Scholarlink

<p>Advanced video streaming platforms increasingly rely on recommendation systems to assist children in navigating vast collections of age-appropriate content. While traditional collaborative filtering (CF) methods effectively utilize past user-item interactions, they often struggle with sparsely connected data. Content-based models, although useful, face limitations in capturing the richness of user preferences. In recent years, graph-based machine learning approaches have demonstrated their ability to address these challenges across various domains. In particular, Graph Neural Networks (GNNs) present a promising avenue for learning from the intricate relationships within user interaction graphs and content features to provide hybrid recommendations. This paper proposes a baseline comparison between different approaches with a GNN-based video recommendation system tailored for children. We conduct our experiments using a comprehensive video dataset, moving beyond mere accuracy to include a range of evaluation metrics. Evaluating our models using large-scale datasets, including MovieLens datasets (25&#xa0;M and 20&#xa0;M), TikTok, and Netflix, demonstrates that GNNs significantly improve the RMSE, Accuracy, Precision, Recall, and NDCG, and relevance of recommendations when enhanced with genre-based preferences. The GNN model effectively mitigates cold-start issues and also surpasses techniques in Top-N recommendations. This method offers a safer and more personalized viewing experience by focusing on genres like animation, adventure, family, and friendly content for children, ensuring that recommendations align with their developmental needs.</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

Genre Preferences Based Personalized Video Recommendations for Children Using Graph Neural Network

  • 期刊论文

  • 发表日期：

    2025年1月30日

查看全文
  • Lucky Harichandan,
  • Sasmita Kumari Nayak,
  • Satyabrata Lenka

摘要

Advanced video streaming platforms increasingly rely on recommendation systems to assist children in navigating vast collections of age-appropriate content. While traditional collaborative filtering (CF) methods effectively utilize past user-item interactions, they often struggle with sparsely connected data. Content-based models, although useful, face limitations in capturing the richness of user preferences. In recent years, graph-based machine learning approaches have demonstrated their ability to address these challenges across various domains. In particular, Graph Neural Networks (GNNs) present a promising avenue for learning from the intricate relationships within user interaction graphs and content features to provide hybrid recommendations. This paper proposes a baseline comparison between different approaches with a GNN-based video recommendation system tailored for children. We conduct our experiments using a comprehensive video dataset, moving beyond mere accuracy to include a range of evaluation metrics. Evaluating our models using large-scale datasets, including MovieLens datasets (25 M and 20 M), TikTok, and Netflix, demonstrates that GNNs significantly improve the RMSE, Accuracy, Precision, Recall, and NDCG, and relevance of recommendations when enhanced with genre-based preferences. The GNN model effectively mitigates cold-start issues and also surpasses techniques in Top-N recommendations. This method offers a safer and more personalized viewing experience by focusing on genres like animation, adventure, family, and friendly content for children, ensuring that recommendations align with their developmental needs.