Analysis of Hoax Information Characteristics and Clustering Based on Komdigi Clarification Data Using TF-IDF and K-Means

Abstract

This study aims to analyze the characteristics of hoax information and identify its thematic clusters in Indonesia based on clarification data from the Ministry of Communication and Digital Affairs (Komdigi), obtained through the Kaggle platform. A quantitative approach based on text mining was applied to 3,954 unique hoax-related articles that had undergone text preprocessing. Textual features were extracted using Term Frequency–Inverse Document Frequency (TF-IDF) with an n-gram range of (1,2), followed by clustering using the K-Means algorithm. Cluster quality was evaluated using the Silhouette Score based on cosine distance and the Davies–Bouldin Index to support the selection of the number of clusters. The results showed that the seven-cluster configuration (K=7) yielded a Silhouette Score of 0.0277. Cluster 5 was the largest, containing 1,600 articles (approximately 40.46%), which, based on the interpretation of dominant terms and representative articles, were primarily associated with financial fraud, social assistance, and phishing links. Meanwhile, Cluster 7 comprised 598 articles (15.12%), with themes related to national politics and figures holding positions of state authority. The evaluation results indicated limited separation between clusters, suggesting that the clustering results should be interpreted with caution. This study provides an overview of the thematic characteristics of hoax information within the analyzed dataset and may support the development of more targeted digital literacy policies and theme-based automated content filtering systems.