Scholarlink

<p>Recommender systems enhance user experience through personalized suggestions, but traditional collaborative filtering methods often sacrifice diversity and novelty for accuracy, leading to user fatigue. Balancing these competing objectives is challenging, and existing multi-objective approaches generate overwhelming Pareto sets that complicate decision-making. We propose S-MOIA-T, a multi-objective recommendation framework that integrates Singular Value Decomposition (SVD) for rating prediction, Multi-Objective Immune Algorithm (MOIA) for Pareto-optimal solution generation, and TOPSIS for structured ranking and selection. This combination effectively balances accuracy, diversity, and novelty while simplifying recommendation selection. Extensive experiments on MovieLens-100K, Netflix, and Book-Crossing datasets demonstrate S-MOIA-T’s superiority, achieving precision up to 0.83, diversity of 0.45, and novelty exceeding 1.2. These results significantly outperform baseline methods including User-CF, Item-CF, NMF, and state-of-the-art multi-objective algorithms (NNIA-RS, PMOEA, MOEA/D, MOPSO, EMO-GP). S-MOIA-T provides a practical, scalable solution for user-centric recommender systems.</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

Multi-objective Immune Computing for Recommender Systems: A TOPSIS-Based Optimization Approach

  • 期刊论文

  • 发表日期：

    2026年4月18日

查看全文
  • Fatima Ezzahra Zaizi,
  • Sara Qassimi,
  • Said Rakrak

摘要

Recommender systems enhance user experience through personalized suggestions, but traditional collaborative filtering methods often sacrifice diversity and novelty for accuracy, leading to user fatigue. Balancing these competing objectives is challenging, and existing multi-objective approaches generate overwhelming Pareto sets that complicate decision-making. We propose S-MOIA-T, a multi-objective recommendation framework that integrates Singular Value Decomposition (SVD) for rating prediction, Multi-Objective Immune Algorithm (MOIA) for Pareto-optimal solution generation, and TOPSIS for structured ranking and selection. This combination effectively balances accuracy, diversity, and novelty while simplifying recommendation selection. Extensive experiments on MovieLens-100K, Netflix, and Book-Crossing datasets demonstrate S-MOIA-T’s superiority, achieving precision up to 0.83, diversity of 0.45, and novelty exceeding 1.2. These results significantly outperform baseline methods including User-CF, Item-CF, NMF, and state-of-the-art multi-objective algorithms (NNIA-RS, PMOEA, MOEA/D, MOPSO, EMO-GP). S-MOIA-T provides a practical, scalable solution for user-centric recommender systems.