IMPLEMENTASI K-MEANS DAN ANALISIS SENTIMEN KRITIK SARAN BERBASIS NLP PADA DATA MONEV BBPSDMP KOMINFO MAKASSAR

Abstract

Manual analysis of large-scale and unstructured textual feedback data is often inefficient and subjective, thereby hindering data-driven decision-making. This study aims to design and implement an integrated analytical workflow to automatically filter, cluster, and classify feedback data consisting of criticisms and suggestions. The research employs a hybrid approach that begins with TF-IDF-based data filtering, followed by dimensionality reduction using Latent Semantic Analysis (LSA), and topic clustering through K-Means clustering optimized with the Silhouette Score. The resulting cluster labels are then used as training data to build a Multinomial Naive Bayes classification model. The results show that this workflow successfully identified two main thematic clusters, namely "Criticism and Expectations" and "Suggestions and Compliments", and the classification model achieved an overall accuracy of 91%. Although class imbalance affected the recall of the minority class (47%), the model demonstrated high precision (95%) for that class. It is concluded that this hybrid approach effectively transforms raw data into structured insights, and utilizing clustering results as training data is an efficient strategy for automating feedback categorization, providing a reliable tool for institutional analysis.