一种基于改进的巴氏系数的协同过滤推荐算法
首发时间:2019-09-23
摘要:基于邻居的协同过滤推荐,是当前推荐系统信息过滤技术中非常常见的推荐方法。但传统基于邻居的协同过滤推荐方法必须完全依赖用户共同评分项,且存在极为稀疏的数据集中预测准确性不高的问题。巴氏系数协同过滤算法可以有效改善上述问题。但该种方法也存在两个很明显的缺陷:一个是未考虑两个用户评分项个数不同时的情况,另一个是对用户偏好没有针对性考虑。该文在巴氏系数协同过滤算法的基础上进行了改进,既能充分利用用户的所有评分信息,又考虑到用户对项目的积极评分偏好。实验结果表明,改进的巴氏系数协同过滤算法在数据集上获得更好的推荐结果,提高了推荐的准确度。
关键词: 协同过滤 巴氏系数协同过滤算法 相似性度量
For information in English, please click here
A Collaborative Filtering Recommendation Algorithm Based on Improved Bhattacharyya Coefficient
Abstract:Neighbor-based collaborative filtering recommendation is a very common recommendation method in current recommendation system information filtering technology. However, the traditional neighbor based collaborative filtering recommendation method have to rely entirely on the common scoring items of users, and the accuracy of prediction in extremely sparse data sets is not high. Bhattacharyya coefficient for collaborative filtering can effectively improve the above problems. But there are two obvious drawbacks to this approach: one is that it fails to consider the case that the number of scoring items of two users is not the same; and the other is thatA Collaborative Filtering Recommendation Algorithm Based on Improved Bhattacharyya Coefficient there is no specific cA Collaborative Filtering Recommendation Algorithm Based on Improved Bhattacharyya Coefficientonsideration for user preferences. In this paper, the Bhattacharyya coefficient collaborative filtering algorithm is improved to make full use of all users\' scoring information. In addition, considering users\' positive scoring preferences for projects, the final user similarity value is calculated by using projects with user scores higher than their own average scores. The experimental results show that the algorithm obtains better recommendation results on the dataset and improves the accuracy of the recommendation.
Keywords: Collaborative filtering Bhattacharyya coefficient for collaborative filtering Similarity measure
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
一种基于改进的巴氏系数的协同过滤推荐算法
评论
全部评论0/1000