摘要
针对用户的偏好推荐需求,提出一种改进的LFM算法BBLFM算法,通过引入隐含特征将稀疏的相关矩阵分解为两个相对稠密的矩阵,减少了空间复杂度,同时实现LFM的隐语义分析功能,深入挖掘了用户的潜在特征,提高了推荐的准确性。具体地,设计了一种基于BM-25的精确用户关注点查找与权重赋值方法,同时引入软概率情感分析方法的结果,合成出一种基于语义的标签体系。此外,还构建了一个基于BERT的用户偏好分析网络,根据用户曾经浏览或点击的历史论坛数据,来为用户画像,给出用户的主题偏好。在真实的百度贴吧数据集上进行的对比实验结果,表明算法在推荐准确性上优于比较的算法。
In this paper,we propose an improved LFM algorithm for user preference recommendation,which decomposes the sparse correlation matrix into two relatively dense matrices by introducing hidden features.Our model greatly reduces the space complexity while realizing the hidden semantic analysis function of LFM,fully excavating the potential characteristics of users,and improving the accuracy of recommendation.Specifically,we design a precise user focus search and weight assignment method based on BM-25,and at the same time introduce the results of the soft probability sentiment analysis method to synthesize a semantic-based labeling system.In addition,we have also built a BERT-based user preference analysis network,which can profile the user and give the user’s theme preference based on the historical forum data that the user has browsed or clicked on.Finally,we conduct comparative experiments on the real Baidu Tieba dataset.The experimental results show that our algorithm is superior to the compared algorithms in terms of recommendation accuracy.
作者
巨星海
周刚
JU Xinghai;ZHOU Gang(Information Engineering University, Zhengzhou 450001, China)
出处
《信息工程大学学报》
2021年第4期433-437,449,共6页
Journal of Information Engineering University
基金
国家自然科学基金资助项目(61702549)。
关键词
舆情分析
偏好分析
LFM算法
BERT
情感分析
public opinion analysis
preference analysis
LFM algorithm
BERT
sentiment analysis
