Automatic Representative News Generation using On-Line Clustering

The increasing number of online news provider has produced large volume of news every day. The large volume can bring drawback in consuming information efficiently because some news contain similar contents but they have different titles that may appear. This paper presents a new system for automatically generating representative news using on-line clustering. The system allows the clustering to be dynamic with the features of centroid update and new cluster creation. Text mining is implemented to extract the news contents. The representative news is obtained from the closest distance to each centroid that calculated using Euclidean distance. For experimental study, we implement our system to 460 news in Bahasa Indonesia. The experiment performed 70.9% of precision ratio. The error is mainly caused by imprecise results from keyword extraction that generates only one or two keywords for an article. The distribution of centroid’s keywords also affects the clustering results.