Indonesian News and Article Clustering with K-Means++

• Stopwords obtained from https://github.com/masdevid/ID-Stopwords
• Dataset comes from private source

Guide

1. Install necessary library
2. Run preprocess.ipynb
3. Run 01.ipynb, 02.ipynb, 03.ipynb & 04.ipynb

Word Representation Parameter

Notebook	Word represenatation	min_df	norm (TF-IDF)	sublinear (TF-IDF)
01.ipynb	Bag Of Words (BOW)	100	-	-
02.ipynb	TF-IDF	25	l1	False
03.ipynb	TF-IDF	25	l2	False
04.ipynb	TF-IDF	5	l2	True

Supervised Benchmark Result

Notebook	Adjusted Rand	Homogeneity	Completeness	V Measure (Beta: 1.0)	Fowlkes-Mallows
01.ipynb	0.16832494978963927	0.3867110347229015	0.5536117350784411	0.42753439777044283	0.4217245815042002
02.ipynb	0.3583653763786496	0.48606249528892637	0.8130287205170411	0.6083988003735386	0.6210833495522297
03.ipynb	0.6217382880435744	0.717235296227637	0.7380120740605376	0.7274753686083797	0.7136882030375186
04.ipynb	0.9313637522720699	0.9170386227916734	0.9167906937890842	0.9169146415306973	0.9474406310870145

Unsupervised Benchmark Result

Notebook	Lowest inertia in training	Calinski-Harabasz	Silhouette (Sample Size 20% doc.)	Davies Bouldin
01.ipynb	13412085.957	676.2524626345348	0.04580253760601429	5.378978998734258
02.ipynb	957.967	669.9463630311741	0.028479348370840964	5.508398316924348
03.ipynb	39233.847	454.4597438515346	0.02209624498094567	7.831057435168252
04.ipynb	39628.660	349.04195674478404	0.01835547854087986	8.14052446235942