K-Means and K-Medoids for Indonesian Text Summarization

Abstract

The purpose of this study is to build automated summation tools, especially in grouping methods such as K-Means and K-Medoids. Finding the best method between the two algorithms, this study focuses on comparing the two methods to summarize thesis report documents. This system is divided into Filtering, Tokenization, TF-IDF, Cosine Similarity, and Clustering. Based on 50 test documents, the average accuracy rate is 51.16% for K-Means and 63.35% for K-Medoids. K-Means has a smaller accuracy value than K-Medoids. The accuracy of the resulting K-Means also depends on the size and center of the initial cluster chosen. So, as the next stage of development, research needs to be done that compares the results of the combination of initial size and center cluster values for K-Means and continue with several other classifications.