A MinMax Item-based Method for Multi-Criteria Recommendation Systems
Noor Ifada, Mochammad Kautsar Sophan, Nur Fitriani Dwi Putri · Procedia Computer Science · 2023
One of the common challenges in multi-criteria recommendation systems is dealing with data normalization. The challenge occurs since criteria can have diverse rating ranges and user rating behaviors are dissimilar. Previous studies on data normalization showed the supremacy of the Decoupling technique in the user-based multi-criteria recommendation system and the MinMax technique in the multi-criteria decision-making system as well as in data mining. A study also showed that the performance of Decoupling is improved in the item-based method than in the user-based. However, no study has been conducted to investigate the performance of MinMax compared to Decoupling in the item-based multi-criteria recommendation system. This study aims to combine the MinMax normalization technique and the item-based modeling approach in a multi-criteria recommendation system. The proposed method is named the MinMax Item-based method (MIB). We conducted a series of experiments using the Yelp Hotel multi-criteria rating dataset to perform a sensitivity analysis of MIB. The best settings are then used to benchmark MIB MIB towards DCMItem, i.e., a method that combines the Decoupling normalization technique and item-based multi-criteria modeling approach. The comparison results show the outperformance of MIB towards DCMItem by 2.30% in Precision and 2.00% in NDCG. Therefore, we can conclude that MixMax is able to improve the performance of the item-based multi-criteria recommendation system better than Decoupling.