一种综合标签和时间因素的个性化推荐方法

针对现存的基于标签的社会化推荐系统在构建用户兴趣模型时存在的缺陷,提出一种综合标签及其时间信息的资源推荐(TTRR)模型。此模型考虑了用户的兴趣具有时间性的特点,即用户兴趣是随着时间而变化的、用户最近新打的标签更能反映用户近期的兴趣这一特性。为此,在借鉴协同过滤思想的基础上,通过利用标签使用频率信息和项目的标注时间来构建用户评分伪矩阵;在此基础上计算目标用户的最近邻集合;最后根据邻居用户给出推荐结果。通过在CiteULike数据集上进行实验,并与传统的基于标注的推荐方法进行比较,实验结果表明,TTRR模型能够更好地反映出用户的偏好,能够显著地提高推荐准确度。

Personalized recommendation method based on both tag and time factors

To solve the shortcomings of available user interests modeling methods in existing social tagging systems, this paper proposed a resources recommendation model integrated tag and its time information(TTRR). In this model, it consided the temporal characteristics of the user's interest, that user's preference for products were drifting over time, and the tags that users recently used play a vital to better reflect the recent interest of the user. So, based on the thought of collaborative filtering method, this paper utilized the tag frequency and item's launch time information to construct a pseudo rating matrix. Then it obtained the nearest neighbor sets of the active user based on the pseudo rating matrix. Finally, according to the nearest neighbor sets, it obtained the recommendation results. The experimental results conducted on the CiteULike data sets show that compared with the traditional recommendation method which are based on user's log behaviors, the TTRR model can effectively reflect the user's preferences and significantly improve the recommendation accuracy.