E-commerce enterprise migration risk identification and early warning system and method based on Internet data

Abstract

The invention discloses an e-commerce enterprise ex-migration risk identification and early warning system and method based on Internet data, belongs to the technical field of risk monitoring and early warning, and aims to solve the technical problem of how to integrate various data, scientifically evaluate e-commerce enterprise ex-migration risks and carry out early warning in advance. Comprising a data processing module for regularly collecting e-commerce enterprise related data from a plurality of data sources; the feature extraction module is used for constructing dichotomous tags and constructing structured features for each type of structured monitoring data; the external migration risk assessment module is used for predicting and outputting an enterprise external migration risk value through an external migration risk assessment model based on the enterprise external migration risk characteristics; an ex-migration risk early warning module which constructs a risk level system and performs ex-migration risk early warning; and the migration analysis module is used for calculating the retail volume of the migrated enterprise network in the current period, the retail volume of the migrated enterprise network in the current period and the net retail volume of the migrated enterprise network, and carrying out migration statistical analysis, migration flow direction analysis and migration detail analysis.

Description

Technical Field

[0001] This invention relates to the field of risk monitoring and early warning technology, specifically to a risk identification and early warning system and method for the relocation of e-commerce enterprises based on Internet data. Background Technology

[0002] In recent years, the e-commerce industry has developed rapidly. Under the backdrop of the digital economy, e-commerce has played an increasingly prominent role in promoting consumption, stabilizing foreign trade, expanding employment, and facilitating the digital transformation of the economy and society. Currently, the identification of risks associated with the relocation of e-commerce enterprises in various regions faces the following issues:

[0003] 1. Limited data sources: Currently, the government mainly relies on traditional channels such as changes in business registration information and tax data. It often focuses on only a single indicator, such as changes in the registered location of the enterprise or changes in tax payment. It does not make sufficient use of internet data such as land acquisition data, foreign investment data, recruitment data, and cooperation agreements. It is difficult to capture the dynamics of e-commerce enterprises' cross-regional operations (such as shipping and recruitment in different locations) and to identify and comprehensively and accurately assess the risks of enterprises relocating.

[0004] 2. Data fragmentation: Currently, there is a lack of effective collaboration mechanisms among departments such as commerce and taxation, resulting in poor information communication and data sharing. This makes it impossible to form a joint force to comprehensively and accurately identify and assess the risks of enterprise relocation.

[0005] 3. Information lag: Most existing methods are post-event analyses, which often only discover risks after the company has completed the change of business address or the transfer of core business, making it impossible to provide early warnings and resulting in delayed response measures.

[0006] How to integrate various data, scientifically assess the risks of e-commerce companies relocating, and provide early warnings is a technical problem that needs to be solved. Summary of the Invention

[0007] The technical objective of this invention is to address the above-mentioned shortcomings by providing a system and method for identifying and warning of the relocation risks of e-commerce enterprises based on internet data, in order to solve the technical problem of how to integrate multiple data sources, scientifically assess the relocation risks of e-commerce enterprises, and provide early warnings.

[0008] In a first aspect, the present invention provides an e-commerce enterprise relocation risk identification and early warning system based on Internet data, comprising a data processing module, a feature extraction module, a relocation risk assessment module, a relocation risk early warning module, and a migration analysis module;

[0009] The data processing module is used to periodically collect relevant data from e-commerce companies from multiple data sources, obtain various types of data as monitoring data, and preprocess the monitoring data to obtain structured monitoring data.

[0010] The feature extraction module is used to construct binary labels with the target variable of whether e-commerce companies have relocated, and to construct structured features for each type of structured monitoring data. The extracted structured features are then optimized and standardized to obtain the relocation risk features of the companies.

[0011] The relocation risk assessment module is equipped with a relocation risk assessment model, which is used to predict and output the relocation risk value of an enterprise based on its relocation risk characteristics. The relocation risk assessment model is a machine learning model constructed through a two-step method of base model training and model fusion.

[0012] The relocation risk early warning module is used to construct a risk level system based on the historical distribution of enterprise relocation risk values, conduct relocation risk early warning based on the current enterprise risk level status, and generate and push early warning signals.

[0013] The migration analysis module is used to periodically collect data on enterprise relocation, including the number of enterprises relocating in the current period, the number of enterprises relocating out in the current period, and the net number of enterprises relocating in. Combined with the collected e-commerce platform data, it calculates the online retail sales of enterprises relocating in the current period, the online retail sales of enterprises relocating out in the current period, and the online retail sales of net enterprises relocating in the current period. It also calculates the corresponding cumulative indicator data and growth rate data, performs migration statistical analysis, migration flow analysis, and migration detail analysis, and displays the results of migration statistical analysis, migration flow analysis, and migration detail analysis.

[0014] As a preferred approach, collecting relevant data from e-commerce companies from multiple data sources includes the following steps:

[0015] We used Python web scraping technology to collect publicly available information from local e-commerce platforms as e-commerce platform data. This included information on the online stores' licenses and permits, the business locations of the online stores, the product information listed on the online stores, and the product shipping locations. We removed duplicate data and invalid fields from the collected e-commerce platform data and split the text address information into three levels of structured data: province, city, and district / county.

[0016] Enterprise business registration information is collected through authorized public websites, including registered capital information, business scope, legal representative, enterprise status, and business registration change records. The registered capital information is corrected and standardized to a unified unit, and the enterprise status is standardized. The timestamps and fields before and after the changes in the business registration change records are extracted to form a change trajectory dataset, including the registered capital before and after the change, the business scope before and after the change, the legal representative before and after the change, and the changes in enterprise status.

[0017] By collecting publicly available industry information, we collect and process data on enterprises' external investments, including time, enterprise name, project name, province of the project, city of the project, district / county of the project, project details, and investment amount, and standardize the unit for investment amount;

[0018] By collecting and processing publicly available information, data on land acquisition by enterprises is collected, including the time, enterprise name, land parcel name, province, city, district / county of the land parcel, land type, construction progress, and area.

[0019] Obtain enterprise investment and factory construction data from authorized platforms, including construction location, scale, investment amount, construction progress, and expected production time;

[0020] We collect and process corporate recruitment data from various recruitment websites, including information such as time, company name, job title, number of openings, province of work location, city of work location, district / county of work location, job nature, and job responsibilities.

[0021] We collected data on relevant strategic cooperation agreements signed by enterprises through various publicly available news sources, and used NLP technology to extract structured data including the partners, provinces, cities, districts and counties, and content of the cooperation.

[0022] Collect and process enterprise qualification data, including the qualifications of high-tech enterprises, specialized and innovative small and medium-sized enterprises, and specialized and innovative little giants.

[0023] As a preferred option, when constructing binary labels with the target variable of whether e-commerce companies have relocated, the label values ​​are determined based on pre-configured judgment rules, and the label values ​​are divided into positive labels and negative labels.

[0024] When optimizing the extracted structured features, a three-level screening strategy of variance filtering, mutual information entropy, and recursive feature elimination is adopted to remove low-variance features and features with low correlation to the target variable, while retaining a predetermined number of core features.

[0025] When standardizing the extracted structured features, Z-Score standardization is used for numerical features, and one-hot encoding is used for categorical variables.

[0026] As a preferred approach, machine learning models constructed using the two-step method of base model training and model fusion include the following:

[0027] Three base models with different learning logics were selected and trained on the training set respectively. The three base models were logistic regression, random forest and gradient boosting tree.

[0028] The output probabilities of the three base models are used as meta-features. Logistic regression is constructed as the meta-model. The output results of the base models are fused. During fusion, the training set is divided into 5 folds. Each fold of data is predicted by the three base models to obtain 5 sets of base model output probabilities, which are used as the training set of the meta-features. The meta-model takes the meta-features as input and the real migration label as the target to learn the weights of the base model results. The test set first obtains the probability output through the three base models, and then inputs it into the meta-model. Finally, the fused enterprise migration risk value is output.

[0029] Accuracy, precision, recall, and F1 score were used as metrics to validate model performance.

[0030] As a preferred option, the relocation risk early warning module is used to perform the following:

[0031] Risk level classification: Based on the risk value distribution of historical migration data, the K-means clustering algorithm is used to determine the early warning threshold and classify the risk into three levels: low risk, medium risk and high risk.

[0032] Early warning triggering and response: When a company's risk level rises from low risk to medium risk or high risk, an early warning message is generated. The early warning message includes the company's basic information, risk level, and risk value for relocation. The early warning message is sent to relevant departments through early warning push channels, and the latest monitoring data of the company is collected regularly to update the risk value and risk level for relocation.

[0033] Secondly, the present invention provides a method for identifying and warning of risks associated with the relocation of e-commerce enterprises based on internet data, comprising the following steps:

[0034] Data processing: Regularly collect relevant data from e-commerce companies from multiple data sources to obtain various types of data as monitoring data, and preprocess the monitoring data to obtain structured monitoring data;

[0035] Feature extraction: Taking whether e-commerce companies have relocated as the target variable, binary labels are constructed, and structured features are constructed for each type of structured monitoring data. The extracted structured features are then optimized and standardized to obtain the relocation risk features of enterprises.

[0036] Relocation Risk Assessment: Based on the characteristics of enterprise relocation risk, the relocation risk assessment model predicts and outputs the enterprise relocation risk value. The relocation risk assessment model is a machine learning model constructed through a two-step method of base model training and model fusion.

[0037] Relocation Risk Warning: Construct a risk level system based on the historical distribution of enterprise relocation risk values, generate and push warning signals based on the current risk level status of enterprises, and conduct relocation risk warnings.

[0038] Migration Analysis: Regularly collect data on enterprise relocation, including the number of enterprises relocating in the current period, the number of enterprises relocating out in the current period, and the net number of enterprises relocating in. Combined with data collected from e-commerce platforms, calculate the online retail sales of enterprises relocating in the current period, the online retail sales of enterprises relocating out in the current period, and the net online retail sales of enterprises relocating in the current period. Calculate the corresponding cumulative indicator data and growth rate data, conduct migration statistical analysis, migration flow analysis, and migration detail analysis, and display the results of migration statistical analysis, migration flow analysis, and migration detail analysis.

[0039] As a preferred approach, collecting relevant data from e-commerce companies from multiple data sources includes the following steps:

[0040] We used Python web scraping technology to collect publicly available information from local e-commerce platforms as e-commerce platform data. This included information on the online stores' licenses and permits, the business locations of the online stores, the product information listed on the online stores, and the product shipping locations. We removed duplicate data and invalid fields from the collected e-commerce platform data and split the text address information into three levels of structured data: province, city, and district / county.

[0041] Enterprise business registration information is collected through authorized public websites, including registered capital information, business scope, legal representative, enterprise status, and business registration change records. The registered capital information is corrected and standardized to a unified unit, and the enterprise status is standardized. The timestamps and fields before and after the changes in the business registration change records are extracted to form a change trajectory dataset, including the registered capital before and after the change, the business scope before and after the change, the legal representative before and after the change, and the changes in enterprise status.

[0042] By collecting publicly available industry information, we collect and process data on enterprises' external investments, including time, enterprise name, project name, province of the project, city of the project, district / county of the project, project details, and investment amount, and standardize the unit for investment amount;

[0043] By collecting and processing publicly available information, data on land acquisition by enterprises is collected, including the time, enterprise name, land parcel name, province, city, district / county of the land parcel, land type, construction progress, and area.

[0044] Obtain enterprise investment and factory construction data from authorized platforms, including construction location, scale, investment amount, construction progress, and expected production time;

[0045] We collect and process corporate recruitment data from various recruitment websites, including information such as time, company name, job title, number of openings, province of work location, city of work location, district / county of work location, job nature, and job responsibilities.

[0046] We collected data on relevant strategic cooperation agreements signed by enterprises through various publicly available news sources, and used NLP technology to extract structured data including the partners, provinces, cities, districts and counties, and content of the cooperation.

[0047] Collect and process enterprise qualification data, including the qualifications of high-tech enterprises, specialized and innovative small and medium-sized enterprises, and specialized and innovative little giants.

[0048] As a preferred option, when constructing binary labels with the target variable of whether e-commerce companies have relocated, the label values ​​are determined based on pre-configured judgment rules, and the label values ​​are divided into positive labels and negative labels.

[0049] When optimizing the extracted structured features, a three-level screening strategy of variance filtering, mutual information entropy, and recursive feature elimination is adopted to remove low-variance features and features with low correlation to the target variable, while retaining a predetermined number of core features.

[0050] When standardizing the extracted structured features, Z-Score standardization is used for numerical features, and one-hot encoding is used for categorical variables.

[0051] As a preferred approach, machine learning models constructed using the two-step method of base model training and model fusion include the following:

[0052] Three base models with different learning logics were selected and trained on the training set respectively. The three base models were logistic regression, random forest and gradient boosting tree.

[0053] The output probabilities of the three base models are used as meta-features. Logistic regression is constructed as the meta-model. The output results of the base models are fused. During fusion, the training set is divided into 5 folds. Each fold of data is predicted by the three base models to obtain 5 sets of base model output probabilities, which are used as the training set of the meta-features. The meta-model takes the meta-features as input and the real migration label as the target to learn the weights of the base model results. The test set first obtains the probability output through the three base models, and then inputs it into the meta-model. Finally, the fused enterprise migration risk value is output.

[0054] Accuracy, precision, recall, and F1 score were used as metrics to validate model performance.

[0055] As a preferred option, relocation risk warning includes the following steps:

[0056] Risk level classification: Based on the risk value distribution of historical migration data, the K-means clustering algorithm is used to determine the early warning threshold and classify the risk into three levels: low risk, medium risk and high risk.

[0057] Early warning triggering and response: When a company's risk level rises from low risk to medium risk or high risk, an early warning message is generated. The early warning message includes the company's basic information, risk level, and risk value for relocation. The early warning message is sent to relevant departments through early warning push channels, and the latest monitoring data of the company is collected regularly to update the risk value and risk level for relocation.

[0058] The e-commerce enterprise relocation risk identification and early warning system and method based on Internet data of the present invention have the following advantages:

[0059] 1. Comprehensive Data Dimensions: Integrates 7 types of publicly available internet data, covering the entire chain of e-commerce enterprises from "operation-investment-construction-recruitment", solving the problem of single data in traditional methods;

[0060] 2. Proactive Risk Identification: By using feature engineering to uncover potential relocation signals (such as recruitment in other locations or changes in shipping locations), and combining this with weekly dynamic monitoring, relocation risks can be identified in advance.

[0061] 3. High model accuracy: The Stacking model fusion strategy is adopted, with an F1 score ≥ 0.85 and a recall rate ≥ 0.8, reducing the risk of "missed detections".

[0062] 4. Highly practical early warning system: Based on risk level matching response strategies, it provides government departments with decision support for "precisely locating risk points and targeted intervention", helping to reduce the impact of e-commerce companies relocating to the regional economy. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] The invention will be further described below with reference to the accompanying drawings.

[0065] Figure 1 This is a flowchart of a method for identifying and warning of risks associated with the relocation of e-commerce enterprises based on internet data. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0067] This invention provides a system and method for identifying and warning of the relocation risks of e-commerce enterprises based on Internet data, which is used to solve the technical problem of how to integrate multiple data, scientifically assess the relocation risks of e-commerce enterprises, and provide early warnings.

[0068] Example 1:

[0069] This invention discloses an e-commerce enterprise relocation risk identification and early warning system based on Internet data, comprising a data processing module, a feature extraction module, a relocation risk assessment module, a relocation risk early warning module, and a migration analysis module.

[0070] The data processing module is used to periodically collect relevant data from e-commerce companies from multiple data sources, obtain various types of data as monitoring data, and preprocess the monitoring data to obtain structured monitoring data.

[0071] As a specific implementation of the data processing module, collecting relevant data from e-commerce enterprises from multiple data sources includes the following steps:

[0072] (1) Using Python web scraping technology, publicly available information from regional e-commerce platforms was collected as e-commerce platform data. This included information on the store's license and business license status, the store's publicly displayed business location information, the product information listed on the store, and the product's shipping location information. Duplicate data and invalid fields (such as null values ​​for shipping location) were removed from the collected e-commerce platform data. The text address information was then split into three levels of structured data: province, city, and district / county. Specifically, based on the store's license and business license status, the business entity's business registration information was obtained, such as company name, unified social credit code, province of registration, city of registration, and district / county of registration. Based on the store's publicly displayed business location information, the store's publicly displayed business location information on the platform was obtained, namely, the store's province, city, and district / county. Based on the product information in the store, the store's shipping location or main shipping location information was obtained, namely, the store's shipping province, city, and district / county.

[0073] (2) Collect enterprise business information through authorized public websites, including registered capital information, business scope, legal representative, enterprise status and business change records. Correct and unify the registered capital information, standardize the enterprise status (such as relocation, cancellation, etc.), and extract the timestamp and fields before and after the change of business change records to form a change trajectory dataset, including the registered capital before and after the change, the business scope before and after the change, the legal representative before and after the change, and the changes in enterprise status.

[0074] (3) Collect and process enterprise foreign investment data by collecting publicly available industry information, including time, enterprise name, project name, province where the project is located, city where the project is located, district / county where the project is located, project details and investment amount, and standardize the unit of investment amount;

[0075] (4) Collect and process enterprise land acquisition data through public information, including time, enterprise name, land parcel name, province where the land parcel is located, city where the land parcel is located, district / county where the land parcel is located, land parcel nature (such as industrial land, commercial land), construction progress (such as planning, under construction, already in production, etc.) and area (uniformly in square meters), etc.

[0076] (5) Obtain enterprise investment and factory construction data from the authorized platform, including construction location, scale, investment amount, construction progress and expected production time;

[0077] (6) Collect and process enterprise recruitment data from various recruitment websites, including time, company name, job title, number of recruits, province of work location, city of work location, district / county of work location, job nature, and job responsibilities, removing non-full-time positions such as "part-time" and "internship". Based on job title and responsibilities, divide the positions into "core positions" (production, R&D, finance, operations) and "non-core positions" (sales, customer service), laying the foundation for subsequent feature extraction;

[0078] (7) Collect relevant strategic cooperation agreement data signed by enterprises through various public news, using keywords such as "strategic cooperation" and "framework agreement", and use NLP technology (such as BERT model) to extract structured data including partners, provinces, cities, districts and counties, and cooperation content;

[0079] (8) Collect and process enterprise qualification data, including the qualifications of high-tech enterprises, specialized and innovative small and medium-sized enterprises and specialized and innovative little giants.

[0080] The feature extraction module is used to construct binary labels with the target variable of whether e-commerce companies have relocated, and to construct structured features for each type of structured monitoring data. The extracted structured features are then optimized and standardized to obtain the risk features of enterprise relocation.

[0081] As a specific implementation of the feature extraction module, taking whether e-commerce companies have relocated as the target variable, when constructing binary labels, the label values ​​are determined based on pre-configured judgment rules, and the label values ​​are divided into positive and negative labels. The label values ​​are determined according to the following rules:

[0082] Positive label (relocation, label = 1): A company can be identified as relocating if it meets any of the following conditions:

[0083] Condition 1: The "business address" in the business registration information has been changed to an administrative region outside the original registered location, and the business has not been moved back within 6 months after the change;

[0084] Condition 2: Data on land acquisition / investment in factory construction shows that the enterprise has purchased ≥50 mu of industrial land outside its original registered location, or invested ≥10 million yuan in building factory buildings, and the project has been put into production;

[0085] Condition 3: Data from the e-commerce platform shows that the main shipping location of the online store is inconsistent with the original registration location for 6 consecutive months, and the proportion of shipments from different locations is ≥80%.

[0086] Negative label (not relocated, label = 0): Enterprises that have not met any of the above relocation conditions within 12 consecutive months and whose business status is "in operation".

[0087] For the seven types of data, structured features were constructed to form the "Enterprise Relocation Risk Feature Database". The specific feature design is shown in Table 1.

[0088]

[0089]

[0090] When optimizing the extracted structured features, a three-level screening strategy of variance filtering, mutual information entropy, and recursive feature elimination (RFE) is adopted to remove low-variance features (such as "company name") and features with low correlation to the target variable (such as "number of non-core positions"), while retaining a predetermined number of core features.

[0091] When standardizing the extracted structured features (such as "percentage of investment amount in other locations" and "scale of recruitment in other locations"), Z-Score standardization is used for numerical features (mean = 0, standard deviation = 1), and one-hot encoding is used for categorical variables to ensure consistent feature magnitude.

[0092] The relocation risk assessment module is equipped with a relocation risk assessment model, which is used to predict and output the relocation risk value of an enterprise based on its relocation risk characteristics. The relocation risk assessment model is a machine learning model constructed through a two-step method of base model training and model fusion.

[0093] As a specific implementation of the relocation risk assessment module, the machine learning model constructed through a two-step method of base model training and model fusion includes the following:

[0094] (1) Select three base models with different learning logics and train them on the training set respectively. The three base models are logistic regression, random forest and gradient boosting tree.

[0095] (2) The output probabilities of the three base models are used as meta-features. Logistic regression is constructed as the meta-model. The output results of the base models are fused. During fusion, the training set is divided into 5 folds. Each fold of data is predicted by the three base models to obtain 5 sets of base model output probabilities, which are used as the training set of meta-features. The meta-model takes the meta-features as input and the real migration label as the target, and learns the weights of the base model results (e.g., RF weight = 0.4, XGBoost weight = 0.35, LR weight = 0.25). The test set (accounting for 30% of the total data) first obtains the probability output through the three base models, and then inputs it into the meta-model. Finally, the fused enterprise migration risk value (denoted as R, 0≤R≤1) is output.

[0096] (3) The model performance was verified by using accuracy, precision, recall and F1 score as indicators. The F1 score of this model was ≥0.85 and the recall was ≥0.8 (ensuring that as few companies at risk of relocation are missed).

[0097] Logistic Regression (LR) employs L1 regularization (Lasso) to suppress collinearity, with a regularization strength parameter C = 0.1, and outputs the probability of firm relocation (0-1).

[0098] Random Forest (RF): Handles non-linear relationships between features. Sets the number of decision trees to 100, the maximum depth to 10, and the minimum number of sample splits to 5. It evaluates feature importance through out-of-bag (OOB) samples, handles the sample imbalance problem, and outputs the migration probability.

[0099] Gradient Boosting Tree (XGBoost): Optimizes the sample imbalance problem (the proportion of samples from relocated enterprises is usually low). It adopts the "class weight balancing" strategy, sets the learning rate to 0.1, the number of trees to 150, the maximum depth to 8, handles the sample imbalance problem, and outputs the relocation probability.

[0100] The relocation risk early warning module is used to construct a risk level system based on the historical distribution of enterprise relocation risk values, and to generate and push early warning signals based on the current risk level status of the enterprise.

[0101] As a specific implementation of the relocation risk early warning module, a multi-level early warning system is constructed based on the risk value R output by the risk assessment model to achieve accurate early warning through "risk classification and strategy matching".

[0102] Risk Level Classification: Based on the risk value distribution of historical migration data, the K-means clustering algorithm is used to determine the early warning threshold, and the risk is divided into 3 levels:

[0103] Low risk: R < 0.3, the company's operations are stable and there are no obvious signs of relocation;

[0104] Medium risk: 0.3≤R<0.6, indicating that the company is showing signs of relocation (such as increased recruitment in other locations or changes in the shipping location), and requires continuous monitoring;

[0105] High risk: R≥0.6, the enterprise has clearly relocated (such as acquiring land in another location or changing its business address), and immediate intervention is required.

[0106] Early warning triggering and response involve early warning information generation, early warning push, and dynamic early warning updates.

[0107] Warning signal generation: When a company's risk level rises from "low" to "medium" or "high", the system automatically generates a warning message, which includes:

[0108] --Basic information of the enterprise (name, unified social credit code, original place of registration);

[0109] --Risk level and risk value;

[0110] --Key risk characteristics (such as "investment of RMB12 million in building factories in other locations" and "80% of core positions are recruited from other locations").

[0111] Warning notification channels: Warnings are sent to regional economic management departments via system pop-ups, SMS, emails, and other means.

[0112] Dynamic early warning updates: The latest data of enterprises is collected every 7 days to update the risk value and level, realizing "weekly dynamic monitoring" and avoiding delayed early warning.

[0113] The migration analysis module is used to periodically collect data on enterprise relocation, including the number of enterprises relocating in the current period, the number of enterprises relocating out in the current period, and the net number of enterprises relocating in. Combined with the collected e-commerce platform data, it calculates the online retail sales of enterprises relocating in the current period, the online retail sales of enterprises relocating out in the current period, and the online retail sales of net enterprises relocating in the current period. It also calculates the corresponding cumulative indicator data and growth rate data, performs migration statistical analysis, migration flow analysis, and migration detail analysis, and displays the results of migration statistical analysis, migration flow analysis, and migration detail analysis.

[0114] As a specific implementation of the migration analysis module, this module collects monthly statistics by region, including the number of companies moving in, moving out, and net inflow of companies in the current month. It also combines this data with collected e-commerce platform data to calculate the monthly online retail sales of companies moving in, moving out, and net inflow of companies. Furthermore, it calculates the corresponding cumulative indicators and growth rates.

[0115] Construct the following analytical model: Migration statistics: Displays the monthly trends of the above indicators for a specific region; Migration flow: Displays how many companies have moved into the local area from which locations and how many companies have moved out of the local area using a GIS map. Supports drill-down viewing of specific company lists and company registration information, etc. Migration details: Data month, company name, migration time, original province, original city, original district / county, current province, current city, current district / county, current month's online retail sales, cumulative online retail sales.

[0116] The system in this embodiment is implemented as follows.

[0117] Hardware deployment: Servers (CPU: Intel Xeon E5-2680 v4, memory: 64GB, hard disk: 2TB SSD) are used as data processing and model running nodes.

[0118] Software development: Based on Python 3.8, the data acquisition module uses the Scrapy web crawling framework, feature engineering uses the Pandas and Scikit-learn libraries, model training uses the XGBoost and LightGBM libraries, and the early warning system uses the Django framework to develop the web interface.

[0119] Data updates: Seven types of data are automatically collected every morning at midnight for preprocessing and feature updates; the risk model is rerun every Monday to update the enterprise's risk level and early warning information.

[0120] Access control: Set up three levels of user permissions (administrator, auditor, viewer) to ensure the secure flow and use of early warning information.

[0121] Taking a regional e-commerce company A (originally registered in Huqiu District, Suzhou City, Jiangsu Province) as an example, the specific application process of this invention is illustrated below: (1) Data collection: Seven types of data were collected from Company A, including: e-commerce platform data showed that "the main shipping location for the past three months has been Xiaoshan District, Hangzhou City, Zhejiang Province, with 92% of shipments originating from other locations"; recruitment data showed that "75% of job postings were from other locations in the past three months, and 100% of core positions (e-commerce operations and warehouse supervisors) were recruited from other locations"; land acquisition data showed that "60 mu of industrial land was purchased in Xiaoshan District, Hangzhou City, and the land is designated for industrial use with the construction progress listed as 'under construction'". (2) Feature engineering: Extract core features, such as “consistency between registered location and shipping location = 0”, “proportion of shipments from other locations = 0.92”, “proportion of job postings from other locations = 0.75”, and “land acquisition area from other locations = 60 mu”. (3) Risk assessment: Input the features into the fusion model and output the risk value R=0.82; (4) Warning trigger: R=0.82≥0.6, triggering a high-risk warning, the system pushes the warning information to the Suzhou Municipal Bureau of Commerce, indicating that "Enterprise A has the risk of relocation, key risk points: high proportion of shipments from other places, recruitment of core positions in other places, and land acquisition and factory construction in other places"; (5) Dynamic monitoring: The data will be updated weekly. If the progress of Enterprise A’s off-site factory changes to “already in production”, the risk value will rise to 0.95. The system will update the warning level to “extremely high risk” and recommend that the competent department conduct enterprise interviews and policy retention efforts.

[0122] This invention provides a method for identifying and warning of relocation risks for e-commerce enterprises based on Internet data, comprising five steps: data processing, feature extraction, relocation risk assessment, relocation risk warning, and migration analysis.

[0123] Step S100 Data Processing: Regularly collect relevant data from e-commerce companies from multiple data sources to obtain various types of data as monitoring data, and preprocess the monitoring data to obtain structured monitoring data.

[0124] As a specific implementation of data processing, collecting relevant data from e-commerce companies from multiple data sources includes the following steps: (1) The publicly available information of the network points of the regional advantageous e-commerce platform is collected as e-commerce platform data by using Python web crawling technology. This includes the display of licenses and permits of the online store, the information on the business location of the online store, the information on the products listed on the online store, and the information on the shipping location of the products. For the collected e-commerce platform data, duplicate data and invalid fields (such as empty shipping location) are removed, and the text address information is split into three-level structured data of province-city-district, as follows: Through the display of licenses and permits of the online store, the business entity's business registration information of the online store is obtained, such as company name, unified social credit code, enterprise registration province, enterprise registration city, and enterprise registration district; Through the information on the business location of the online store, the information on the business location of the online store published on the platform is obtained, namely the province where the store is located, the city where the store is located, and the district where the store is located; Through the product information in the online store, the shipping location or main shipping location information of the online store is obtained, namely the shipping province, the shipping city, and the shipping district. (2) Collect enterprise business information through authorized public websites, including registered capital information, business scope, legal representative, enterprise status and business change records. Correct and unify the registered capital information, standardize the enterprise status (such as relocation, cancellation, etc.), and extract the timestamp and fields before and after the change of business change records to form a change trajectory dataset, including the registered capital before and after the change, the business scope before and after the change, the legal representative before and after the change, and the changes in enterprise status. (3) Collect and process enterprise foreign investment data by collecting publicly available industry information, including time, enterprise name, project name, province where the project is located, city where the project is located, district / county where the project is located, project details and investment amount, and standardize the unit of investment amount; (4) Collect and process enterprise land acquisition data through public information, including time, enterprise name, land name, province where the land is located, city where the land is located, district / county where the land is located, land nature (such as industrial land, commercial land), construction progress (such as planning, construction, already in production, etc.) and area (uniformly in square meters), etc. (5) Obtain enterprise investment and factory construction data from the authorized platform, including construction location, scale, investment amount, construction progress and expected production time; (6) Collect and process enterprise recruitment data from various recruitment websites, including time, company name, job title, number of recruits, province of work location, city of work location, district / county of work location, job nature, and job responsibilities, and remove non-full-time positions such as "part-time" and "internship". Based on the job title and responsibilities, the positions are divided into "core positions" (production, R&D, finance, operations) and "non-core positions" (sales, customer service) to lay the foundation for subsequent feature extraction; (7) Collect relevant strategic cooperation agreement data signed by enterprises through various public news, using keywords such as "strategic cooperation" and "framework agreement", and use NLP technology (such as BERT model) to extract structured data including the cooperating party, cooperating province, cooperating city, cooperating district / county, and cooperation content; (8) Collect and process enterprise qualification data, including the qualifications of high-tech enterprises, specialized and innovative small and medium-sized enterprises and specialized and innovative little giants.

[0125] Step S200 Feature Extraction: Using whether e-commerce companies have relocated as the target variable, construct binary labels, and construct structured features for each type of structured monitoring data. Then, perform feature optimization and standardization on the extracted structured features to obtain the relocation risk features of enterprises.

[0126] As a specific implementation of the feature extraction module, taking whether e-commerce companies have relocated as the target variable, when constructing binary labels, the label values ​​are determined based on pre-configured judgment rules, and the label values ​​are divided into positive and negative labels. The label values ​​are determined according to the following rules: Positive label (relocation, label=1): A company can be identified as relocating if it meets any of the following conditions: Condition 1: The "business address" in the business registration information has been changed to an administrative region outside the original registered location, and the business has not been moved back within 6 months after the change; Condition 2: Data on land acquisition / investment in factory construction shows that the enterprise has purchased ≥50 mu of industrial land outside its original registered location, or invested ≥10 million yuan in building factory buildings, and the project has been put into production; Condition 3: Data from the e-commerce platform shows that the main shipping location of the online store is inconsistent with the original registration location for six consecutive months, and the proportion of shipments from different locations is ≥80%. Negative label (not relocated, label=0): Enterprises that have not met any of the above relocation conditions within 12 consecutive months and whose business status is "in operation".

[0127] For the seven types of data, structured features were constructed to form a "corporate relocation risk feature database".

[0128] When optimizing the extracted structured features, a three-level screening strategy of variance filtering, mutual information entropy, and recursive feature elimination (RFE) is adopted to remove low-variance features (such as "company name") and features with low correlation to the target variable (such as "number of non-core positions"), while retaining a predetermined number of core features.

[0129] When standardizing the extracted structured features (such as "percentage of investment amount in other locations" and "scale of recruitment in other locations"), Z-Score standardization (mean = 0, standard deviation = 1) is used for numerical features, and one-hot encoding is used for categorical variables to ensure consistent feature magnitude.

[0130] Step S300 Relocation Risk Assessment: Based on the enterprise's relocation risk characteristics, the relocation risk value of the enterprise is predicted and output through the relocation risk assessment model. The relocation risk assessment model is a machine learning model constructed through a two-step method of base model training and model fusion.

[0131] As a specific implementation of relocation risk management, the machine learning model constructed through the two-step method of base model training and model fusion includes the following: (1) Select three base models with different learning logics and train them on the training set respectively. The three base models are logistic regression, random forest and gradient boosting tree; (2) The output probabilities of the three base models are used as meta-features. Logistic regression is constructed as the meta-model. The output results of the base model are fused. During fusion, the training set is divided into 5 folds. Each fold of data is predicted by the three base models to obtain 5 sets of base model output probabilities, which are used as the training set of meta-features. The meta-model takes the meta-features as input and the real migration label as the target, and learns the weights of the base model results (e.g., RF weight = 0.4, XGBoost weight = 0.35, LR weight = 0.25). The test set (accounting for 30% of the total data) first obtains the probability output through the three base models, and then inputs it into the meta-model. Finally, the fused enterprise migration risk value (denoted as R, 0≤R≤1) is output. (3) The model performance was verified by using accuracy, precision, recall and F1 score as indicators. The F1 score of this model was ≥0.85 and the recall was ≥0.8 (ensuring that as few relocation risk enterprises are missed).

[0132] Logistic Regression (LR) employs L1 regularization (Lasso) to suppress collinearity, with a regularization strength parameter C=0.1, and outputs the probability of firm relocation (0-1).

[0133] Random Forest (RF): Handles non-linear relationships between features. Sets the number of decision trees to 100, the maximum depth to 10, and the minimum number of sample splits to 5. It evaluates feature importance through out-of-bag (OOB) samples, handles the sample imbalance problem, and outputs the migration probability.

[0134] Gradient Boosting Tree (XGBoost): Optimizes the sample imbalance problem (the proportion of samples from relocated enterprises is usually low). It adopts the "class weight balancing" strategy, sets the learning rate to 0.1, the number of trees to 150, the maximum depth to 8, handles the sample imbalance problem, and outputs the relocation probability.

[0135] Step S400 Relocation Risk Warning: Construct a risk level system based on the historical distribution of enterprise relocation risk values, issue relocation risk warnings based on the current risk level status of enterprises, and generate and push warning signals.

[0136] As a concrete implementation of relocation risk early warning, a multi-level early warning system is constructed based on the risk value R output by the risk assessment model to achieve accurate early warning through "risk classification and strategy matching".

[0137] Risk Level Classification: Based on the risk value distribution of historical migration data, the K-means clustering algorithm is used to determine the early warning threshold, and the risk is divided into 3 levels: Low risk: R < 0.3, the company's operations are stable and there are no obvious signs of relocation; Medium risk: 0.3≤R<0.6, indicating that the company is showing signs of relocation (such as increased recruitment in other locations or changes in the place of shipment), and requires continuous monitoring; High risk: R≥0.6, the enterprise has clear relocation behavior (such as acquiring land in another location or changing its business address), and immediate intervention is required.

[0138] Early warning triggering and response involve early warning information generation, early warning push, and dynamic early warning updates.

[0139] Warning signal generation: When a company's risk level rises from "low" to "medium" or "high", the system automatically generates a warning message, which includes: --Basic information of the enterprise (name, unified social credit code, original place of registration); --Risk level and risk value; --Key risk characteristics (such as "investment of RMB12 million in building factories in other locations" and "80% of core positions are recruited from other locations").

[0140] Warning notification channels: Warnings are sent to regional economic management departments via system pop-ups, SMS, emails, and other means.

[0141] Dynamic early warning updates: The latest data of enterprises is collected every 7 days to update the risk value and level, realizing "weekly dynamic monitoring" and avoiding delayed early warning.

[0142] Step S500 Migration Analysis: Regularly collect data on enterprise relocation, including the number of enterprises relocating in the current period, the number of enterprises relocating out in the current period, and the net number of enterprises relocating in. Combined with the collected e-commerce platform data, calculate the online retail sales of enterprises relocating in the current period, the online retail sales of enterprises relocating out in the current period, and the net online retail sales of enterprises relocating in. Calculate the corresponding cumulative indicator data and growth rate data, conduct migration statistical analysis, migration flow analysis, and migration detail analysis, and display the results of the migration statistical analysis, migration flow analysis, and migration detail analysis.

[0143] As a specific implementation of migration analysis, this step involves compiling monthly statistics by region, including the number of companies moving in, moving out, and net moving in for the month. Simultaneously, it combines this data with collected e-commerce platform data to calculate the monthly online retail sales of companies moving in, moving out, and net moving in. Furthermore, it calculates the corresponding cumulative indicators and growth rates.

[0144] Construct the following analytical model: Migration statistics: Displays the monthly trends of the above indicators for a specific region; Migration flow: Displays how many companies have moved into the local area from which locations and how many companies have moved out of the local area using a GIS map. Supports drill-down viewing of specific company lists and company registration information, etc. Migration details: Data month, company name, migration time, original province, original city, original district / county, current province, current city, current district / county, current month's online retail sales, cumulative online retail sales.

[0145] The method in this embodiment can be implemented based on the system disclosed in Embodiment 1.

[0146] The above provides a detailed description of the e-commerce enterprise relocation risk identification and early warning system and method based on Internet data provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A risk identification and early warning system for the relocation of e-commerce enterprises based on internet data, characterized in that, It includes a data processing module, a feature extraction module, a relocation risk assessment module, a relocation risk early warning module, and a migration analysis module; The data processing module is used to periodically collect relevant data from e-commerce companies from multiple data sources, obtain various types of data as monitoring data, and preprocess the monitoring data to obtain structured monitoring data. The feature extraction module is used to construct binary labels with the target variable of whether e-commerce companies have relocated, and to construct structured features for each type of structured monitoring data. The extracted structured features are then optimized and standardized to obtain the relocation risk features of the companies. The relocation risk assessment module is equipped with a relocation risk assessment model, which is used to predict and output the relocation risk value of an enterprise based on its relocation risk characteristics. The relocation risk assessment model is a machine learning model constructed through a two-step method of base model training and model fusion. The relocation risk early warning module is used to construct a risk level system based on the historical distribution of enterprise relocation risk values, conduct relocation risk early warning based on the current enterprise risk level status, and generate and push early warning signals. The migration analysis module is used to periodically collect data on enterprise relocation, including the number of enterprises relocating in the current period, the number of enterprises relocating out in the current period, and the net number of enterprises relocating in. Combined with the collected e-commerce platform data, it calculates the online retail sales of enterprises relocating in the current period, the online retail sales of enterprises relocating out in the current period, and the online retail sales of net enterprises relocating in the current period. It also calculates the corresponding cumulative indicator data and growth rate data, performs migration statistical analysis, migration flow analysis, and migration detail analysis, and displays the results of migration statistical analysis, migration flow analysis, and migration detail analysis.

2. The e-commerce enterprise relocation risk identification and early warning system based on internet data according to claim 1, characterized in that, Collecting relevant data from e-commerce companies from multiple data sources includes the following steps: We used Python web scraping technology to collect publicly available information from local e-commerce platforms as e-commerce platform data. This included information on the online stores' licenses and permits, the business locations of the online stores, the product information listed on the online stores, and the product shipping locations. We removed duplicate data and invalid fields from the collected e-commerce platform data and split the text address information into three levels of structured data: province, city, and district / county. Enterprise business registration information is collected through authorized public websites, including registered capital information, business scope, legal representative, enterprise status, and business registration change records. The registered capital information is corrected and standardized to a unified unit, and the enterprise status is standardized. The timestamps and fields before and after the changes in the business registration change records are extracted to form a change trajectory dataset, including the registered capital before and after the change, the business scope before and after the change, the legal representative before and after the change, and the changes in enterprise status. By collecting publicly available industry information, we collect and process data on enterprises' external investments, including time, enterprise name, project name, province of the project, city of the project, district / county of the project, project details, and investment amount, and standardize the unit for investment amount; By collecting and processing publicly available information, data on land acquisition by enterprises is collected, including the time, enterprise name, land parcel name, province, city, district / county of the land parcel, land type, construction progress, and area. Obtain enterprise investment and factory construction data from authorized platforms, including construction location, scale, investment amount, construction progress, and expected production time; We collect and process corporate recruitment data from various recruitment websites, including information such as time, company name, job title, number of openings, province of work location, city of work location, district / county of work location, job nature, and job responsibilities. We collected data on relevant strategic cooperation agreements signed by enterprises through various publicly available news sources, and used NLP technology to extract structured data including the partners, provinces, cities, districts and counties, and content of the cooperation. Collect and process enterprise qualification data, including the qualifications of high-tech enterprises, specialized and innovative small and medium-sized enterprises, and specialized and innovative little giants.

3. The e-commerce enterprise relocation risk identification and early warning system based on Internet data according to claim 1, characterized in that, Using whether e-commerce companies have relocated as the target variable, when constructing binary labels, the label values ​​are determined based on pre-configured judgment rules, and the label values ​​are divided into positive labels and negative labels. When optimizing the extracted structured features, a three-level screening strategy of variance filtering, mutual information entropy, and recursive feature elimination is adopted to remove low-variance features and features with low correlation to the target variable, while retaining a predetermined number of core features. When standardizing the extracted structured features, Z-Score standardization is used for numerical features, and one-hot encoding is used for categorical variables.

4. The e-commerce enterprise relocation risk identification and early warning system based on Internet data according to claim 1, characterized in that, Machine learning models constructed using the two-step method of base model training and model fusion include the following: Three base models with different learning logics were selected and trained on the training set respectively. The three base models were logistic regression, random forest and gradient boosting tree. The output probabilities of the three base models are used as meta-features. Logistic regression is constructed as the meta-model. The output results of the base models are fused. During fusion, the training set is divided into 5 folds. Each fold of data is predicted by the three base models to obtain 5 sets of base model output probabilities, which are used as the training set of the meta-features. The meta-model takes the meta-features as input and the real migration label as the target to learn the weights of the base model results. The test set first obtains the probability output through the three base models, and then inputs it into the meta-model. Finally, the fused enterprise migration risk value is output. Accuracy, precision, recall, and F1 score were used as metrics to validate model performance.

5. The e-commerce enterprise relocation risk identification and early warning system based on Internet data according to claim 1, characterized in that, The relocation risk early warning module is used to perform the following: Risk level classification: Based on the risk value distribution of historical migration data, the K-means clustering algorithm is used to determine the early warning threshold and classify the risk into three levels: low risk, medium risk and high risk. Early warning triggering and response: When a company's risk level rises from low risk to medium risk or high risk, an early warning message is generated. The early warning message includes the company's basic information, risk level, and risk value for relocation. The early warning message is sent to relevant departments through early warning push channels, and the latest monitoring data of the company is collected regularly to update the risk value and risk level for relocation.

6. A method for identifying and warning of the risk of relocation of e-commerce enterprises based on Internet data, characterized in that, Includes the following steps: Data processing: Regularly collect relevant data from e-commerce companies from multiple data sources to obtain various types of data as monitoring data, and preprocess the monitoring data to obtain structured monitoring data; Feature extraction: Taking whether e-commerce companies have relocated as the target variable, binary labels are constructed, and structured features are constructed for each type of structured monitoring data. The extracted structured features are then optimized and standardized to obtain the relocation risk features of enterprises. Relocation Risk Assessment: Based on the characteristics of enterprise relocation risk, the relocation risk assessment model predicts and outputs the enterprise relocation risk value. The relocation risk assessment model is a machine learning model constructed through a two-step method of base model training and model fusion. Relocation Risk Warning: Construct a risk level system based on the historical distribution of enterprise relocation risk values, generate and push warning signals based on the current risk level status of enterprises, and conduct relocation risk warnings. Migration Analysis: Regularly collect data on enterprise relocation, including the number of enterprises relocating in the current period, the number of enterprises relocating out in the current period, and the net number of enterprises relocating in. Combined with data collected from e-commerce platforms, calculate the online retail sales of enterprises relocating in the current period, the online retail sales of enterprises relocating out in the current period, and the net online retail sales of enterprises relocating in the current period. Calculate the corresponding cumulative indicator data and growth rate data, conduct migration statistical analysis, migration flow analysis, and migration detail analysis, and display the results of migration statistical analysis, migration flow analysis, and migration detail analysis.

7. The method for identifying and warning of the risk of relocation of e-commerce enterprises based on Internet data according to claim 6, characterized in that, Collecting relevant data from e-commerce companies from multiple data sources includes the following steps: We used Python web scraping technology to collect publicly available information from local e-commerce platforms as e-commerce platform data. This included information on the online stores' licenses and permits, the business locations of the online stores, the product information listed on the online stores, and the product shipping locations. We removed duplicate data and invalid fields from the collected e-commerce platform data and split the text address information into three levels of structured data: province, city, and district / county. Enterprise business registration information is collected through authorized public websites, including registered capital information, business scope, legal representative, enterprise status, and business registration change records. The registered capital information is corrected and standardized to a unified unit, and the enterprise status is standardized. The timestamps and fields before and after the changes in the business registration change records are extracted to form a change trajectory dataset, including the registered capital before and after the change, the business scope before and after the change, the legal representative before and after the change, and the changes in enterprise status. By collecting publicly available industry information, we collect and process data on enterprises' external investments, including time, enterprise name, project name, province of the project, city of the project, district / county of the project, project details, and investment amount, and standardize the unit for investment amount; By collecting and processing publicly available information, data on land acquisition by enterprises is collected, including the time, enterprise name, land parcel name, province, city, district / county of the land parcel, land type, construction progress, and area. Obtain enterprise investment and factory construction data from authorized platforms, including construction location, scale, investment amount, construction progress, and expected production time; We collect and process corporate recruitment data from various recruitment websites, including information such as time, company name, job title, number of openings, province of work location, city of work location, district / county of work location, job nature, and job responsibilities. We collected data on relevant strategic cooperation agreements signed by enterprises through various publicly available news sources, and used NLP technology to extract structured data including the partners, provinces, cities, districts and counties, and content of the cooperation. Collect and process enterprise qualification data, including the qualifications of high-tech enterprises, specialized and innovative small and medium-sized enterprises, and specialized and innovative little giants.

8. The method for identifying and warning of the risk of relocation of e-commerce enterprises based on Internet data according to claim 6, characterized in that, Using whether e-commerce companies have relocated as the target variable, when constructing binary labels, the label values ​​are determined based on pre-configured judgment rules, and the label values ​​are divided into positive labels and negative labels. When optimizing the extracted structured features, a three-level screening strategy of variance filtering, mutual information entropy, and recursive feature elimination is adopted to remove low-variance features and features with low correlation to the target variable, while retaining a predetermined number of core features. When standardizing the extracted structured features, Z-Score standardization is used for numerical features, and one-hot encoding is used for categorical variables.

9. A method for identifying and warning of the risk of relocation of e-commerce enterprises based on Internet data, as described in claim 6, is characterized in that... Machine learning models constructed using the two-step method of base model training and model fusion include the following: Three base models with different learning logics were selected and trained on the training set respectively. The three base models were logistic regression, random forest and gradient boosting tree. The output probabilities of the three base models are used as meta-features. Logistic regression is constructed as the meta-model. The output results of the base models are fused. During fusion, the training set is divided into 5 folds. Each fold of data is predicted by the three base models to obtain 5 sets of base model output probabilities, which are used as the training set of the meta-features. The meta-model takes the meta-features as input and the real migration label as the target to learn the weights of the base model results. The test set first obtains the probability output through the three base models, and then inputs it into the meta-model. Finally, the fused enterprise migration risk value is output. Accuracy, precision, recall, and F1 score were used as metrics to validate model performance.

10. A method for identifying and warning of the risk of relocation of e-commerce enterprises based on Internet data, as described in claim 6, is characterized in that... Relocation risk warning includes the following operations: Risk level classification: Based on the risk value distribution of historical migration data, the K-means clustering algorithm is used to determine the early warning threshold and classify the risk into three levels: low risk, medium risk and high risk. Early warning triggering and response: When a company's risk level rises from low risk to medium risk or high risk, an early warning message is generated. The early warning message includes the company's basic information, risk level, and risk value for relocation. The early warning message is sent to relevant departments through early warning push channels, and the latest monitoring data of the company is collected regularly to update the risk value and risk level for relocation.

Images