Quantifying the Credibility of E-Learning Systems Using the BERT Model

Abstract

Trust has become crucial in a digitized world for the sustainability of online platforms. As the web interfaces have been playing a vital role in day-to-day activities, ethical design has become essential to protect user autonomy and promote informed decision-making by them. Dark patterns are deceptive design strategies that can harm users’ trust and are dangerous, especially in the e-learning environment. This paper presents a refined Bidirectional Encoder Representations of Transformers (BERT) classifier to automatically detect dark patterns on educational online systems. The framework starts with web scraping of the digital interface followed by organized preprocessing, such as content extraction, text cleansing, normalization, and tokenization. An algorithm to calculate a Credibility Index (CI) of e-learning systems is proposed based on the frequency and the severity of perceived dark patterns. The online systems are categorized into one of the three threat levels—Safe, Moderate, or Critical, which gives clear indication to users regarding the trustworthiness of the site. Using the proposed framework, a customized educational website called SkillNest was developed to predict user trust. It was classified as Moderate due to its CI value of 0.64. This work may help developers in enhancing transparency and trust in educational technology by reducing the manipulative practices for developing e-learning systems.