摘要
通过评分数值提取反映每个用户主要兴趣特征的高兴趣项目特征,并利用用户高概率感兴趣的项目内容进行评分数据填充,形成用户评分数据的有效稠密序列,并进一步按照二值视图思想进行用户评分子序列的二值评分转换。文中所提出的算法普遍优于其他对比实验算法,随着最近邻居数量的不断增加,RMSE最优值达到0.8988,准确率和F值提高最为明显,其中准确度最高提高8.66%,F值最高提高33.96%。使用基于有效稠密序列提取的用户评分数据增强策略要比传统协同过滤方法表现更为优异,并且在表达用户兴趣特征的准确性和一致性方面,使用二值评分数据方法要明显优于原始评分数据方法。
Through extraction of valid ratings,the item features reflecting each user's main interest characteristics are obtained,and the rating data are filled with the content of the user's high probability of interest to form an effective dense sequence of user's rating data,and the binary rating conversion of user's rating sub⁃sequence is further carried out according to the idea of binary view.The proposed algorithm is generally better than other algorithms.With the increase of the number of nearest neighbors,the optimal RMSE is 0.8988,and the accuracy and the F score are improved most obviously.The highest accuracy and the F value are improved by 8.66%and 33.96%,respectively.The user rating data enhancement strategy based on an effective dense sequence extraction performs better than the traditional collaborative filtering method,and the accuracy and the consistency of expressing user interest features using binary rating data method are significantly better than that of the original scoring data method。
作者
崔北亮
周小康
李树青
CUI Beiliang;ZHOU Xiaokang;LI Shuqing(Library,Nanjing Tech University,Nanjing 210009,China;School of Information Engineering,Nanjing University of Finance and Economics,Nanjing 210023,China)
出处
《南京邮电大学学报（自然科学版）》
北大核心
2021年第6期57-65,共9页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
江苏省高等学校自然科学研究重大项目(19KJA510011)资助项目。
关键词
稠密序列
二值视图
数据增强
数据稀疏
推荐系统算法
dense sequence
binary view
data enhancement
data sparsity
recommender system algorithm
