logo
CanyamAI summaries for academic research
language
MobileMobile
  • Phone FeatureWeChat Mini Program
EnglishEnglish
noticeNotice
BlogBlog
delete
arrow
Return

Improved Collaborative Filtering for Recommendations: Integrating Fuzzy Clustering and Temporal Dynamics Analysis

delete2025-10-23
delete0
PRE
AI
S
Senlin Mao
J
Ji Zhang *
T
Tianzi Zang
Z
Zhao Li
李
李娟 (Juan Li)
DOI：10.34133/icomputing.0188delete
deleteOriginal
deleteOriginal request for help
deleteShare
deleteSave
Abstract

Abstract

En 中文
Collaborative filtering plays a vital role in recommendation systems but faces challenges such as data sparsity, scalability, and evolving user preferences. This research introduces an advanced fuzzy clustering method for user categorization that enhances the prediction of user preferences within clusters, even in sparse data scenarios. Our approach can leverage the collective preferences of users within each cluster to generate more accurate recommendations, even when individual user data are limited, to alleviate the data sparsity problem. By limiting similarity assessments to user clusters, our approach improves the scalability of the recommendation algorithm. To address the dynamic nature of user interests, we propose a time decay function that extends the base-level learning function, assigning higher weights to recent ratings for a more accurate representation of current user preferences. The performance of the recommendation algorithm is considerably influenced by the accuracy of the clustering process. Traditional fuzzy clustering methods often prioritize distance during iterations, potentially disregarding the overall consistency of data points, while random initialization of cluster centers can reduce stability and increase susceptibility to noise. We tackle these issues by employing a density-sensitive distance measure for proximity calculations and integrating fuzzy entropy into the clustering process, ensuring global consistency and improving the handling of complex data structures. Moreover, we introduce a mixed-distance approach for selecting initial cluster centers, enhancing the robustness of the clustering algorithm. Experimental results across multiple datasets validate the effectiveness of our proposed method in improving recommendation accuracy and user satisfaction.
Keywords:
ATTRIBUTES

Journal

I
Intelligent Computing
IF:
3.7
Papers:
13
Citations:
0

Organization

No organization information available