Classification of public complaint report types on social crimes using a chatbot for law enforcement agencies

Abstract

Social crime is a complex problem that occurs every day and requires a quick response. The large number of reports with language variations makes the manual classification process difficult. This research aims to develop an AI-based chatbot to classify types of social crime reports automatically using the IndoBERT model. Data was obtained from East Denpasar Police, LAPOR website, and X social media. The initial data set of 250 reports was augmented to 6,250 data using synonym augmentation technique. The data was then divided into 70:20:10 training scenarios to produce the best model. The evaluation showed high performance with accuracy 0.999200, precision 0.999203, recall 0.999200, and F1-score 0.999200. Validation was also done through confusion matrix and accuracy-loss graph. The chatbot is able to receive reports from the public and classify them into five main categories, namely theft, maltreatment, embezzlement, domestic violence, and murder. The results show that IndoBERT is effective in understanding and classifying Indonesian text reports accurately. The system is expected to assist law enforcement agencies in improving efficiency and speed in handling community reports as well as supporting the digitisation of the social crime complaint process.