Scholarlink

<p>In the era of digital content overload, delivering timely, personalized, and emotionally aligned news recommendations is increasingly critical. Traditional recommender systems, though effective in modeling short-term click behavior, often fall short in capturing long-term user engagement and affective relevance. To address these limitations, we propose a biologically inspired reinforcement learning framework based on membrane computing principles. This research introduces a biologically inspired reinforcement learning system for personalized news recommendation, incorporating sentiment-aware reward modeling within a modular membrane computing paradigm. The proposed architecture breaks down the learning pipeline into independent components-contrastive representation learning, supervised click prediction, and Proximal Policy Optimisation (PPO)-each regulated by specific optimisation protocols and inter-component communication. Using membrane computing concepts, the system enables concurrent signal processing and dynamic adaption. A new sentiment-aligned reward function is proposed to connect behavioral relevance with emotional resonance, improving user engagement by incentivising emotional alignment between user preferences and content polarity. A sentiment-guided gating system regulates content embeddings before policy inference. Experimental validation is performed on the MIND-small dataset utilizing Sentence-BERT for news encoding, with thorough assessment across Precision@5, Recall@5, NDCG@5, and Sentiment Alignment Rate. The results indicate that the proposed framework surpasses both baseline and ablated variations, exhibiting substantial enhancements in rating accuracy and emotional personalisation. This study presents a systematic integration of membrane-inspired modularity with policy-based reinforcement learning, providing a scalable, interpretable, and sentiment-aware method for real-time personalized content distribution.</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 membrane-inspired PPO framework with supervised and contrastive learning for sentiment-aligned news recommendation

  • 期刊论文

  • 发表日期：

    2026年2月17日

查看全文
  • Swathi Edem,
  • B. V. Ram Naresh Yadav

摘要

In the era of digital content overload, delivering timely, personalized, and emotionally aligned news recommendations is increasingly critical. Traditional recommender systems, though effective in modeling short-term click behavior, often fall short in capturing long-term user engagement and affective relevance. To address these limitations, we propose a biologically inspired reinforcement learning framework based on membrane computing principles. This research introduces a biologically inspired reinforcement learning system for personalized news recommendation, incorporating sentiment-aware reward modeling within a modular membrane computing paradigm. The proposed architecture breaks down the learning pipeline into independent components-contrastive representation learning, supervised click prediction, and Proximal Policy Optimisation (PPO)-each regulated by specific optimisation protocols and inter-component communication. Using membrane computing concepts, the system enables concurrent signal processing and dynamic adaption. A new sentiment-aligned reward function is proposed to connect behavioral relevance with emotional resonance, improving user engagement by incentivising emotional alignment between user preferences and content polarity. A sentiment-guided gating system regulates content embeddings before policy inference. Experimental validation is performed on the MIND-small dataset utilizing Sentence-BERT for news encoding, with thorough assessment across Precision@5, Recall@5, NDCG@5, and Sentiment Alignment Rate. The results indicate that the proposed framework surpasses both baseline and ablated variations, exhibiting substantial enhancements in rating accuracy and emotional personalisation. This study presents a systematic integration of membrane-inspired modularity with policy-based reinforcement learning, providing a scalable, interpretable, and sentiment-aware method for real-time personalized content distribution.