Scholarlink

<p>Collaborative Filtering (CF), a cornerstone of modern recommender systems, often suffers from significant challenges related to data sparsity and scalability, which can impair recommendation quality. Clustering algorithms offer a potent solution by grouping users with similar preferences, thereby creating denser, more manageable subspaces for prediction. This study provides a systematic comparative analysis of five prominent clustering algorithms—K-means, K-medoids, Fuzzy C-means (FCM), Self-Organizing Maps (SOM), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN)—within the CF framework. To further enhance performance, we introduce and evaluate a novel integration where these clustering techniques are coupled with an Autoencoder, a deep learning model designed to learn powerful latent feature representations from user rating data. We conduct a comprehensive experimental evaluation on two standard benchmark datasets, MovieLens 1&#xa0;M and Epinions, across four distinct CF configurations that vary by similarity measure (Pearson Correlation and Jaccard) and prediction method (Similarity-Based Prediction and Maximizing Average Satisfaction). Performance is assessed through a dual-pronged approach, measuring both cluster quality (using Davies-Bouldin and Dunn indices) and recommendation performance (using Accuracy, Precision, and Recall). Our findings consistently reveal that integrating an Autoencoder leads to significant improvements in both cluster quality and prediction accuracy across all tested algorithms. Notably, the Autoencoder-based FCM approach consistently outperforms all other methods, underscoring the substantial potential of deep feature learning to refine user groupings and ultimately deliver more accurate and effective recommendations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Improving collaborative filtering performance: a comparative analysis of clustering algorithms with and without autoencoders

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • Reza Barzegar Nozari

摘要

Collaborative Filtering (CF), a cornerstone of modern recommender systems, often suffers from significant challenges related to data sparsity and scalability, which can impair recommendation quality. Clustering algorithms offer a potent solution by grouping users with similar preferences, thereby creating denser, more manageable subspaces for prediction. This study provides a systematic comparative analysis of five prominent clustering algorithms—K-means, K-medoids, Fuzzy C-means (FCM), Self-Organizing Maps (SOM), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN)—within the CF framework. To further enhance performance, we introduce and evaluate a novel integration where these clustering techniques are coupled with an Autoencoder, a deep learning model designed to learn powerful latent feature representations from user rating data. We conduct a comprehensive experimental evaluation on two standard benchmark datasets, MovieLens 1 M and Epinions, across four distinct CF configurations that vary by similarity measure (Pearson Correlation and Jaccard) and prediction method (Similarity-Based Prediction and Maximizing Average Satisfaction). Performance is assessed through a dual-pronged approach, measuring both cluster quality (using Davies-Bouldin and Dunn indices) and recommendation performance (using Accuracy, Precision, and Recall). Our findings consistently reveal that integrating an Autoencoder leads to significant improvements in both cluster quality and prediction accuracy across all tested algorithms. Notably, the Autoencoder-based FCM approach consistently outperforms all other methods, underscoring the substantial potential of deep feature learning to refine user groupings and ultimately deliver more accurate and effective recommendations.