基于交互式图传递模型的Top-N推荐

摘要

This paper proposes an interactive graph propagation model for Top-N recommendation, in which the user graph propagation and item graph propagation in the 'user-item'rating space are well unified in an interactive way. Considering the imbalance between the original ratings given by user and those predicted ones via graph propagation, we propose an integrated method to measure the similarity between graph vertices. Furthermore, to boost the potential of forming more reliable predicted ratings on items to be recommended for active user, the rating predictions via user graph propagation and item graph propagation are lineally combined. Experimental results on MovieLens and EachMovie data sets demonstrate that the proposed method outperforms the states of the user graph model and item graph model a lot.%本文提出了一种基于交互式图传递模型的Top-N推荐方法.在‘用户-项目’评分信息空间中通过图模型交互传递用户关系和项目关系,实现用户关系和项目关系的有效融合.针对用户和项目描述空间的相似性度量问题,本文提出了一种鲁棒的组合相似性度量方法,以平衡原始评分信息和经传递得到的预测分值信息间之间的不对称性.最后,通过融合从用户图与项目图视角得到的活跃用户对每个项目的预测分值,形成其对项目评价分值更加可靠的预测.在MovieLens和EachMovie两个数据库上的实验结果表明,同基于用户图模型与项目图模型的协同滤波推荐方法相比,本文提出的方案取得了更好的推荐性能.