A Preliminary Study of Similarity Measure Implementation on Hybrid Memory-based Recommendation Model

Hybrid memory-based (HMB) combines two memory-based (MB) models using a linear fusion function. Their performances depend on the similarity measures used. Many studies have compared the performances of the MB-based methods using various similarity measures. However, no study is on HMB other than using Pearson Coefficient Correlation (PCC) and Cosine. This paper conducts a preliminary study of implementing similarity measures on HMB. It investigates whether the superior performance of similarity measures when implemented on MB-based methods is also consistent with HMB-based methods. We use the Bhattacharyya Coefficient (BC) and Improved Triangle Similarity with User Rating Preference (ITR) measures that have shown superiority when implemented on the MB-based methods. Experiment results show that the implementation of BC outperforms PCC, but ITR is inferior to PCC. Hence, the superiority of similarity measures is inconsistent when implemented on MB-based and HMB-based methods. Further study using other measures on HMB is worthwhile.