Abstract: 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.

Key words: Graph-based recommendation,Random walk,Purchase interests,Repeated recommendation