Public Complaint Text Classification in the Wargaku Application Using Natural Language Processing

Abstract

The Wargaku application is utilized by Surabaya residents to submit complaints concerning population administration services. With the increasing number of complaints, manual categorization becomes inefficient and susceptible to errors. This research aims to create an automatic classification system utilizing Natural Language Processing (NLP) and machine learning techniques. The dataset comprises 2,303 complaints divided into 18 categories. During preprocessing, text data was converted into numerical form using the Term Frequency–Inverse Document Frequency (TF-IDF) approach. Three machine learning models were tested: Support Vector Machine (SVM), Random Forest (RF), and Neural Network (NN), with evaluations based on accuracy and F1-score. Hyperparameter tuning was applied to enhance model performance. The SVM model yielded the best outcome with a training-to-testing data ratio of 85:15, resulting in a training accuracy of 93.96%, an F1-score of 96.08%, and a testing F1-score of 94.15%. This model was deployed in a web-based application via Streamlit to automatically categorize public complaints. The findings confirm the effectiveness of combining NLP and SVM in improving the efficiency of digital public service systems.