IMPLEMENTASI NATURAL LANGUAGE PROCESSING (NLP) UNTUK KLASIFIKASI PENGADUAN MASYARAKAT PADA APLIKASI WARGAKU DI DISPENDUKCAPIL SURABAYA

IMPLEMENTATION OF NATURAL LANGUAGE PROCESSING (NLP) FOR PUBLIC COMPLAINT CLASSIFICATION IN THE WARGAKU APPLICATION AT DISPENDUKCAPIL SURABAYA

Aplikasi Wargaku digunakan oleh masyarakat Surabaya untuk menyampaikan pengaduan layanan kependudukan. Seiring meningkatnya jumlah pengaduan, klasifikasi manual menjadi tidak efisien dan rawan kesalahan. Penelitian ini bertujuan mengembangkan sistem klasifikasi otomatis berbasis Natural Language Processing (NLP) dan machine learning. Data yang digunakan terdiri dari 2.303 pengaduan dalam 18 kategori. Setelah dilakukan preprocessing, data diubah menjadi representasi numerik menggunakan metode Term Frequency–Inverse Document Frequency (TF-IDF). Tiga algoritma yang dibandingkan adalah Support Vector Machine (SVM), Random Forest (RF), dan Neural Network (NN), dengan evaluasi menggunakan metrik accuracy dan F1-score. Proses hyperparameter tuning dilakukan untuk mengoptimalkan performa model. Hasil terbaik diperoleh dari SVM pada rasio data 85:15, dengan akurasi 93,96%, F1-score latih 96,08%, dan F1-score uji 94,15%. Model SVM terbaik diimplementasikan dalam aplikasi web berbasis Streamlit untuk melakukan klasifikasi pengaduan secara otomatis. Hasil penelitian ini menunjukkan efektivitas pendekatan NLP dan SVM dalam meningkatkan efisiensi layanan publik digital.

The Wargaku application is used by the people of Surabaya to submit complaints related to population administration services. As the number of complaints increases, manual classification becomes inefficient and prone to errors. This study aims to develop an automatic classification system based on Natural Language Processing (NLP) and machine learning. The dataset consists of 2,303 complaints categorized into 18 service types. After preprocessing, the text data was transformed into numerical representations using the Term Frequency–Inverse Document Frequency (TF-IDF) method. Three classification algorithms were compared: Support Vector Machine (SVM), Random Forest (RF), and Neural Network (NN), evaluated using accuracy and F1-score metrics. Hyperparameter tuning was conducted to optimize model performance. The best results were obtained from the SVM model with an 85:15 train-test ratio, achieving an accuracy of 93.96%, training F1-score of 96.08%, and testing F1-score of 94.15%. The best SVM model was implemented into a web-based application using Streamlit to automatically classify public complaints. The results of this study demonstrate the effectiveness of the NLP and SVM approach in enhancing the efficiency of digital public services.