Scholarlink

<p>Collaborative filtering represents the most prevalent form of recommender systems, which is based on the interests and ratings of users. Data sparsity is a significant challenge for this type of system, and we need to predict the unavailable ratings in the user-item matrix. Similarity functions are used to predict the ratings and find the similarity between users or items. In this paper, an improved collaborative filtering-based recommender system called LA-NSGA-II-RS is presented to solve the data sparsity problem. In the proposed system, a new hybrid method is presented to find the similarity between users to overcome the problems of existing similarity methods in scenarios with high sparsity. Also, the problem of finding the list of recommended items is modeled as a multi-objective optimization problem in which, in addition to the accuracy objective, the coverage and novelty objectives are also considered simultaneously. To solve this model, an improved NSGA-II optimization algorithm called LA-NSGA-II is presented, in which a learning automaton is used to dynamically tune the rates of mutation and crossover to improve convergence of the algorithm and prevent getting stuck in a local optimum. Implementation of the proposed LA-NSGA-II optimization algorithm on standard ZDT functions shows that this algorithm obtains higher values for the Hyper Volume (HV) and IGD criteria compared to some other optimization algorithms. Also, the results obtained from implementing the proposed LA-NSGA-II-RS recommender system on two datasets, ML 100K and ML 1M indicate its effectiveness compared to the compared state-of-the-art recommender systems.</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

A collaborative filtering recommender system based on a new hybrid similarity measure and improved NSGA-II algorithm

  • 期刊论文

  • 发表日期：

    2026年2月5日

查看全文
  • Atena Torkashvand,
  • Seyed Mahdi Jameii,
  • Akram Reza

摘要

Collaborative filtering represents the most prevalent form of recommender systems, which is based on the interests and ratings of users. Data sparsity is a significant challenge for this type of system, and we need to predict the unavailable ratings in the user-item matrix. Similarity functions are used to predict the ratings and find the similarity between users or items. In this paper, an improved collaborative filtering-based recommender system called LA-NSGA-II-RS is presented to solve the data sparsity problem. In the proposed system, a new hybrid method is presented to find the similarity between users to overcome the problems of existing similarity methods in scenarios with high sparsity. Also, the problem of finding the list of recommended items is modeled as a multi-objective optimization problem in which, in addition to the accuracy objective, the coverage and novelty objectives are also considered simultaneously. To solve this model, an improved NSGA-II optimization algorithm called LA-NSGA-II is presented, in which a learning automaton is used to dynamically tune the rates of mutation and crossover to improve convergence of the algorithm and prevent getting stuck in a local optimum. Implementation of the proposed LA-NSGA-II optimization algorithm on standard ZDT functions shows that this algorithm obtains higher values for the Hyper Volume (HV) and IGD criteria compared to some other optimization algorithms. Also, the results obtained from implementing the proposed LA-NSGA-II-RS recommender system on two datasets, ML 100K and ML 1M indicate its effectiveness compared to the compared state-of-the-art recommender systems.