GraphShop: A graph neural network model for shop-type recommendation

Abstract

It is essential to predict the popularity when the investors want to decide which type of shops to open at a given location. Existing shop-type recommender systems solve this problem based on collaborative filtering. However, most of existing collaborative filtering based methods make recommendation for each region, and have difficulty to analyze each shop specifically. To tackle this problem, we propose GraphShop, a new deep learning based shop-type recommender system. The GraphShop represents every shop as a node in a graph, learns its embedding vector, and finally predicts the popularity with a type. We also propose three aggregation functions to gather the neighborhood information of each shop from three different perspectives. Because of lack of an open dataset for the shop-type recommendation, we have constructed a new large shop-type dataset, that can be accessed through . Experimental results show that our proposed method outperforms other existing state-of-the-art methods.