摘要： 目前许多基于社会化标签的推荐均忽视用户的兴趣变化及反复性,影响了推荐质量。针对该问题,提出一种将指数遗忘权重和时间窗口相结合的算法,既突出了近期兴趣的重要性,又强调了反复出现的早期数据。建立基准标签集,根据指数偏移后的标签向量选出目标用户的最近邻居,通过目标用户时间窗内标记的资源计算其所有资源的推荐权重向量,结合推荐权重和资源相似度给出最近邻居标记资源的推荐分数,取分数最高的前K 个资
源做出推荐。仿真实验结果表明,改进后的算法能动态地跟踪、学习用户的兴趣变化,提高推荐精度。
关键词:
协同过滤,
标签,
兴趣变化,
指数遗忘,
时间窗,
推荐
Abstract: Many recommendation methods based on social tagging ignore the change and repeatability of user interests,which
may lead to unsatisfactory results. In order to solve these problems,a new method which efficiently combines exponential forgetting-based data weight and time windows is proposed. The method not only highlights the importance of recent interest, but also stresses the recurring early data. Based on standard tag set of the target user,the nearest neighbour set can be gained according to exponential offset tag vectors,and then calculates weight vectors via items within time windows. Recommendation values of the nearest neighbour set are computed by weight vectors and similarity. Finally,it makes recommendation of items within the top K predicted values. Simulation experimental results show that the proposed algorithm for recommendation can
dynamically track the changes in user`s interest and has high quality of precision to some extent.
Key words:
collaborative filtering,
tag,
interest change,
exponential forgetting,
time window,
recommendation
中图分类号: