Leveraging Hybrid Semantic Ontology-based Model for Library Book Recommendation System

This paper presents a study on leveraging the Hybrid Semantic Ontology-based (HSO) model for the library book recommendation system (RS). HSO consists of Collaborative Filtering (CF), Content-based (CB), and Hybrid (HB) modules. HSO can also optionally add a clustering process prior to the CB module to improve its recommendation quality. Despite the promising results of current HSO methods for library book RS, the prospect of further exploitation is still possible. The CF module of HSO implements a matrix factorization, and its choice of algorithm can be different, e.g., Non- Negative Matrix Factorization (NMF) or Singular Value Decomposition (SVD). Meanwhile, the clustering process implements a categorical data clustering technique, and its selected technique can also diverge, e.g., Set-Valued k-Modes (SV-k-Modes) or Density Rough k-Modes (DRk-M). We propose to improve the recommendation quality of library book RS by implementing NMF -based DCG Optimization for Collaborative Ranking (DO-NMF) matrix factorization algorithm and Modified Fuzzy k- Partition (MFkP) clustering technique. Experiment results using the UTM Library Dataset confirm that employing the clustering technique can always improve the HSO model's performance for library book RS. The average performance increase percentage from Top-1 to Top-20 is 60.57%. Meanwhile, both of our proposed methods surpass the performance of all benchmarking methods. The results show that DO-NMF is a superior matrix factorization algorithm to NMF and SVD. Furthermore, MFkP is a better categorical data clustering technique than SV -k-Modes and DRk-M.