MADANI, Faiq and Adi, Kusworo and Farikhin, Farikhin (2024) PENGARUH DISTANCE METRIKS TERHADAP PERFORMANSI WEIGHTED K-NEAREST NEIGHBOR (WKNN) PADA ANALISIS SENTIMEN DI SOSIAL MEDIA. Masters thesis, UNIVERSITAS DIPONEGORO.

	Text Cover-1 (1).pdf Download (45kB)
	Text Cover.pdf Restricted to Repository staff only Download (466kB)
	Text Bab - 1.pdf Download (729kB)
	Text Bab - 2.pdf Download (1MB)
	Text Bab - 3.pdf Restricted to Repository staff only Download (1MB)
	Text Bab - 4.pdf Restricted to Repository staff only Download (2MB)
	Text Bab - 5.pdf Restricted to Repository staff only Download (718kB)
	Text Daftar Pustaka.pdf Download (596kB)

Abstract

Penelitian ini berisi tentang pembahasan mengenai pengaruh berbagai distance metric terhadap performansi algoritma Weighted K-Nearest Neighbor (WKNN)dalam klasifikasi analisis sentimen pada salah satu platform sosial media yaitu X. Penelitian ini didasarkan pada salah satu kelemahan pada metode K-Nearest Neighbor yang hanya memperhitungkan jarak tanpa mempertimbangkan bobot tetangga. Selain itu juga pada algoritma tersebut terdapat perhitungan distance metrics yang bervariasi dan memiliki perhitungan yang berbeda. Distance metricyang digunakan meliputi Euclidean, Cosine Similarity, Minkowski, dan Jaccard. Di dalam penelitian ini nantinya akan dapat mengklasifikasikan postingan pada platform X ini akan bersifat positif, negatif atupun netral. Hasil penelitian menunjukkan bahwa pemilihan distance metric berpengaruh signifikan terhadap kinerja WKNN. Euclidean distance memberikan performa terbaik dengan akurasi, presisi, recall, dan f1-score masing-masing sebesar 76%, diikuti oleh Cosine Similarity dengan performa sedikit lebih rendah, yakni 74% di semua metrik evaluasi. Minkowski menghasilkan akurasi 68% dan presisi 72%, sedangkanJaccard menunjukkan performa yang lebih stabil dengan hasil 70% di seluruh metrik evaluasi. Berdasarkan hasil komparasi ini, Euclidean distance terbukti menjadi metric yang paling optimal dalam klasifikasi meskipun Cosine Similaritydan Jaccard juga memberikan hasil yang kompetitif.
Kata Kunci : klasifikasi, analisis sentimen, weighted k-nearest neighbor, distance metrics

This research discusses the impact of various distance metrics on the performanceof the Weighted K-Nearest Neighbor (WKNN) algorithm in sentiment analysis on one of the social media platforms, X. This research is based on one of the weaknesses of the K-Nearest Neighbor method, which only considers distance without taking into account the weight of neighbors. Additionally, the algorithm involves various distance metrics, each with different calculations. The distance metrics used include Euclidean, Cosine Similarity, Minkowski, and Jaccard. This study aims to classify posts on the X platform as positive, negative, or neutral. The research results show that the choice of distance metric significantly affects the performance of WKNN. Euclidean distance delivers the best performance with an accuracy, precision, recall, and f1-score of 76%, followed by Cosine Similarity, which has slightly lower performance, achieving 74% in all metrics. Minkowski produces an accuracy of 68% and precision of 72%, while Jaccard demonstrates more stable performance with 70% across all evaluation metrics. Based on this comparison, Euclidean distance proves to be the most optimal metric for classification tasks, although Cosine and Jaccard also deliver competitive results.
Keywords : classification, sentiment analysis, weighted k-nearest neighbor, distance metrics

Actions (login required)

View Item