摘要
产生推荐列表是基于用户的协同过滤推荐算法的重要步骤,也是最终的结果。针对在基于用户的协同过滤推荐算法中,"产生推荐列表"环节的研究相对较少的这一现象,为了改进推荐算法的性能,通过权重计算和共同评分项方法来选定推荐项目,即首先将项目按照评分的近邻用户数量的多少进行排序,然后对排序的项目进行综合权重计算,将其结果由高到低进行再次排序,从而产生推荐列表。该算法经MovieLens数据集测试,在测试中使用"平均绝对误差"作为实验测评指标,结果表明,在目标用户的相似用户数为60时,该算法相较于不考虑共同评分项或综合权重计算因素的算法,有着更低的平均绝对误差,其值为0.77。该算法能够在一定程度上提高推荐系统的准确度。
The recommended list is an important step of the user-based collaborative filtering recommendation algorithm and is also the fi- nal result. According to the phenomenon of less research on the "generation of recommendation list" in collaborative filtering recommen- dation algorithm based on user, in order to improve the performance of it, the recommended items are selected by weight calculation and the method of co-rating number. Firstly the co-rating items is sorted by the number of nearest neighbor. Then, the ranking items are cal- culated by comprehensive weight, and the results are sorted by high to low, and the recommended list is generated. The algorithm is tested by MovieLens data set. It uses the "Mean Absolute Error" as the evaluation index in the test. The results show that when the target user' s similar user number is 60, the algorithm has a lower mean absolute error compare with those calculation algorithms which don' t consider the factors of common rating items or comprehensive weight, and the value is 0.77. The algorithm can improve the accuracy of the rec- ommendation system to a certain extent.
出处
《计算机技术与发展》
2016年第9期69-72,共4页
Computer Technology and Development
基金
2014年福建省教育科技计划项目(JB14129)
关键词
协同过滤算法
评分项
综合权重
准确度
collaborative filtering algorithm
co-rating
comprehensive weight
accuracy