Peningkatan Performa Clustering pada Large Text Dataset Menggunakan Stamantic Spherical K-Means

Abstract

Document clustering can’t avoid the problem of high dimensionality, which can be overcome by combining the advantage of statistical and semantic features. This study aims to determine the performance of clustering with the Stamantic (statistical and semantic) feature extraction technique compared to the several Bag Of Words Model (Bag Of All Word, Bag Of Noun, Bag Of Noun and Adjective) as well as a comparison between Spherical K-Means and K-Means++ clustering algorithm. Stamantic feature extraction use the Wordnet (Wn) database to form semantic features, while statistical features are obtained from TF-IDF (Term Frequency Inverse Document Frequency) word sense. Evaluation were carried out on clustering results with several metrics. The highest Silhouette score is 0.162213 on the BONA feature from Pubmed dataset which clustered with K-Means++ algorithm. The highest Purity score around 0.949643 on the BONA feature from Scopus dataset with Spherical K-Means algorithm. The highest AMI (Adjusted Mutual Information) score is 0.880835 on the BONA feature from Scopus dataset with Spherical K-Means clustering algorithm. The test results show that the Stamantic feature loses to all BOW features. Due to the loss data information from the effect of using Wn library and disambiguation process which inappropriate.