一种综合用户和项目因素的协同过滤推荐算法

Collaborative filtering recommendation algorithm based on both user and item

针对用户评分数据极端稀疏情况下传统协同过滤推荐算法的不足,提出了一种综合用户和项目因素的最近邻协同过滤推荐(HCFR)算法.该算法首先以一种改进的相似性度量方法(ISIM)为基础,根据当前评分数据的稀疏情况,动态调节相似度的计算值,真实地反映彼此之间的相似性.然后,在产生推荐时综合考虑用户和项目的影响因素,分别计算目标用户和目标项目的最近邻集合.最后,根据评分数据的稀疏情况,自适应地调节目标用户和目标项目的最近邻对最终推荐结果的影响权重,并给出推荐结果.实验结果表明,与传统的只基于用户或基于项目的推荐算法相比,HCFR算法在用户评分数据极端稀疏情况下仍能显著地提高推荐系统的推荐质量.

To solve the shortcomings of the traditional collaborative filtering recommendation algorithms in the situation of extreme sparsity of user＇s rating data,a hybrid collaborative filtering recommendation（HCFR） algorithm for the nearest neighbors based on users and items is proposed.First,on the basis of correlation similarity,this algorithm adopts an improved similarity measure method（ISIM） which can dynamically adjust the value of similarity according to the current state of sparse rating data and truly reflect the real situation.Then,in the process of generating recommendation results,both user factors and item factors are considered and the nearest neighbor sets of the active user and the active item are obtained.Finally,according to the sparsity of the user＇s rating data,different self-adaptive influence weights of the neighbor sets of the active user and the active item are adjusted,and the final recommendation results are obtained.The experimental results show that compared with the traditional recommendation algorithms which are only based on user or item,the HCFR algorithm can effectively improve the recommendation quality even in the situation of extreme sparsity of user＇s rating data.