基于时序逆影响的随机游走推荐算法*

相似性计算是协同过滤推荐的关键步骤，针对传统相似性计算认为相似关系是对等的且没有考虑消费顺序和时间间隔的问题，提出了基于时序逆影响的随机游走推荐算法。首先，基于用户时序关联图提出一种新的称为时序逆影响的相似性度量，利用随机游走得到了目标用户近邻集合；其次，利用随机游走在项目时序关联图上进一步改进推荐的多样性和覆盖率。它不但认为用户间相似是不对称的，考虑了用户消费项目的顺序和时间间隔，获得了用户全局的直接和间接近邻，而且考虑了项目间的时序逆影响。通过在真实数据集上的大量试验结果表明，与其他随机游走方法相比，不但能提高推荐性能、缓解数据稀疏，而且通过提高多样性和覆盖率解决了过拟合的问题。

A Temporal Inverse Influence Based Recommendation method by Using Random Walk

Similarity computation was a very critical step in traditional collaborative filtering (CF) recommendation. However, traditional similarity always thought that the relationships were symmetric and rarely considered the order of common consumed items and the time interval of them. To get better recommendation performance, a random walk based recommendation method based on temporal inverse influence was proposed. It firstly proposed an new similarity measurement called temporal inverse influence on user temporal correlation graph and selected the similar users who most influenced on the target user using Random Walk with Restart. Lastly, it used the item temporal correlation graph to further improve the diversity and coverage of CF recommendation results. It not only thinks that the influence is asymmetric, takes the order of common consumed items of users and the time interval of them into account to find direct and indirect neighbors from a global perspective, but also consider the inverse influence between items to improve the recommendation results. The experimental results on a real dataset show that our proposed method can not only achieve better recommendation performance, but also alleviate data sparsity and tackle the over fitting problem by getting better coverage and diversity compared to several random-walk-based methods.