ABSTRAKSI: Collaborative Filtering (CF) merupakan salah satu metode pada sistem pemberi rekomendasi yang dipelajari untuk mengatasi masalah “information overload”. Collaborative filtering memberikan rekomendasi berdasarkan kedekatan karakteristik user dalam memberikan informasi.
Tugas akhir ini mengimplementasikan dan menganalisis Nearest-Bislusters yang merupakan salah satu Collaborative Filtering yang membangun kedekatan antar user berdasarkan persamaan antar user maupun antar item. Tugas akhir ini menggunakan algoritma biclustering agar dapat mengelompokkan baik user maupun item dalam satu kelompok secara simultan. Tujuan dari tugas akhir ini adalah mengukur akurasi hasil rekomendasi yang dihasilkan system setelah diimplementasikan algoritma Nearest-Bicluster. Parameter yang digunakan dalam analisis adalah ukuran bicluster , prosentase overlapping bicluster, prosentase splitting test set, ukuran neighborhood (k), ukuran daftar rekomendasi (N), dan ukuran training set dan test set.
Pada algoritma Nearest-Bicluster, proses biclustering merupakan proses yang sangat penting dalam menghasilkan rekomendasi yang berkualitas. Saat ukuran bicluster n=4 dan m=10 merupakan ukuran bicluster yang paling baik untuk menghasilkan rekomendasi yang optimum. Hasil pengujian menunjukkan bahwa semakin besar prosentase overlapping bicluster dan ukuran neighborhood (k) maka semakin kecil nilai FMeasure. Nilai Precision akan semakin menurun saat prosentase overlapping test set semakin besar. Dan nilai FMeasure akan semakin besar untuk ukuran daftar rekomendasi (N) yang semakin besar dan ukuran training set yang semakin besar.Kata Kunci : Collaborative filtering, Nearest-Bicluster, BiclusteringABSTRACT: Collaborative filtering (CF) is one method in the recommender systems that have been studied to overcome the problem of "information overload". Collaborative filtering provides recommendations based on the preference of user characteristics in providing information.
This final project implement and analyze the Nearest-Bislusters Algorithm that one of Collaborative filtering method that builds closeness between users based on similarities between users or between items. This final project using biclustering algorithm that can classify both users and items in one group simultaneously. The aim of this final project is to measure the accuracy of the results of recommendations after implemented by Nearest-Bicluster algorithm. The parameters used in the analysis are bicluster size, percentage of overlapping bicluster, the percentage of splitting the test set, the size of neighborhood (k), the size of a list of recommendations (N), and the size of the training set and test sets.
On Nearest-Bicluster algorithms, biclustering process is an important process in producing high quality of recommendations. Bicluster size n = 4 and m = 10 is the best bicluster size to produce an optimum recommendation. The test results showed that the greater percentage of overlapping bicluster and neighborhood size (k) the smaller value of FMeasure. Precision value will decrease when the percentage of overlapping test sets bigger. And FMeasure value will be increase for the size of a list of recommendations (N) are getting larger and the size of the larger training set.Keyword: Collaborative filtering, Nearest-Bicluster, Biclustering