Unsupervised Text Mining of Employee Feedback for Identifying Organizational Strengths and Improvement Areas

Abstrak

Employee feedback provides rich signals about organizational performance, yet its free-text format makes systematic analysis at scale difficult. This study proposes an unsupervised text mining workflow in Orange Data Mining to extract actionable themes from continuous employee comments by separating two semantic polarities: strength feedback (“What went well?”) and improvement feedback (“What could be improved?”). After cleaning and Indonesian-language preprocessing (Sastrawi stemming, custom stopwords), 3,406 strength and 3,172 improvement entries were represented using TF–IDF. Improvement feedback was clustered using K-Means and assessed with silhouette-based validation, while both feedback types were explored using LDA topic modeling supported by topic coherence checks for interpretability. The results reveal recurring organizational themes related to goal execution and performance, supervision, communication/coordination, and motivation, with notable vocabulary overlap between strengths and areas for improvement. Scientifically, this work demonstrates how polarity-aware unsupervised analytics improves interpretability compared to treating feedback as a single corpus, and practically, it provides a scalable way for managers to transform unstructured feedback into structured insights for targeted improvement initiatives.