Scholarlink

As video streaming platforms expand, delivering personalized and safe content recommendations for children becomes increasingly vital. Traditional collaborative filtering (CF) and content-based models struggle with data sparsity and cold-start issues, particularly in low-interaction settings. This paper introduces a novel model Hybrid Genre-Preference GNN (HGP-GNN), that integrates semantic content embedding, temporal user engagement modeling, and hierarchical preference learning to enhance recommendation quality for children. Unlike conventional GNN-based recommendation systems, HGP-GNN constructs multi-relational graphs that incorporate user history, implicit content attributes, and evolving genre interests. An adaptive attention mechanism dynamically adjusts genre preferences based on engagement patterns, while a reinforcement learning (RL)-based reward system fine-tunes recommendations for sustained engagement and content diversity. To ensure responsible AI practices, we integrate age-appropriate filtering and explainable AI (XAI) techniques, enhancing transparency and enabling effective parental controls. Extensive evaluations on MovieLens (25M, 20M), TikTok, and Netflix datasets demonstrate significant improvements over state-of-the-art models in RMSE, Accuracy, Precision, Recall, NDCG, and recommendation relevance. Our genre-adaptive GNN enhances Top-N recommendation precision for child-friendly genres such as animation, adventure, and family entertainment, effectively mitigating cold-start challenges. HGP-GNN establishes a safer, more engaging, and developmentally appropriate recommendation framework for children.

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

Graph Neural Networks for Personalized Children’s Video Recommendations: A Genre-Aware Hybrid Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • G. R. Ramya,
  • B. Abhiram,
  • P. Venu Teja,
  • T. B. Bhargavi,
  • M. Lokesh Chowdary

摘要

As video streaming platforms expand, delivering personalized and safe content recommendations for children becomes increasingly vital. Traditional collaborative filtering (CF) and content-based models struggle with data sparsity and cold-start issues, particularly in low-interaction settings. This paper introduces a novel model Hybrid Genre-Preference GNN (HGP-GNN), that integrates semantic content embedding, temporal user engagement modeling, and hierarchical preference learning to enhance recommendation quality for children. Unlike conventional GNN-based recommendation systems, HGP-GNN constructs multi-relational graphs that incorporate user history, implicit content attributes, and evolving genre interests. An adaptive attention mechanism dynamically adjusts genre preferences based on engagement patterns, while a reinforcement learning (RL)-based reward system fine-tunes recommendations for sustained engagement and content diversity. To ensure responsible AI practices, we integrate age-appropriate filtering and explainable AI (XAI) techniques, enhancing transparency and enabling effective parental controls. Extensive evaluations on MovieLens (25M, 20M), TikTok, and Netflix datasets demonstrate significant improvements over state-of-the-art models in RMSE, Accuracy, Precision, Recall, NDCG, and recommendation relevance. Our genre-adaptive GNN enhances Top-N recommendation precision for child-friendly genres such as animation, adventure, and family entertainment, effectively mitigating cold-start challenges. HGP-GNN establishes a safer, more engaging, and developmentally appropriate recommendation framework for children.