Scholarlink

<p>This paper solves the problem of expressing nonlinear user-item interactions in collaborative filtering for product recommendations in the cultural industry and resulting low hit rates. To music works as our focus, we introduce a personalized recommendation system via an improved neural collaborative filtering model, where the MLP layer is substituted with a convolutional neural network (CNN) to more efficiently extract the intricate, nonlinear relationships between users and music works. In addition, an attention mechanism is integrated to weigh interactions among users and music and automatically determine the most informative features to recommend. The outputted convolutional neural matrix factorization (CNMF) model integrates CNN and attention mechanisms to improve both accuracy and diversity of recommendations. Experimental results show that CNMF performs better than baseline models, with a hit rate of 0.92 when K = 30, beating NeuMF, with significant improvements in Recall@K (+ 0.15) and MRR@K (+ 0.10). These results reflect the power of CNMF in breaking the bottleneck of conventional collaborative filtering models in the cultural field, providing more individualized and rich recommendations.</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

CNN and attention mechanism-based convolutional neural matrix factorization for music recommendation in the cultural industry

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Zhuo-Kai Ma,
  • Xiao-Jun Zhu,
  • Si-Qi Hao

摘要

This paper solves the problem of expressing nonlinear user-item interactions in collaborative filtering for product recommendations in the cultural industry and resulting low hit rates. To music works as our focus, we introduce a personalized recommendation system via an improved neural collaborative filtering model, where the MLP layer is substituted with a convolutional neural network (CNN) to more efficiently extract the intricate, nonlinear relationships between users and music works. In addition, an attention mechanism is integrated to weigh interactions among users and music and automatically determine the most informative features to recommend. The outputted convolutional neural matrix factorization (CNMF) model integrates CNN and attention mechanisms to improve both accuracy and diversity of recommendations. Experimental results show that CNMF performs better than baseline models, with a hit rate of 0.92 when K = 30, beating NeuMF, with significant improvements in Recall@K (+ 0.15) and MRR@K (+ 0.10). These results reflect the power of CNMF in breaking the bottleneck of conventional collaborative filtering models in the cultural field, providing more individualized and rich recommendations.