RankPro-M Method to Alleviate the Sparsity Problem in Collaborative Filtering

Abstract

The rapid shift from conventional commerce to online platforms has been driven by evolving consumer behavior that demands fast, accurate, and personalized services. Consequently, e-commerce has become a primary channel for product marketing and service delivery without temporal or spatial constraints. However, the continuous expansion of e-commerce platforms has led to a substantial increase in both the volume and diversity of available products, thereby complicating the task of delivering personalized recommendations aligned with user preferences. Recommender systems offer an effective solution to this challenge, with Collaborative Filtering (CF) being among the most widely adopted techniques. Despite its popularity, CF suffers from a critical limitation known as the data sparsity problem, which adversely affects recommendation accuracy and system reliability. This study proposes RankPro-M, a ranking-oriented imputation approach designed to mitigate the impact of sparsity in recommender systems. RankPro-M operates by identifying items with high rating frequency and imputing missing ratings using mode values as representations of dominant user preferences. The imputed rating matrix is subsequently processed through ranking aggregation mechanisms (Borda, Copeland, and WP-Rank) to generate item recommendations. Experimental results demonstrate that the application of RankPro-M consistently improves recommendation quality, as indicated by increased Normalized Discounted Cumulative Gain (NDCG) values across multiple evaluation scenarios. These findings confirm that RankPro-M effectively addresses data sparsity and enhances the performance of ranking-based recommender systems.