基于边界矩阵低阶近似和近邻模型的协同过滤算法

发布时间：2019-01-13 08:58

【摘要】：为解决矩阵分解应用到协同过滤算法的局限性和准确率等问题,提出基于边界矩阵低阶近似(BMA)和近邻模型的协同过滤算法(BMAN-CF)来提高物品评分预测的准确率。首先,引入BMA的矩阵分解算法,挖掘子矩阵的隐含特征信息,提高近邻集合查找的准确率;然后,根据传统基于用户和基于物品的协同过滤算法分别预测出目标用户对目标物品的评分,利用平衡因子和控制因子动态平衡两个预测结果,得到目标用户对物品的评分;最后,利用MapReduce计算框架的特点,对数据进行分块,将该算法在Hadoop环境下并行化。实验结果表明,BMAN-CF比其他矩阵分解算法有更高的评分预测准确率,且加速比实验验证了该算法具有较好的可扩展性。
[Abstract]:In order to solve the problem of the limitation and accuracy of matrix decomposition applied to collaborative filtering algorithm, a collaborative filtering algorithm (BMAN-CF) based on low-order approximate boundary matrix (BMA) and nearest neighbor model is proposed to improve the accuracy of item scoring prediction. Firstly, BMA's matrix decomposition algorithm is introduced to mine the implicit feature information of the submatrix to improve the accuracy of nearest neighbor set search. Then, according to the traditional collaborative filtering algorithm based on users and articles, the target user's score of the target item is predicted, and the target user's score of the object is obtained by using the balance factor and the control factor to dynamically balance the two prediction results. Finally, using the characteristics of MapReduce computing framework, the data is divided into blocks, and the algorithm is parallelized in Hadoop environment. The experimental results show that BMAN-CF has a higher score prediction accuracy than other matrix decomposition algorithms, and the speedup experiment shows that the algorithm has better scalability.
【作者单位】：东北大学计算机科学与工程学院;
【基金】：国家自然科学基金资助项目(61572123) 国家杰出青年科学基金资助项目(61225012,71325002) 辽宁省百千万人才工程项目(2013921068) 赛尔网络下一代互联网技术创新项目(NGII20160616)~~
【分类号】：TP391.3

【相似文献】