Analyzing Topic Trends in HRIS User Feedback Using BERTopic for Data-Driven System Evaluation

Abstract

This study examines temporal changes in Human Resource Information System (HRIS) user feedback and differences between complaints and suggestions. We audited 5,228 raw records and analyzed 4,509 anonymized, deduplicated records dated 7 December 2022 to 5 October 2026. We modeled topics with BERTopic using multilingual sentence embeddings. Category labels were normalized into complaint (3,136) and suggestion (1,373). The study identified 22 topics and 1,022 outliers (22.7%). Its largest cluster contained 1,729 records (38.3%) and mixed clock and attendance terms, so it was not assigned a narrow issue label. Among more interpretable clusters, inability to record attendance (topic 1) increased from 5.8% of feedback in 2023 to 9.0% in 2025, while face detection (topic 10) increased from 0.7% to 3.8%; slow loading (topic 2) declined from 3.3% to 1.3%. These descriptive shares suggest issues for review, not evidence that a product change caused them. The broad largest cluster and lack of human label validation limit interpretation.