Research On Shopping Mall Choice Prediction Of Consumers Based On Spatiotemporal Data

With the changes in domestic anti-epidemic policies,the human flow in urban mall areas has burst into growth.The consumers who shop in mall areas can be broadly classified as real customers who complete the purchase behavior and potential buyers who contribute to the pedestrian flow of the mall area.All individuals whose trip ends within the mall area can be defined as mall consumers.By analyzing the historical information of consumers,we can predict their future shopping choices,which is of vital importance for improving the quality of mall marketing,urban comprehensive governance capacity,and consumers’ travel and shopping experience.The research on individual travel destination prediction based on spatio-temporal data with deep neural networks has achieved lots of achievement,but there are still significant challenges in predicting the mall choices of consumers.On the one hand,collecting consumer information through questionnaires is fragmentary and timelagged.On the other hand,previous studies have focused primarily on improving the structure of neural networks while ignoring the spatio-temporal social characteristics of the trajectory data inherently.Considering above,this thesis completes the following research with the goal of accurately predicting consumers’ mall choices:(1)Dividing urban functional areas considering high-density poi distribution.To acquire consumers’ trajectory information completely and efficiently,we can retain individual trajectory data whose destination is within the mall area.However,this approach requires the boundary of commercial functional areas.The current method of using POI(Point Of Interests)static spatio-temporal data for dividing urban functional areas is limited by the high-density distribution of POI,which results in a low accuracy rate.To resolve this problem,the thesis proposed a long text classification model that considers the distribution of high-density POIs in space.Firstly,the model has designed an algorithm to extract spatial information to make the disordered POI data textualized while retaining the distribution feature.Subsequently,the model employs the sliding window to divide lengthy POI text into shorter partial texts which match with the input requirements of the ERNIE(Enhanced Representation through k Nowledge Int Egration)model,and obtain the feature embedding matrix.Further,the model utilizes multi-scale fusion Text CNN(Text Convolutional Neural Network)for secondary feature extraction of the embedding matrix,and uses a Bi-LSTM(Bi-directional Long Short Term Memory)network to serialize the multiple partial short texts.Finally,the model employs an attention mechanism which can assigns more classification weights to crucial POIs,thereby ensuring accurate classification of urban functional areas.According to the experiments conducted,the proposed model can achieve a classification accuracy of 70.59%,which is 3.53% higher than the accuracy achieved by the best baseline model.(2)Consumer shopping mall choice prediction based on trajectory data.There are associations between consumers’ preferences for mall choice and the multidimensional spatio-temporal social features of the area where they are located.However,existing studies have not fully considered these important factors.To resolve this problem,this thesis proposes a method for predicting consumers’ mall choice based on trajectory data.The method obtains the trajectory representation with multidimensional features by jointly embedding the spatial co-occurrence semantics,social features and deep temporal features of consumers’ trajectories.Nextly,the method utilizes a two-layer GRU(Gate Recurrent Unit)to extract features and uses spatio-temporal attention mechanism to distinguish crucial trajectory points.This method ultimately leads to accurate predictions of consumers’ mall choices.Compared with the baseline algorithm,the prediction accuracy of this method is 1.06% higher and the average distance error is 0.33 KM lower,enabling better prediction of consumers’ shopping district choices.