中文摘要：

推荐系统以用户购买行为相似性为基础,而用户购买不仅包括是与否的选择信息,还有其购买时间和购买后对产品的评价信息作为反馈结果.满意商品能正确反映用户兴趣偏好,而很久以前购买和负面评价的商品,则将误导用户兴趣的分析.因此,在传统二部图推荐的基础上加入用户评价和时间衰减因素,提出一种基于用户反馈的时序推荐方法,经过多个数据集上的实验证明,提出方法在不同推荐列表长度的命中率指标上均有较大幅度的提升.

英文摘要：

Bipartite Graph recommendation methods are based on the similarity of usersshopping behavior,and the resource allocation algorithm is employed to output the recommendation results.In fact,the shopping behavior of users does not simply answer yes or no for the recommendation,but indicates more information.For example,items which users were satisfied with can indicate their interests correctly,as well as the awful shopping experiences mean the users make a wrong decision with the items they have bought.It was the same when the recommendation results came from the items they have purchased several years ago.If we do not consider their feedback information in our recommendation,the results can be confused with users.A recommendation method combined with temporal and rating information based on bipartite graph was proposed in the paper.In the proposed method,the initial weights of items can be allocated adaptively by the usersfeedback,so the recommendation can be correctly and timely.Experiments on several real life datasets show notable improvement on the Top-N hits metric.