Scholarlink

<p>As stock markets become increasingly complex and volatile, investors require intelligent decision-support systems to mitigate risk and enhance portfolio performance. Recommender systems can provide valuable insights for stock selection. Nonetheless, current models face challenges such as sparse data and the cold-start problem, which limit their effectiveness. This paper presents a modular hybrid recommender system comprising: (1) a recommendation module that combines Risk-Level Reputation Matrix Factorization (RRMF) with Self-Attention Matrix Factorization (SAMF), and (2) an optimization module that constructs individualized portfolios based on user objectives and risk tolerance. RRMF addresses data sparsity by incorporating users’ risk-level reputations, while SAMF captures dependencies using attention mechanisms to improve cold-start performance. The optimization module applies statistical methods to generate risk-aware portfolios. Experimental results on synthetic data provide preliminary validation, indicating that the hybrid model outperforms standalone RRMF and SAMF, achieving a MAE of 0.6408 and RMSE of 0.851, while improving other evaluation metrics.</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

A hybrid risk-aware recommender system for personalized stock portfolios with risk-level reputation and self-attention matrix factorization

  • 期刊论文

  • 发表日期：

    2026年6月3日

查看全文
  • Sara Javadi,
  • Seyyed Mohsen Hashemi

摘要

As stock markets become increasingly complex and volatile, investors require intelligent decision-support systems to mitigate risk and enhance portfolio performance. Recommender systems can provide valuable insights for stock selection. Nonetheless, current models face challenges such as sparse data and the cold-start problem, which limit their effectiveness. This paper presents a modular hybrid recommender system comprising: (1) a recommendation module that combines Risk-Level Reputation Matrix Factorization (RRMF) with Self-Attention Matrix Factorization (SAMF), and (2) an optimization module that constructs individualized portfolios based on user objectives and risk tolerance. RRMF addresses data sparsity by incorporating users’ risk-level reputations, while SAMF captures dependencies using attention mechanisms to improve cold-start performance. The optimization module applies statistical methods to generate risk-aware portfolios. Experimental results on synthetic data provide preliminary validation, indicating that the hybrid model outperforms standalone RRMF and SAMF, achieving a MAE of 0.6408 and RMSE of 0.851, while improving other evaluation metrics.