Scholarlink

Recommender systems are essential for personalization in digital services but often rely on correlations that introduce spurious biases, impairing quality, robustness, and fairness. Causal inference offers a promising solution, yet a systematic synthesis of debiasing methods–particularly with emerging tools like large language models (LLMs) is still lacking. This paper fills this gap by reviewing three core causal frameworks (potential outcomes, SCM, and counterfactual models) and four debiasing paradigms (propensity score, causal graph, SCM, and counterfactual methods). It highlights current trends and future directions, including LLMs integration, to advance causal debiasing in recommender systems.

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

Causal Insights for Debiasing Recommender Systems: Future Directions and Challenges

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hui Wang,
  • Xi Zhou,
  • Mingming Li,
  • Hongyue Zhang,
  • Jiao Dai,
  • Jizhong Han,
  • Tao Guo

摘要

Recommender systems are essential for personalization in digital services but often rely on correlations that introduce spurious biases, impairing quality, robustness, and fairness. Causal inference offers a promising solution, yet a systematic synthesis of debiasing methods–particularly with emerging tools like large language models (LLMs) is still lacking. This paper fills this gap by reviewing three core causal frameworks (potential outcomes, SCM, and counterfactual models) and four debiasing paradigms (propensity score, causal graph, SCM, and counterfactual methods). It highlights current trends and future directions, including LLMs integration, to advance causal debiasing in recommender systems.