基于条件型游走二部图协同过滤算法

发布时间：2018-01-25 00:08

本文关键词： 电子商务协同过滤条件型游走二部图稀疏性　出处：《计算机应用研究》2017年12期 　论文类型：期刊论文

【摘要】：针对拥有少量评分的新用户采用传统方法很难找到目标用户的最近邻居集的问题,提出了一种条件型游走二部图协同过滤算法。首先根据复杂网络理论的二部图网络,将用户—项目评分矩阵转换为用户—项目二部图,采用条件型游走计算目标用户与其他用户之间的相似性;然后根据协同过滤算法预测未评分项目,产生推荐。研究结果表明,在同样的数据稀疏性情况下,基于条件型游走二部图协同过滤算法在MAE和准确率都要优于其他两种传统的协同过滤算法,从而提高了算法的推荐精度;而且当训练值的比例很低时,即数据稀疏程度越大时,算法推荐质量的提高程度越大。
[Abstract]:It is difficult to find the nearest neighbor set of target users by traditional methods for new users with small scores. A conditional walking bipartite graph collaborative filtering algorithm is proposed. Firstly, according to the bipartite graph network of complex network theory, the user-item scoring matrix is transformed into the user-item bipartite graph. The similarity between the target user and other users is calculated by conditional walk. Then the ungraded items are predicted according to the collaborative filtering algorithm and the recommendations are generated. The results show that the same data sparsity is the case. The collaborative filtering algorithm based on conditional walking bipartite graph is superior to the other two traditional collaborative filtering algorithms in MAE and accuracy, so the recommendation accuracy of the algorithm is improved. Moreover, when the ratio of training value is very low, that is, the more sparse the data, the greater the improvement of the algorithm recommendation quality.
【作者单位】：上海财经大学数学学院;上海财经大学信息管理与工程学院;上海财经大学实验中心;
【分类号】：TP391.3
【正文快照】： 0引言近年来,推荐系统技术中应用了很多有关复杂网络的二部图,获得了诸多丰富的科研成果。例如资源分配和热传导这两种算法,两者主要运用了二部图网络对推荐系统进行诠释,提高了算法的工作效率,解决了数据稀疏性等传统难题,对提升推荐系统的性能起到了至关重要的作用[1~3]。在

【相似文献】