Lexicon-enhanced aspect-based sentiment analysis for evaluating Indonesian learning applications

Abstract

This study evaluates user perceptions of two Indonesian learning applications, Ruangguru and Pahamify, through lexicon-enhanced aspect-based sentiment analysis of 14,768 Google Play reviews published between 2020 and 2025. EduAppLex-ID, a pilot domain lexicon containing 46 multiword expressions, was developed as an extension of the Indonesian Sentiment Lexicon. Overall sentiment toward both applications was favorable, with Pahamify showing a small but statistically significant advantage. Pedagogical quality received the most positive evaluation, whereas system quality was evaluated negatively for both applications. Monetization produced the largest difference between the applications. A lexicon audit identified eighteen high-frequency entries whose polarity assignments were unsuitable for the educational-application domain. These entries accounted for approximately one-fifth of the corpus’s sentiment mass, and their removal increased the observed between-application effect size. Because no human-annotated validation dataset was available, interpret the findings as exploratory evidence. The results demonstrate the importance of domain-specific lexicon calibration and aspect-level analysis when examining Indonesian educational-application reviews.