Store site selection model training method and apparatus

Abstract

The application discloses a shop site selection model training method and device, the method comprises the following steps: determining the position information corresponding to a plurality of target shop information of a target site selection industry; determining the position information of at least one other industry shop in a specific area range corresponding to each target shop information, obtaining the position information of a plurality of other industry shops; determining the space-time feature data corresponding to each target shop information or other industry shop; taking the position information and space-time feature data corresponding to all target shop information as positive sample training data set, taking the position information and space-time feature data corresponding to all other industry shops as negative sample training data set, inputting into a site selection score training algorithm model for training until convergence, and obtaining a site selection score algorithm model. It can be seen that the application can provide industry-accurate-oriented site selection service when the algorithm model obtained by training is used for site selection prediction, and improve site selection accuracy and efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a store site selection model training method and device. BACKGROUND

[0002] With the acceleration of the update iteration of offline stores and the development of data analysis technology, enterprises begin to consider more factors when selecting the location of offline stores, and gradually start to use some neural network model prediction technology to assist enterprises in selecting offline store locations with higher profits and more audiences. However, the existing technology does not consider how to train the model using industry target store information and non-industry target store information in the region when providing store site selection technical services using neural network model prediction technology, so the trained model cannot obtain accurate features of the corresponding industry, and therefore the site selection effect is poor and the efficiency is low. It can be seen that the existing store site selection model training method has defects and needs to be solved. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a store site selection model training method and device, which can enable the trained algorithm model to provide industry-accurate site selection services for users when performing site selection prediction, and improve the accuracy and efficiency of site selection.

[0004] To solve the above technical problems, the first aspect of the present application discloses a store site selection model training method, which comprises:

[0005] determining the position information corresponding to a plurality of target store information of a target site selection industry;

[0006] determining the position information of at least one other industry store in a specific area range corresponding to each target store information according to the position information corresponding to each target store information, to obtain the position information of a plurality of other industry stores;

[0007] determining the spatio-temporal feature data corresponding to each target store information or other industry store according to the position information corresponding to each target store information or other industry store;

[0008] taking the position information and spatio-temporal feature data corresponding to all target store information as positive sample training data set, and taking the position information and spatio-temporal feature data corresponding to all other industry stores as negative sample training data set;

[0009] inputting the positive sample training data set and the negative sample training data set into a site selection score training algorithm model for training until convergence, to obtain a site selection score algorithm model.

[0010] As an optional implementation, in the first aspect of the present application, the determining of the spatio-temporal feature data corresponding to each of the target store information or the other industry store according to the location information corresponding to each of the target store information or the other industry store comprises:

[0011] determining the area information corresponding to each of the target store information or the other industry store according to the location information corresponding to each of the target store information or the other industry store and a preset location area mapping relationship;

[0012] determining the spatio-temporal feature data corresponding to each of the target store information or the other industry store according to the area information corresponding to each of the target store information or the other industry store.

[0013] As an optional implementation, in the first aspect of the present application, the spatio-temporal feature data comprises at least one of geographic feature data, commercial feature data and user feature data; the geographic feature data comprises at least one of point site density degree feature, public transportation convenience feature and road accessibility feature; the commercial feature data comprises at least one of proportion feature of each commercial category, competitive level feature, commercial diversity level feature and commercial reputation distribution feature; and the user feature data comprises at least one of resident population distribution feature, working population distribution feature, user age distribution feature and aggregated trajectory feature in different time periods.

[0014] As an optional implementation, in the first aspect of the present application, the determining of the spatio-temporal feature data corresponding to each of the target store information or the other industry store according to the area information corresponding to each of the target store information or the other industry store comprises:

[0015] determining the area spatio-temporal data corresponding to each of the target store information or the other industry store from a preset area spatio-temporal database according to the area information corresponding to each of the target store information or the other industry store;

[0016] inputting each of the target store information or the other industry store into a computing model based on a Spark framework to calculate the spatio-temporal feature data corresponding to each of the target store information or the other industry store.

[0017] As an optional implementation, in the first aspect of the present application, the area spatio-temporal data comprises at least one of POI point site data, traffic station data, road network structure data, industry distribution data, population statistics data, store review data, people flow trajectory data, logistics trajectory data and vehicle flow trajectory data.

[0018] As an optional implementation, in the first aspect of the present application, the site selection scoring training algorithm model and / or the site selection scoring algorithm model comprises at least one of a GBDT algorithm model, an MLP algorithm model, a ResNet algorithm model, a NODE algorithm model, a TabNet algorithm model and an algorithm model based on a Transformer architecture.

[0019] As an optional implementation, in the first aspect of the present application, the method further comprises:

[0020] In the process of inputting the positive sample training data set and the negative sample training data set into the site selection scoring training algorithm model for training until convergence to obtain the site selection scoring algorithm model, a feature evaluation algorithm model in the site selection scoring training algorithm model is used to evaluate and screen the spatio-temporal feature data in the positive sample training data set and / or the negative sample training data set to obtain preferred feature data with a higher importance level.

[0021] In addition, the inputting of each target store information or the other industry store corresponding to the input into the calculation model based on the Spark framework to calculate the spatio-temporal feature data corresponding to each target store information or the other industry store comprises:

[0022] Inputting each target store information or the other industry store corresponding to the input into the calculation model based on the Spark framework to calculate a plurality of candidate spatio-temporal feature data corresponding to each target store information or the other industry store;

[0023] The candidate spatio-temporal feature data corresponding to each target store information or the other industry store corresponding to the preferred feature data with the same data parameters are determined as the spatio-temporal feature data corresponding to each target store information or the other industry store.

[0024] As an optional implementation, in the first aspect of the present application, the inputting of the positive sample training data set and the negative sample training data set into the site selection scoring training algorithm model for training until convergence to obtain the site selection scoring algorithm model comprises:

[0025] Inputting the positive sample training data set and the negative sample training data set into the site selection scoring training algorithm model for training until convergence to obtain a trained site selection scoring teacher algorithm model;

[0026] Model distillation is performed on the site selection scoring teacher algorithm model to obtain the site selection scoring algorithm model.

[0027] The second aspect of the present application discloses a store site selection model training device, the device comprises:

[0028] a first determining module configured to determine position information corresponding to a plurality of target store information of a target industry;

[0029] a second determining module configured to determine position information of at least one other industry store within a specific area range corresponding to each of the target store information according to the position information corresponding to each of the target store information, to obtain a plurality of position information of the other industry store;

[0030] a feature determining module configured to determine spatio-temporal feature data corresponding to each of the target store information or the other industry store according to the position information corresponding to each of the target store information or the other industry store;

[0031] a sample determining module configured to take the position information and the spatio-temporal feature data corresponding to all the target store information as positive sample training data set, and take the position information and the spatio-temporal feature data corresponding to all the other industry store as negative sample training data set;

[0032] a model training module configured to input the positive sample training data set and the negative sample training data set into a site selection scoring training algorithm model for training until convergence, to obtain a site selection scoring algorithm model.

[0033] As an optional implementation, in the second aspect of the present application, the specific manner in which the feature determining module determines the spatio-temporal feature data corresponding to each of the target store information or the other industry store according to the position information corresponding to each of the target store information or the other industry store includes:

[0034] determining area information corresponding to each of the target store information or the other industry store according to the position information corresponding to each of the target store information or the other industry store, and a preset position area mapping relationship;

[0035] determining spatio-temporal feature data corresponding to each of the target store information or the other industry store according to the area information corresponding to each of the target store information or the other industry store.

[0036] As an optional implementation, in the second aspect of the present invention, the spatiotemporal feature data includes at least one of geographic feature data, commercial feature data, and user feature data; the geographic feature data includes at least one of location density features, public transportation convenience features, and road accessibility features; the commercial feature data includes at least one of the proportion features of various commercial categories, competitiveness level features, commercial diversity level features, and commercial reputation distribution features; the user feature data includes at least one of the resident population distribution features, working population distribution features, user age group distribution features, and aggregated trajectory features for different time periods.

[0037] As an optional implementation, in a second aspect of the invention, the feature determination module determines the specific method of spatiotemporal feature data corresponding to each target store or other industry store based on the regional information corresponding to each target store or other industry store, including:

[0038] Based on the regional information corresponding to each target store or other industry store, determine the regional spatiotemporal data corresponding to each target store or other industry store from a preset regional spatiotemporal database;

[0039] The inputs for each of the target store information or other industry stores are fed into a Spark-based computational model to calculate the spatiotemporal feature data for each of the target store information or other industry stores.

[0040] As an optional implementation, in the second aspect of the present invention, the regional spatiotemporal data includes at least one of POI location data, transportation station data, road network structure data, business distribution data, demographic data, store review data, pedestrian flow trajectory data, logistics trajectory data, and vehicle flow trajectory data.

[0041] As an optional implementation, in the second aspect of the present invention, the location scoring training algorithm model and / or the location scoring algorithm model includes at least one of the following: GBDT algorithm model, MLP algorithm model, ResNet algorithm model, NODE algorithm model, TabNet algorithm model, and algorithm model based on Transformer architecture.

[0042] As an optional implementation, in the second aspect of the present application, the device further comprises a feature evaluation module, which is configured to, when the model training module performs the inputting of the positive sample training data set and the negative sample training data set into the site scoring training algorithm model for training until convergence to obtain the site scoring algorithm model, utilize a feature evaluation algorithm model in the site scoring training algorithm model to evaluate and screen the spatio-temporal feature data in the positive sample training data set and / or the negative sample training data set to obtain preferred feature data with a higher importance level.

[0043] In addition, the specific manner in which the feature determination module inputs each of the target store information or the other industry store into a computing model based on a Spark framework to calculate the spatio-temporal feature data corresponding to each of the target store information or the other industry store includes:

[0044] inputting each of the target store information or the other industry store into a computing model based on a Spark framework to calculate a plurality of candidate spatio-temporal feature data corresponding to each of the target store information or the other industry store;

[0045] determining, as the spatio-temporal feature data corresponding to each of the target store information or the other industry store, the candidate spatio-temporal feature data in the plurality of candidate spatio-temporal feature data corresponding to each of the target store information or the other industry store that has the same data parameters as the preferred feature data.

[0046] As an optional implementation, in the second aspect of the present application, the specific manner in which the model training module inputs the positive sample training data set and the negative sample training data set into the site scoring training algorithm model for training until convergence to obtain the site scoring algorithm model includes:

[0047] inputting the positive sample training data set and the negative sample training data set into the site scoring training algorithm model for training until convergence to obtain a trained site scoring teacher algorithm model;

[0048] performing model distillation on the site scoring teacher algorithm model to obtain the site scoring algorithm model.

[0049] A third aspect of the present application discloses another store site selection model training device, which comprises:

[0050] a memory storing executable program codes;

[0051] a processor coupled to the memory;

[0052] The processor invokes the executable program code stored in the memory to execute part or all of the steps of the store site selection model training method disclosed in the first aspect of the application.

[0053] The fourth aspect of the application discloses a computer readable storage medium storing a computer program for electronic data exchange, wherein the computer program causes a computer to execute part or all of the steps of the store site selection model training method disclosed in the first aspect of the application.

[0054] Compared with the prior art, the embodiments of the application have the following beneficial effects:

[0055] In the embodiments of the application, a store site selection model training method and device are disclosed, and the method comprises the following steps: determining position information corresponding to a plurality of target store information of a target site selection industry; determining position information of at least one other industry store in a specific area range corresponding to each target store information, to obtain position information of a plurality of other industry stores; determining spatio-temporal feature data corresponding to each target store information or the other industry store; inputting the position information and the spatio-temporal feature data corresponding to all the target store information as positive sample training data set, and inputting the position information and the spatio-temporal feature data corresponding to all the other industry stores as negative sample training data set into a site selection score training algorithm model for training until convergence, to obtain a site selection score algorithm model. It can be seen that, in the embodiments of the application, the target store information of a target industry and the target store information of other industries in the position range thereof are screened to obtain positive and negative sample sets, so as to train the algorithm model, so that the algorithm model obtained by training can provide industry-precise site selection services for users when performing site selection prediction, and the accuracy and efficiency of site selection are improved. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative effort.

[0057] Figure 1 is a flowchart of a store site selection model training method disclosed by the embodiments of the application.

[0058] Figure 2 is a structural schematic diagram of a store site selection model training device disclosed by the embodiments of the application.

[0059] Figure 3 is a structural schematic diagram of another store site selection model training device disclosed by the embodiments of the application. DETAILED DESCRIPTION

[0060] In order to make the person skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor are within the scope of protection of the present application.

[0061] The terms “second”, “second”, etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish different objects, not to describe a specific order. In addition, the terms “include” and “have” and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or equipment including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or equipment.

[0062] In this paper, the term “embodiment” means that the specific features, structures or characteristics described in conjunction with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment to other embodiments. The person skilled in the art explicitly and implicitly understands that the embodiments described herein can be combined with other embodiments.

[0063] The present application discloses a store site selection model training method and device, which can obtain positive and negative sample sets by filtering target store information of a target industry and target store information of other industries within the location range of the target store, to train an algorithm model, so that the trained algorithm model can provide industry-accurate site selection services for users when predicting site selection, and improve the accuracy and efficiency of site selection. The following will be described in detail.

[0064] Embodiment one

[0065] Please refer to Figure 1 , Figure 1 is a flowchart of a store site selection model training method disclosed by the embodiments of the present application. Among them, Figure 1 The described store site selection model training method is applied in a store site selection model training chip, a training terminal or a training server (wherein the server can be a local server or a cloud server). As Figure 1 shown, the store site selection model training method can include the following operations:

[0066] 101、Determine the position information corresponding to the plurality of target store information of the target site industry.

[0067] Optionally, the site industry can be an industry type divided by different standards, such as catering industry, hotel industry, or retail industry, etc., which is not limited by the present application. Optionally, the site area can be an area selected by the site user to open a store, which can be a specific area under a preset area division rule, such as a specific city, county, district, etc. under the administrative area division rule, or a specific area under a specific spatial expression relationship restriction, such as an area within a specific radius with XX position point as the center, which is not limited by the present application. Optionally, the position information can be GPS position information, latitude and longitude position information, or position information based on other positioning rules, which is not limited by the present application.

[0068] 102、According to the position information corresponding to each target store information, determine the position information of at least one other industry store within the specific area range corresponding to each target store information, and obtain the position information of the plurality of other industry stores.

[0069] Optionally, the specific area range can be a circular range or other shape range with each target store information corresponding position information as the center or center, and a specific distance as the radius or other extension distance. Specifically, the size of the specific area range can be determined according to the sales influence of the target store information, which can be proportional to the sales influence. Optionally, the sales influence can be determined by calculating the weighted sum value of the business circle influence representation value of the location of the target store information, the average sales of the target store in a specific time period, and the marketing advertising investment cost of the target store.

[0070] In a specific embodiment, for the industry to be modeled, all brands and their store location information in the industry are queried in the self-owned industry database, and the geographical location of the store opened in the current industry is taken as a positive example. For each positive example sample, we filter the store location of non-current industry within its radius range as a negative example. The selection range of the radius is an adjustable hyperparameter, and the positive and negative example samples and corresponding feature data can be used as a training set to train the model.

[0071] 103、According to the position information corresponding to each target store information or other industry store, determine the spatio-temporal feature data corresponding to each target store information or other industry store.

[0072] 104、Use the position information and spatio-temporal feature data corresponding to all target store information as positive sample training data set, and use the position information and spatio-temporal feature data corresponding to all other industry stores as negative sample training data set.

[0073] 105、inputting the positive sample training data set and the negative sample training data set into the site selection score training algorithm model for training until convergence, to obtain a site selection score algorithm model.

[0074] Optionally, the site selection score algorithm model can include at least one of a GBDT algorithm model, an MLP algorithm model, a ResNet algorithm model, a NODE algorithm model, a TabNet algorithm model, and an algorithm model based on a Transformer architecture.

[0075] In one specific embodiment, a multi-source algorithm library is designed to model the site selection score algorithm model, wherein the multi-source algorithm library contains machine learning algorithms and deep learning algorithms. Among the machine learning algorithms, we choose the current mainstream GBDT series algorithms, including XGBoost, LightGBM and CatBoost, each of which has its own characteristics. For the GBDT series algorithms, we respectively adopt the Bayesian parameter search based on the Gaussian process and the TPE process. For the rapidly developing deep learning algorithms in recent years, we select several representative algorithms from different angles. First, the multi-layer perception MLP algorithm model with a relatively simple model architecture, second, the ResNet algorithm which plays an important role in the field of computer vision, and the NODE (Neural Oblivious Decision Ensembles) algorithm and the TabNet algorithm which integrate decision trees through general deep learning. In recent years, algorithms based on the Transformer architecture have achieved great success in various fields, breaking multiple SOTA indicators. In this invention, the Transformer algorithm is also included in the multi-source algorithm library. Since the original Transformer is designed for seq2seq structure, we choose the FT-Transformer (Feature Tokenizer + Transformer) algorithm based on the Transformer modification to adapt to our data, in which numerical and categorical variables are converted into Embedding (embedding) input into the Transformer network through the Feature Tokenizer network. Further, after the training of multiple models is completed, the algorithms with better performance can be selected for further processing, for example, the algorithm model with better performance is determined as the final model for prediction.

[0076] Optionally, further, in actual prediction, the site selection industry input by the site selection user and the plurality of candidate location points can be acquired, and the spatio-temporal feature data of the site selection industry corresponding to each candidate location point is determined, the spatio-temporal feature data of the site selection industry corresponding to each candidate location point is input into the trained site selection scoring algorithm model to obtain the site selection scoring parameter corresponding to each candidate location point, and finally the target store site corresponding to the site selection user is determined according to the site selection scoring parameters corresponding to all candidate location points.

[0077] Optionally, all candidate location points can be arranged in descending order of site selection scoring parameters to obtain a location list, and the location list is displayed to the site selection user for selection of the target store site, or the first preset number of candidate location points in the location list is directly determined as the target store site corresponding to the site selection user.

[0078] In a specific embodiment, the user inputs the store to be site-selected and the region where the user is located. The region can be a series of candidate address points or a travel region such as a street, a county, or a city. The selected region is mapped to the corresponding spatial grid according to the geographical position, and then the corresponding features of the grid are extracted from the industry feature library to form the spatio-temporal feature data to be predicted. All grids and corresponding spatio-temporal feature data are input into the site selection scoring algorithm model to obtain the evaluation scores and the ranking list of all grids in the region. Finally, the position information with high scores is displayed on a real-time large screen to show the user.

[0079] As can be seen, the above embodiment can filter the target store information of the target industry and the target store information of other industries in the location range of the target industry to obtain the positive and negative sample sets, train the algorithm model, so that the trained algorithm model can provide the user with industry-precise site selection services when performing site selection prediction, and improve the accuracy and efficiency of site selection.

[0080] As an optional implementation, in the step 103, the spatio-temporal feature data corresponding to each target store information or other industry store is determined according to the location information corresponding to each target store information or other industry store, and includes:

[0081] According to the location information corresponding to each target store information or other industry store and a preset location region mapping relationship, the region information corresponding to each target store information or other industry store is determined;

[0082] According to the region information corresponding to each target store information or other industry store, the spatio-temporal feature data corresponding to each target store information or other industry store is determined.

[0083] Optionally, the location region mapping relationship can be calculated in the following way:

[0084] obtaining store information of a plurality of stores of the industry;

[0085] determining location information and area information corresponding to each store information;

[0086] determining each area information and the location information of the store information of all industries in the corresponding area as a location-area mapping relationship.

[0087] One of the purposes of the embodiment is to collect the location information of the store information of a plurality of industries and determine the corresponding area information to obtain the location-area mapping relationship, so that the corresponding area of the location can be determined during training.

[0088] Optionally, the spatio-temporal feature data includes at least one of geographic feature data, commercial feature data, and user feature data, wherein the geographic feature data includes at least one of point density feature, public transportation convenience feature, and road accessibility feature. The commercial feature data includes at least one of proportion feature of each commercial category, competitive level feature, commercial diversity level feature, and commercial reputation distribution feature. The user feature data includes at least one of resident population distribution feature, working population distribution feature, user age distribution feature, and aggregated trajectory feature in different time periods.

[0089] It can be seen that by implementing the optional embodiment, the area information corresponding to each target store information or other industry store can be determined, and the spatio-temporal feature data corresponding to each target store information or other industry store can be determined accordingly, so that the spatio-temporal feature data accurately representing the characteristics of the area and the store can be determined, and a model with excellent prediction effect can be quickly and efficiently trained during training, thereby improving the training efficiency.

[0090] As an optional embodiment, in the above step, determining the spatio-temporal feature data corresponding to each target store information or other industry store according to the area information corresponding to each target store information or other industry store includes:

[0091] determining the area spatio-temporal data corresponding to each target store information or other industry store from a preset area spatio-temporal database according to the area information corresponding to each target store information or other industry store;

[0092] inputting each target store information or other industry store into a computing model based on a Spark framework to calculate the spatio-temporal feature data corresponding to each target store information or other industry store.

[0093] Optionally, the regional spatio-temporal data includes at least one of POI point data, traffic station data, road network structure data, industry distribution data, population statistics data, store review data, people flow trajectory data, logistics trajectory data, and vehicle flow trajectory data.

[0094] In one specific embodiment, the regional spatio-temporal data corresponding to the grid is extracted from the multi-source data warehouse by the Hive data warehouse tool. The regional spatio-temporal data can be mainly divided into static spatio-temporal data and dynamic spatio-temporal data. The static spatio-temporal data mainly includes POI point data, traffic station data, road network structure data, industry distribution data, population statistics data, and store review data. The dynamic spatio-temporal data mainly includes trajectory information of people flow, logistics flow, and vehicle flow. After obtaining the spatio-temporal data on the grid, a distributed feature engineering pipeline is constructed by using the Spark framework to process the feature data and construct high-order features. Specifically, three categories of spatio-temporal feature data can be calculated based on the regional spatio-temporal data by using the feature engineering pipeline. First, geographical space-related features, including point density, public transportation convenience, and road accessibility. Second, business features, including the proportion of each business category, competitive level, business diversity level, and business reputation distribution. Third, people-related features, including resident population distribution, working population distribution, age distribution, and aggregated trajectory in different time periods. After feature engineering, the industry dataset required for model prediction or training is obtained, and the entire data preparation process is completed in an offline distributed state.

[0095] As can be seen, by implementing the optional embodiment, the input of each target store information or other industry store corresponding to the calculation model based on the Spark framework can be realized to calculate the spatio-temporal feature data corresponding to each target store information or other industry store. Thus, the spatio-temporal feature data accurately representing the characteristics of the region and the store can be determined. Subsequently, a model with excellent prediction effect can be quickly and efficiently trained during training, thereby improving the training efficiency.

[0096] As an optional embodiment, the method further includes:

[0097] When the positive sample training dataset and the negative sample training dataset are input into the site selection score training algorithm model for training until convergence, the site selection score training algorithm model is obtained, and the feature evaluation algorithm model in the site selection score training algorithm model is used to evaluate and filter the spatio-temporal feature data in the positive sample training dataset and / or the negative sample training dataset to obtain preferred feature data with higher importance level.

[0098] Optionally, the feature evaluation algorithm model includes at least one of a GBDT algorithm model and a TabNet algorithm model.

[0099] Correspondingly, in the above step, each target store information or other industry store corresponding input into the calculation model based on the Spark framework, to calculate each target store information or other industry store corresponding spatio-temporal feature data, including:

[0100] Input each target store information or other industry store corresponding into the calculation model based on the Spark framework, to calculate each target store information or other industry store corresponding multiple candidate spatio-temporal feature data;

[0101] The candidate spatio-temporal feature data with the same data parameters as the preferred feature data in the multiple candidate spatio-temporal feature data corresponding to each target store information are determined as the spatio-temporal feature data corresponding to each target store information or other industry store.

[0102] In a specific embodiment, for algorithms such as GBDT series algorithm or TabNet algorithm that can evaluate the importance of original features, the data features generated by the training model are analyzed by using the algorithm, and the features with high importance level are screened out for further guiding feature engineering and optimizing the model.

[0103] As can be seen, through the optional implementation, the preferred feature data with higher importance level can be obtained by using the feature evaluation algorithm model, and in the subsequent step, the candidate spatio-temporal feature data with the same data parameters as the preferred feature data in the multiple candidate spatio-temporal feature data corresponding to each store information or other industry store are determined as the spatio-temporal feature data corresponding to each store information or other industry store, so that the feature evaluation operation can be performed by using part of the algorithm model to finally obtain the spatio-temporal feature data with higher importance level, and the prediction accuracy of the trained model is improved.

[0104] As an optional implementation, in the above step 105, the positive sample training data set and the negative sample training data set are input into the site selection score training algorithm model for training until convergence, and a site selection score algorithm model is obtained, including:

[0105] The positive sample training data set and the negative sample training data set are input into the site selection score training algorithm model for training until convergence, and a trained site selection score teacher algorithm model is obtained;

[0106] The site selection score teacher algorithm model is subjected to model distillation to obtain a site selection score algorithm model.

[0107] In a specific embodiment, for deep models with complex structure and large number of parameters, such as FT-Transformer, after training the model, the model distillation method can be used to distill the original model as a teacher model into a relatively lightweight student model, so as to make fast prediction online.

[0108] It can be seen that by implementing the optional embodiment, model distillation can be performed on the site selection score teacher algorithm model trained according to the positive sample training data set and the negative sample training data set to obtain a site selection score algorithm model, so as to minimize the parameter volume of the model while retaining the model prediction capability, and then the prediction operation amount can be effectively reduced and the prediction efficiency can be improved during subsequent prediction.

[0109] Embodiment two

[0110] Please refer to Figure 2 , Figure 2 is a structural schematic diagram of a shop site selection model training device disclosed by an embodiment of the application. Wherein, Figure 2 The shop site selection model training device described is applied in a shop site selection model training chip, a training terminal or a training server (wherein, the server can be a local server or a cloud server). As shown in Figure 2 The shop site selection model training device can include:

[0111] The first determination module 201 is configured to determine the position information corresponding to the plurality of target shop information of the target site selection industry.

[0112] Optionally, the site selection industry can be an industry type divided by different standards, such as the catering industry, the hotel industry or the retail industry, etc., which is not limited by the application. Optionally, the site selection area can be an area selected by a site selection user who wants to open a shop, which can be a specific area under a preset area division rule, such as a specific city, county or district under an administrative area division rule, or a specific area under a specific spatial expression relationship restriction, such as an area within a specific radius with XX position point as the center, which is not limited by the application. Optionally, the position information can be GPS position information, latitude and longitude position information or position information based on other positioning rules, which is not limited by the application.

[0113] The second determination module 202 is configured to determine the position information of at least one other industry shop within the specific area range corresponding to each target shop information according to the position information corresponding to each target shop information, and obtain the position information of the plurality of other industry shops.

[0114] Optionally, the specific area range can be a circular range or other shape range with the position information corresponding to each target store information as the center of the circle or the center, and a specific distance as the radius or other extension distance. Specifically, the size of the specific area range can be determined according to the sales influence of the target store information, which can be proportional to the sales influence. Optionally, the sales influence can be determined by calculating the weighted sum of the business circle influence representation value of the location of the target store information, the average sales of the target store in a specific time period, and the marketing advertising investment cost of the target store.

[0115] In a specific embodiment, for the industry to be modeled, the location information of all brands and their stores in the industry is queried in the self-owned industry database, and the geographical location where the stores of the current industry have been opened is taken as a positive example. For each positive example sample, we filter the store locations of non-current industries within its radius range as negative examples. The selection range of the radius is an adjustable hyperparameter, and the positive and negative example samples and the corresponding feature data can be used as a training set to train the model.

[0116] The feature determination module 203 is configured to determine the spatio-temporal feature data corresponding to each target store information or other industry store according to the position information corresponding to each target store information or other industry store.

[0117] The sample determination module 204 is configured to take the position information and spatio-temporal feature data corresponding to all target store information as positive sample training data set, and take the position information and spatio-temporal feature data corresponding to all other industry stores as negative sample training data set.

[0118] The model training module 205 is configured to input the positive sample training data set and the negative sample training data set into the site selection score training algorithm model for training until convergence, to obtain the site selection score algorithm model.

[0119] Optionally, the site selection score algorithm model can include at least one of GBDT algorithm model, MLP algorithm model, ResNet algorithm model, NODE algorithm model, TabNet algorithm model, and algorithm model based on Transformer architecture.

[0120] In one specific embodiment, a multi-source algorithm library is designed to model the site selection scoring algorithm model, wherein the multi-source algorithm library includes machine learning algorithms and deep learning algorithms. In the machine learning algorithms, we choose the current mainstream GBDT series algorithms, including XGBoost, LightGBM and CatBoost, which have different characteristics. For the GBDT series algorithms, we use Gaussian process-based and TPE process-based Bayesian parameter search respectively. For the rapidly developing deep learning algorithms in recent years, we select several representative algorithms from different angles. First, the multi-layer perception MLP algorithm model with a relatively simple model architecture, second, the ResNet algorithm which plays an important role in computer vision, and the NODE (Neural Oblivious Decision Ensembles) algorithm and TabNet algorithm which are general deep learning to integrate decision trees. In recent years, algorithms based on the Transformer architecture have achieved great success in various fields, breaking many SOTA indicators. The Transformer algorithm is also included in the multi-source algorithm library in the present application. Since the original Transformer is designed for seq2seq structure, we choose the FT-Transformer (Feature Tokenizer+Transformer) algorithm based on the modification of Transformer to adapt to our data, in which numerical and categorical variables are converted into Embedding input into the Transformer network through the Feature Tokenizer network. Further, after training multiple models, the better performing algorithms can be selected for further processing, for example, the better performing algorithm model is determined as the final prediction model.

[0121] Optionally, further, during actual prediction, the site selection industry input by the site selection user and the plurality of candidate location points can be obtained, and the spatio-temporal feature data of the site selection industry corresponding to each candidate location point is determined. The spatio-temporal feature data of the site selection industry corresponding to each candidate location point is input into the trained site selection scoring algorithm model to obtain the site selection scoring parameter corresponding to each candidate location point. Finally, the target store site corresponding to the site selection user is determined according to the site selection scoring parameters corresponding to all candidate location points.

[0122] Optionally, all candidate location points can be arranged in descending order of site selection scoring parameters to obtain a location list, and the location list is displayed to the site selection user for selection of the target store site, or the first pre-set number of candidate location points in the location list are directly determined as the target store site corresponding to the site selection user.

[0123] In a specific embodiment, the user inputs the store and the region where the user wants to locate the store. The region can be a series of candidate address points or a travel region such as a street, a county, or a city. The selected region is mapped to a corresponding spatial grid according to the geographical position, and then the corresponding features of the grid are extracted from the industry feature library to form the spatio-temporal feature data to be predicted. All the grids and the corresponding spatio-temporal feature data are input into the site selection scoring algorithm model to obtain the evaluation scores and the ranking list of all the grids in the region. Finally, the position information with the highest score is displayed on a real-time large screen to show the user.

[0124] It can be seen that the above embodiment can filter the target store information of the target industry and the target store information of other industries in the location range of the target store information to obtain the positive and negative sample sets, so as to train the algorithm model. Thus, the algorithm model obtained by training can provide the industry-precise site selection service for the user when performing site selection prediction, and improve the accuracy and efficiency of site selection.

[0125] As an optional implementation, the feature determination module 203 determines the specific manner of determining the spatio-temporal feature data corresponding to each target store information or other industry store according to the location information corresponding to each target store information or other industry store, which includes:

[0126] According to the location information corresponding to each target store information or other industry store and the preset location region mapping relationship, the region information corresponding to each target store information or other industry store is determined.

[0127] According to the region information corresponding to each target store information or other industry store, the spatio-temporal feature data corresponding to each target store information or other industry store is determined.

[0128] Optionally, the location region mapping relationship can be calculated in the following manner:

[0129] For any industry, a plurality of store information of the industry is obtained.

[0130] The location information and the region information corresponding to each store information are determined.

[0131] Each region information and the location information of the store information of all industries in the corresponding region are determined as the location region mapping relationship.

[0132] One of the purposes of the embodiment is to collect the location information of the store information of a plurality of industries and determine the corresponding region information to obtain the location region mapping relationship, so that the corresponding region of the location can be determined during training.

[0133] Optionally, the spatio-temporal feature data comprises at least one of geographic feature data, commercial feature data and user feature data, wherein the geographic feature data comprises at least one of point density feature, public transportation convenience feature and road accessibility feature; the commercial feature data comprises at least one of proportion feature of each commercial category, competitive level feature, commercial diversity level feature and commercial reputation distribution feature; and the user feature data comprises at least one of resident population distribution feature, working population distribution feature, user age distribution feature and aggregated trajectory feature in different time periods.

[0134] It can be seen that, by implementing the optional embodiment, the region information corresponding to each target store information or other industry store can be determined, and the spatio-temporal feature data corresponding to each target store information or other industry store can be determined accordingly, so that the spatio-temporal feature data accurately representing the characteristics of the region and the store can be determined, and a model with excellent prediction effect can be quickly and efficiently trained in subsequent training, thereby improving the training efficiency.

[0135] As an optional embodiment, the specific manner in which the feature determination module 203 determines the spatio-temporal feature data corresponding to each target store information or other industry store according to the region information corresponding to each target store information or other industry store comprises:

[0136] According to the region information corresponding to each target store information or other industry store, region spatio-temporal data corresponding to each target store information or other industry store is determined from a preset region spatio-temporal database.

[0137] Each target store information or other industry store is input into a computing model based on a Spark framework, so as to calculate the spatio-temporal feature data corresponding to each target store information or other industry store.

[0138] Optionally, the region spatio-temporal data comprises at least one of POI point data, traffic station data, road network structure data, industry distribution data, population statistics data, store review data, people flow trajectory data, logistics trajectory data and vehicle flow trajectory data.

[0139] In a specific embodiment, the region space-time data corresponding to the grid is extracted from the multi-source data warehouse by the Hive data warehouse tool, and the region space-time data can be mainly divided into static space-time data and dynamic space-time data. The static space-time data mainly includes POI point data, traffic station data, road network structure data, industry distribution data, population statistics data, and store review data, and the dynamic space-time data mainly includes trajectory information such as people flow, material flow, and vehicle flow. After obtaining the space-time data on the grid, a distributed feature engineering pipeline is constructed by using the Spark framework to process the feature data and construct high-order features. Specifically, based on the region space-time data, three categories of space-time feature data can be calculated by using the feature engineering pipeline. First, the geographical space-related features mainly include point density, public transportation convenience, and road accessibility. Second, the business features mainly include the proportion of various business categories, the level of competition, the level of business diversity, and the distribution of business reputation. Finally, the people-related features mainly include the distribution of resident population, the distribution of working population, the distribution of age groups, and aggregated trajectories in different time periods. After feature engineering, the industry data set required for model prediction or training is obtained, and the entire data preparation process is completed in an offline distributed state.

[0140] It can be seen that by implementing the optional embodiment, the input of each target store information or other industry store corresponding to the Spark framework-based computing model can be calculated to obtain the space-time feature data corresponding to each target store information or other industry store, so as to determine the space-time feature data accurately representing the characteristics of the region and the store. Subsequently, a model with excellent prediction effect can be quickly and efficiently trained during training, thereby improving the training efficiency.

[0141] As an optional embodiment, the device further includes a feature evaluation module for evaluating and screening the space-time feature data in the positive sample training data set and / or the negative sample training data set by using a feature evaluation algorithm model in the site selection score training algorithm model when the model training module 205 executes the input of the positive sample training data set and the negative sample training data set into the site selection score training algorithm model for training until convergence, to obtain preferred feature data with higher importance level.

[0142] Optionally, the feature evaluation algorithm model includes at least one of a GBDT algorithm model and a TabNet algorithm model.

[0143] Correspondingly, the specific way in which the feature determination module 203 inputs each target store information or other industry store into the Spark framework-based computing model to calculate the space-time feature data corresponding to each target store information or other industry store includes:

[0144] The input corresponding to each target store information or other industry store is input into a computing model based on a Spark framework to calculate a plurality of candidate spatio-temporal feature data corresponding to each target store information or other industry store;

[0145] The candidate spatio-temporal feature data corresponding to each target store information and having the same data parameters as the preferred feature data are determined as the spatio-temporal feature data corresponding to each target store information or other industry store.

[0146] In a specific embodiment, for algorithms such as GBDT series algorithms or TabNet algorithms that can evaluate the importance of original features, the data features generated by the training model are analyzed by using the algorithms to filter out features with higher importance levels for further guiding feature engineering and optimizing the model.

[0147] As can be seen, through the optional implementation, the preferred feature data with higher importance levels can be obtained by using the feature evaluation algorithm model for evaluation and screening, and in the subsequent process, the candidate spatio-temporal feature data corresponding to each store information or other industry store and having the same data parameters as the preferred feature data are determined as the spatio-temporal feature data corresponding to each store information or other industry store, so that the feature evaluation operation can be performed by using part of the algorithm model to finally obtain the spatio-temporal feature data with higher importance levels and improve the prediction accuracy of the trained model.

[0148] As an optional implementation, the model training module 205 inputs the positive sample training data set and the negative sample training data set into the site selection score training algorithm model for training until convergence to obtain the specific manner of the site selection score algorithm model, including:

[0149] The positive sample training data set and the negative sample training data set are input into the site selection score training algorithm model for training until convergence to obtain the trained site selection score teacher algorithm model.

[0150] The site selection score teacher algorithm model is subjected to model distillation to obtain the site selection score algorithm model.

[0151] In a specific embodiment, for deep models with complex structures and large parameters, such as FT-Transformer, after the model is trained, the model distillation method can be used to distill the original model as a teacher model into a relatively lightweight student model to make fast predictions online.

[0152] It can be seen that by implementing the optional embodiment, the site selection score algorithm model can be obtained by model distillation on the site selection score teacher algorithm model trained according to the positive sample training data set and the negative sample training data set, so as to minimize the parameter volume of the model while retaining the model prediction capability, and then the prediction operation amount can be effectively reduced and the prediction efficiency can be improved during subsequent prediction.

[0153] Embodiment three

[0154] Please refer to Figure 3 , Figure 3 The shop site selection model training device disclosed in the embodiment of the application is another shop site selection model training device. Figure 3 The shop site selection model training device is applied to a shop site selection model training chip, a training terminal or a training server (wherein the server can be a local server or a cloud server). As shown in Figure 3 The shop site selection model training device can include:

[0155] a memory 301 storing executable program codes;

[0156] a processor 302 coupled with the memory 301;

[0157] The processor 302 invokes the executable program codes stored in the memory 301, and is configured to execute the steps of the shop site selection model training method described in embodiment one.

[0158] Embodiment four

[0159] The embodiment of the application discloses a computer readable storage medium storing a computer program for electronic data exchange, wherein the computer program causes a computer to execute the steps of the shop site selection model training method described in embodiment one.

[0160] Embodiment five

[0161] The embodiment of the application discloses a computer program product, which includes a non-transitory computer readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute the steps of the shop site selection model training method described in embodiment one.

[0162] The above describes certain embodiments of the present application, and other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order and still achieve desirable results. Additionally, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0163] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer-readable storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0164] The apparatus, device, non-volatile computer-readable storage medium and method provided in the embodiments of this specification are corresponding. Therefore, the apparatus, device and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, device and non-volatile computer storage medium will not be repeated here.

[0165] In the 1990s, it was quite obvious to distinguish whether an improvement in a technology was in hardware (e.g., improvement in circuit structures of diodes, transistors, switches, etc.) or in software (improvement in method flow). However, as technology has evolved, many improvements in method flow today can be considered as direct improvements in hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structures by programming the improved method flow into hardware circuits. Therefore, it cannot be said that an improvement in a method flow cannot be implemented by hardware entity modules. For example, a programmable logic device (PLD) (e.g., a field programmable gate array (FPGA)) is an integrated circuit whose logic function is determined by user programming of the device. A digital system is "integrated" on a PLD by the designer programming it, rather than by asking a chip manufacturer to design and fabricate a custom integrated circuit chip. Moreover, instead of manually fabricating integrated circuit chips, this programming is now mostly implemented by "logic compiler" software, which is similar to software compilers used in program development, and the original code to be compiled is written in a specific programming language, which is called a hardware description language (HDL), and there are many such languages, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., and the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should be aware that, as long as the method flow is logically programmed in the above-mentioned hardware description languages and programmed into an integrated circuit, a hardware circuit implementing the logical method flow can be easily obtained.

[0166] The controller can be implemented in any suitable way, for example, the controller can take the form of a microprocessor or processor and a computer readable medium storing computer readable program code, such as software or firmware, executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that, in addition to being implemented in pure computer readable program code, the controller can equally well be implemented to perform the same functions using logic gates, switches, an application specific integrated circuit, a programmable logic controller and an embedded microcontroller, etc. by means of a logical programming of the method steps. The controller can thus be considered as a hardware component, and the means comprised therein for performing the various functions can be considered as structures within the hardware component. Alternatively, the means for performing the various functions can even be considered as both a software module implementing the method and a structure within the hardware component.

[0167] The systems, apparatuses, modules or units illustrated by the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0168] For the sake of description, the above apparatuses are described in functional division and are described respectively. Of course, the functions of the units can be implemented in the same or multiple software and / or hardware when implementing the present specification.

[0169] Those skilled in the art will understand that the embodiments of the present specification can be provided as a method, a system or a computer program product. Therefore, the embodiments of the present specification can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0170] The specification is presented with reference to flow diagrams and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. It will be understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing element or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the specification can present a method, apparatus or computer program product according to embodiments of the specification. Flow diagrams and / or block diagrams can also present a method, apparatus or computer program product to achieve functions specified in flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the specification can present a method, apparatus or computer program product according to embodiments of the specification. Flow diagrams and / or block diagrams can also present a method, apparatus or computer program product to achieve functions specified in flow diagrams and / or block diagrams block or blocks.

[0171] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the specification can present a method, apparatus or computer program product according to embodiments of the specification. Flow diagrams and / or block diagrams can also present a method, apparatus or computer program product to achieve functions specified in flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the specification can present a method, apparatus or computer program product according to embodiments of the specification. Flow diagrams and / or block diagrams can also present a method, apparatus or computer program product to achieve functions specified in flow diagrams and / or block diagrams block or blocks.

[0172] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the specification can present a method, apparatus or computer program product according to embodiments of the specification. Flow diagrams and / or block diagrams can also present a method, apparatus or computer program product to achieve functions specified in flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the specification can present a method, apparatus or computer program product according to embodiments of the specification. Flow diagrams and / or block diagrams can also present a method, apparatus or computer program product to achieve functions specified in flow diagrams and / or block diagrams block or blocks.

[0173] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0174] The memory can include non-persistent memory and / or storage mechanisms such as, for example, random access memory (RAM), non-volatile memory (NVM), and / or a persistent memory such as, for example, read-only memory (ROM) or flash memory. The memory is an example of computer-readable media.

[0175] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0176] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to encompass non-exclusive inclusion, such that processes, methods, articles or devices that comprise a list of elements not only include those elements, but also include other elements not expressly listed or inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0177] The specification can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The specification can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0178] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0179] Finally, it should be noted that: the shop site selection model training method and device disclosed in the embodiment of the application disclosed only for the preferred embodiment of the application, only for the description of the technical solutions of the application, not to limit; although the application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand; it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A method for training a store location selection model, characterized in that, The method comprises: determining position information corresponding to a plurality of target store information of a target site selection industry; determining position information of at least one other industry store within a specific area range corresponding to each of the target store information according to the position information corresponding to each of the target store information, to obtain position information of a plurality of the other industry stores; determining area information corresponding to each of the target store information or the other industry store according to the position information corresponding to each of the target store information or the other industry store and a preset position area mapping relationship; determining area space-time data corresponding to each of the target store information or the other industry store from a preset area space-time database according to the area information corresponding to each of the target store information or the other industry store; the area space-time data comprises at least one of POI point data, traffic station data, road network structure data, industry distribution data, population statistics data, store review data, people flow trajectory data, logistics trajectory data and vehicle flow trajectory data; inputting the area space-time data corresponding to each of the target store information or the other industry store into a computing model based on a Spark framework to calculate space-time feature data corresponding to each of the target store information or the other industry store; the space-time feature data comprises geographical space-related features, business features and people-related features; the geographical space-related features comprise point density, public transportation convenience and road accessibility; the business features comprise proportion of each business category, competitiveness level, business diversity level and business reputation distribution; the people-related features comprise resident population distribution, working population distribution, age distribution and aggregated trajectory in different time periods; taking the position information and the space-time feature data corresponding to all the target store information as positive sample training data set and taking the position information and the space-time feature data corresponding to all the other industry stores as negative sample training data set; inputting the positive sample training data set and the negative sample training data set into a site selection score training algorithm model for training until convergence to obtain a site selection score algorithm model. 2.The store location modeling training method of claim 1, wherein, The space-time feature data comprises at least one of geographical feature data, business feature data and user feature data; the geographical feature data comprises at least one of point density feature, public transportation convenience feature and road accessibility feature; the business feature data comprises at least one of proportion feature of each business category, competitiveness level feature, business diversity level feature and business reputation distribution feature; the user feature data comprises at least one of resident population distribution feature, working population distribution feature, user age distribution feature and aggregated trajectory feature in different time periods. 3.The store location modeling training method of claim 1, wherein, The site selection score training algorithm model and / or the site selection score algorithm model comprises at least one of GBDT algorithm model, MLP algorithm model, ResNet algorithm model, NODE algorithm model, TabNet algorithm model and algorithm model based on Transformer architecture. 4.The store location modeling training method of claim 1, wherein, The method further comprises: In the execution of the inputting the positive sample training data set and the negative sample training data set into the site scoring training algorithm model for training until convergence to obtain the site scoring algorithm model, the feature evaluation algorithm model in the site scoring training algorithm model is used to evaluate and screen the spatio-temporal feature data in the positive sample training data set and / or the negative sample training data set to obtain preferred feature data with higher importance level. In addition, the inputting the regional spatio-temporal data corresponding to each of the target store information or the other industry stores into the computing model based on the Spark framework to calculate the spatio-temporal feature data corresponding to each of the target store information or the other industry stores comprises: Inputting the regional spatio-temporal data corresponding to each of the target store information or the other industry stores into the computing model based on the Spark framework to calculate a plurality of candidate spatio-temporal feature data corresponding to each of the target store information or the other industry stores; Determining the candidate spatio-temporal feature data with the same data parameters as the preferred feature data from the plurality of candidate spatio-temporal feature data corresponding to each of the target store information or the other industry stores as the spatio-temporal feature data corresponding to each of the target store information or the other industry stores. 5.The store location modeling training method of claim 1, wherein, The inputting the positive sample training data set and the negative sample training data set into the site scoring training algorithm model for training until convergence to obtain the site scoring algorithm model comprises: Inputting the positive sample training data set and the negative sample training data set into the site scoring training algorithm model for training until convergence to obtain a trained site scoring teacher algorithm model; Performing model distillation on the site scoring teacher algorithm model to obtain the site scoring algorithm model. 6.A shop site selection model training apparatus, comprising: The device is used to execute the store site selection model training method according to any one of claims 1-5, and the device comprises: A first determination module configured to determine position information corresponding to a plurality of target store information of a target site selection industry; A second determination module configured to determine, according to the position information corresponding to each of the target store information, position information of at least one other industry store within a specific area range corresponding to each of the target store information, to obtain position information of a plurality of the other industry stores; A feature determination module configured to determine, according to the position information corresponding to each of the target store information or the other industry store, spatio-temporal feature data corresponding to each of the target store information or the other industry store; A sample determination module configured to take the position information and the spatio-temporal feature data corresponding to all the target store information as a positive sample training data set, and take the position information and the spatio-temporal feature data corresponding to all the other industry stores as a negative sample training data set; A model training module configured to input the positive sample training data set and the negative sample training data set into a site scoring training algorithm model for training until convergence to obtain a site scoring algorithm model. 7.A shop site selection model training apparatus, characterized by comprising: The device comprises: A memory storing executable program codes; A processor coupled with the memory; The processor invokes the executable program code stored in the memory to execute the store site selection model training method according to any one of claims 1-5.

Images