LAUSR

In order to accelerate the transformation of offline retailers and improve sales by using big data technology, this paper proposes a data-driven customer profile modeling method based on the collected historical purchase records of offline consumers. This method is mainly divided into three aspects: (1) an incremental RFM model is designed to classify the value of historical consumers and support the dynamic update of the model, which is more efficient than the traditional RFM model; (2) the commodity preference of different types of customers is analyzed by the TGI model, so as to guide the retail terminal to optimize the marketing strategy; (3) a commodity purchase behavior prediction model based on LSTM is proposed, which can predict the commodity that each customer may purchase in the future, so as to optimize the retail strategy. According to extensive experiments based on a true tobacco dataset, the incremental RFM model can save 80% more time than the traditional method, and our proposed prediction model can achieve 59.32% accuracy, which is better than other baselines.