Scholarlink

Recommender Systems (RSs) are essential tools for avoiding information overload on the World Wide Web (or simply the Web). Providing accurate and trustworthy recommendations will increase the confidence of the end-users in an RS and make them more satisfied and engaged. Neighbourhood-based methods are the most popular and widely used technique for recommendations. In order to provide more accurate recommendations, the effect of neighbourhood size is investigated in User-based Collaborative Filtering (UBCF) recommendations. In particular, this work examines an open-source and scalable collaborative recommendation framework, Apache Mahout. Firstly, brief functionalities of the Mahout library are discussed along with its API classes. Secondly, several offline experiments are conducted with various MovieLens datasets, including the latest one, MovieLens-25M, to analyze the functioning of UBCF methods as implemented in the Mahout. Finally, the accuracy of UBCF methods is measured in terms of two metrics: Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Empirically obtained results demonstrate that an RS can provide trustworthy recommendations to its end users by increasing the neighbourhood size in UBCF methods.

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

An Empirical Analysis of Neighborhood-Based Approaches for Trustworthy Recommendations with Apache Mahout

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Vijay Verma

摘要

Recommender Systems (RSs) are essential tools for avoiding information overload on the World Wide Web (or simply the Web). Providing accurate and trustworthy recommendations will increase the confidence of the end-users in an RS and make them more satisfied and engaged. Neighbourhood-based methods are the most popular and widely used technique for recommendations. In order to provide more accurate recommendations, the effect of neighbourhood size is investigated in User-based Collaborative Filtering (UBCF) recommendations. In particular, this work examines an open-source and scalable collaborative recommendation framework, Apache Mahout. Firstly, brief functionalities of the Mahout library are discussed along with its API classes. Secondly, several offline experiments are conducted with various MovieLens datasets, including the latest one, MovieLens-25M, to analyze the functioning of UBCF methods as implemented in the Mahout. Finally, the accuracy of UBCF methods is measured in terms of two metrics: Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Empirically obtained results demonstrate that an RS can provide trustworthy recommendations to its end users by increasing the neighbourhood size in UBCF methods.