基于项目聚类和时间衰减的动态协同过滤算法

Dynamic Collaborative Filtering Algorithm Based on Item Clustering and Time Decay

传统协同过滤推荐算法侧重于用户兴趣和项目的关系,目的是向用户推荐符合其兴趣的项目。但忽略了用户兴趣随时间的变化,将不同时间段的项目评分同等对待,降低了推荐的准确率。另一方面,基于项目的协同过滤算法在寻找目标项目的最近邻居时,因需要遍历整个项目空间而导致开销较大。为了解决上述问题,设计了一种基于项目聚类和时间衰减的动态协同过滤推荐算法ITDCF。该算法适用于基于项目的协同过滤,首先根据用户的评分对项目进行聚类,以快速找出目标项目的最近邻。接着,在计算项目相似度和预测评分阶段都引入时间衰减因子,以客观反映用户兴趣,提高推荐精度。最后,将前N个项目推荐给用户。在MovieLens数据集上对Popular、ItemCF、ITDCF算法的准确率、召回率和F 1值的测试结果表明,ITDCF算法在准确性和效率上都有所提高。

The traditional collaborative filtering recommendation algorithm focuses on the relationship between user interests and items,with the aim of recommending users to items that match their interests.However,the change of user interest over time is ignored,and the item scores of different time periods are treated equally,which reduces the accuracy of the recommendation.In addition,the item-based collaborative filtering algorithm leads to a large overhead because it needs to traverse the entire item space when searching for the nearest neighbor of the target item.In order to solve the above problems,we design a dynamic collaborative filtering recommendation algorithm ITDCF based on item clustering and time decay,which is suitable for item-based collaborative filtering.Firstly,the item is clustered according to the user’s score to quickly find the nearest neighbor of the target item.Then,the time decay factor is introduced in both the calculation item similarity and the prediction score stage to objectively reflect the user’s interest and improve the recommendation accuracy.Finally,the top N items are recommended to the user.The precision,recall and F 1 values of Popular,ItemCF and ITDCF algorithms are tested on the MovieLens dataset.The results show that the ITDCF algorithm has improved accuracy and efficiency.