Topic-Based Clustering and Social Network Analysis of News Headlines Using LDA and DBSCAN

Abstract

The rapid growth of online news has made it difficult for readers to find information that suits their interests and understand current issues in society. Therefore, automatic clustering is needed to facilitate efficient information analysis. The purpose of this study is to apply Latent Dirichlet Allocation (LDA) to extract topic representations from news headlines and evaluate its effectiveness for clustering using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) method. The data used are only news headlines from three online media outlets: detik.com, kompas.com, and CNNIndonesia. The research stages include data preprocessing, text weighting, vector representation, LDA application for topic formation, and DBSCAN for density-based cluster formation. Model parameters are determined through experiments to obtain the optimal number of topics and the appropriate epsilon and MinPts values. Evaluation is carried out using a coherence score for topic quality and a silhouette score for cluster quality. In addition, Social Network Analysis (SNA) is used to analyze the relationship structure between news headlines within each cluster. The results of the study show that LDA is effective in producing topic representations, DBSCAN is able to cluster data, and SNA provides an understanding of the relationship structure within the cluster.