Soojung Lee, Journal of Internet Computing and Services, Vol. 26, No. 6, pp. 101-108, Dec. 2025
10.7472/jksii.2025.26.6.101, Full Text:  HTML
Keywords: Recommender System, Collaborative Filtering, Similarity Measure, data sparsity, singularity

Abstract

This study proposes a novel similarity calculation method that leverages the strengths of memory-based collaborative filtering techniques while effectively mitigating data sparsity. Memory-based CF is easy to implement and interpret in relatively small-scale data environments, does not require complex model training or large computational resources, and is suitable for real-time recommendation updates. However, in environments with sparse rating data, finding users with similar preferences can be challenging. To address this, this study extends the traditional Jaccard coefficient structure to leverage multi-genre information for items and incorporates the concept of singularity, which considers the distribution of each user's genre preferences. This design enhances semantic connectivity by securing genre-based indirect similarity even when direct common rating items are lacking. Furthermore, we propose a novel method that integrates the proposed similarity measure with existing calculation methods to reflect user ratings. Experimental results using public datasets demonstrate improved prediction accuracy compared to existing methods, particularly in terms of coverage, which reflects the diversity and scope of recommended items.

Statistics
Cite this article