Perbandingan Aplikasi Algoritma Kernel K-Means pada Graf Bipartit dan K-Means pada Matriks Dokumen- Istilah dalam Dataset Penelitian Covid-19 RISTEKBRIN

Abstract

The outbreak concerning the Covid-19 case in Indonesia has raised various topics of research carried out by researchers in diverse fields and from many institutions. Based on data compiled by the Sinta Ristekbrin portal, 351 research topics have been uploaded by researchers. This study is aimed to analyze and map research topics that are being and/or have been conducted during the period of the Covid-19 pandemic in Indonesia. Analysis and mapping are accomplished by applying the kernel k-means algorithm for document clustering based on bipartite graphs and k-means on document term matrix. Ristekbrin s Covid-19 research dataset is modeled as a bipartite graph between terms and documents. Furthermore, the similarity score is calculated using the kernel method. The kernel matrix value that represents the similarity score between documents is used as input for the kernel k-means clustering algorithm, which provides the results of mapping research topics. As comparison, we applied original k-means algorithm on the document-term matrix of the dataset. From these two methods, the clustering results were validated using Dunn index as an internal validation. The Dunn index was used because the dataset did not provide initial information regarding the label or name of each clusters..The comparison Dunn index shows that the kernel k-means algorithm outperforms than the k-means algorithm. This study is expected to provide additional information that supports government programs in accelerating the handling of Covid-19 in Indonesia..