摘要
传统的推荐模型存在两大问题:一是用户的偏好发生变化,而推荐模型没有随之发生改变;二是对异常数据的处理能力有限,导致推荐结果的单一性使得多样性推荐不足.用户行为数据是近年来推荐研究领域热点,利用用户行为数据能够挖掘用户更深层次的信息.对从评论文本和用户源信息提取用户偏好,提出UICTM模型.在此基础上运用LDA技术得到UIFT算法,融合时间因素优化UIFT算法,对CF、TMF、UIFT、UIFT+进行实验数据分析比较.结果表明:UIFT+算法在均方差MSE和推荐度ACC上整体优于其他推荐算法.
The traditional recommendation model has two major problems:one is that the user’s preference changes,but the recommendation model does not change with it;Second,the processing capacity of abnormal data is limited,resulting in the lack of singleness and diversity of recommendation results.User generated content(UGC)is a hot spot in the field of recommendation research in recent years.Using UGC data can mine users’deeper information.In this paper,UICTM model is proposed to extract user preferences from comment text and user source information.On this basis,the UIFT algorithm is obtained by using LDA technology,the UIFT algorithm is optimized by integrating time factors,and the experimental data of CF,TMF,UIFT and UIFT+are analyzed and compared.The results show that UIFT+algorithm is better than other recommendation algorithms in mean square error MSE and recommendation ACC.
作者
丁丽
方晓
DING Li;FANG Xiao(Department of Information Engineering,Bozhou Vocational and Technical College,Bozhou 236800,China)
出处
《青海师范大学学报（自然科学版）》
2022年第1期14-23,96,共11页
Journal of Qinghai Normal University(Natural Science Edition)
基金
安徽省教育厅自然科学重大项目(KJ2021ZD0162)
安徽省高校优秀青年人才支持计划项目(gxyq2019226)
亳州职业技术学院重点项目(ykIId002、BYK2025)
安徽省质量工程项目(2020jxtd173)
