Toward Inclusive Low-Code Development: Detecting Accessibility Issues in User Reviews

2025-04-27

AI Summary

Low-code platforms heavily rely on graphical user interfaces, posing significant accessibility barriers for users with visual impairments such as color blindness or low vision. To address this, we propose the first automated accessibility issue identification method tailored to low-code development: a hybrid detection framework that jointly leverages dual Transformer-based semantic modeling (BERT/RoBERTa) and domain-specific keyword rules, balancing deep semantic understanding with robustness under low-resource conditions. Evaluated on a manually annotated, fine-grained dataset of low-code accessibility feedback, our approach achieves 78% accuracy and 78% F1-score—substantially outperforming single-model baselines. This work delivers a practical, deployable feedback analysis tool for auditing and improving low-code platform accessibility, thereby advancing inclusive low-code development practices.

Technology Category

• Humans and AI: AI for Accessibility
• Natural Language Processing: Safety and Robustness
• Computer Vision: Low Level & Physics-based Vision

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Low-code applications are gaining popularity across various fields, enabling non-developers to participate in the software development process. However, due to the strong reliance on graphical user interfaces, they may unintentionally exclude users with visual impairments, such as color blindness and low vision. This paper investigates the accessibility issues users report when using low-code applications. We construct a comprehensive dataset of low-code application reviews, consisting of accessibility-related reviews and non-accessibility-related reviews. We then design and implement a complex model to identify whether a review contains an accessibility-related issue, combining two state-of-the-art Transformers-based models and a traditional keyword-based system. Our proposed hybrid model achieves an accuracy and F1-score of 78% in detecting accessibility-related issues.

Problem

• Detecting accessibility issues in low-code applications
• Addressing exclusion of visually impaired users in low-code development
• Identifying accessibility-related reviews using hybrid AI models

Innovation

• Combines Transformers and keyword-based models
• Detects accessibility issues in user reviews
• Achieves 78% accuracy with hybrid approach