Scholarlink

<p>Recommendation systems play a very important role and are therefore essential in the modern digital platforms that aid users in identifying content aligned to their preferences based on historical interactions. The Collaborative Filtering (CF) is adopted widely as it provides relationships between the user and item that are used to generate the recommendations. The proposed work develops and assesses a hybrid CF model and compares it with two baseline methods: User-user CF and Neural CF. The study analyzes the accuracy of the recommendations generated by the hybrid model across different domains through hybridization. Two datasets used in this experiment are Food.com recipe ratings and MovieLens, representing distinct domains. The former dataset is experimented with raw and mean-centered Item-item CF as well as Singular Value Decomposition (SVD) based Matrix Factorization and a weighted hybrid model. The Hybrid model achieved the best performance with an RMSE of 0.5015, improving prediction accuracy by 12.4% over mean-centered CF (0.5725 RMSE). For the MovieLens dataset, Item-Item CF, SVD, and Neural Collaborative Filtering (NCF) were evaluated, with NCF yielding the lowest prediction error (0.7138 RMSE) and outperforming SVD (0.8019) and CF (0.9368). Bias analysis indicated minor but consistent variations in residual error across user demographics. Overall, the Hybrid CF model proved more effective for sparse food-rating data, while NCF better captured complex preference structures in dense movie-rating contexts.</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

Product recommendation framework with enhanced item-based collaborative filtering

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月10日

查看全文
  • Archana Praveen Kumar,
  • Chinmai,
  • B. Priya Kamath,
  • Roopashri Shetty

摘要

Recommendation systems play a very important role and are therefore essential in the modern digital platforms that aid users in identifying content aligned to their preferences based on historical interactions. The Collaborative Filtering (CF) is adopted widely as it provides relationships between the user and item that are used to generate the recommendations. The proposed work develops and assesses a hybrid CF model and compares it with two baseline methods: User-user CF and Neural CF. The study analyzes the accuracy of the recommendations generated by the hybrid model across different domains through hybridization. Two datasets used in this experiment are Food.com recipe ratings and MovieLens, representing distinct domains. The former dataset is experimented with raw and mean-centered Item-item CF as well as Singular Value Decomposition (SVD) based Matrix Factorization and a weighted hybrid model. The Hybrid model achieved the best performance with an RMSE of 0.5015, improving prediction accuracy by 12.4% over mean-centered CF (0.5725 RMSE). For the MovieLens dataset, Item-Item CF, SVD, and Neural Collaborative Filtering (NCF) were evaluated, with NCF yielding the lowest prediction error (0.7138 RMSE) and outperforming SVD (0.8019) and CF (0.9368). Bias analysis indicated minor but consistent variations in residual error across user demographics. Overall, the Hybrid CF model proved more effective for sparse food-rating data, while NCF better captured complex preference structures in dense movie-rating contexts.