Implementing Set-Valued k-Modes Clustering Technique to Enhance the Quality of Library Book Recommendation System using Hybrid Semantic Ontology-based model
Imam Fadhkur Rokhim, Noor Ifada, Fika Hastarita Rachman · 2023
This research implements the Set-Valued k-Modes (SV-k-Modes) clustering technique on the Hybrid Semantic Ontology-based (HSO) model for the library book recommendation system. Note that HSO originally consisted of three algorithms: Collaborative Filtering (CF) Matrix Factorization, Content-based (CB) ONTO Semantic Similarity (ONTO), and Hybrid. In this case, we cluster the library books before the implementation of the CB ONTO algorithm of HSO such that the calculation of books’ semantic similarities is based on the books of each cluster instead of all books in the system. SV-k-Modes technique consists of three main consecutive algorithms. First is the algorithm for generating initial cluster centers which depends on the density and mutual Jaccard coefficient distance of each book. The second algorithm is for clustering the books into k clusters which computation is based on the dissimilarity or distance measure between the books with their set-valued keywords. Third is the algorithm for updating the centers in the second algorithm which is based on the highest frequency of keywords in the clusters. The second and third algorithms are repeated until convergence. A series of experiments using a real-world library dataset shows the outperformance of our proposed method compared to the original HSO method. The average increase percentage of our proposed method from Top-1 to Top-20 in terms of the Normalized Discounted Cumulative Gain (NDCG) is 1.002%. The increase is not very significant but it confirms that adding the SV-k-Modes technique on the HSO model can increase the performance of the library book recommendation system.