SiMoI New Method to Solve the Sparsity Problem in Collaborative Filtering

Abstract

Sparsity data is a major challenge in collaborative recommendation systems, characterized by the predominance of missing values within the user-item matrix. When a substantial portion of data is unavailable, the estimation process becomes hindered, and prediction accuracy declines due to limited usable information. To address this issue, this study introduces a novel method called SiMoI (Similarity, Mode, and Minimum Imputation), which is adaptively designed to handle high levels of sparsity. The SiMoI method combines user similarity with imputation strategies based on mode and minimum values. By leveraging subsets of the most informative users and items, the method efficiently fills missing entries while maintaining prediction stability. Evaluation was conducted using both real and synthetic datasets with varying sizes and degrees of sparsity, including an extreme scenario with 93.7% missing data. Experimental results show that SiMoI consistently produces more accurate predictions than baseline methods. Under high-sparsity conditions, SiMoI achieved an RMSE as low as 0.823, outperforming KNNI (0.947) and MEAN (1.021). Moreover, SiMoI demonstrated resilience across different data scales and sparsity distributions, indicating its flexibility and scalability in diverse contexts. These findings suggest that SiMoI is an effective and stable approach for addressing sparsity and holds strong potential for implementation in user-based recommendation systems, particularly in real-world scenarios where data availability is frequently limited.