融入信任的变权重相似度模型在线学习协同推荐算法  

Collaborative Recommendation Algorithm of Online Learning Based on Trust-combined Similarity Model with Variable Weight

作　　者：谢修娟[1] 陈永 李香菊[1] 莫凌飞[3] 

机构地区：[1]东南大学成贤学院,计算机工程系,南京210000 [2]金斗云信息科技有限公司,南京210000 [3]东南大学仪器科学与工程学院,南京210000

出　　处：《小型微型计算机系统》2018年第3期525-528,共4页Journal of Chinese Computer Systems

基　　金：江苏高校哲学社会科学基金项目(2016SJD880186)资助;江苏省现代教育技术研究课题项目(2016-R-46509)资助;“十二五”国家科技支撑计划项目(2013BAJ05802-2)资助.

摘　　要：针对传统的基于用户的协同推荐算法存在的数据稀疏以及对用户评分的强依赖问题,提出基于社交信任机制的在线学习协同推荐算法.利用学习行为日志数据,采取主动评分和被动评分相结合的综合评分方法,构建学习者-学习资源评分矩阵.并根据学习者间的关注和评论两种社交数据,创建一种变权重相似度模型,进而生成近邻集实现推荐.使用历史学习数据作为本文的实验数据集,确定最优的调节因子值,并与传统算法进行对比实验.实验结果表明,数据稀疏度明显改善,是改进前的近11倍,且平均绝对偏差(MAE)减少了2%左右,推荐质量更高.Focusing on the problem of data sparsity and strong dependence on user＇s assessment, a collaborative recommendation algo- rithm based on social trust mechanism is proposed. Then, used the log data of learning behavior, the comprehensive evaluation method combined with the active and passive score is adoped to bulid a learner-learning resource score matrix. Afterwards, according to the learner＇s two types of social contact data, such as learner＇s attention and comments, a similarity model with variable weight was devel- oped. Furthermore, the nearest neighbor set is generated as well as the recommendation. At the same time, the history learning data is used as the experimental data set, which will determine the optimal adjustment factor value and make comparison with the traditional algorithm. The experimental results show that the data sparsity improves significantly, nearly 11 times higher than the traditional meth- od. And the mean absolute error I MAE} reduces by about 2% ,the quality of the recommendation is more effectively.

关 键 词：协同过滤 在线学习推荐 数据稀疏 社交信任 学习行为 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]