一种基于时间特性的二部图推荐算法

发布时间：2018-12-06 07:15

【摘要】：随着互联网的发展,大量商品信息不断涌现,从而产生了信息过载问题。推荐系统作为解决此问题的有效手段,近年来得到快速发展。现存方法大多以用户行为和商品内容相似性为基础,利用用户购买记录和商品描述信息来产生推荐结果。事实上,用户的购买行为与时间也有着密切的联系。例如,最近购买的商品往往更能体现用户的当前兴趣。因此,在传统基于相似性推荐的基础上,本文提出一种基于时间特性的二部图推荐方法,通过调整初始资源权重分布体现用户兴趣随时间的变化趋势。实验证明,本文提出的方法在面向时间的Top-N命中率上有较大幅度提升。本文工作不仅对现有推荐算法的效果提高具有实际意义,对推荐系统在真实商业环境中的应用也有很大促进作用。
[Abstract]:With the development of the Internet, a large number of commodity information has been emerging, resulting in the problem of information overload. As an effective means to solve this problem, recommendation system has been developed rapidly in recent years. Most of the existing methods are based on the similarity of user behavior and commodity content, and make use of user purchase records and product description information to produce recommendation results. In fact, the user's purchase behavior and time are also closely related. Recent purchases, for example, tend to better reflect the current interest of the user. Therefore, based on the traditional similarity recommendation, this paper proposes a bipartite graph recommendation method based on time characteristics, which reflects the trend of user interest with time by adjusting the initial resource weight distribution. Experimental results show that the proposed method can greatly improve the hit ratio of time oriented Top-N. The work of this paper not only has practical significance for improving the effect of existing recommendation algorithms, but also promotes the application of recommendation system in real business environment.
【作者单位】：电子科技大学互联网科学中心;
【基金】：国家自然科学基金资助项目(61103109,11105024,61003231) 中央高校基本科研业务费基金资助项目(ZYGX2011J057,ZYGX2012J071,ZYGX2012J085) 四川省科技基金资助项目(2010HH0002,2011GZ0106,20112Z0001,2012RZ0002,2012RZ0003) 高等学校博士学科点专项科研基金资助项目(20120185120017)
【分类号】：TP393.0

【参考文献】