Data processing method and system based on neural network model

Abstract

The invention discloses a data processing method and system based on a neural network model. The method comprises the following steps: establishing a standardized salary information table; receiving a salary adjustment request, extracting feature data in the salary adjustment request based on a neural network model, and adding the feature data to the salary information table; based on a preset salary calculation rule, splitting the salary calculation task into a plurality of sub-calculation tasks; determining the complexity of the sub-calculation tasks, and distributing the plurality of sub-calculation tasks to a plurality of calculation nodes according to the complexity; and executing the calculation program by the plurality of calculation nodes to obtain a salary calculation result. According to the method, an information table is established to store salary data, feature data of different types of newly added data are extracted based on a neural network and added into the information table, meanwhile, the complexity of different task flows and the difference of calling databases in salary calculation are considered, task allocation is conducted on calculation nodes with different performances through sub-calculation task division, and the calculation efficiency is improved. And the current residual resources are effectively and reasonably applied to the maximum extent.

Description

Technical Field

[0001] The present invention relates to the field of computer data processing, and in particular to a data processing method and system based on a neural network model. Background Art

[0002] With the development of the intelligent data era, salary management is transforming and upgrading to digitalization. At present, the human resources management system is established by integrating employees' daily performance, work quality and performance results through scientific data and information management, which can effectively improve the level of human resources information management and strengthen business management construction.

[0003] However, in human resources management work, there is usually a considerable amount of data information, such as personnel rosters, attendance assessments, resumes, performance information, etc. Different information streams are repeatedly called during the salary calculation process. They have different demands for service resources, and the types, quantities and duration of service resources required are different. The reading and processing of data is becoming more and more complicated.

[0004] In order to achieve efficient and coordinated human resource management, it is necessary to integrate existing computing resources to establish a salary calculation system for comprehensive processing of multiple types of data, so as to uniformly collect and analyze data from different business systems and improve the efficiency of the salary management process. Summary of the invention

[0005] In response to the problems existing in the above-mentioned prior art, the present invention provides a data processing method and system based on a neural network model, the purpose of which is to pre-process different types of data through a neural network, integrate complex salary-related information, and adaptively distribute it to multiple computing nodes for parallel processing, thereby improving the data analysis efficiency of the salary calculation process and reducing labor costs and operational difficulty.

[0006] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, the present invention provides a data processing method based on a neural network model, comprising: Obtain employee salary-related information and establish a standardized salary information table; receiving an uploaded salary adjustment request, extracting feature data in the salary adjustment request based on a neural network model, and adding the feature data to the salary information table; Based on preset salary calculation rules, the salary calculation task is split into multiple sub-calculation tasks that can be executed in parallel; the complexity of the sub-calculation tasks is determined, and the multiple sub-calculation tasks are allocated to multiple computing nodes in the salary calculation system according to the complexity; The plurality of computing nodes respectively call corresponding data in the salary information table, and execute a computing program to obtain a salary calculation result.

[0007] Preferably, the extracting characteristic data in the salary adjustment request based on a neural network model includes: When the salary adjustment request is text data, extracting semantic features in the text data based on a semantic analysis model; matching the semantic features with attributes in the salary information table, and after matching, adding feature data corresponding to the semantic features in the salary adjustment request to the matching attributes in the salary information table; When the salary adjustment request is image data, multiple text data blocks in the image data are obtained based on a text detection model and a text recognition model, and a subject word is obtained based on a text data block with the greatest importance among the multiple text data blocks; and feature data of the subject word in the salary adjustment request is added to an attribute matching the subject word in the salary information table.

[0008] Preferably, the semantic analysis model adopts the TextRank4ZH model, the text detection model adopts the DBNet network model, and the text recognition model adopts the SVTR network model.

[0009] Preferably, the acquiring of a plurality of text data blocks in the image data based on a text detection model and a text recognition model comprises: Performing horizontal correction on the image data to obtain a corrected image; Obtain multiple text lines in the image data based on the text detection model; Recognize text information in multiple text lines based on a text recognition model; After combining the upper and lower adjacent text lines, the text information is segmented to obtain a phrase sequence; The text lines are reassembled into a plurality of text data blocks according to the pixel similarity of the image backgrounds of the plurality of text lines and the continuity of the phrase sequence.

[0010] Preferably, the acquiring of the subject words based on the text data block with the greatest importance among the plurality of text data blocks comprises: Extracting multiple keywords from multiple text data blocks based on a semantic analysis model; Calculating the importance of the plurality of text data blocks based on the positions of the plurality of text data blocks in the image data and the word frequencies of the keywords of the text data blocks; The keywords of the most important text data blocks among the plurality of text data blocks are used as subject words.

[0011] Preferably, determining the complexity of the sub-computing task includes: Based on the preset salary calculation rule, determine the number of code lines and execution time step required to execute each of the sub-computing tasks, and obtain the first complexity of the sub-computing tasks; Obtain a data group corresponding to the data called by each of the sub-computing tasks, and for each data group, calculate the similarity between the data in the data group to obtain the second complexity of the sub-computing task; The first complexity and the second complexity constitute a multi-dimensional vector to obtain a complexity vector of the sub-computing task.

[0012] Preferably, allocating the plurality of sub-computing tasks to a plurality of computing nodes in the salary calculation system according to the complexity comprises: (1) Performing a first clustering of the plurality of sub-computing tasks according to the complexity vector to obtain a plurality of task levels; wherein the task levels are arranged in descending order according to the eigenvalues ​​of the complexity vector: the first task level, the second task level, ..., the nth task level, where n is the total number of task levels, to form a task sequence with the complexity of the sub-tasks arranged from large to small; recorded as the first sequence.

[0013] (2) Based on the currently available computing nodes, the currently available computing nodes are sorted according to the load occupancy rate of the computing nodes, which is recorded as the second sequence. The computing node with a smaller load occupancy rate is ranked higher in the second sequence. (3) For each sub-computing task in the first sequence, assign it to the computing nodes in the second sequence in the order in the first sequence. Let the mth sub-computing task in the first sequence be assigned to the nth computing node in the second sequence, then the m+1th sub-computing task in the first sequence be assigned to the n+1th computing node in the second sequence, and so on. (4) When all sub-computing tasks in the first sequence are assigned, a new first sequence is generated, where the new first sequence is initially an empty sequence; (5) Obtaining the load occupancy rate of the computing node after the sub-computing task is assigned; (6) If the load occupancy rate of the computing node is greater than a preset threshold, remove the sub-computing task of the smallest task level in the computing node, and add the removed sub-computing task to the new first sequence; return to step (5); (7) If the load occupancy rate of the computing node is not greater than the preset threshold, all sub-computing tasks in the new first sequence are sorted to form a task sequence with sub-task complexity from large to small, and then return to step (2).

[0014] In a second aspect, the present invention provides a data processing system based on a neural network model, comprising: The information table establishment module is used to obtain employee salary-related information and establish a standardized salary information table; A data updating module, configured to receive an uploaded salary adjustment request, extract characteristic data in the salary adjustment request based on a neural network model, and add the characteristic data to the salary information table; A task allocation module is used to split the salary calculation task into multiple sub-calculation tasks that can be executed in parallel based on preset salary calculation rules; determine the complexity of the sub-calculation tasks, and allocate the multiple sub-calculation tasks to multiple computing nodes in the salary calculation system according to the complexity; The salary calculation module is used to use the plurality of computing nodes to respectively call the corresponding data in the salary information table, and execute the calculation program to obtain the salary calculation result.

[0015] In a third aspect, the present invention provides an electronic device, comprising: A memory for storing executable instructions; The processor is used to implement a data processing method based on a neural network model as described above when running the executable instructions stored in the memory.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, implement a data processing method based on a neural network model as described above.

[0017] The data processing method and system based on the neural network model of the present invention have the following beneficial effects: The present invention establishes an information table to store salary data, extracts features from text data and image data based on a neural network, identifies feature data therein, and adds feature data of different types of newly added data to the information table, thereby achieving standardized storage and logical organization of complex data. At the same time, the complexity of different task flows and differences in calling databases in salary calculation are taken into consideration, and tasks are assigned to computing nodes of different performances through sub-computing task division, so that limited computing resources are quickly allocated to the maximum extent, greatly reducing the computing power requirements for salary management. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of a data processing method based on a neural network model of the present invention; Figure 2 It is a structural block diagram of a data processing system based on a neural network model of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] To facilitate understanding of this embodiment, a data processing method based on a neural network model disclosed in an embodiment of the present invention is first introduced in detail.

[0021] like Figure 1 As shown, the method includes: Obtain employee salary-related information and establish a standardized salary information table; receiving an uploaded salary adjustment request, extracting feature data in the salary adjustment request based on a neural network model, and adding the feature data to the salary information table; Based on the preset salary calculation rules, the salary calculation task is split into multiple sub-calculation tasks that can be executed in parallel; for example, salary can include basic salary, performance salary, seniority allowance, other subsidies, welfare insurance and other components. According to the company's preset salary calculation rules, the salary calculation can be split into multiple sub-calculation tasks and executed simultaneously to improve the efficiency of salary calculation.

[0022] Determining the complexity of the sub-computation tasks, and assigning a plurality of the sub-computation tasks to a plurality of computing nodes in the salary calculation system according to the complexity; The plurality of computing nodes respectively call corresponding data in the salary information table, and execute a computing program to obtain a salary calculation result.

[0023] This embodiment first establishes a standardized data table to effectively store a large amount of salary-related information. The salary information table includes attribute values ​​and data values ​​corresponding to the attribute values. When new materials are received and the data in the salary information table needs to be modified, different data types are processed differently through the neural network model, and the characteristic data is extracted and then filled into the attribute column corresponding to the salary information table, so as to achieve timely data update and ensure the accuracy and effectiveness of new data entry. Considering that there are different task flows in salary calculation, they have different requirements for service resources, and there are differences in the types, quantities and occupancy time of the functions called, this embodiment allocates tasks to computing nodes with different performances according to the complexity of the sub-computing tasks, and effectively maximizes the application of the current remaining resources.

[0024] The salary adjustment request can be additional supporting materials such as award certificates uploaded by the employee, or it can be subtractive supporting materials such as penalty notices.

[0025] Preferably, the extracting characteristic data from the salary adjustment request based on a neural network model includes: When the salary adjustment request is text data, semantic features in the text data are extracted based on a semantic analysis model; the semantic features are the subject and keywords of the text.

[0026] Matching the semantic feature with the attribute in the salary information table, and after the matching is completed, adding the feature data corresponding to the semantic feature in the salary adjustment request to the matching attribute in the salary information table; When the salary adjustment request is image data, multiple text data blocks in the image data are obtained based on a text detection model and a text recognition model, and a subject word is obtained based on a text data block with the greatest importance among the multiple text data blocks; and feature data of the subject word in the salary adjustment request is added to an attribute matching the subject word in the salary information table.

[0027] Preferably, the semantic analysis model adopts the TextRank4ZH model, the text detection model adopts the DBNet network model, and the text recognition model adopts the SVTR network model.

[0028] TextRank4ZH is a Python implementation of the TextRank algorithm for Chinese text. It can calculate the importance of each word based on the co-occurrence relationship between words and obtain important sentence and keyword features of text data.

[0029] The DBNet network is a segmentation-based text detection network that contains an FPN structure that can handle text detection of different scales and can handle text with complex backgrounds. The uploaded salary adjustment request can be an award certificate, a leave slip, a reimbursement list, etc. For example, in an award certificate, the text box size of the title and the text box size of the body are quite different, so segmentation of different scales is required, and the text will be surrounded by an image background; leave slips and reimbursement lists may contain different degrees of graffiti pollution, which requires high-precision boundary recognition capabilities. The DBNet network can well meet the detection needs of text boxes of different scales.

[0030] The SVTR network model introduces local and global hybrid blocks to extract stroke features and inter-character correlations respectively, and combines a multi-scale backbone to form a multi-granular feature description. At the same time, the downsampling operation makes the computational complexity lower than that of recognition models such as CRNN, which can better meet the requirements of improving computational efficiency in the present invention.

[0031] Since the image data in the uploaded salary adjustment request may include both text and non-text elements, for example, an award certificate may include non-text elements such as special logos and seals and text elements, and text elements may be embedded in the text elements, which may easily cause interference in text analysis. The Global Mixing structure in the SVTR network model can establish long-term dependencies between components from different characters, accurately extract the boundaries between non-text and text, and weaken the impact of non-text components to increase the importance of text components.

[0032] It should be noted that the above three neural network models are all existing technologies, and their specific network structures are not described in detail in the embodiments of the present invention.

[0033] Preferably, the acquiring of a plurality of text data blocks in the image data based on a text detection model and a text recognition model comprises: The image data is horizontally corrected to obtain a corrected image; since the uploader cannot ensure that the shooting angle is perpendicular to the material, the text in it has a certain tilt and distortion, so correction is required before dividing the text line. In some embodiments, the image data can be corrected by using a perspective transformation method, which is not limited here.

[0034] Obtain multiple text lines in the image data based on the text detection model; Recognize text information in multiple text lines based on a text recognition model; After combining the upper and lower adjacent text lines, the text information is segmented to obtain a phrase sequence; According to the pixel similarity of the image background of the multiple text lines and the continuity of the phrase sequence, the text lines are reassembled to form multiple text data blocks. That is, after the text detection model detects the text in multiple regions, it also includes the merging of large connected regions. It can be understood that the greater the pixel similarity of the image background of two text lines, the greater the probability that the two text lines are merged into the same text data block, and the greater the continuity of the position distribution of the phrase sequences of the two text lines, the greater the probability that the two text lines are merged into the same text data block. Considering that the position distribution of the phrase sequences of the two text lines is very different or the position distance is very far, the greater the probability that the two text lines are merged into the same text data block. In the embodiment of the present application, the pixel similarity of the image background of multiple text lines and the continuity of the phrase sequence are fused to analyze whether a text data block is formed.

[0035] In one implementation, three text data blocks are obtained by dividing according to the above method, namely, a title data block, a body data block, and a signature data block.

[0036] Preferably, the acquiring of the subject words based on the text data block with the greatest importance among the plurality of text data blocks comprises: Extracting multiple keywords from multiple text data blocks based on a semantic analysis model; The importance of the multiple text data blocks is calculated based on the positions of the multiple text data blocks in the image data and the word frequencies of the keywords of the text data blocks; specifically, the importance of the multiple text data blocks is calculated based on the size of the characters in the text blocks, the distance between the current text block and other text blocks, the position of the current text block in the layout of all text blocks, and the word frequency analysis of the keywords of the text data blocks.

[0037] The keywords in the title data block have stronger representation ability and can be used as a summary of the entire image data. The word frequency refers to the proportion of the number of occurrences of the keyword in the text data block.

[0038] The keywords of the most important text data blocks among the plurality of text data blocks are used as subject words.

[0039] Preferably, determining the complexity of the sub-computing task includes: Based on the preset salary calculation rule, determine the number of code lines and execution time step required to execute each of the sub-computing tasks, and obtain the first complexity of the sub-computing tasks; The first complexity obtained in this step represents the execution complexity of the sub-computing task.

[0040] Obtain a data group corresponding to the data called by each of the sub-computing tasks, and for each data group, calculate the similarity between the data in the data group to obtain the second complexity of the sub-computing task; The first complexity obtained in this step represents the data complexity of the sub-computing task.

[0041] The present invention uses the similarity between training sample data as the criterion for judging data complexity. The higher the similarity, the lower the data complexity. For example, the data volume of attendance rate is days, and the data features of attendance days of different employees are relatively concentrated, so the data complexity is relatively low.

[0042] The first complexity and the second complexity constitute a multi-dimensional vector to obtain a complexity vector of the sub-computing task.

[0043] Preferably, allocating the plurality of sub-computing tasks to a plurality of computing nodes in the salary calculation system according to the complexity comprises: (1) Performing a first clustering of the plurality of sub-computing tasks according to the complexity vector to obtain a plurality of task levels; wherein the task levels are arranged in descending order according to the eigenvalues ​​of the complexity vector: the first task level, the second task level, ..., the nth task level, where n is the total number of task levels, to form a task sequence with the complexity of the sub-tasks arranged from large to small; recorded as the first sequence.

[0044] (2) Based on the currently available computing nodes, the currently available computing nodes are sorted according to the load occupancy rate of the computing nodes, which is recorded as the second sequence. The computing node with a smaller load occupancy rate is ranked higher in the second sequence. (3) For each sub-computing task in the first sequence, assign it to the computing nodes in the second sequence in the order in the first sequence. Let the mth sub-computing task in the first sequence be assigned to the nth computing node in the second sequence, then the m+1th sub-computing task in the first sequence be assigned to the n+1th computing node in the second sequence, and so on. (4) When all sub-computing tasks in the first sequence are assigned, a new first sequence is generated, where the new first sequence is initially an empty sequence; (5) Obtaining the load occupancy rate of the computing node after the sub-computing task is assigned; (6) If the load occupancy rate of the computing node is greater than a preset threshold, remove the sub-computing task of the smallest task level in the computing node, and add the removed sub-computing task to the new first sequence; return to step (5); (7) If the load occupancy rate of the computing node is not greater than the preset threshold, all sub-computing tasks in the new first sequence are sorted to form a task sequence with the complexity of the sub-tasks decreasing from large to small, and then return to step (2); In the embodiment of the present application, between the above steps (1) and (2), step (11) is also included: (11) obtaining a call database of each of the sub-computing tasks, performing a second clustering on the plurality of sub-computing tasks according to the category and quantity of the call database of each of the sub-computing tasks to obtain a plurality of call categories, and marking the call category of each sub-task in the first sequence; The called database refers to a database in the human resources system, such as an attendance management database, a sales management database, a procurement management database, etc. For example, when calculating salary, working days and meal allowances are related to the attendance management database, and attendance-related data need to be called.

[0045] Categorize multiple sub-computing tasks into multiple task groups with similar database call requirements, reduce changes in resource allocation, avoid different computing nodes repeatedly reading the same data from the same database, reduce database call congestion in the salary calculation system, and improve the stability of the calculation process.

[0046] Furthermore, when reallocating tasks in step (8), if the number of sub-computing tasks to be allocated, i.e., the number of sub-computing tasks in the new first sequence, is less than the preset reallocation threshold, then for the sub-computing tasks in the new first sequence, an adapted computing node is selected from all computing nodes according to the calling category of the sub-computing tasks in the new first sequence. Among them, the adaptation is determined based on the comprehensive situation of the calling category of the sub-computing task, the calling category of the task on the computing node, and the load occupancy rate of the computing node. The preset reallocation threshold is determined based on the number of available computing nodes. It can be understood that in an embodiment of the present application, after multiple sub-computing tasks are allocated, for multiple sub-computing tasks on the same computing node, the execution order of the sub-computing tasks on the current computing node is set according to the calling category, and the execution order of the sub-computing tasks of the same calling category is continuous.

[0047] It should be understood that in the same computing node, calling the same data processing can effectively reduce the complexity of calling the function and the execution time of the steps.

[0048] Since there is no dependency between the sub-computing tasks of the present invention, all unallocated sub-computing tasks can be reallocated to the computing node when there are still resources remaining in the current computing node.

[0049] If the load occupancy rate of the computing node is not greater than a preset threshold and the first sub-computing task of the minimum task level of the computing node has no unselected sub-computing tasks of the same calling category, the sub-computing task allocation to the computing node is terminated.

[0050] The present invention preferentially removes sub-computing tasks at the minimum task level, gradually reduces the computational burden of computing nodes, and reasonably utilizes spare resources as much as possible. At the same time, when the number of sub-computing tasks to be allocated, i.e., the number of sub-computing tasks in the new first sequence, is less than the preset reallocation threshold, an adapted computing node is selected from all computing nodes according to the calling category of the sub-computing task. It can be understood that the design of the same computing node calling the same or similar database within a period of time only requires obtaining the most recent calculation result corresponding to the content of the target calculation dimension obtained by the data type calculation corresponding to the identifier of the target calculation mode and data type, which can reduce the calls of different complex functions. The salary calculation process is effectively optimized, the computing power of the computing nodes is maximized, and an efficient level of human resource information management is achieved.

[0051] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0052] Based on the same inventive concept, the embodiment of the present application also provides a data processing system based on a neural network model for implementing the above-mentioned. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the embodiment of the data processing system based on the neural network model provided below can refer to the limitations of the method above, and will not be repeated here.

[0053] like Figure 2 As shown, the present invention also provides a data processing system based on a neural network model, comprising: The information table establishment module is used to obtain employee salary-related information and establish a standardized salary information table; A data updating module, configured to receive an uploaded salary adjustment request, extract characteristic data in the salary adjustment request based on a neural network model, and add the characteristic data to the salary information table; A task allocation module is used to split the salary calculation task into multiple sub-calculation tasks that can be executed in parallel based on preset salary calculation rules; determine the complexity of the sub-calculation tasks, and allocate the multiple sub-calculation tasks to multiple computing nodes in the salary calculation system according to the complexity; The salary calculation module is used to use the plurality of computing nodes to respectively call the corresponding data in the salary information table, and execute the calculation program to obtain the salary calculation result.

[0054] It should be noted that: the data processing system based on the neural network model provided in this embodiment only uses the division of the above-mentioned functional modules as an example to illustrate when processing the salary calculation task. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules. Each functional module can be composed of a single execution unit, or two or more execution units can be integrated into one functional module to realize all the functions of the functional module.

[0055] Those skilled in the art will appreciate that the above modules may be implemented in whole or in part by software, hardware, or a combination thereof. The above modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor may call and execute operations corresponding to the above modules.

[0056] An embodiment of the present invention further provides an electronic device, including: A memory for storing executable instructions; The processor is used to implement a data processing method based on a neural network model as described above when running the executable instructions stored in the memory. The processor can be a central processing unit, other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0057] The specific details of the salary calculation performed by the above-mentioned computer device can be understood by referring to the relevant descriptions and effects corresponding to the previous methods, and will not be repeated here.

[0058] Another embodiment of the present invention further provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, implement a data processing method based on a neural network model as described above.

[0059] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk, or any other computer-readable storage medium that can store program codes.

[0060] The present invention is not limited to the above-mentioned specific implementation modes. Various changes made by ordinary technicians in this field based on the above-mentioned concepts without creative work are all within the protection scope of the present invention.

Claims

1. A data processing method based on a neural network model, characterized in that: The steps include: Obtain employee salary-related information and establish a standardized salary information table; receiving an uploaded salary adjustment request, extracting feature data in the salary adjustment request based on a neural network model, and adding the feature data to the salary information table; Based on preset salary calculation rules, the salary calculation task is split into multiple sub-calculation tasks that can be executed in parallel; the complexity of the sub-calculation tasks is determined, and the multiple sub-calculation tasks are allocated to multiple computing nodes in the salary calculation system according to the complexity; The plurality of computing nodes respectively call corresponding data in the salary information table, and execute a computing program to obtain a salary calculation result.

2. The data processing method based on the neural network model according to claim 1 is characterized in that: The extracting characteristic data from the salary adjustment request based on the neural network model includes: When the salary adjustment request is text data, extracting semantic features in the text data based on a semantic analysis model; matching the semantic features with attributes in the salary information table, and after matching, adding feature data corresponding to the semantic features in the salary adjustment request to the matching attributes in the salary information table; When the salary adjustment request is image data, multiple text data blocks in the image data are obtained based on a text detection model and a text recognition model, and a subject word is obtained based on a text data block with the greatest importance among the multiple text data blocks; and feature data of the subject word in the salary adjustment request is added to an attribute matching the subject word in the salary information table.

3. The data processing method based on the neural network model according to claim 2 is characterized in that: The semantic analysis model adopts the TextRank4ZH model, the text detection model adopts the DBNet network model, and the text recognition model adopts the SVTR network model.

4. The data processing method based on the neural network model according to claim 3 is characterized in that: The acquiring of a plurality of text data blocks in the image data based on a text detection model and a text recognition model comprises: Performing horizontal correction on the image data to obtain a corrected image; Obtain multiple text lines in the image data based on the text detection model; Recognize text information in multiple text lines based on a text recognition model; After combining the upper and lower adjacent text lines, the text information is segmented to obtain a phrase sequence; The text lines are reassembled into a plurality of text data blocks according to the pixel similarity of the image backgrounds of the plurality of text lines and the continuity of the phrase sequence.

5. The data processing method based on the neural network model according to claim 4 is characterized in that: The acquiring of a subject word based on the text data block with the greatest importance among the plurality of text data blocks comprises: Extracting multiple keywords from multiple text data blocks based on a semantic analysis model; Calculating the importance of the plurality of text data blocks based on the positions of the plurality of text data blocks in the image data and the word frequencies of the keywords of the text data blocks; The keywords of the most important text data blocks among the plurality of text data blocks are used as subject words.

6. The data processing method based on the neural network model according to claim 1, characterized in that: Determining the complexity of the sub-computing task includes: Based on the preset salary calculation rule, determine the number of code lines and execution time step required to execute each of the sub-computing tasks, and obtain the first complexity of the sub-computing tasks; Obtain a data group corresponding to the data called by each of the sub-computing tasks, and for each data group, calculate the similarity between the data in the data group to obtain the second complexity of the sub-computing task; The first complexity and the second complexity constitute a multi-dimensional vector to obtain a complexity vector of the sub-computing task.

7. The data processing method based on the neural network model according to claim 6 is characterized in that: The allocating the plurality of sub-computing tasks to the plurality of computing nodes in the salary calculation system according to the complexity comprises: (1) performing a first clustering of the plurality of sub-computing tasks according to the complexity vector to obtain a plurality of task levels; wherein the task levels are arranged in descending order according to the eigenvalues ​​of the complexity vector: the first task level, the second task level, ..., the nth task level, where n is the total number of task levels, to form a task sequence with the complexity of the sub-tasks decreasing from large to small; recorded as the first sequence; (2) Based on the currently available computing nodes, the currently available computing nodes are sorted according to the load occupancy rate of the computing nodes, which is recorded as the second sequence. The computing node with a smaller load occupancy rate is ranked higher in the second sequence. (3) For each sub-computing task in the first sequence, assign it to the computing nodes in the second sequence in the order in the first sequence. Let the mth sub-computing task in the first sequence be assigned to the nth computing node in the second sequence, then the m+1th sub-computing task in the first sequence be assigned to the n+1th computing node in the second sequence, and so on. (4) When all sub-computing tasks in the first sequence are assigned, a new first sequence is generated, where the new first sequence is initially an empty sequence; (5) Obtaining the load occupancy rate of the computing node after the sub-computing task is assigned; (6) If the load occupancy rate of the computing node is greater than a preset threshold, remove the sub-computing task of the smallest task level in the computing node, and add the removed sub-computing task to the new first sequence; return to step (5); (7) If the load occupancy rate of the computing node is not greater than the preset threshold, all sub-computing tasks in the new first sequence are sorted to form a task sequence with sub-task complexity from large to small, and then return to step (2).

8. A data processing system based on a neural network model, characterized in that: include: The information table establishment module is used to obtain employee salary-related information and establish a standardized salary information table; A data updating module, configured to receive an uploaded salary adjustment request, extract characteristic data in the salary adjustment request based on a neural network model, and add the characteristic data to the salary information table; A task allocation module is used to split the salary calculation task into multiple sub-calculation tasks that can be executed in parallel based on preset salary calculation rules; determine the complexity of the sub-calculation tasks, and allocate the multiple sub-calculation tasks to multiple computing nodes in the salary calculation system according to the complexity; The salary calculation module is used to use the plurality of computing nodes to respectively call the corresponding data in the salary information table, and execute the calculation program to obtain the salary calculation result.

9. An electronic device, characterized in that: The electronic device comprises: A memory for storing executable instructions; A processor is used to implement a data processing method based on a neural network model as described in any one of claims 1 to 7 when running the executable instructions stored in the memory.

10. A computer-readable storage medium storing executable instructions, characterized in that: When the executable instructions are executed by the processor, a data processing method based on a neural network model as described in any one of claims 1 to 7 is implemented.

Images