Scholarlink

<p>Recommender Systems (RS) are an indispensable tool that provide relevant information to the user. However, RS is vulnerable to shilling attacks. At present, there exist independent works to detect fake profiles and generate recommendations. However, there is an absence of a unified work that provides algorithms for both detection of fake profiles as well as generation of recommendations. To this, in this work we propose a bio-inspired optimization model that filters group shilling or fake profiles as well as generates reliable recommendations. The proposed model works in two phases. In the first phase, the correlation among users is computed using random forest and Pearson correlation coefficient. Then, group fake profiles are identified using swarm intelligence as a dimensionality reduction problem. In the second phase, a three-layer recommendation algorithm using an evolutionary approach is proposed to generate recommendations by creating a lattice of bi-clusters. Extensive experiments on Movielens, Amazon, and Netflix data demonstrated notable performance of proposed hybrid approach with group fake profile detection rate of 98% and generating recommendations with an accuracy of 80%.</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

Hybridizing grey wolf optimization and memetic algorithms to mitigate shilling attacks for ensuring reliable recommendations

  • 期刊论文

  • 发表日期：

    2025年7月22日

查看全文
  • Saumya Bansal,
  • Niyati Baliyan,
  • Rakhee Sharma

摘要

Recommender Systems (RS) are an indispensable tool that provide relevant information to the user. However, RS is vulnerable to shilling attacks. At present, there exist independent works to detect fake profiles and generate recommendations. However, there is an absence of a unified work that provides algorithms for both detection of fake profiles as well as generation of recommendations. To this, in this work we propose a bio-inspired optimization model that filters group shilling or fake profiles as well as generates reliable recommendations. The proposed model works in two phases. In the first phase, the correlation among users is computed using random forest and Pearson correlation coefficient. Then, group fake profiles are identified using swarm intelligence as a dimensionality reduction problem. In the second phase, a three-layer recommendation algorithm using an evolutionary approach is proposed to generate recommendations by creating a lattice of bi-clusters. Extensive experiments on Movielens, Amazon, and Netflix data demonstrated notable performance of proposed hybrid approach with group fake profile detection rate of 98% and generating recommendations with an accuracy of 80%.