A method for selecting industry codes based on Bayesian classification

Through Bayesian classification method, the priors and conditional probability of the business scope of the enterprise is calculated, and industry coding options are recommended, which solves the problem of inconsistency between the business scope of the enterprise and the industry data in the banking financial institutions, and improves the data checksum repair efficiency. Reduces the risk and cost of manual adjustments.

CN115358309BActive Publication Date: 2024-11-22JIANGSU SUNING BANK CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210973020.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-15
Publication Date
2024-11-22
Estimated Expiration
2042-08-15

AI Technical Summary

Technical Problem

In the processing of regulatory data reporting by banking financial institutions, it is difficult to automatically solve the problem of inconsistency between the enterprise's "business scope" and "industry" data, resulting in high risk and high cost of manual adjustments, which affects data operation and management.

Method used

Using the Bayesian classification-based industry coding selection method, by obtaining basic information of the enterprise, analyzing business scope, calculating prior probability and conditional probability, recommending the most likely industry coding options, improving customer selection efficiency and accuracy, suitable for new Industry code classification for old customers.

Benefits of technology

Using probability calculation of historical data, we automatically solve the industry coding matching problem, improve data checksum repair efficiency, reduce the risks and costs of manual adjustments, and improve the efficiency of data operation and management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115358309B_ABST
    Figure CN115358309B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for selecting industry codes based on Bayesian classification. The method includes obtaining relevant application materials submitted by users; extracting basic information of the enterprise from the application materials; judging whether the enterprise is a new industry customer, if so, listing all industries within the business scope of the enterprise, and allowing the customer to select the industry to which it belongs; otherwise, by analyzing the content of the business scope of the enterprise, identifying the data set of the business scope of the enterprise, and providing the four-level industry code recommendation data option with the highest probability; if the four-level industry code recommendation data option is not recognized by the user, expanding the recommendation scope until the selection is confirmed successfully, or allowing the customer to verify and confirm the industry code recommended by the system, and storing it in the library if it is correct. The present invention solves the main industry classification calculation problem of enterprises operating across industries and the verification and repair problem of existing historical data, improves the efficiency and accuracy of customers' positioning of industry codes, and gradually improves the industry code classification capabilities for new customer groups.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of industry code selection methods based on Bayesian classification, and in particular to an industry code selection method based on Bayesian classification. Background Art

[0002] my country's banking and financial institutions are responsible for reporting to the national financial regulatory authorities on a regular basis. As government departments have higher and higher requirements for the accuracy of regulatory reporting, banks need to invest more and more manpower and material resources in the verification of regulatory reporting data. Due to the complexity of the data source processing process and different processing logic, after the data is summarized into a comprehensive report, it is easy to have inconsistent or inaccurate data.

[0003] In the processing of regulatory submission data, the submission content involving enterprise business data includes data in two dimensions: "business scope" and "industry". Among them, "business scope" belongs to multiple parallel relationship data, that is, the business scope of an enterprise may be relatively broad, spanning multiple "industries". However, the enterprise can only fill in one "industry", and some system data filling introduces manual filling content, which easily causes the problem of mismatch between the enterprise's "business scope" and "industry".

[0004] When data inconsistency occurs, if the data scale is small, it can be adjusted manually. However, if the data involved is too large, manual adjustment is risky and cannot fundamentally solve the problem of data inconsistency. Furthermore, data repair takes a long time and involves high manual repair costs, which is not conducive to subsequent data operation management. Summary of the invention

[0005] The purpose of the present invention is to provide an industry code selection method based on Bayesian classification in view of the deficiencies in the prior art.

[0006] To achieve the above object, the present invention provides an industry code selection method based on Bayesian classification, comprising:

[0007] Step 1: Obtain relevant application materials submitted by the user when applying for relevant services;

[0008] Step 2: extracting basic information of the enterprise from the application materials;

[0009] Step 3: Based on the extracted basic information of the enterprise, determine whether the enterprise is a new industry customer. If it is a new industry customer, jump to step 6; otherwise, go to step 4;

[0010] Step 4: Identify the enterprise business scope data set by analyzing the content of the enterprise business scope, and give the recommended data options of the four-level industry codes with the highest probability;

[0011] Step 5: If the four-level industry code recommended data option is not recognized by the user, based on the business scope of the enterprise, the four-level industry code recommended data options with the highest probability after filtering are given for the customer to choose. If it is still not recognized by the user, it will fall back to the three-level industry code recommended data option for the customer to choose. If it is still not recognized by the user, the user will choose to continue to fall back until the selection is confirmed successfully or jump to step 6;

[0012] Step 6. If the customer is a new industry customer, all industries within the business scope of the enterprise will be listed, and the customer will select the industry to which he / she belongs. Otherwise, the customer will verify and confirm the industry code recommended by the system.

[0013] Step 7: After the user confirms that the selected or verified industry code is correct, it will be stored in the warehouse after going through the entire business processing flow.

[0014] Furthermore, the application materials include a business license, and step 2 parses the content of the business license based on OCR.

[0015] Furthermore, the step 4 specifically includes:

[0016] After data cleaning and extraction of each business scope record in the current historical data, count the i-th business scope record r i The number of items in a specific category within a row Calculate the i-th business scope record r i The prior probability p(r i )for:

[0017]

[0018] Where M represents the total number of business record items;

[0019] Count the number of each four-level industry code in the current historical data, and map the frequency of occurrence to the probability of the four-level classification code, as follows:

[0020]

[0021] Where N represents the total number of four-level industry codes. Represents the i-th industry code h i The number of occurrences in the current dataset;

[0022] According to the content analysis result of the enterprise's business scope, the business scope of the enterprise is given as r1, r2, ..., r n , where n is a natural number greater than 2;

[0023] When the industry code is fixed, calculate the current given business scope r1, r2, ..., rn The conditional probability p(r j |h i ), where j∈[1,…,n];

[0024] According to the current given business scope r1, r2, ..., r n The conditional probability p(r j |h i ) Calculate the probability p(h) of industry code i |r1,r2,...,r n ), as follows:

[0025]

[0026] Among them, p(r1, r2, ..., r n |h i ) is to confirm that the industry code is h i In the case of r1, r2, ..., r n The conditional probability of this combination, p(r1, r2, ..., r n |h l ) is to confirm that the industry code is h l In the case of r1, r2, ..., r n The conditional probability of this combination, p(h l ) is the probability of the four-level classification code of the l-th business scope, l∈[1,…,N];

[0027] Calculate argmax(p(h i |r1,r2,…,r n )) and is used as the recommended data option for the fourth-level industry code with the highest probability. argmax(.) is the function for finding the maximum independent variable.

[0028] Furthermore, the data cleaning and extraction method is specifically as follows:

[0029] The business scope is divided into several major categories based on "general business items" and "franchise items";

[0030] Use ";" as a separator to decompose each business item and remove the "." symbol to form a business scope list busi_scope_list.

[0031] Furthermore, the currently given business range r1, r2, ..., r n The conditional probability p(r j |h i ) is calculated as follows:

[0032] Extract each industry code h iThe business scope set of all associated enterprises {r1, r2, ..., r m}, where m is a natural number greater than 2;

[0033] Statistics for each business scope k Frequency And calculate the business scope r k The conditional probability p(r k |h i )for:

[0034]

[0035] Among them, r k ∈{r1, r2, ..., r m};

[0036] Find the probability set S of the subset of the business scope of the enterprise, where S∈{r1, r2, ..., r m};

[0037] Repeat until all industry codes are traversed.

[0038] Furthermore, when calculating the probability p(h i |r1,r2,...,r n ) and then i The probability of the industry is summarized and sorted into three levels of industry coding probability The details are as follows:

[0039]

[0040] in, is the probability of the third-level industry code corresponding to the fourth-level industry code of the i-th category;

[0041] Coding probability of the three industries Accumulate and construct the secondary industry coding probability The details are as follows:

[0042]

[0043] in, is the probability of the second-level industry code corresponding to the fourth-level industry code of the i-th category.

[0044] Furthermore, the method for determining whether the enterprise is a new industry customer is as follows:

[0045] Query the prior probability p(r i ), if its prior probability p(r i) is 0, then the enterprise is judged to be a new industry customer, otherwise, the enterprise is judged to be not a new industry customer.

[0046] Furthermore, it also includes:

[0047] Regularly update the prior probabilities, conditional probabilities, and information about new industries and new customers that are pre-calculated and stored in the system.

[0048] Furthermore, the application materials are submitted through an online browser page or App page.

[0049] Furthermore, the value of A is 3 or 5.

[0050] Beneficial effects: 1. The present invention makes full use of the probability calculation of existing historical data to solve the main industry classification calculation problem of enterprises operating across industries;

[0051] 2. The present invention classifies data with empty industry codes in historical data based on a big data statistics method, thereby solving the problem of verifying and repairing the existing historical data.

[0052] 3. Compared with the traditional method of listing and selecting all industry information, the present invention improves the efficiency and accuracy of customers' positioning of industry codes through probability calculation and sorting;

[0053] 4. The present invention can be repeatedly iterated and optimized, and can gradually improve the industry coding classification capabilities for new customer groups as the size of the customer group changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flowchart of the industry code selection method based on Bayesian classification;

[0055] Figure 2 This is a schematic diagram after a preliminary decomposition based on common keywords;

[0056] Figure 3 It is a schematic diagram after decomposing the content into sibling nodes through punctuation marks. DETAILED DESCRIPTION

[0057] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. The embodiments are implemented based on the technical solution of the present invention. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0058] like Figure 1 As shown, an embodiment of the present invention provides an industry code selection method based on Bayesian classification, comprising:

[0059] Step 1: Obtain the relevant application materials submitted by the user when applying for relevant services. For example, when applying for a corporate loan, the user needs to upload the business license of the company, the ID card of the applicant and other documents. The above application materials can generally be submitted through the online browser page or App page.

[0060] Step 2: Extract the basic information of the enterprise from the application materials. The basic information of the business license can be parsed through OCR, thereby converting the image content into text content, such as company name, business scope, etc. In addition, according to regulatory requirements, each enterprise needs to give a four-level industry code based on national standards, but the business license itself does not list the detailed four-level standard code. Therefore, it is necessary to list the industry code to which the enterprise belongs based on its business scope.

[0061] Step 3: Based on the extracted basic information of the enterprise, determine whether the enterprise is a new industry customer. If it is a new industry customer, jump to step 6 and guide the customer to fill in the industry code by themselves (based on the mapping relationship between the business scope and the national standard industry classification code). Otherwise, go to step 4 and enter the industry classification engine link to calculate the maximum possible industry code.

[0062] Step 4: Identify the enterprise business scope data set by analyzing the content of the enterprise business scope, and give the recommended data options for the four-level industry codes with the highest probability. The details are as follows:

[0063] After data cleaning and extraction of each business scope record in the current historical data, count the i-th business scope record r i The number of items in a specific category within a row Calculate the i-th business scope record r i The prior probability p(r i )for:

[0064]

[0065] Among them, M represents the total number of business record details.

[0066] Count the number of each four-level industry code in the current historical data, and map the frequency of occurrence to the probability of the four-level classification code, as follows:

[0067]

[0068] Where N represents the total number of four-level industry codes (fixed value according to national standards). Represents the i-th industry code h i The number of occurrences in the current dataset;

[0069] According to the results of the content analysis of the business scope of the enterprise, the business scope of the enterprise is given as r1, r2, ..., r n , where n is a natural number greater than 2.

[0070] When the industry code is fixed, calculate the current given business scope r1, r2, ..., r n The conditional probability p(r j |h i ), where j∈[1,…,n]. The details are as follows:

[0071] Extract each industry code h i The business scope set of all associated enterprises {r1, r2, ..., r m}, where m is a natural number greater than 2;

[0072] Statistics for each business scope k Frequency And calculate the business scope r k The conditional probability p(r k |h i )for:

[0073]

[0074] Among them, r k ∈{r1, r2, ..., r m};

[0075] Find the probability set S of the subset of the business scope of the enterprise, where S∈{r1,r2,...,r m};

[0076] Repeat until all industry codes are traversed.

[0077] According to the current given business scope r1, r2, ..., r n The conditional probability p(r j |h i ) Calculate the probability p(h) of industry code i |r1,r2,...,r n ), since the industry coding probability should satisfy ∑p(h i |r1,r2,...,r n )=1 (i.e. the sum of all industry coding probabilities under conditional probability is 1), and since r1, r2, ..., r n They are independent of each other. The probability of calculating industry codes according to the Bayesian formula is as follows:

[0078]

[0079] Since the total probability formula (p(r) = p(r|h)*p(h)+p(r|h c )*p(h c )) and the data in the industry h set are independent variables (only one industry code is allowed), the above formula can be further transformed into:

[0080]

[0081] Among them, p(r1, r2, ..., r n |h i ) is to confirm that the industry code is h i In the case of r1, r2, ..., r n The conditional probability of this combination, p(r1, r2, ..., r n |h l ) is to confirm that the industry code is h l In the case of r1, r2, ..., r n The conditional probability of this combination, p(h l ) is the probability of the four-level classification code of the l-th business scope, l∈[1,…,N];

[0082] Calculate argmax(p(h i |r1,r2,…,r n )) and is used as the recommended data option for the fourth-level industry code with the highest probability. argmax(.) is the function for finding the maximum independent variable.

[0083] Step 5: If the four-level industry code recommended data option is not recognized by the user, based on the business scope of the enterprise, the four-level industry code recommended data options with the highest probability after filtering are given for the customer to choose. If it is still not recognized by the user, it will fall back to the three-level industry code recommended data option for the customer to choose. If it is still not recognized by the user, the user can choose to continue to fall back until the selection is confirmed successfully or jump to step 6. The value of A above can be 3 or 5. After calculating the probability p(h i |r1,r2,…,r n ) can be used i The probability of the industry is summarized and sorted into three levels of industry coding probability The details are as follows:

[0084]

[0085] in, is the probability of the third-level industry code corresponding to the fourth-level industry code of the i-th category;

[0086] Coding probability of three-level industries Accumulate and construct the secondary industry coding probability The details are as follows:

[0087]

[0088] in, is the probability of the second-level industry code corresponding to the fourth-level industry code of the i-th category.

[0089] Step 6. If the customer is a new industry customer, all industries within the business scope of the enterprise will be listed, and the customer will select the industry to which he / she belongs. Otherwise, the customer will verify and confirm the industry code recommended by the system.

[0090] Step 7: After the user confirms that the selected or verified industry code is correct, it will be stored in the database after going through the entire business processing process. The stored data can be used for future regular regulatory reporting and statistics.

[0091] The specific method of data cleaning and extraction in step 4 above is as follows:

[0092] For details, please refer to Figure 2 First, we conduct preliminary decomposition based on common keywords in the system. For example, we can decompose the business scope into several major categories based on "general business items" and "franchise items".

[0093] See also Figure 3 Then, for each major subcategory, the content is decomposed into brother nodes by punctuation. Specifically, ";" is used as a separator to decompose each business item, and the "." symbol is removed to form a business scope list busi_scope_list.

[0094] The above-mentioned method of determining whether an enterprise is a new industry customer is as follows:

[0095] Query the prior probability p(r i ), if its prior probability p(r i ) is 0, the enterprise is judged to be a new industry customer, otherwise it is judged to be a non-new industry customer.

[0096] As business development and customer groups gradually change, it is necessary to regularly update the pre-calculated results of various prior probabilities and the calculation results of conditional probabilities in the system. The pre-calculated prior probabilities, conditional probabilities, and information on new industries and new customers in the system can be regularly updated on a weekly, monthly, or quarterly basis.

[0097] The above is only a preferred embodiment of the present invention. It should be noted that for ordinary technicians in this technical field, other parts not specifically described belong to the prior art or common knowledge. Several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention.

Claims

1. A method for selecting industry codes based on Bayesian classification, characterized in that: include: Step 1: Obtain relevant application materials submitted by the user when applying for relevant services; Step 2: Extracting basic information of the enterprise from the application materials; Step 3: Based on the extracted basic information of the enterprise, determine whether the enterprise is a new industry customer. If it is a new industry customer, jump to step 6; otherwise, go to step 4; Step 4: Identify the enterprise business scope dataset by analyzing the content of the enterprise business scope and provide the recommended data options for the four-level industry codes with the highest probability; Step 5: If the fourth-level industry code recommended data option is not approved by the user, then based on the business scope of the enterprise, the filtered A fourth-level industry code recommended data options with the highest probability are provided for the customer to choose. If the user still does not approve, the user is given back to the third-level industry code recommended data option for selection. If the user still does not approve, the user can choose to continue to backtrack until the selection is successfully confirmed or jump to step 6. Step 6. If the client is a new industry customer, all industries within the business scope of the enterprise will be listed, and the client will select the industry to which they belong. Otherwise, the client will verify and confirm the industry code recommended by the system. Step 7: After the user confirms that the selected or verified industry code is correct, the entire business process is completed and the code is stored in the warehouse; The step 4 specifically includes: After cleaning and extracting each business scope record in the current historical data, count the i-th business scope record The number of specific category items in a row , calculate the i-th business scope record Prior probability under the current data set for: ; Where M represents the total number of business record items; Count the number of each level 4 industry code in the current historical data, and map the frequency of occurrence to the probability of the level 4 classification code, as follows: ; Among them, N represents the total number of four-level industry codes. Represents the i-th industry code The number of occurrences in the current dataset; The business scope of the enterprise is given according to the content analysis results of the enterprise's business scope , where n is a natural number greater than 2; Calculate the current business scope when the industry code is fixed The conditional probability of ,in ; According to the current business scope The conditional probability of Calculating the probability of industry codes , as follows: ; in, To confirm the industry code In the case of The conditional probability of this combination is To confirm the industry code In the case of The conditional probability of this combination is For the The probability of the four-level classification code of the business scope, ; calculate , and as the recommended data option for the fourth-level industry coding with the highest probability, To find the function with the largest independent variable; Calculating the probability of industry coding Afterwards, The probability of the industry is summarized and sorted into three levels of industry coding probability , as follows: ; in, The probability of the third-level industry code corresponding to the fourth-level industry code of the i-th category; Coding probability of the three-level industries Accumulate and construct the secondary industry coding probability , as follows: ; in, is the probability of the second-level industry code corresponding to the fourth-level industry code of the i-th category.

2. The method for selecting industry codes based on Bayesian classification according to claim 1, characterized in that: The application materials include a business license, and step 2 is based on OCR to parse the business license content.

3. The method for selecting industry codes based on Bayesian classification according to claim 1, characterized in that: The data cleaning and extraction method is as follows: The business scope is divided into several major categories based on "general business items" and "franchise items"; Use ";" as a separator to break down each business item and remove the "." symbol to form the business scope list busi_scope_list.

4. The method for selecting industry codes based on Bayesian classification according to claim 1, characterized in that: The current business scope The conditional probability of is calculated as follows: Extract each industry code The business scope of all related companies , where m is a natural number greater than 2; Statistics for each business scope Frequency , and calculate the business scope The conditional probability of for: ; in, ; Find the probability set of the subset of business scope of the enterprise ,in, ; Repeat until all industry codes are traversed.

5. The method for selecting industry codes based on Bayesian classification according to claim 1, characterized in that: The specific method for determining whether the enterprise is a new industry customer is as follows: Query the prior probability of each business scope of the enterprise , if its prior probability If it is 0, the enterprise is judged to be a new industry customer; otherwise, the enterprise is judged to be a non-new industry customer.

6. The method for selecting industry codes based on Bayesian classification according to claim 1, characterized in that: Also includes: Regularly update the prior probabilities, conditional probabilities, and information about new industries and new customers that are pre-calculated and stored in the system.

7. The method for selecting industry codes based on Bayesian classification according to claim 1, characterized in that: The application materials are submitted through the online browser page or App page.

8. The method for selecting industry codes based on Bayesian classification according to claim 1, characterized in that: The value of A is 3 or 5.