WO2021232595A1

WO2021232595A1 - Enterprise state supervision method, apparatus, and device, and computer readable storage medium

Info

Publication number: WO2021232595A1
Application number: PCT/CN2020/106230
Authority: WO
Inventors: 刘春�
Original assignee: 平安国际智慧城市科技股份有限公司
Priority date: 2020-05-22
Filing date: 2020-07-31
Publication date: 2021-11-25
Also published as: CN111798352A

Abstract

An enterprise state supervision method, apparatus, and device, and a computer readable storage medium. Said method comprises: transmitting preset sample data to an initial model, and training the initial model on the basis of a federated learning algorithm, so as to generate a state data model (S10); acquiring structured data corresponding to an enterprise state in an enterprise to be checked, and transmitting the structured data to the state data model, so as to generate data score values of the structured data (S20); and according to the data score values, determining whether the enterprise state of said enterprise is valid (S30).

Description

Enterprise state supervision method, device, equipment and computer readable storage medium

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on May 22, 2020, the application number is 202010445603.8, and the invention title is "Enterprise State Supervision Methods, Devices, Equipment, and Computer-readable Storage Media", and its entire contents Incorporated in this application by reference.

Technical field

This application relates to the field of data processing technology, and in particular to a method, device, equipment and computer-readable storage medium for monitoring the state of an enterprise.

Background technique

At present, there are a large number of domestic catering companies, a large number of catering companies rise up every year, and there are also a large number of catering companies that have ceased operations. For catering companies that have ceased operations, if they fail to complete the cancellation of their business licenses in time, the business status of their industrial and commercial registration will lag behind, and they need to be supervised to update their business status. At present, the supervision of the business status of catering enterprises by supervisors depends on the realization of inquiries on industrial and commercial registration. The inventor realizes that supervisors usually make inquiries according to the set time limit, and the catering companies may have been closed for a long time at the time of inquiries, which may easily lead to untimely supervision. At the same time, there is still a problem of inefficiency in the inquiries of industrial and commercial registration, which also affects the efficiency of supervision.

Technical solutions

The main purpose of this application is to provide a method, device, equipment, and computer-readable storage medium for monitoring the status of an enterprise, aiming to solve the technical problems of untimely and inefficient monitoring of the operating status of catering enterprises in the prior art.

In order to achieve the above-mentioned objective, an embodiment of the present application provides a method for monitoring the state of an enterprise, and the method for monitoring the state of an enterprise includes the following steps:

Transmitting the preset sample data to the initial model, and training the initial model based on a federated learning algorithm to generate a state data model;

Acquiring structured data corresponding to the state of the enterprise in the enterprise to be checked, and transmitting the structured data to the state data model to generate a data score of the structured data;

According to the data score, whether the enterprise status of the enterprise to be checked is effective is supervised.

To achieve the above objective, this application also provides an enterprise state monitoring device, the enterprise state monitoring device includes:

A generating module, used for transmitting preset sample data to the initial model, and training the initial model based on a federated learning algorithm to generate a state data model;

An obtaining module is used to obtain structured data corresponding to the state of the enterprise in the enterprise to be checked, and transmit the structured data to the state data model to generate a data score of the structured data;

The supervision module is used to supervise whether the enterprise status of the enterprise to be checked is valid according to the data score.

Further, in order to achieve the above-mentioned purpose, the present application also provides an enterprise state monitoring device, the enterprise state monitoring device including a memory, a processor, and an enterprise state monitoring program stored on the memory and running on the processor, When the enterprise state monitoring program is executed by the processor, the following steps are implemented:

In addition, in order to achieve the above-mentioned object, the present application also provides a computer-readable storage medium with an enterprise state monitoring program stored on the computer-readable storage medium, and when the enterprise state monitoring program is executed by a processor, the following steps are implemented:

This application provides a method, device, equipment, and computer-readable storage medium for monitoring the state of an enterprise. The preset sample data is first transmitted to the initial model, and the initial model is trained based on the federated learning algorithm to generate the state data model; Check the structured data corresponding to the state of the enterprise in the enterprise, and transfer the structured data to the state data model to generate the data score of the structured data; and then, according to the data score, supervise whether the enterprise state of the enterprise to be verified is valid. Among them, the preset sample data is various types of data representing their respective states in each enterprise, and is the real and effective data of each enterprise. The federated learning algorithm is combined with the preset sample data of a large number of enterprises for training, which enriches the training sample size and makes all The generated state data model is more accurate. Therefore, the status data model is used to monitor the effectiveness of the state of the enterprise to be verified, which combines various real data of the enterprise to be verified to reflect the state of the enterprise, avoids relying on the inspection of industrial and commercial registration for supervision, and ensures the authenticity of the state of the supervised enterprise. While ensuring the effectiveness and accuracy of supervision, it also ensures the timeliness of supervision, which is conducive to timely supervision and efficient supervision.

Description of the drawings

FIG. 1 is a schematic diagram of the structure of an enterprise state monitoring device in a hardware operating environment involved in a solution according to an embodiment of the application;

FIG. 2 is a schematic flowchart of the first embodiment of the method for monitoring the state of an applicant enterprise;

FIG. 3 is a schematic diagram of functional modules of a preferred embodiment of the enterprise state monitoring device of this application.

Embodiments of the present invention

It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

As shown in FIG. 1, FIG. 1 is a schematic diagram of the structure of an enterprise state monitoring device in a hardware operating environment involved in the solution of an embodiment of the present application.

In the following description, the use of suffixes such as “module”, “part” or “unit” used to indicate elements is only for the description of the present application, and has no specific meaning in itself. Therefore, "module", "part" or "unit" can be used in a mixed manner.

The enterprise state monitoring device in the embodiment of the present application may be a PC, or a portable terminal device such as a tablet computer and a portable computer.

As shown in FIG. 1, the enterprise state monitoring device may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to implement connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a magnetic disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

Those skilled in the art can understand that the structure of the enterprise state monitoring device shown in FIG. 1 does not constitute a limitation on the enterprise state monitoring device, and may include more or less components than those shown in the figure, or a combination of certain components, or different components. The layout of the components.

As shown in FIG. 1, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a detection program.

In the device shown in Figure 1, the network interface 1004 is mainly used to connect to the back-end server and communicate with the back-end server; the user interface 1003 is mainly used to connect to the client (user side) to communicate with the client; and the processor 1001 can be used to call the detection program stored in the memory 1005 and perform the following operations:

Further, the step of obtaining the structured data corresponding to the state of the enterprise in the enterprise to be checked includes:

Collect enterprise text data of the enterprise to be checked, and extract the text data corresponding to the enterprise status from each of the enterprise text data for classification, and obtain multiple types of status text data;

Extracting state keywords in multiple types of the state text data respectively, and performing format conversion on the extracted multiple types of state keywords according to a preset data format, to obtain the structured data.

Further, the step of separately extracting state keywords in the multiple types of state text data includes:

Performing segmentation processing and sentence processing on multiple types of the state text data respectively, generating multiple types of to-be-recognized clauses, and eliminating invalid clauses in the multiple types of the to-be-recognized clauses;

Perform word segmentation processing on the multiple types of the to-be-recognized clauses after the invalid clauses are eliminated, and generate multiple types of to-be-recognized word segmentation;

The noise words that are irrelevant to the state of the enterprise in the multiple types of word segmentation to be recognized are eliminated to obtain the state keywords in the multiple types of state text data.

Further, the step of transmitting the structured data to the state data model to generate the data score of the structured data includes:

Transmitting the structured data to the state data model, and determining target sample data respectively matching various types of sub-data in the structured data;

Determine the sub-scores of the various types of sub-data according to the scores and weights respectively corresponding to each of the target sample data;

The data score of the structured data is generated according to the sub-score of the various types of the sub-data.

Further, the step of supervising whether the enterprise status of the enterprise to be checked is valid according to the data score includes:

Determine the target state corresponding to the combination formed by the maximum value, the minimum value and the average value in the data score according to the preset correspondence between the combined score and the state;

Find the registration status corresponding to the company to be checked, and supervise whether the company status of the company to be checked is valid according to the consistency between the target status and the registration status.

Further, after the step of supervising whether the enterprise state of the enterprise to be checked is valid according to the data score, the processor 1001 may be used to call the detection program stored in the memory 1005 and perform the following operations:

Transmitting the research and judgment score corresponding to the enterprise to be checked to the state data model, and determine whether the research and judgment score matches the data score;

If it matches the data score, store the data score and the structured data correspondingly;

If it does not match the data score, searching for target sample data that matches the structured data in the preset sample data;

Removing the target sample data and the score label corresponding to the target sample data, and generating the research and judgment score as the to-be-trained score label of the structured data;

According to the structured data and the score label to be trained, the preset sample data is updated, and the state data model is optimized for training based on the updated preset sample data.

Further, the step of transmitting preset sample data to the initial model, and training the initial model based on a federated learning algorithm, and generating a state data model includes:

Obtain the positive sample data corresponding to the preset positive field name and the negative sample data corresponding to the preset negative field name, and use each of the positive sample data and each of the negative sample data as the The preset sample data is transmitted to the initial model, the initial model is trained, and the model gradient is generated;

The model gradient is transmitted to the coordinating party corresponding to the federated learning algorithm, so that the coordinating party aggregates the model gradient and at least one other model gradient generated based on the federated learning algorithm to generate a return gradient ；

Receive the return gradient returned by the coordinator, and continuously train the initial model according to the return gradient until the initial model converges to obtain the state data model.

The specific implementation of the enterprise state supervision device of this application is basically the same as the following embodiments of the enterprise state supervision method, and will not be repeated here.

In order to better understand the above technical solutions, exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the drawings show exemplary embodiments of the present disclosure, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

In order to better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the accompanying drawings of the specification and specific implementations.

2, the first embodiment of the present application provides a schematic flowchart of a method for monitoring the state of an enterprise. In this embodiment, the enterprise state monitoring method includes the following steps:

Step S10, transmitting preset sample data to an initial model, and training the initial model based on a federated learning algorithm to generate a state data model;

The method for monitoring the state of an enterprise in this embodiment is applied to a monitoring server, and is suitable for monitoring the state of an enterprise through the monitoring server. Among them, the state of an enterprise is the state of operation of the enterprise. A large amount of data in various aspects such as procurement, arrears, information disclosure, industrial and commercial registration, supervision, sales, training, etc. is used to determine whether the enterprise is in an operating state and then supervise it. The enterprise may be various types of enterprises such as catering, clothing, travel, finance, construction, etc. This embodiment preferably takes a catering enterprise as an example for description. Specifically, the initial model is trained through a large amount of data in various aspects of various catering companies that have determined the state of the enterprise, and the state data model for enterprise supervision is obtained and deployed to the supervision server. Pre-set data indicators that characterize the operating conditions of catering companies, such as procurement, industrial and commercial registration, sales, training, bright kitchens and other indicators, and then obtain a large number of such indicators from various catering companies with determined corporate status Data, that is, to obtain data such as procurement data, industrial and commercial registration data, sales data, training data and bright kitchens of multiple catering companies. Different weights and scores are set for the acquired data according to their respective degree of influence on the effectiveness of the state of the enterprise. For example, compared with training data, procurement data has a higher degree of influence on the effectiveness of the state of the enterprise. High weight and score. After each acquired data is set with its own weight and score, all kinds of data are transmitted as preset sample data to the initial model for training.

It should be noted that the initial model is a preset network model, and the training is implemented based on a federated learning algorithm. The federated learning algorithm is a way to continue machine learning under the premise of protecting data privacy and meeting legal compliance requirements. It uses technical algorithms to encrypt the model built, and the federated parties can also conduct model training without providing their own data. After obtaining the model parameters, federated learning protects user data privacy through parameter exchange under the encryption mechanism. The data and the model itself will not be transmitted, nor can they guess the other party’s data. Therefore, there is no possibility of leakage at the data level, and it does not violate stricter rules. Therefore, the data protection act of the People’s Republic of China can protect data privacy while maintaining data integrity to a high degree.

In this embodiment, regions of the same level that need to be supervised on the state of the enterprise are regarded as the two sides of the federation, such as two different counties, two different cities, and so on. Different regions set their own initial models, and conduct federated training to their respective initial models through the data of the catering companies that have determined their corporate status, that is, the combination of their respective preset sample data, and then obtain the federated training for their respective initial models A state data model for monitoring the corporate state of catering companies in the local area. That is, each area of the alliance has been trained to obtain a local state data model to supervise the business state of local catering companies. Training with data from other regions has enriched the training sample size and made the state data model more accurate while ensuring data security.

Step S20: Obtain structured data corresponding to the state of the enterprise in the enterprise to be checked, and transmit the structured data to the state data model to generate a data score of the structured data;

Furthermore, the catering company that needs to be verified as the company to be verified, and the structured data corresponding to the status of the company is obtained. The structured data is the various types of data indicators that represent the business status of the catering company. After structured processing, data in a specific data structure is obtained. Among them, various types of data corresponding to data indicators form various types of sub-data in structured data; for example, sales data and purchase data corresponding to data indicators form two types of sub-data in structured data, and these two types of sub-data The data structure is the same.

Furthermore, the structured data is transferred to the state data model, and the type prediction of various sub-data in the structured data is performed through the state data model. Then according to the type of prediction, the scores and weights of various types of sub-data are searched, and the data scores of structured data are generated from each score and weight to indicate whether the business status of the enterprise is normal or not. Specifically, the steps of transmitting structured data to the state data model and generating the data score of the structured data include:

Step S21, transmitting the structured data to the state data model, and determining target sample data respectively matching various types of sub-data in the structured data;

Further, after the structured data is transmitted to the state data model, the model parameters obtained after the state data model are trained are used to classify the various sub-data in the structured data, and search for various types of data in the preset sample data. The sub-data respectively match the target sample data. Among them, the matching is determined by the size of the similarity. When the similarity between a certain type of sub-data and a certain data in the preset sample data is greater than the preset threshold, it is determined that the type of sub-data matches the data, and the This data is used as the target sample data. For example, the structured data contains sub-data a and b, and the preset sample data in the state data model includes three types of data p1, p2, and p3; then both a and b are transmitted to the state data model, and based on its model parameters, calculate The similarity between a and p1, p2, and p3, respectively; if the similarity with p3 is greater than the preset threshold, then p3 is used as the target sample data matching a. Similarly, calculate the similarity between b and p1, p2, and p3 respectively; if the similarity between b and p1 is greater than the preset threshold, then p1 is used as the target sample data matching b.

Step S22: Determine the sub-scores of the various types of sub-data according to the scores and weight values respectively corresponding to each of the target sample data;

Understandably, each data constituting the preset sample data has its own score and weight. After the target sample data is obtained, the score and weight corresponding to each target sample data can be searched. Furthermore, the scores and weights are calculated, and the scores and weights corresponding to the same target sample data are multiplied to obtain sub-scores of various sub-data. For example, for the aforementioned target sample data p1 and p3, if the score and weight value of p1 are w1 and k1, and the score and weight value of p3 are w3 and k3, then the sub-score of sub-data a is k3 *w3, the sub-score of sub-data b is k1*w1.

Step S23: Generate data scores of the structured data according to the sub-scores of the various types of sub-data.

Further, after all the sub-scores of various types of sub-data are calculated, the data scores of the structured data can be obtained according to each sub-score. In order to accurately characterize the business status of the enterprise through data scores, the data scores can be set as a collection of multiple scores, including at least the minimum, maximum, and average values of each sub-score. By comparing the sub-scores of various sub-data, the minimum and maximum values are selected; meanwhile, the average value of each sub-score is processed to obtain the average; and then the minimum, maximum, and average values obtained are used as Data score for structured data. For example, for the above structured data containing sub-data a and b, compare the respective sub-scores k3*w3 and k1*w1 to determine the maximum value k1*w1 and minimum value k3*w3. Perform average processing between the sub-scores to get the average value (k1*w1+k1*w1)/2; then k1*w1, k3*w3 and (k1*w1+k1*w1)/2 are formed into structured data Data points.

Step S30, according to the data score, supervise whether the enterprise status of the enterprise to be checked is valid.

Furthermore, the corresponding relationship of different operating states is set in advance for different data scores, and the obtained data score of structured data is compared with the corresponding relationship, and the corresponding relationship is determined to be consistent with the data score of structured data. According to the target data score of the target data, the actual business status of the enterprise to be checked is determined based on the operating status of the target data score in the corresponding relationship. The actual enterprise status represents the current true operating status of the enterprise to be checked. Furthermore, according to the actual enterprise status, whether the registered enterprise status of the enterprise to be checked is effective is supervised, and it is determined whether there is an update lag in the registered enterprise status. If there is an update lag, the enterprise status is determined to be invalid; otherwise, if it does not exist The lag in the update determines that the state of the enterprise is valid. Specifically, according to the data score, the steps to supervise whether the enterprise status of the enterprise to be checked is effective include:

Step S31: Determine the target state corresponding to the combination formed by the maximum value, the minimum value and the average value in the data score according to the preset correspondence between the combined score and the state;

Understandably, because the data score is a collection of multiple scores including the maximum value, the minimum value, and the average value, when setting the correspondence between the data score and the operating status, the multiple score ranges are formed as The combined score is preset to correspond to the state. After determining the data score, call the corresponding relationship, and compare the combination of the maximum, minimum, and average of the data scores with the combined scores in the corresponding relationship to determine whether the values of the combined combination are in the combination Within the range of the score; if each value exists in the range of each score of a certain combination, that is, the maximum, minimum and average values exist in the range of the maximum of a certain combination of scores, The minimum range and the average range are used to find the state corresponding to the combined score in the corresponding relationship, as the target state corresponding to the data score, which represents the current actual operating state of the enterprise to be checked.

Step S32, searching for the registration status corresponding to the enterprise to be checked, and supervising whether the enterprise status of the enterprise to be checked is valid according to the consistency between the target state and the registration status.

Further, the registration status of the enterprise to be checked is searched, and the searched registration status is compared with the target status, and the consistency of the two is judged. If it is determined that the two are consistent after comparison, it means that the registered status of the company to be verified is consistent with its actual operating status, and it is determined that the company status of the company to be verified is valid. On the contrary, if the registration status and the target status are inconsistent after comparison, it means that the status of the company to be verified is inconsistent with the status of actual operations, and the status of the registered company has a lag in updating, and it is determined that the status of the company to be verified is invalid. In this way, the real data of all aspects of the enterprise to be checked can be combined to supervise its operating status, and the authenticity and effectiveness of supervision can be ensured. At the same time, supervision can be achieved by obtaining structured data, and the timeliness and efficiency of supervision can be ensured.

The enterprise state monitoring method of this embodiment first transmits preset sample data to the initial model, and trains the initial model based on the federated learning algorithm to generate a state data model; then obtains structured data corresponding to the state of the enterprise in the enterprise to be checked , And transfer the structured data to the status data model to generate the data score of the structured data; and then, according to the data score, supervise whether the enterprise status of the enterprise to be checked is valid. Among them, the preset sample data is various types of data representing their respective states in each enterprise, and is the real and effective data of each enterprise. The federated learning algorithm is combined with the preset sample data of a large number of enterprises for training, which enriches the training sample size and makes all The generated state data model is more accurate. Therefore, the status data model is used to monitor the effectiveness of the state of the enterprise to be verified, which combines various real data of the enterprise to be verified to reflect the state of the enterprise, avoids relying on the inspection of industrial and commercial registration for supervision, and ensures the authenticity of the state of the supervised enterprise. While ensuring the effectiveness and accuracy of supervision, it also ensures the timeliness of supervision, which is conducive to timely supervision and efficient supervision.

Further, based on the first embodiment of the enterprise state supervision method of the present application, a second embodiment of the enterprise state supervision method of the present application is proposed. The steps include:

Step S24: Collect enterprise text data of the enterprise to be checked, and extract the text data corresponding to the enterprise status from each of the enterprise text data for classification, to obtain multiple types of status text data;

Understandably, there are a lot of data involved in the business process of an enterprise. In the supervision process, in order to obtain structured data that characterizes the business status of the enterprise to be checked, it is necessary to obtain all kinds of data generated in the business process first, and then Data related to business status is extracted from various types of data for processing to obtain structured data. Specifically, the supervisory server connects with the company to be verified to collect its corporate text data from the company to be verified. The text data of the company is the various types of data involved in the business process of the company to be verified, including at least purchases, arrears, and information. Disclosure, business registration, supervision, sales, training, corporate structure, corporate employee composition, and other data in text form.

Further, extract the collected corporate text data based on the data indicators representing the operating conditions of the catering company, extract the text data corresponding to the corporate status, and classify the extracted text data to obtain multiple types of status text corresponding to the data indicators data. That is, to determine the data index to which the extracted text data belongs, and divide the data belonging to the same type of data index into the same type to form multiple types of status text data such as sales and purchases.

Step S25, extracting state keywords in multiple types of the state text data respectively, and performing format conversion on the extracted multiple types of state keywords according to a preset data format to obtain the structured data.

Furthermore, the status keywords in various status text data are extracted to characterize the operating status of the enterprise to be checked in various aspects through various types of status keywords. At the same time, a preset data format is set in advance according to the required data structure. According to the preset data format, the extracted multiple types of status keywords are formatted separately, and each type of status keyword is converted into the preset data format. In the form of data, multiple types of structured data are obtained. For example, for purchasing data, the preset data format is: purchasing category-purchasing time-purchasing data, after extracting the state keywords of each purchase in the purchase text data, the state keywords of each purchase are in accordance with the preset format Arrange the data, and check whether the time keyword in the status keyword is consistent with the purchase time format required in the preset data format. If the required purchase time format is XXXX year-XX month-XX day, and the time of the time keyword If the format is XX.XX.XX, it is judged that the time format is inconsistent. While arranging the status keywords according to the preset format data, the time format of the time keywords is converted to meet the requirements of the preset data format. Structured data.

Step S251, performing segmentation processing and sentence processing on the multiple types of state text data respectively, generating multiple types of to-be-recognized clauses, and removing invalid clauses in the multiple types of the to-be-recognized clauses;

Further, in this implementation, extracting state keywords in various state text data is a process of processing each state text data separately, and the separate processing may be serial processing or parallel processing. Specifically, first perform segmentation processing on each type of state text data to obtain a text data segment of the state text data, and then perform segmentation processing on each text data segment to obtain multiple text sentences as the to-be-recognized clauses. After that, search for sentences that are not related to the operating status in multiple clauses to be identified, and remove the searched sentences as invalid clauses to ensure that the status keywords extracted from the clauses to be identified are all related to the operating status .

Step S252, performing word segmentation processing on the multiple types of the to-be-recognized clauses after removing the invalid clauses, respectively, to generate multiple types of to-be-recognized word segmentation;

Furthermore, after eliminating invalid clauses, each type of sentence to be recognized is subjected to word segmentation processing, and the sentence to be recognized is divided into multiple words according to the language logic, and the word to be recognized in each type of state text data is obtained.

Step S253: Eliminate noise words that are irrelevant to the state of the enterprise among the multiple types of word segmentation to be recognized, and obtain multiple types of state keywords in the state text data.

Further, pre-set words related to the business status form a dictionary, and compare the divided words to be recognized with the words in the dictionary one by one to determine whether the word to be recognized exists in the dictionary. If it exists in the dictionary, it is determined that the segmented word to be recognized is a valid word related to the business status, and if it does not exist in the dictionary, it is determined that the segmented word to be recognized is an invalid word that has nothing to do with the business status. After finding out all the invalid words that are not related to the business status in each type of word segmentation to be recognized, all the invalid words are eliminated as noise words that are not related to the state of the enterprise, and the status keywords in the text data of each type of status are obtained. Furthermore, various status keywords are formatted according to the preset data format to obtain structured data representing the actual state of the enterprise from various aspects. Each sub-data in the structured data exists in the same preset data format, which is convenient The processing of each sub-data in the same way is conducive to the improvement of processing efficiency.

In this embodiment, status keywords are extracted from various corporate text data of the company to be checked, and are generated as structured data to represent the actual status of the company. Because all kinds of enterprise text data are the real data of the enterprise, and represent the operating status of the enterprise from various aspects, the structured data generated according to it can reflect the true state of the enterprise from many aspects, and improve the effectiveness of the actual state of the enterprise. Sex and accuracy.

Further, based on the first embodiment or the second embodiment of the enterprise state supervision method of the present application, a third embodiment of the enterprise state supervision method of the present application is proposed. In the third embodiment, according to the data score, the supervision office After describing the steps to verify whether the enterprise status of the enterprise is valid, the following include:

Step S40, transmitting the research and judgment score corresponding to the enterprise to be checked to the state data model, and judge whether the research and judgment score matches the data score;

This embodiment is provided with an optimization mechanism for the state data model. Specifically, the enterprise to be checked is scored manually based on the state of the enterprise represented by the structured data, and the research and judgment scores of the enterprise to be checked are obtained and transmitted to the supervision server. The supervisory server transmits the research judgment score to the state data model, and judges whether the data score generated by the state data model matches the research judgment score. Among them, matching is not required to be completely consistent. When the numerical difference between the data score and the research score is within a certain range, it means that the two are relatively close, and the two can be considered to match. On the contrary, it shows that the two are far apart, and it is determined that the two do not match.

Step S50, if it matches the data score, store the data score and the structured data correspondingly;

Further, if it is determined that the research score matches the data score, it means that the state data model can accurately process structured data at present, and optimization is not necessary. At this time, the data scores and structured data are formed into a corresponding relationship and then stored as the basis for corporate state supervision.

Step S60, if it does not match the data score, search for target sample data that matches the structured data in the preset sample data;

Furthermore, if it is determined that there is a mismatch between the research score and the data score, it means that the state data model currently has low accuracy in processing structured data, and it needs to be optimized. Because the data score is generated based on the preset sample data similar to the structured data of the enterprise to be checked in the state data model, and the data score generated based on the similar preset sample data is inaccurate, the optimization processing of the state data model That is to process the similar preset sample data. Specifically, the structured data is compared with the data in the preset sample data, and the data whose similarity with the structured data is greater than the preset similarity threshold is searched, and the data obtained by the search is used as the target of matching with the structured data Sample data, which is similar to the structured data of the verification enterprise, is used to generate the preset sample data of the data score.

Step S70, removing the target sample data and the score label corresponding to the target sample data, and generating the research and judgment score as the to-be-trained score label of the structured data;

Further, the state data model generates data scores according to the score labels that represent the scores and weights carried by the target sample data. In the process of optimizing the state data model due to the inaccuracy of the generated data scores, the target sample data The score label carried by it is removed from the preset sample data and is not used as a training sample for the state data model. At the same time, because the research and judgment scores are accurate scores, the research and judgment scores are generated as the to-be-trained score labels of the structured data, which are used to train the state data model and optimize the state data model.

In step S80, the preset sample data is updated according to the structured data and the score label to be trained, and the state data model is optimized for training based on the updated preset sample data.

Furthermore, according to the data index, the structured data is converted into sample data, and the converted sample data and the score label to be trained are used as new preset sample data to update the preset sample data. Thereafter, the state data model is optimized and trained based on the updated preset sample data to improve the accuracy of the state data model.

It should be noted that this embodiment can also reset the score label for the target sample data for training, that is, only remove the score label corresponding to the target sample data, and retain the target sample data. And set the new score label of the target sample data according to the research and judgment score, and then use the target sample data and its new score label as the new preset sample data, optimize the training of the state data model, and improve the performance of the state data model. accuracy.

Further, based on the first embodiment, the second embodiment or the third embodiment of the enterprise state supervision method of the present application, the fourth embodiment of the enterprise state supervision method of the present application is proposed. In the fourth embodiment, the preset sample The data is transmitted to the initial model, and the initial model is trained based on the federated learning algorithm. The steps of generating the state data model include:

Step S11: Obtain the positive sample data corresponding to the preset positive field name and the negative sample data corresponding to the preset negative field name, and combine each of the positive sample data and each of the negative sample data Transmitting to the initial model as the preset sample data, training the initial model, and generating model gradients;

This embodiment is based on a federated learning algorithm to perform federated training on the initial model to generate a state data model. The federated training involves at least two regions, each region has its own initial model, and the presets used for training between the regions The sample data are independent of each other. All regions have the same training process for their respective initial models, and this embodiment uses any one of them for description. Specifically, the preset sample data includes positive sample data indicating that the state of the enterprise is valid, and negative sample data indicating that the state of the enterprise is invalid. A preset positive field name representing a positive sample and a preset negative field name representing a negative sample are preset. After collecting a large amount of data on the data indicators of various catering companies that have determined the enterprise status of the party, the large amount of data collected is filtered according to the preset positive field name and negative field name to obtain the The positive sample data corresponding to the field name, and the negative sample data corresponding to the preset negative field name. Then set different scores and weights for each positive sample data, and set different scores and weights for each negative sample data, and then set each positive sample data and each negative sample data as the default The sample data is transmitted to the initial model for training, and the model gradient used by the party to update the model parameters is generated.

Step S12: Transmit the model gradient to the coordinator corresponding to the federated learning algorithm, so that the coordinator can aggregate the model gradient and at least one other model gradient generated based on the federated learning algorithm to generate Return gradient

Further, in order to coordinate the initial model training process of the various regions, a coordinator corresponding to the federated learning algorithm is set in the federal training process. The coordinator can be any party in the various regions, or it can be independent of each party. Third party in the region. The generated model gradient is transmitted to the coordinating party, and the coordinating party aggregates the model gradient and other model gradients generated by other parties based on the federated learning algorithm. The aggregation can be set to mean aggregation or weighted aggregation according to requirements. Generate a return gradient and return it to the supervisory server in each region.

Step S13: Receive the return gradient returned by the coordinator, and continuously train the initial model according to the return gradient until the initial model converges to obtain the state data model.

Furthermore, after receiving the return gradient returned by the coordinator, the initial model is continuously trained according to the return gradient. After each training, it is judged whether the initial model converges. If it converges, it means that the trained initial model can accurately generate data scores, and the trained initial model is used as the state data model. On the contrary, if it does not converge, continue training until it converges to obtain the state data model.

It should be noted that the convergence of the initial model can be determined by the convergence function in the initial model. After each training of the initial model, the test sample data is processed according to the model parameters obtained through training in the initial model to obtain the processing result. The convergence function is used to calculate the loss value between the processing result and the expected result. After the loss value continues to be less than the preset value for many times, it is determined that the initial model has converged, and the training is stopped. Otherwise, the training is continued.

In this embodiment, a federated learning algorithm is used to perform federated training on the initial model to obtain a state data model. The preset sample data in various regions is not transmitted outside, which protects data privacy while enriching the number of training samples and optimizes the training effect of the state data model. , Which makes the enterprise state supervision based on the state data model more accurate.

Furthermore, this application also provides an enterprise state monitoring device.

Referring to Fig. 3, Fig. 3 is a schematic diagram of the functional modules of the first embodiment of the enterprise state monitoring device of this application. The enterprise state monitoring device includes:

The generating module 10 is used for transmitting preset sample data to the initial model, and training the initial model based on a federated learning algorithm to generate a state data model;

The obtaining module 20 is configured to obtain structured data corresponding to the state of the enterprise in the enterprise to be checked, and transmit the structured data to the state data model to generate a data score of the structured data;

The supervision module 30 is configured to supervise whether the enterprise status of the enterprise to be checked is valid according to the data score.

In the enterprise state monitoring device of this embodiment, the generation module 10 first transmits the preset sample data to the initial model, and trains the initial model based on the federated learning algorithm to generate the state data model; The structured data corresponding to the state of the enterprise is transmitted to the state data model to generate the data score of the structured data; and the supervisory module 30 supervises whether the enterprise state of the enterprise to be checked is valid according to the data score. Among them, the preset sample data is various types of data representing their respective states in each enterprise, and is the real and effective data of each enterprise. The federated learning algorithm is combined with the preset sample data of a large number of enterprises for training, which enriches the training sample size and makes all The generated state data model is more accurate. Therefore, the status data model is used to monitor the effectiveness of the state of the enterprise to be verified, which combines various real data of the enterprise to be verified to reflect the state of the enterprise, avoids relying on the inspection of industrial and commercial registration for supervision, and ensures the authenticity of the state of the supervised enterprise. While ensuring the effectiveness and accuracy of supervision, it also ensures the timeliness of supervision, which is conducive to timely supervision and efficient supervision.

Further, the acquisition module 20 includes:

The collection unit is used to collect enterprise text data of the enterprise to be checked, and extract the text data corresponding to the enterprise status from each of the enterprise text data for classification, and obtain multiple types of status text data;

The conversion unit is used to extract state keywords in multiple types of the state text data, and perform format conversion on the extracted multiple types of state keywords according to a preset data format to obtain the structured data.

Further, the conversion unit is also used for:

Further, the acquisition module 20 further includes:

The first transmission unit is configured to transmit the structured data to the state data model, and determine target sample data respectively matching various types of sub-data in the structured data;

The first determining unit is configured to determine the sub-scores of various types of the sub-data according to the scores and weights respectively corresponding to each of the target sample data;

The generating unit is configured to generate the data score of the structured data according to the sub-score of the various types of sub-data.

Further, the monitoring module 30 further includes:

The second determining unit is configured to determine the target state corresponding to the combination formed by the maximum value, the minimum value and the average value in the data score according to the preset correspondence between the combined score and the state;

The supervision unit is configured to search for the registration status corresponding to the enterprise to be checked, and supervise whether the enterprise status of the enterprise to be checked is valid according to the consistency between the target state and the registration status.

Further, the enterprise state monitoring device further includes:

The judgment module is configured to transmit the research judgment score corresponding to the enterprise to be checked to the state data model, and judge whether the research judgment score matches the data score;

A storage module, configured to store the data score and the structured data correspondingly if it matches the data score;

A search module, configured to search for target sample data matching the structured data in the preset sample data if the score does not match the data;

A rejecting module, configured to reject the target sample data and the score label corresponding to the target sample data, and generate the research and judgment score as the to-be-trained score label of the structured data;

The update module is configured to update the preset sample data according to the structured data and the score label to be trained, and optimize the training of the state data model based on the updated preset sample data.

Further, the generating module 10 further includes:

The acquiring unit is configured to acquire positive sample data corresponding to the preset positive field name and negative sample data corresponding to the preset negative field name, and combine each of the positive sample data and each of the negative The sample data is transmitted to the initial model as the preset sample data, and the initial model is trained to generate model gradients;

The first transmission unit is configured to transmit the model gradient to the coordinating party corresponding to the federated learning algorithm, so that the coordinating party can transfer the model gradient and at least one other model gradient generated based on the federated learning algorithm Perform aggregation to generate a return gradient;

The receiving unit is configured to receive the return gradient returned by the coordinating party, and continuously train the initial model according to the return gradient until the initial model converges to obtain the state data model.

The specific implementation of the enterprise state monitoring device of this application is basically the same as the foregoing embodiments of the enterprise state monitoring method, and will not be repeated here.

In addition, the embodiment of the present application also proposes a computer-readable storage medium. The computer-readable storage medium may be non-volatile or volatile.

The computer-readable storage medium stores an enterprise state monitoring program, and the enterprise state monitoring program is executed by a processor to implement the steps of the enterprise state monitoring method as described above.

The specific implementation of the computer-readable storage medium of this application is basically the same as the foregoing embodiments of the enterprise state supervision method, and will not be repeated here.

It should be noted that in this article, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, It also includes other elements that are not explicitly listed, or elements inherent to the process, method, article, or system. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, article, or system that includes the element.

The serial numbers of the foregoing embodiments of the present application are for description only, and do not represent the superiority or inferiority of the embodiments.

Through the description of the above implementation manners, those skilled in the art can clearly understand that the above-mentioned embodiment method can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。 Based on this understanding, the technical solution of this application essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a computer-readable storage medium as described above (such as The ROM/RAM, magnetic disk, optical disk) includes several instructions to make a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in the various embodiments of the present application.

The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims

An enterprise state supervision method, wherein the enterprise state supervision method includes the following steps:

Transmitting the preset sample data to the initial model, and training the initial model based on a federated learning algorithm to generate a state data model;

Acquiring structured data corresponding to the state of the enterprise in the enterprise to be checked, and transmitting the structured data to the state data model to generate a data score of the structured data;

According to the data score, whether the enterprise status of the enterprise to be checked is effective is supervised.
The method for monitoring the state of an enterprise according to claim 1, wherein the step of obtaining structured data corresponding to the state of the enterprise in the enterprise to be checked comprises:

Collect enterprise text data of the enterprise to be checked, and extract the text data corresponding to the enterprise status from each of the enterprise text data for classification, and obtain multiple types of status text data;

Extracting state keywords in multiple types of the state text data respectively, and performing format conversion on the extracted multiple types of state keywords according to a preset data format, to obtain the structured data.
3. The method for monitoring the state of an enterprise according to claim 2, wherein the step of extracting state keywords in multiple types of state text data respectively comprises:

Performing segmentation processing and sentence processing on multiple types of the state text data respectively, generating multiple types of to-be-recognized clauses, and eliminating invalid clauses in the multiple types of the to-be-recognized clauses;

Perform word segmentation processing on the multiple types of the to-be-recognized clauses after the invalid clauses are eliminated, and generate multiple types of to-be-recognized word segmentation;

The noise words that are irrelevant to the state of the enterprise in the multiple types of word segmentation to be recognized are eliminated to obtain the state keywords in the multiple types of state text data.
The method for monitoring the state of an enterprise according to claim 1, wherein the step of transmitting the structured data to the state data model to generate a data score of the structured data comprises:

Transmitting the structured data to the state data model, and determining target sample data respectively matching various types of sub-data in the structured data;

Determine the sub-scores of the various types of sub-data according to the scores and weights respectively corresponding to each of the target sample data;

The data score of the structured data is generated according to the sub-score of the various types of the sub-data.
The method for monitoring the state of an enterprise according to claim 1, wherein the step of monitoring whether the enterprise state of the enterprise to be checked is valid according to the data score comprises:

Determine the target state corresponding to the combination formed by the maximum value, the minimum value and the average value in the data score according to the preset correspondence between the combined score and the state;

Find the registration status corresponding to the company to be checked, and supervise whether the company status of the company to be checked is valid according to the consistency between the target status and the registration status.
The method for monitoring the state of an enterprise according to any one of claims 1 to 5, wherein the step of supervising whether the enterprise state of the enterprise to be checked is valid according to the data score afterwards comprises:

Transmitting the research and judgment score corresponding to the enterprise to be checked to the state data model, and determine whether the research and judgment score matches the data score;

If it matches the data score, store the data score and the structured data correspondingly;

If it does not match the data score, searching for target sample data that matches the structured data in the preset sample data;

Removing the target sample data and the score label corresponding to the target sample data, and generating the research and judgment score as the to-be-trained score label of the structured data;

According to the structured data and the score label to be trained, the preset sample data is updated, and the state data model is optimized for training based on the updated preset sample data.
The method for monitoring the state of an enterprise according to any one of claims 1-5, wherein the step of transmitting preset sample data to an initial model, and training the initial model based on a federated learning algorithm to generate a state data model include:

Obtain the positive sample data corresponding to the preset positive field name and the negative sample data corresponding to the preset negative field name, and use each of the positive sample data and each of the negative sample data as the The preset sample data is transmitted to the initial model, the initial model is trained, and the model gradient is generated;

The model gradient is transmitted to the coordinating party corresponding to the federated learning algorithm, so that the coordinating party aggregates the model gradient and at least one other model gradient generated based on the federated learning algorithm to generate a return gradient ；

Receive the return gradient returned by the coordinator, and continuously train the initial model according to the return gradient until the initial model converges to obtain the state data model.
An enterprise state monitoring device, wherein the enterprise state monitoring device includes:

A generating module, used for transmitting preset sample data to the initial model, and training the initial model based on a federated learning algorithm to generate a state data model;

An obtaining module is used to obtain structured data corresponding to the state of the enterprise in the enterprise to be checked, and transmit the structured data to the state data model to generate a data score of the structured data;

The supervision module is used to supervise whether the enterprise status of the enterprise to be checked is valid according to the data score.
An enterprise state monitoring device, wherein the enterprise state monitoring device includes a memory, a processor, and an enterprise state monitoring program stored on the memory and running on the processor, and the enterprise state monitoring program is The following steps are implemented when the processor is executed:

Transmitting the preset sample data to the initial model, and training the initial model based on a federated learning algorithm to generate a state data model;

Acquiring structured data corresponding to the state of the enterprise in the enterprise to be checked, and transmitting the structured data to the state data model to generate a data score of the structured data;

According to the data score, whether the enterprise status of the enterprise to be checked is effective is supervised.
9. The enterprise state monitoring device according to claim 9, wherein the step of obtaining structured data corresponding to the state of the enterprise in the enterprise to be checked comprises:

Collect enterprise text data of the enterprise to be checked, and extract the text data corresponding to the enterprise status from each of the enterprise text data for classification, and obtain multiple types of status text data;

Extracting state keywords in multiple types of the state text data respectively, and performing format conversion on the extracted multiple types of state keywords according to a preset data format, to obtain the structured data.
10. The enterprise state monitoring device according to claim 10, wherein the step of extracting state keywords in multiple types of state text data respectively comprises:

Performing segmentation processing and sentence processing on multiple types of the state text data respectively, generating multiple types of to-be-recognized clauses, and eliminating invalid clauses in the multiple types of the to-be-recognized clauses;

Perform word segmentation processing on the multiple types of the to-be-recognized clauses after the invalid clauses are eliminated, and generate multiple types of to-be-recognized word segmentation;

The noise words that are irrelevant to the state of the enterprise in the multiple types of word segmentation to be recognized are eliminated to obtain the state keywords in the multiple types of state text data.
9. The enterprise state monitoring device according to claim 9, wherein the step of transmitting the structured data to the state data model to generate a data score of the structured data comprises:

Transmitting the structured data to the state data model, and determining target sample data respectively matching various types of sub-data in the structured data;

Determine the sub-scores of the various types of sub-data according to the scores and weights respectively corresponding to each of the target sample data;

The data score of the structured data is generated according to the sub-score of the various types of the sub-data.
9. The enterprise state monitoring device according to claim 9, wherein the step of monitoring whether the enterprise state of the enterprise to be checked is valid according to the data score comprises:

Determine the target state corresponding to the combination formed by the maximum value, the minimum value and the average value in the data score according to the preset correspondence between the combined score and the state;

Find the registration status corresponding to the company to be checked, and supervise whether the company status of the company to be checked is valid according to the consistency between the target status and the registration status.
The enterprise state monitoring device according to any one of claims 9-13, wherein the step of monitoring whether the enterprise state of the enterprise to be checked is valid according to the data score afterwards comprises:

Transmitting the research and judgment score corresponding to the enterprise to be checked to the state data model, and determine whether the research and judgment score matches the data score;

If it matches the data score, store the data score and the structured data correspondingly;

If it does not match the data score, searching for target sample data that matches the structured data in the preset sample data;

Removing the target sample data and the score label corresponding to the target sample data, and generating the research and judgment score as the to-be-trained score label of the structured data;

According to the structured data and the score label to be trained, the preset sample data is updated, and the state data model is optimized for training based on the updated preset sample data.
The enterprise state monitoring device according to any one of claims 9-13, wherein the step of transmitting preset sample data to an initial model, and training the initial model based on a federated learning algorithm to generate a state data model include:

Obtain the positive sample data corresponding to the preset positive field name and the negative sample data corresponding to the preset negative field name, and use each of the positive sample data and each of the negative sample data as the The preset sample data is transmitted to the initial model, the initial model is trained, and the model gradient is generated;

The model gradient is transmitted to the coordinating party corresponding to the federated learning algorithm, so that the coordinating party aggregates the model gradient and at least one other model gradient generated based on the federated learning algorithm to generate a return gradient ；

Receive the return gradient returned by the coordinator, and continuously train the initial model according to the return gradient until the initial model converges to obtain the state data model.
A computer-readable storage medium, wherein an enterprise state monitoring program is stored on the computer-readable storage medium, and the following steps are implemented when the enterprise state monitoring program is executed by a processor:

Transmitting the preset sample data to the initial model, and training the initial model based on a federated learning algorithm to generate a state data model;

Acquiring structured data corresponding to the state of the enterprise in the enterprise to be checked, and transmitting the structured data to the state data model to generate a data score of the structured data;

According to the data score, whether the enterprise status of the enterprise to be checked is effective is supervised.
15. The computer-readable storage medium of claim 16, wherein the step of obtaining structured data corresponding to the state of the enterprise in the enterprise to be checked comprises:

Collect enterprise text data of the enterprise to be checked, and extract the text data corresponding to the enterprise status from each of the enterprise text data for classification, and obtain multiple types of status text data;

Extracting state keywords in multiple types of the state text data respectively, and performing format conversion on the extracted multiple types of state keywords according to a preset data format, to obtain the structured data.
17. The computer-readable storage medium of claim 17, wherein the step of extracting state keywords in the plurality of types of state text data respectively comprises:

Performing segmentation processing and sentence processing on multiple types of the state text data respectively, generating multiple types of to-be-recognized clauses, and eliminating invalid clauses in the multiple types of the to-be-recognized clauses;

Perform word segmentation processing on the multiple types of the to-be-recognized clauses after the invalid clauses are eliminated, and generate multiple types of to-be-recognized word segmentation;

The noise words that are irrelevant to the state of the enterprise in the multiple types of word segmentation to be recognized are eliminated to obtain the state keywords in the multiple types of state text data.
15. The computer-readable storage medium of claim 16, wherein the step of transmitting the structured data to the state data model to generate a data score of the structured data comprises:

Transmitting the structured data to the state data model, and determining target sample data respectively matching various types of sub-data in the structured data;

Determine the sub-scores of the various types of sub-data according to the scores and weights respectively corresponding to each of the target sample data;

The data score of the structured data is generated according to the sub-score of the various types of the sub-data.
15. The computer-readable storage medium according to claim 16, wherein the step of monitoring whether the enterprise status of the enterprise to be checked is valid according to the data score comprises:

Determine the target state corresponding to the combination formed by the maximum value, the minimum value and the average value in the data score according to the preset correspondence between the combined score and the state;

Find the registration status corresponding to the company to be checked, and supervise whether the company status of the company to be checked is valid according to the consistency between the target status and the registration status.