SMART:一种面向电商平台快速消费品的图推荐算法

SMART:A Graph-based Recommendation Algorithm for Fast Moving Consumer Goods in E-commerce Platform

提出一种针对电商平台快速消费品的图推荐算法SMART。该算法在传统二部图推荐算法的基础上增加商品种类节点及其与用户和商品两类节点的关联边,且利用每个用户对商品种类和单个商品的兴趣倾向设置无向边的权重,通过节点间转移概率的差异实现有倾向性的随机游走;经过多次迭代,用户节点到所有节点的游走概率会收敛到稳定值,收敛后的游走概率能够在一定程度上体现用户对商品的购买概率;最后考虑每个用户对商品所属商家的兴趣偏好,对用户节点到各商品节点的游走概率进行调整,并根据调整后的游走概率计算每个用户的TOP-N推荐列表。在京东生鲜类商品的评论数据集上对所提出的推荐算法进行性能评价,实验结果表明该算法的确能够提供高质量的推荐,与基本二部图推荐算法相比,准确率提高了1.32%,召回率提高了1.48%。

This paper proposed a new graph-based recommendation algorithm for fast consuming goods in e-commerce platform,called SMART.Different from the traditional recommendation based on user-item bigraph,there are four types of nodes including users,items,categories,and their corresponding edges.In the new user-item-category graph,the weight of each undirected edge is set according to each user's interests in the items and their categories,so we can run biased random walks on the weighted graph.After many iterations,the probabilities from each user node walking to all other nodes would converge to stable values.It is believed that the convergence probabilities can reflect the probabilities with which users purchase the goods.At last,we explored the user's preference for shops to adjust the convergence probabilities and compute the TOP-N recommendation list.We evaluated the performance of the proposed algorithm based on the dataset of comment records in JD fresh goods,and the results show that our algorithm can provide high quality recommendation.Compared with the basic bigraph-based recommendation,the accuracy and the recall of our algorithm have increased by 1.32％ and 1.48％ respectively.