Method and system for full life cycle neural network labeling based on ERI

Through the ERI-based full-life cycle neural network labeling method, the convolutional neural network algorithm and multi-code labels are used to solve the correlation difficulties of cross-data sources in the existing technology, and efficient data traceability and retrieval are achieved, which improves the The efficiency of enterprise data management.

CN117455508BActive Publication Date: 2024-11-22ZHEJIANG ANGV SOFTWARE
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202311360787.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-10-19
Publication Date
2024-11-22
Estimated Expiration
2043-10-19

AI Technical Summary

Technical Problem

Existing tag technology cannot achieve efficient data association and data traceability across tables, databases, and servers, and the retrieval algorithm is inefficient.

Method used

Using the full-life cycle neural network label method based on ERI, the learning index tag is generated through the convolutional neural network learning algorithm, and the address code, application code, data code, life code and verification code are combined to realize the positioning and association of data. And record the creation, survival and demise of tags, and complete the memory and traceability of business processes.

Benefits of technology

It realizes rapid definition and traceability of enterprise data, opens up the correlation of heterogeneous data, and improves data retrieval efficiency and memory capabilities of business processes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117455508B_ABST
    Figure CN117455508B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for full life cycle neural network labeling based on ERI, and relates to the field of big data application information technology. The method includes step S1: establishing a label composed of multiple characters based on ERI, the label consists of an address code, an application code, a data code, a life code and a check code, and defining a unique identifier for the data by encoding and parsing the label; S2: using a convolutional neural network learning algorithm to generate a learning index label, and locate and associate the data; S3: recording the creation, survival and extinction status of each label, managing the entire production cycle of the label, and completing the memory and tracing of the entire business process by associating the identifier with the business data. The method can associate various types of heterogeneous data in the enterprise, and through the algorithm of the neural network and the full life cycle management of the label, realize the rapid definition and tracing of enterprise data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of big data application information technology, and in particular to a method and system for full life cycle neural network labeling based on ERI. Background Art

[0002] With the continuous growth and diversification of enterprise data, labeling has become a common processing method. Labeling is to classify and mark data so that enterprises can analyze and use data. At the same time, because digital marketing has become the main direction of all industries in the future, big data labeling has become an inevitable trend.

[0003] The commonly used labeling technology on the market is that the system generates random codes as labels. Generally, commonly used relational data has the function of automatically generating identification codes, such as auto-increment and seed to randomly generate unique identification. This type of technology is convenient and easy to use, but has poor readability and efficiency. Users define labeling rules by themselves. Users will define labels according to sources, business scenarios, data types, etc., but this type of labeling has certain limitations, such as the inability to achieve cross-table, cross-database, and cross-server positioning. It is less efficient in terms of retrieval algorithms and data tracing. Summary of the invention

[0004] The problem solved by the present invention is how to associate various types of heterogeneous data in an enterprise, and realize the rapid definition and tracing of enterprise data through the algorithm of neural network and the full life cycle management of tags.

[0005] To solve the above problems, the present invention provides a method for full life cycle neural network labeling based on ERI, comprising the steps of:

[0006] S1: A label consisting of multiple characters is established based on ERI. The label consists of an address code, an application code, a data code, a life code, and a check code. The unique identifier is defined for the data by encoding and parsing the label;

[0007] S2: Use convolutional neural network learning algorithm to generate learning index labels to locate and associate data;

[0008] S3: Records the creation, existence, and extinction status of each tag, manages the entire production cycle of the tag, and completes the memory and traceability of the entire business process by associating the identifier with business data.

[0009] In the above method, the encoding and decoding algorithm of the tag includes the definition of the address code, application code, data code, life code and check code. For the deep learning algorithm of the neural network, a learning tag will be generated, which can quickly identify, define and associate a large number of tags. The life code of each tag records the generation, validity and expiration of the tag and the time, and is associated with the business data. In this way, the recording and tracing of various types of data in the business process can be implemented.

[0010] Furthermore, the step S2 comprises:

[0011] S21: Based on the definition of the tag in step S1, the address code and life code are directly parsed using static data representation, and the application code and data code are parsed using a comparison table;

[0012] S22: Use the deep learning algorithm of neural network to generate label index identification according to the hit rate of data.

[0013] Furthermore, the step S22 includes:

[0014] S221: Periodically insert a pooling layer between consecutive convolutional layers;

[0015] S222: Use MAX to perform operations in the aggregation layer, and independently operate each depth slice of the input data volume to change the spatial size;

[0016] S223: Using the back propagation method, when forward propagating through the convergence layer, the index of the largest element in the pool is recorded to generate an index identifier.

[0017] Furthermore, the step S3 comprises:

[0018] S31: The identifier of the business data and the identifier of the index tag have a life code, and the validity period of the identifier is recorded;

[0019] S32: When locating and associating data, the life code is parsed to determine whether the identifier is valid, and different life cycles are defined for different identifiers;

[0020] S33: Associating the three states of generation, validity and expiration of each tag recorded in the life code of each tag with time and business data.

[0021] A full life cycle neural network labeling system based on ERI, including:

[0022] Establishment module: used to establish a label consisting of multiple characters based on ERI. The label consists of an address code, an application code, a data code, a life code and a check code;

[0023] Definition module: defines a unique identifier for data by encoding and parsing tags;

[0024] Generation module: used to generate learning index labels using neural network learning algorithms to locate and associate data;

[0025] Association module: used to record the creation, existence and extinction status of each label, manage the entire production cycle of the label, and complete the memory and traceability of the entire business process by associating the identification with the business data.

[0026] Furthermore, the generating module comprises:

[0027] Parsing unit: used to directly parse the address code and life code represented by static data, and parse the application code and data code by using a comparison table;

[0028] Generation unit: used to generate label index identifiers according to the hit rate of data through the deep learning algorithm of convolutional neural network.

[0029] Furthermore, the generating unit comprises:

[0030] Insertion subunit: used to periodically insert a pooling layer between consecutive convolutional layers;

[0031] Operation subunit: used to use MAX to operate in the aggregation layer, independently operate each depth slice of the input data volume, and change the spatial size;

[0032] Recording subunit: used to record the index of the largest element in the pool and generate an index identifier when forward propagating through the pooling layer by means of back propagation.

[0033] Furthermore, the association module includes:

[0034] Recording unit: used to record the validity period of the life code in the identifier of the business data and the identifier of the index tag;

[0035] Judgment unit: used to locate and associate data, parse the life code to determine whether the identification is valid, and define different life cycles for different identifications;

[0036] Association unit: used to associate the three states of label generation, validity and expiration recorded in the life code of each label with time and business data.

[0037] The present invention adopts the above technical solution to achieve the following beneficial effects:

[0038] The present invention can build a data identification system for enterprises through the identification method of full-life cycle neural network tags, open up the association of various types of heterogeneous data for enterprises through a unified and comprehensive ERI identification process, and realize the memory and tracing of various types of data in the business processes of enterprises through the algorithm of convolutional neural network and the full life cycle management of tags. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The method flow of the full life cycle neural network label based on ERI provided in the first embodiment of the present invention Figure 1 ;

[0040] Figure 2 The method flow of the full life cycle neural network label based on ERI provided in the first embodiment of the present invention Figure 2 ;

[0041] Figure 3 The system structure of the ERI-based full life cycle neural network tag provided in the second embodiment of the present invention Figure 1 ;

[0042] Figure 4 The system structure of the ERI-based full life cycle neural network tag provided in the second embodiment of the present invention Figure 2 . DETAILED DESCRIPTION

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0044] The following are specific embodiments of the present invention and the accompanying drawings to further describe the technical solution of the present invention, but the present invention is not limited to these embodiments.

[0045] Embodiment 1

[0046] This embodiment provides a method for full life cycle neural network labeling based on ERI, such as Figure 1 and Figure 2 As shown, the method comprises the steps of:

[0047] S1: A label consisting of multiple characters is established based on ERI. The label consists of an address code, an application code, a data code, a life code, and a check code. The unique identifier is defined for the data by encoding and parsing the label;

[0048] S2: Use convolutional neural network learning algorithm to generate learning index labels to locate and associate data;

[0049] S3: Records the creation, existence, and extinction status of each tag, manages the entire production cycle of the tag, and completes the memory and traceability of the entire business process by associating the identifier with business data.

[0050] Specifically, a 36-bit label ERI (Enterprise Resources Identifier) ​​is established. The label consists of five parts: address code, application code, data code, life code, and check code. It is a non-random code, and the meaning of each code can be quickly understood through encoding and parsing technology.

[0051] The composition of Enterprise Resources Identifier (ERI):

[0052] Logo example: composed of uppercase and lowercase letters and numbers 0016EEAE3C40 T2fH 16a9Pz06NOwtUG9D3W5g.

[0053] The tag defines a unique identifier for data within the system and data outside the system. ERI consists of 36 characters and is divided into 5 sections: Site+APP Category+DataBase+Lifecycle+Verification.

[0054] Part 1: 1-12 bits (12 bits): Site address code: The IPV6 address is used as the identification part. For example, 0016EAAE3C40 indicates the hardware source of the data, such as the service address, machine equipment or intelligent device. It can be the database data on the server or the data collected from the machine equipment.

[0055] Part 2: 13-16 digits (4 digits): APP Category application code, indicating various applications based on the site, such as ERP applications, MES applications, SCM applications, etc. For example, T2fH indicates ERP applications. A total of 16777216 applications can be supported.

[0056] Part 3: 17-26 bits (10 digits) DataBase data code, such as: R6a9Pz06NO, composed of library number + table number + record number.

[0057] Among them: library number is 2 digits, table number is 2 digits, and record number is 6 digits.

[0058] Database is 2 digits, for example, the 0001 account set database is represented by R6. A single device can support 4096 databases.

[0059] Table number is 2 digits: For example, the PRODUCT table is represented by a9. A single database supports 4096 tables.

[0060] Record number 6 digits: record number in the positioning table, such as Pz06NO. A single table supports 6.8 billion records.

[0061] Part 4: 27-32 bits (8 bits), Lifecycle code: consists of two segments: creation time and end time.

[0062] For example, in wtUG9D, the first 5 digits represent the creation time, and the last 5 digits represent the end time, accurate to 0.1 milliseconds.

[0063] Part 5: 33-36 bits (4 bits), Verification, uses the Md5 algorithm to irreversibly encrypt the first 32 characters to verify whether the content is lost or tampered with.

[0064] See also Figure 2 , wherein step S2 comprises:

[0065] S21: Based on the definition of the tag in step S1, the address code and life code are directly parsed using static data representation, and the application code and data code are parsed using a comparison table;

[0066] S22: Use the deep learning algorithm of neural network to generate label index identification according to the hit rate of data.

[0067] Specifically, based on the definition rules of data tags, any physical location can be defined. Since it is composed of pure letters, some information is represented by pure static data, such as address code and life code, which can be directly parsed. For application codes and data codes, a comparison table is needed for parsing. In order to speed up the parsing of tags, a neural network algorithm learning algorithm is used to generate a tag index identifier based on the hit rate of the data. The index identifier uses the same structure as the data identifier.

[0068] Index identification generation technology, the index identification adopts the ERI unified identification format, which consists of five parts: address code, application code, data code, life code, and check code. This type of identification is generated according to the convolutional neural network algorithm to improve the location and association speed of data and form an index database, as shown in Table 1 below:

[0069] Index ID Source Identification Target identification 91F00000OA913H1291E0000EQX2X52QO 91E0000EEEFSU5U191E0000EEEFSU581 91E0000EEEFSV5UZ91E0000EEEFSU581

[0070] Table 1

[0071] The source identifier is the purchase order S001 for example. The details of the purchase of 100 items of material A are as shown in Table 2:

[0072] Logo Purchase order number Material Code quantity 91E0000EEEFSU5U191E0000EEEFSU581 S001 A 100

[0073] Table 2

[0074] The corresponding incoming order details are incoming order number R001, and 50 details of material A are received, as shown in Table 3:

[0075] Logo Purchase order number Material Code quantity 91E0000EEEFSV5UZ91E0000EEEFSU581 R001 A 50

[0076] Table 3

[0077] According to the index in Table 1, the association between the two documents can be quickly established. The two pieces of data may be generated by different servers and different applications.

[0078] Wherein, step S22 comprises:

[0079] S221: Periodically insert a pooling layer between consecutive convolutional layers;

[0080] S222: Use MAX to perform operations in the aggregation layer, and independently operate each depth slice of the input data volume to change the spatial size;

[0081] S223: Using the back propagation method, when forward propagating through the convergence layer, the index of the largest element in the pool is recorded to generate an index identifier.

[0082] Specifically, for a large number of index tags generated, in order to speed up the positioning and association speed, a convolutional neural network algorithm is used, and the life code of the index standard is updated to reduce the number of valid indexes and speed up the speed. The specific methods are as follows:

[0083] Usually a pooling layer is periodically inserted between consecutive convolutional layers. Its function is to gradually reduce the spatial size of the data body, which can reduce the number of parameters in the network, reduce the consumption of computing resources, and effectively control overfitting. The pooling layer uses the MAX operation to operate independently on each depth slice of the input data body to change its spatial size. The most common form is that the pooling layer uses a filter of size 2x2 to downsample each depth slice with a step size of 2, discarding 75% of the activation information. Each MAX operation takes the maximum value from 4 numbers, that is, in a 2x2 area in the depth slice, the depth remains unchanged.

[0084] Some formulas for the pooling layer: Input data volume size, there are two hyperparameters: spatial size FF and step size SS; Output data volume size, where:

[0085] W2 = (W1-F) / S+1;

[0086] H2=(H1-F) / S+1;

[0087] D2=D1.

[0088] The main differences from the previous convolution size calculation are two: first, basically no additional supplementation is performed during the pooling process; second, the depth before and after pooling remains unchanged.

[0089] In practice, there are usually only two forms of max pooling layers: one is F = 3, S = 2F = 3, S = 2, also called overlapping pooling, and the other more commonly used is F = 2, S = 2F = 2, S = 2. Pooling for a larger receptive field requires a larger pooling size and is often destructive to the network.

[0090] Specifically, the back propagation method is adopted: the back propagation of the max(x,y) function can be simply understood as passing the gradient back only along the largest number. Therefore, when forward propagating through the pooling layer, the index of the largest element in the pool is usually recorded, an index identifier is generated, and a certain life cycle is given according to the set effective time, so that the gradient routing is very efficient during the back propagation.

[0091] Wherein, step S3 comprises:

[0092] S31: The identifier of the business data and the identifier of the index tag have a life code, and the validity period of the identifier is recorded;

[0093] S32: When locating and associating data, the life code is parsed to determine whether the identifier is valid, and different life cycles are defined for different identifiers;

[0094] S33: Associating the three states of generation, validity and expiration of each tag recorded in the life code of each tag with time and business data.

[0095] Specifically, the business identifier and the index identifier have a life code, which records the validity period of the identifier. The life code is described as follows:

[0096] Reference to the fourth part: 27-36 bits (12 bits), Lifecycle code: consists of two segments: creation time + end time. For example, wtUG9D3W5g, the first 5 bits represent the creation time, and the last 5 bits are the end time, accurate to 0.1 milliseconds.

[0097] When locating and associating a tag, the life code will be parsed first to determine whether the identification is valid. Different life cycles can be defined for different identification codes.

[0098] Specifically, for business identification, the life cycle will be defined according to various intelligences of different businesses. For example, for sales orders, we can give the validity period of the sales order identification through the average time from the generation to the closing of the sales order. For example, the life cycle of the sales order platform is defined as 100 days, and the average period of time from the generation of the shipping order to the generation of the account closing is 30 days, then the shipping order identification can be defined as 30 days.

[0099] Specifically, index identifiers are randomly generated based on the frequency of index hits. For example, if the system detects that the hit rate of a sales order relationship associated with shipments is greater than 3 times per day, the validity period can be defined as 3 days. If the hit rate is less than 1 time per 3 days, the validity period is defined as 1 day. The system can analyze the hit rate report through reports to give the most reasonable life cycle recommendations, continuously improve the system's operating efficiency, and achieve the effect of longer running time and higher efficiency.

[0100] This method builds a data identification system for enterprises through the identification method of full-life cycle neural network tags, connects various types of heterogeneous data for enterprises through a unified and comprehensive ERI identification process, and realizes the memory and traceability of various types of data in the enterprise business process through the algorithm of convolutional neural network and the full life cycle management of tags.

[0101] Embodiment 2

[0102] This embodiment provides a system for full life cycle neural network labeling based on ERI, such as Figure 3 and Figure 4 As shown, the system includes:

[0103] Establishment module: used to establish a label consisting of multiple characters based on ERI. The label consists of an address code, an application code, a data code, a life code and a check code;

[0104] Definition module: defines a unique identifier for data by encoding and parsing tags;

[0105] Generation module: used to generate learning index labels using neural network learning algorithms to locate and associate data;

[0106] Association module: used to record the creation, existence and extinction status of each label, manage the entire production cycle of the label, and complete the memory and traceability of the entire business process by associating the identification with the business data.

[0107] See also Figure 4 , wherein the generation module includes:

[0108] Parsing unit: used to directly parse the address code and life code represented by static data, and parse the application code and data code by using a comparison table;

[0109] Generation unit: used to generate label index identifiers according to the hit rate of data through the deep learning algorithm of convolutional neural network.

[0110] The generation unit includes:

[0111] Insertion subunit: used to periodically insert a pooling layer between consecutive convolutional layers;

[0112] Operation subunit: used to use MAX to operate in the aggregation layer, independently operate each depth slice of the input data volume, and change the spatial size;

[0113] Recording subunit: used to record the index of the largest element in the pool and generate an index identifier when forward propagating through the pooling layer by means of back propagation.

[0114] Among them, the associated modules include:

[0115] Recording unit: used to record the validity period of the life code in the identifier of the business data and the identifier of the index tag;

[0116] Judgment unit: used to locate and associate data, parse the life code to determine whether the identification is valid, and define different life cycles for different identifications;

[0117] Association unit: used to associate the three states of label generation, validity and expiration recorded in the life code of each label with time and business data.

[0118] The system's encoding and decoding algorithms for tags include the definition of address codes, application codes, data codes, life codes, and check codes. For the deep learning algorithm of the neural network, learning tags will be generated, which can quickly identify, define, and associate a large number of tags. The life code of each tag records the generation, validity, and expiration of the tag and the time, and associates it with the business data. In this way, the recording and tracing of various types of data in the business process can be implemented.

[0119] Although the disclosure is disclosed as above, the protection scope of the disclosure is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the disclosure, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A method for identifying neural network labels based on ERI throughout the life cycle. It is characterized in that Includes steps: S1: A label consisting of multiple characters is established based on ERI. The label consists of an address code, an application code, a data code, a life code, and a check code. The unique identifier is defined for the data by encoding and parsing the label; S2: Use convolutional neural network learning algorithm to generate learning index labels to locate and associate data; S3: Records the creation, existence, and extinction status of each tag, manages the entire life cycle of the tag, and completes the memory and traceability of the entire business process by associating the identifier with business data; The step S2 comprises: S21: Based on the definition of the tag in step S1, the address code and life code are directly parsed using static data representation, and the application code and data code are parsed using a comparison table; S22: Using a neural network deep learning algorithm, a label index identifier is generated based on the hit rate of the data; The step S22 comprises: S221: Periodically insert a pooling layer between consecutive convolutional layers; S222: Use MAX to perform operations in the aggregation layer, and independently operate each depth slice of the input data volume to change the spatial size; S223: Using the back propagation method, when forward propagating through the convergence layer, the index of the largest element in the pool is recorded to generate an index identifier.

2. The identification method of the ERI-based full life cycle neural network tag according to claim 1, It is characterized in that The step S3 comprises: S31: The identifier of the business data and the identifier of the index tag have a life code, and the validity period of the identifier is recorded; S32: When locating and associating data, the life code is parsed to determine whether the identifier is valid, and different life cycles are defined for different identifiers; S33: Associating the three states of generation, validity and expiration of each tag recorded in the life code of each tag with time and business data.

3. A full life cycle neural network tagging system based on ERI, It is characterized in that include: Establishment module: used to establish a label consisting of multiple characters based on ERI. The label consists of an address code, an application code, a data code, a life code and a check code; Definition module: defines a unique identifier for data by encoding and parsing tags; Generation module: used to generate learning index labels using neural network learning algorithms to locate and associate data; Association module: used to record the creation, existence and extinction status of each tag, manage the entire life cycle of the tag, and complete the memory and traceability of the entire business process by associating the identifier with the business data; The generation module comprises: Parsing unit: used to directly parse the address code and life code represented by static data, and parse the application code and data code by using a comparison table; Generation unit: used to generate label index identification according to the hit rate of data through the deep learning algorithm of convolutional neural network; The generating unit comprises: Insertion subunit: used to periodically insert a pooling layer between consecutive convolutional layers; Operation subunit: used to use MAX to operate in the aggregation layer, independently operate each depth slice of the input data volume, and change the spatial size; Recording subunit: used to record the index of the largest element in the pool and generate an index identifier when forward propagating through the pooling layer by means of back propagation.

4. The identification system of the ERI-based full life cycle neural network tag according to claim 3, It is characterized in that The association module includes: Recording unit: used to record the validity period of the life code in the identifier of the business data and the identifier of the index tag; Judgment unit: used to locate and associate data, parse the life code to determine whether the identification is valid, and define different life cycles for different identifications; Association unit: used to associate the three states of label generation, validity and expiration recorded in the life code of each label with time and business data.

Citation Information

Patent Citations

  • Personnel information tagging system of military industrial group based on clustering

    CN109472490A

  • Data processing method and device

    CN112100645A

  • Big data decision analysis system and method based on industrial Internet identifier analysis

    CN112165391A

  • Data full life cycle marking method and system

    CN116012027A

  • Product full life cycle management system and method based on industrial Internet of Things technology

    CN116151845A