To improve the forecast accuracy of the sales of stores in shopping malls, this paper proposes a prediction method based on deep learning that comprehensively considers the external data, such as online review data of shopping mall stores, weather data, weekday/weekend data, and historical transaction data of the stores. To begin with, the online review data of the stores are pre-trained with BERT (Bidirectional Encoder Representations from Transformers) to complete the multi-label sentiment classification and obtain the intensity index of perceived sentiment of reviews. The index, together with other external data, such as online ratings, weather, weekday/weekend differences, and historical transactions of the stores, is pre-processed. At last, the Long Short-Term Memory (LSTM) and the Attention models are used to predict the sales volume of stores in a certain shopping mall. The results show that the addition of external data – weather, weekday/weekend, online ratings and intensity index of sentiment of reviews – to the historical sales data-based model can effectively improve the forecast accuracy of store sales.
Ye, Qiongwei; Yuan, Qiang; Ding, Dan; Liao, Qichen; and Chan, Yiu Ching, "Sales forecasting of stores in shopping malls: A study based on external data and transaction data" (2022). ICEB 2022 Proceedings (Bangkok, Thailand). 40.