Akademik Veri Yönetim Sistemi
COMPUTER STANDARDS & INTERFACES, cilt.95, 2026 (SCI-Expanded, Scopus)
In today's world, where users are surrounded by a multitude of products, recommender systems are employed to assist users in finding products of interest. Clustering methods are frequently utilized in recommender systems to suggest relevant products. Fuzzy clustering techniques, one of the most commonly used clustering methods, determine the degree of relevance of each product to a cluster through the membership matrix it generates. However, determining the number of clusters in these methods poses a challenge. This study proposes an Adaptive Fuzzy C-Means Jensen Shannon (AFCM-JS) algorithm, a fuzzy and interest-based clustering method that estimates the number of clusters. The proposed AFCM-JS algorithm is implemented on an artificial dataset consisting of 6 clusters and 1000 elements. The results of the study are compared with Fuzzy C-Means (FCM), Probabilistic C-Means (PCM), and Probabilistic Fuzzy C-Means (PFCM) methods, which are fuzzy-based clustering algorithms, and the interest-based method JS. To evaluate the comparison results, 7 different cluster validity indices and an accuracy metric are employed. AFCM-JS method consistently and accurately predicted the number of clusters when tested with different maximum cluster numbers. When the clustering ability of the method is tested with cluster validity indices and the accuracy metric, AFCM-JS is found to be successful. The performance of the AFCM-JS method is tested on a dataset created for a movie recommendation system with the aim of recommending movies to users. For this purpose, movie data is weighted with a Dirichlet function for action, adventure, comedy, drama, and horror genres, creating a dataset that includes the characteristics of these 5 movie genres. The AFCM-JS method is compared with 3 different fuzzy clustering methods using 7 different cluster validity indices with this created movie dataset. Additionally, the AFCM-JS algorithm is compared with the other 3 fuzzy clustering methods based on the accuracy metric. As a result of this comparison, the AFCM-JS method achieves the highest performance among the methods with 81.9366%. Furthermore, when the performance of the proposed method is compared in terms of cluster validity indices, the AFCM-JS method successfully predicts the appropriate number of clusters and effectively groups similar movies according to their genres, accomplishing the purpose.