基于模糊偏好向量和物品质量的推荐算法  

Recommendation Algorithm Based on Fuzzy Preference Vector and Item Quality

作　　者：艾均[1] 孙浩[1] 苏湛[1] 周越 Jun Ai;Hao Sun;Zhan Su;Yue Zhou(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海

出　　处：《运筹与模糊学》2024年第2期218-233,共16页Operations Research and Fuzziology

基　　金：国家自然科学基金项目(61803264)。

摘　　要：推荐系统能够快速有效地帮助人们在海量信息中进行筛选和过滤,为用户智能化地推荐其感兴趣的物品,因此具有重要的理论意义和广泛的应用价值。针对现有研究中推荐误差和推荐列表排序可以进一步优化的需求。本文应用模糊隶属度函数以计算了多维物品品类向量上的用户偏好,并基于用户偏好特征和物品品类向量计算用户间相似性,然后,进一步基于相似性和用户相关性筛选构建用户相似性网络,利用用户相似性网络中的社区信息,对预测过程进行加权。同时,利用统计方法过滤物品的异常评分以度量物品质量排序,利用物品质量排序的优化结果辅助预测的误差修正。基于MovieLens的小型数据集进行了对比试验,与时下流行的几个算法进行对比,本文设计的算法有效地提高了预测准确性及推荐排序准确性,实验结果表明了利用用户模糊偏好、物品质量排序、相似性网络及误差修正手段,可以进一步挖掘推荐系统地潜在信息,在仅增加少量计算量的情况下,进一步提升了系统性能,并且保持了系统的可解释性。Recommendation systems can quickly and effectively help people filter and sift through massive amounts of information,intelligently recommending items of interest to users.Therefore,they have significant theoretical importance and wide application value.This paper addresses the need for further optimization of recommendation errors and recommendation list ranking in existing research.The fuzzy membership function is applied to calculate user preferences on multidimensional item category vectors.Based on user preference features and item category vectors,user similarity is calculated.Then,further based on similarity and user relevance,a user similarity network is constructed by filtering and selection.The community information in the user similarity network is used to weight the prediction process.At the same time,statistical methods are used to filter out abnormal ratings of items to measure item quality ranking,and the optimized results of item quality ranking are used to assist in the correction of prediction errors.A comparative experiment was conducted on a small dataset from MovieLens,and the algorithm designed in this paper was compared with several popular algorithms.The results show that the designed algorithm effectively improves prediction accuracy and recommendation ranking accuracy.The experimental results demonstrate that by using user fuzzy preferences,item quality ranking,similarity network,and error correction methods,the potential information of the recommendation system can be further mined.With only a small amount of additional computation,system performance is further improved,and the system’s interpretability is maintained.

关 键 词：推荐系统 用户模糊偏好 物品质量排序 相似性网络 误差修正 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]