Classification of customer complaints on social media for e-commerce in Indonesia

Abstract

The e-commerce industry in Indonesia has experienced rapid growth, especially during the COVID-19 pandemic, which accelerated the shift to online platforms. The market is expected to grow by 105.5% from 2025 to 2030 due to increased internet and smartphone use. As e-commerce expands, companies must improve how they handle customer complaints to build trust and loyalty. Social media is a crucial channel for customer interactions, but it also includes non-complaint messages like positive comments, general questions, and spams that need to be filtered out. This research proposes a machine learning model to automatically classify social media interactions into complaints and non-complaints, focusing on Indonesian-language content. The modeling process utilized 10,600 data points collected from social media X. The best model, a bidirectional encoder representation from transformers (BERT) based classifier, achieved an F1-score of 98.3%. The McNemar test revealed significant performance differences between several models, with the BERT-based model outperforming others. This demonstrates that it is highly effective in distinguishing between complaints and non-complaints, making it a valuable tool for enhancing customer service in Indonesia's e-commerce sector.