Abstract:Accurate prediction of commodity sales is the premise and basis of improving the efficiency of commodity marketing. In order to better predict commodity sales, existing researchers have proposed methods based on deep neural networks (DNNs), convolutional neural networks (CNNs), time series analysis, etc., nevertheless most of these methods only take into account the temporal or spatial characteristics of the commodity sales process unilaterally. At the same time, based on the modeling analysis of commodity sales data, it is found that there is a great correlation between the commodity sales and the spatial location and sales time of corresponding retail merchants. In order to more accurately predict the sales of a certain commodity in a specific store and at a specific time, this study first constructs a large-scale knowledge graph system based on merchants. Through the data model of a graph, the study describes related interaction scenarios of commodity sales and corresponding business circles, merchants, and users. At the same time, the spatial and data characteristics of merchant data are added to the graph model to describe the spatial and temporal characteristics of merchants. Finally, based on the constructed merchant knowledge graph, the study aggregates information based on a graph convolutional neural network (GCN) to obtain spatial features and then uses long short-term memory (LSTM) to extract temporal features. Furthermore, the study performs a weighted combination on the two features to predict commodity sales. Preliminary research results show that the commodity sales predicted by the hybrid model algorithm based on the graph and LSTM model is the closest to the actual sales. In addition, compared with that of traditional neural network algorithms, the average prediction accuracy of the model is 89%. Finally, by constructing an assembly line workflow, the whole intelligent prediction system of commodity sales is deployed in the production environment, which provides intelligent decision-making for realizing accurate commodity marketing.