摘要
为提高基于时间加权的协同过滤推荐算法的相似度精度,提出一种在基于时间加权的协同过滤算法基础上结合评分预测的方法。利用评分预测方法得出相应的评分数据,根据总体相似性赋予部分预测评分一个合理的用户评分时的时间刻度,在经过上述两步处理的评分矩阵上利用时间加权方法计算相似度。使用MovieLens数据集对该算法、传统协同过滤算法和基于时间加权的协同过滤算法进行对比,对比实验结果表明,相比传统协同过滤算法和基于时间加权的协同过滤算法,该算法的平均误差分别降低了近5%和3%,当邻居个数为80时,其准确率最高达到29.24%,在一定程度上降低了计算相似度时数据稀疏性对相似度精度的不利影响,有效提高了相似度精度。
To improve the accuracy of similarity of collaborative filtering recommendation algorithm based on time weighted,the method combining the algorithm based on time weighted with the method of rating prediction was proposed.The method of the rating prediction was used to get score data.The overall similarity was come up with to give these scores reasonable time marks. The algorithm based on time weighted was applied to calculate similarity on the rating matrix that proceeded through the prior two steps.The data set of MovieLens was used to compare the proposed algorithm,traditional collaborative filtering recommen-dation algorithm and collaborative filtering recommendation algorithm based on time weighted.The results show this algorithm reduces about 5% and 3% of mean absolute error relative to traditional collaborative filtering recommendation algorithm and col-laborative filtering recommendation algorithm based on time weighted respectively.Also,the precision of this algorithm come to a head equals to 29.24 % when the neighbor equals to eighty.This algorithm can decrease bad influence of data sparse on simi-larity partly and improve the accuracy of similarity effectively.
出处
《计算机工程与设计》
北大核心
2016年第7期1827-1830,1872,共5页
Computer Engineering and Design
基金
贵州省科学技术基金项目(黔科合J字LKS[2013]29号)
关键词
时间加权
协同过滤
相似度
评分预测
稀疏性
time weighted
collaborative filtering
similarity
rating prediction
data sparse
