students prioritize task efficiencies, whereas educators emphasize pedagogical uses for diverse needs and require training for stronger integration.Reported student priorities include brainstorming (37%) and summarizing or quick answers (33% each); educators prioritize lesson ideas (31%) and lesson plans (29%).Source: §II · p. 2
- 66% of surveyed teachers were non-users of AI tools, while 34% were active users spanning sporadic, proactive, and regular engagement.The cited survey reports 44% aware but unused in teaching, 13% using AI outside teaching, and 9% unaware.Source: §II · p. 2
- Studies of higher-education use report benefits in writing and lesson design alongside concerns about integrity, misinformation, authenticity, and equity.Other cited work explores inquiry-based learning, teacher and student views, and convenience-driven student use.Source: §II · p. 2
- Large-scale sentiment-driven evaluations of AI education apps remain scarce, despite app-store reviews offering authentic, voluminous evidence beyond surveys and interviews.The paper applies review analysis to top Google Play apps and distills themes to inform future ecosystems.Source: §II · pp. 2–3
III. METHODOLOGY
The study uses a sentiment-driven NLP and large-language-model pipeline to evaluate user perceptions of AI education apps. Its workflow spans data collection through theme synthesis.
III. METHODOLOGY
- The study conducts a systematic, sentiment-driven evaluation of Google Play reviews to assess perceptions of AI-integrated education tools.The stated goal is to gather authentic, real-time feedback.Source: §III · p. 3
- The pipeline automates sentiment extraction and synthesis to scale analysis while minimizing manual biases.Source: §III · p. 3
- Figure 3 presents a multi-stage workflow running from data collection to theme synthesis.Source: §III · p. 3, Figure 3 · p. 3
A. App Selection and Categorization
The study selected prominent AI education apps and grouped them by function, allowing overlap for multifunctional designs. The app set spans homework support, quizzes, language practice, and learning-management functions.
A. App Selection and Categorization
- 22 prominent AI education apps were selected based on high ratings, download counts, and explicit generative AI integration.Source: A. App Selection and Categorization · p. 3
- The selected apps span homework and math assistance, quiz tools, language practice, and mobile LMS functions.Examples include Gauth’s step-by-step math and science help, Teacher AI’s language practice, and Blackboard’s course access and progress tracking.Source: A. App Selection and Categorization · p. 3
- Apps were categorized into seven functional types, with overlaps marking multifunctional designs.The categorization reflects demand for homework and math aids and identifies growth potential in LMS and language tools.Source: A. App Selection and Categorization · p. 3
B. Data Collection
The study scraped Google Play app metadata and verbatim user reviews to build a large corpus from popular AI education apps. Collection covered reviews through November 2025.
B. Data Collection
- Automated Python-based web scraping extracted Google Play app metadata, including descriptions and ratings, alongside verbatim user reviews.Source: B. Data Collection · p. 3
- The corpus included reviews through November 2025, ranging from tens of thousands to millions per app.Source: B. Data Collection · p. 3
- Selecting popular apps was intended to mirror real-world adoption and capture diverse educational experiences.Source: B. Data Collection · p. 3
C. Sentiment Analysis and Theme Extraction
The pipeline classifies review sentiment, extracts recurring points, synthesizes the most frequent themes, and aggregates sentiment percentages for comparisons across apps and categories.
Sentiment classification
- RoBERTa labeled reviews positive or negative, supporting measurement of satisfaction while capturing nuances such as sarcasm or mixed sentiments.The model is described as a transformer-based NLP framework pretrained on extensive social media data.Source: C. Sentiment Analysis and Theme… · p. 4
Key point extraction and theme synthesis
- GPT-4o extracted recurring themes and pain points from labeled reviews without fine-tuning, then GPT-5 summarized each app’s top five positive and negative points.GPT-4o prioritized frequency and contextual relevance; GPT-5 prioritized frequency across reviews and alignment with educational goals.Source: C. Sentiment Analysis and Theme… · p. 4
Aggregation and categorization
- Sentiment percentages were aggregated by app and category for cross-app comparisons; Table 1 marks apps appearing in multiple functional categories in red.The table caption identifies the red highlighting as indicating overlapping functionalities.Source: C. Sentiment Analysis and Theme… · p. 4, Table 1 · p. 4
IV. RESULTS
Review sentiment was predominantly positive, with homework helpers and multifunctional apps among the strongest performers, while language and LMS apps showed greater negativity.
Scope and app mix
- The review analysis covered 20 top AI educational apps and found a market emphasis on homework and math aids, with fewer dedicated tools in other areas.The paper links this pattern to demand for targeted, immediate assistance in e-learning.Source: §IV · p. 4
Overall sentiment
- 95.9% positive for Edu AI, 92.7% for Answer.AI, and 21.8% for Teacher AI; most apps exceeded 50% positive.Figure 4 orders the apps from highest to lowest positive sentiment.Source: Figure 4 · p. 4, §IV · p. 4
Review themes
- Praise emphasized efficiency, accuracy, personalization, quick problem-solving, and engaging customization, while complaints included incomplete answers, paywalls, ads, crashes, scanning problems, and billing errors.Lower-performing Tutor AI and Teacher AI were also criticized for outdated AI, poor speech recognition, and missing offline access.Source: §IV · pp. 4–5
Cross-app and category trends
- Homework helpers averaged >80% positive sentiment for speed and versatility; quiz generators were ~mid-70s, while language/LMS apps were <60% due to instability.The broader comparison reports ~85% positive for homework tools versus ~75% for specialized tools.Source: §IV · p. 5
V. DISCUSSION
The discussion frames review findings as both evidence of personalized-learning momentum and a basis for addressing adoption barriers through hybrid models, immersive technologies, and a practical roadmap.
Findings and implications
- Real-world reviews point to strong momentum in personalized learning, especially for homework and math tools, alongside monetization barriers, ads, crashes, and equity concerns.The discussion reports 85% positive sentiment for homework/math tools and notes that multifunctional designs drive retention while amplifying monetization complaints.Source: §V · p. 5
Hybrid AI-human models
- Hybrid AI-human models would pair real-time AI assistance with teacher oversight to mitigate inaccuracies and dependency, while positioning AI as support rather than replacement.The discussion proposes AI handling routine tasks so educators can focus on deeper interactions and critical thinking.Source: §V · p. 5
VR/AR opportunities
- VR/AR is proposed for immersive scenarios such as virtual labs and interactive quizzes, potentially addressing engagement and multimodal-input gaps in current apps.The discussion connects this proposal to persistent notation and speech-recognition failures in STEM and language domains.Source: §V · p. 5
Developer and policy roadmap
- The roadmap asks developers to improve personalization, collaboration, bias mitigation, learning supports, gamification, and equitable monetization, while policymakers address free tiers, multilingual support, accuracy, and privacy.The proposed measures aim to bridge digital divides and make e-teaching platforms more equitable and innovative.Source: §V · p. 5
VI. CONCLUSION
The conclusion finds positive feedback for homework assistance, personalization, and engagement, while identifying accuracy, monetization, and stability challenges to equitable adoption.
Conclusion
- Homework helpers averaged 85% positive reviews, with feedback emphasizing homework efficiency, personalization, and engagement.The conclusion also identifies inaccuracies, monetization barriers, and technical instability as challenges to equitable adoption.Source: §VI · p. 5
- The paper proposes hybrid AI-human models, VR/AR integration, adaptive developer features, and inclusive policymaking as directions for refining e-teaching ecosystems.The conclusion presents these proposals as ways to support ethical refinement and mitigate dependency and digital divides.Source: §VI · p. 5
- The authors call for future research on longitudinal impacts and cross-platform integrations to support high-quality, more widely accessible e-learning.Source: §VI · p. 5