Recognition of Stores' Relationship Based on Constrained Spectral Clustering

Mobile Internet has gradually penetrated into all aspects of the daily life. Ever explosive growth recently hit the New Retail, which is closely integrated Internet online advantages with the offline stores-based facilities. Users can choose the most convenient stores for online or offline consumption, which determines that there are common users among stores, and the sales of stores could interact with each other. To make stores' operation network more efficient, the relationships among stores are explored and most efficient store clusters are identified, considering the geographical positions and business dependencies of different stores. In this paper, we first build business correlation matrix based on common user among stores respectively. Second, a constrained spectral clustering model is established to correct the outliers in each unsupervised iteration. Finally, the business data of Luckin Coffee are collected to validate our model. The results show that our method outperforms pure K-means and pure Spectral Clustering, which achieves an appropriate balance between spatial aggregation and business aggregation. This method can be applied to other new retail scenarios where stores have businesses interaction with each other.