Search

CN-122022753-A - Multimode professional image construction and intelligent post matching method based on simulation model and federal learning

CN122022753ACN 122022753 ACN122022753 ACN 122022753ACN-122022753-A

Abstract

The invention relates to the technical field of professional portraits construction, in particular to a multimode professional portraits construction and intelligent post matching method based on a simulation model and federal learning. The method comprises the following steps of multi-mode professional data acquisition and preprocessing, professional capability asymmetric federal simulation modeling, multi-mode professional portrait construction, intelligent post matching and feedback. Based on national professional skill standards, the professional skill spectrum and the transfer matrix are constructed, the skill association strength is calculated through historical skill co-occurrence data, and the acceptable threshold of the enterprise skill gap is quantized by combining a dynamic tolerance function, so that standardized support is provided for post matching. In the federal training, a weighted federal average algorithm is combined with an entropy weight method to aggregate weight updating patterns and matrixes, and model parameters, pattern structures and feature dimensions are optimized through a full-link feedback mechanism, so that dynamic iteration is realized, and the suitability and reliability of post matching are improved.

Inventors

  • ZHANG JIAN

Assignees

  • 北京齐绘未来教育科技有限公司

Dates

Publication Date
20260512
Application Date
20260202

Claims (10)

  1. 1. The multi-mode professional image construction and intelligent post matching method based on the simulation model and federal learning is characterized by comprising the following steps of: S1, acquiring and preprocessing multi-mode professional data through a multi-source heterogeneous data acquisition technology, preprocessing original data by adopting a data cleaning, feature labeling and normalization technology, generating a standardized data subset comprising a personal data subset and an enterprise data subset, and outputting the standardized data subset to S2 through a data interaction interface; S2, carrying out asymmetric federal simulation modeling on occupational capacity, namely receiving a standardized data subset output by S1, constructing an asymmetric federal simulation framework based on occupational capacity dynamics characteristics, initializing a personal terminal side capacity evolution simulation model, an enterprise terminal side post demand simulation model and corresponding modeling logic, respectively carrying out simulation processing on the personal terminal side and the enterprise terminal side by the personal terminal side and the enterprise terminal side through the personal terminal side capacity evolution simulation model and the enterprise terminal side post demand simulation model to generate capacity vectors representing dynamic changes of skill states and characterization containing post skill demand vectors and dynamic tolerance functions, quantifying acceptable threshold values of enterprises on skill gaps and service elasticity adjustment ranges, maintaining an occupational capacity spectrum and a transfer matrix constructed based on industry standard skill frames, aggregating associated parameter increment and correction signals output by a personal simulation end and an enterprise simulation end in the federal training iteration process, updating the occupational capacity spectrum and the transfer matrix, respectively according to the updated calibration capacity simulation model, optimizing the post simulation model and the updated post demand simulation model, and optimizing the updated post demand simulation model, and the analysis result being simultaneously matched with the state demand simulation model and the updated by the personal terminal side simulation end, and the step 4; S3, constructing a multi-mode professional portrait, namely receiving the capability vector, the post skill requirement vector and the updated professional capability spectrum output by the S2, constructing a personal multi-mode professional portrait and an enterprise post requirement portrait, and outputting the personal multi-mode professional portrait and the enterprise post requirement portrait to the S4; and S4, intelligent post matching and feedback, namely receiving the personal multi-mode professional portrait and the enterprise post requirement portrait output in the step S3, calculating the matching degree, outputting a matching result, and feeding back the matching result to the step S2 as a matching result feedback signal.
  2. 2. The method for constructing and matching multi-modal occupation images with intelligent posts based on simulation models and federal learning according to claim 1, wherein the multi-modal occupation data collection and preprocessing in S1 comprises the following steps: S11, acquiring multi-mode professional data of a personal end and an enterprise end, and performing format conversion and unified storage on the acquired original data; S12, cleaning the stored original data, removing abnormal values in the personal skill state data by adopting a 3 sigma principle, filtering extreme values in the enterprise post demand data by adopting a box graph method, supplementing the missing data by adopting a K nearest neighbor filling method, and de-duplicating the data based on the time stamp priority; S13, based on a preset industry standard skill dictionary, automatically marking the characteristics of the cleaned data, wherein the personal data marking dimension comprises skill types, skill levels and application scenes, the enterprise data marking dimension comprises post core skills, skill requirement priorities and service association degrees, and the marking errors are corrected by manual sampling rechecking; S14, processing the marked data by adopting a differential normalization algorithm, mapping the personal data to a [0,1] interval by adopting Min-Max normalization, and converting enterprise data into standard normal distribution data by adopting Z-Score normalization; S15, dividing the normalized data into blocks according to the attribution and the time dimension of the data, generating a standardized data subset comprising a personal data subset and an enterprise data subset, and storing the standardized data subset according to Parquet format and attaching a data acquisition time stamp and a preprocessing flow tracing identifier; s16, outputting the standardized data subset to S2 through an encryption RESTfulAPI interface, encrypting data transmission by adopting an SSL protocol and a TLS protocol, and carrying out integrity check; And S17, receiving the atlas iteration feedback signal from the S2, extracting atlas node association degree parameters, retaining the characteristics with the association degree higher than a set value with the professional energy spectrum nodes when the parameter variation amplitude exceeds a preset threshold, and removing redundant characteristics to adjust the dimension of the preprocessing characteristics.
  3. 3. The method for matching multi-modal occupation image construction with intelligent posts based on simulation models and federal learning according to claim 2, wherein the process of asymmetric federal simulation modeling of occupation ability in S2 comprises the following steps: S21, receiving the standardized data subset output by the S1, completing integrity and format verification, building an asymmetric federal simulation framework based on occupational capacity dynamics characteristics, initializing a personal terminal side capacity evolution simulation model, an enterprise terminal side post demand simulation model and corresponding modeling logic, and determining core parameters as skill evolution rate thresholds Coefficient of demand fluctuation ; S22, respectively carrying out simulation processing on the personal data subset and the enterprise data subset at the corresponding terminal side through the personal terminal side capability evolution simulation model and the enterprise terminal side post requirement simulation model to generate capability vectors representing dynamic changes of skill states And include a post skill requirement vector, a dynamic tolerance function Is characterized by (2); S23, maintaining a professional ability map and a transfer matrix based on an industry standard skill frame By skill association strength Defining a map rule, initializing matrix state conversion logic based on probability distribution, and providing basic support for a personal terminal side capacity evolution simulation model and an enterprise terminal side post demand simulation model; S24, in federal training iteration, processing parameter increment of personal terminal side capability evolution simulation model by adopting weighted aggregation algorithm Parameter increment of enterprise terminal side post demand simulation model And updating occupational capacity map and transfer matrix according to map correction signals Distributing the post encryption to a personal terminal side capacity evolution simulation model and an enterprise terminal side post demand simulation model; S25, calibrating personal terminal side capacity evolution simulation model logic based on the updated map, optimizing enterprise terminal side post demand simulation model logic, outputting core data to S3, receiving S4 a matching result feedback signal, and transmitting the fluctuation amplitude feedback signal to the user terminal And adjusting parameters of the personal terminal side capacity evolution simulation model and the enterprise terminal side post demand simulation model to form a self-adaptive closed loop.
  4. 4. The method for matching multi-modal professional image construction and intelligent post based on simulation model and federal learning according to claim 3, wherein the process of receiving the standardized data subset and completing verification, constructing the asymmetric federal simulation frame and initializing the model in S2.1 comprises the following steps: S21.1, checking the integrity of the standardized data subset by adopting an SHA-256 hash algorithm, performing format check on the data subset based on JSONSchema specifications, and returning to S1 for re-preprocessing if the check is not passed; s21.2, after the asymmetric federal simulation framework is built, determining the attribution of core parameters, namely skill evolution rate threshold value Corresponding to personal terminal side capability evolution simulation model, requirement fluctuation coefficient Corresponding to the enterprise terminal side post demand simulation model, respectively associating historical professional skill evolution data and enterprise demand change data with parameter values; S21.3, respectively fitting the double data sources through a time sequence fitting algorithm, and combining time attenuation weights Determination of And (3) with Synchronously initializing basic structure parameters of a personal terminal side capacity evolution simulation model and an enterprise terminal side post demand simulation model.
  5. 5. The method for matching multi-modal professional image construction and intelligent post based on simulation models and federal learning according to claim 4, wherein the process of simulating data and generating capacity vectors and requirement characterization by the corresponding models in S2.2 comprises the following steps: S22.1, a personal terminal side invokes a personal terminal side capability evolution simulation model, time sequence modeling is carried out on skill state change data of a personal data subset, and skill proficiency and evolution trend core features are extracted; S22.2 generating a capability vector based on the extracted features Vector data are synchronously stored to a personal terminal side capacity evolution simulation model; s22.3, the enterprise terminal side invokes an enterprise terminal side post demand simulation model to fit enterprise historical skill gap adaptation data and business fluctuation data Constructing a dynamic tolerance function ; S22.4 by dynamic tolerance function Quantizing skill gap acceptable threshold and business elasticity adjusting range to generate a post skill demand vector and a post skill demand vector Is characterized by (3).
  6. 6. The method for matching multi-modal occupation image construction with intelligent posts based on simulation models and federal learning according to claim 5, wherein the process of maintaining the occupation energy spectrum and the transfer matrix and adapting the corresponding model in S2.3 comprises the following steps: s23.1, adopting national professional skill standards as industry standard skill frameworks, and defining professional energy spectrum nodes as skill IDs, levels and application scenes; S23.2, counting historical skill co-occurrence data, and calculating the correlation strength between skills through a similarity algorithm Based on Defining a map edge attribute with a skill evolution path; S23.3, counting historical skill conversion data, and initializing a transfer matrix based on data probability distribution State transition logic of (a); S23.4, completing the transfer matrix After initialization, synchronizing the map and matrix basic data to a personal terminal side capacity evolution simulation model and an enterprise terminal side post demand simulation model.
  7. 7. The method for matching multi-modal professional image construction and intelligent post based on simulation model and federal learning according to claim 6, wherein the process of aggregating model parameter increment and correction signals and updating distribution map matrix in S2.4 comprises the following steps: S24.1, calling a weighted federal average algorithm to increase the parameter of the personal terminal side capability evolution simulation model Parameter increment of enterprise terminal side post demand simulation model Performing aggregation treatment on the map correction signals; S24.2, calculating the data volume weight and model contribution weight of the personal terminal side capacity evolution simulation model and the data volume weight and model contribution weight of the enterprise terminal side post demand simulation model through an entropy weight method, and carrying out weighted combination on the data volume weight and model contribution weight of the personal terminal side capacity evolution simulation model and the enterprise terminal side post demand simulation model to obtain a final aggregation weight 、 ; S24.3, based on aggregation weight 、 Correction coefficient Updating occupational energy map spectrum association strength And transfer matrix ; S24.4, encrypting the updated map and matrix parameters by adopting an SSL protocol and a TLS protocol, and distributing the encrypted map and matrix parameters to a personal terminal side capacity evolution simulation model and an enterprise terminal side post demand simulation model after hash verification.
  8. 8. The method for matching multi-modal professional image construction with intelligent post based on simulation models and federal learning according to claim 7, wherein the process of calibrating the optimization model logic and receiving feedback adjustment parameters in S2.5 comprises the following steps: S25.1, calibrating a capacity evolution simulation model logic of the personal terminal side based on the updated map by the personal terminal side, and adjusting structural parameters of the capacity evolution simulation model of the personal terminal side when the error exceeds a preset threshold value by adopting an average absolute error MAE and root mean square error RMSE double-index verification model to output precision; S25.2, optimizing the enterprise terminal side post demand simulation model logic by the enterprise terminal side based on the updated map, counting demand matching accuracy, and adjusting a dynamic tolerance function when the accuracy is lower than a reference value Coefficients of (2); S25.3, capability vector Outputting the post skill demand vector and the updated map to S3, and simultaneously receiving a matching result feedback signal of S4, wherein feedback data are synchronized to a personal terminal side capacity evolution simulation model and an enterprise terminal side post demand simulation model; S25.4, calculating fluctuation amplitude of the matching result by adopting a sliding window algorithm When the fluctuation amplitude exceeds a preset threshold, automatically adjusting the dynamic parameters of the personal terminal side capability evolution simulation model Dynamic parameters of enterprise terminal side post demand simulation model Federal aggregate weights 、 。
  9. 9. The method for building and matching multi-modal occupation pictures with intelligent posts based on simulation models and federal learning according to claim 8, wherein the process of receiving S2 output data and building the personal multi-modal occupation pictures and the business post requirement pictures in S3 comprises the following steps: s31.1, receiving the capability vector of S2 output Post skill demand vector and updated professional ability map; s31.2, vector of capability Mapping to a professional energy spectrum, associating the spectrum node attribute, and extracting the individual multi-modal characteristics; s31.3, constructing a personal multi-mode professional portrait based on the extracted personal multi-mode features; s31.4, associating post skill demand vectors with skill association strength of professional ability maps Extracting multi-mode characteristics of the enterprise post requirements; s31.5, constructing an enterprise post requirement portrait based on the extracted enterprise post requirement multi-mode features; s31.6, outputting the personal multi-mode occupation portrait and the enterprise post requirement portrait to S4.
  10. 10. The method for matching multi-modal professional image construction and intelligent post based on simulation model and federal learning according to claim 9, wherein the process of receiving the S3 output image and calculating the matching degree, the output result and the feedback signal in S4 comprises the following steps: s41.1, receiving the personal multi-mode occupation portrait and the enterprise post requirement portrait output by the step S3; s41.2, skill association strength based on professional ability map Calculating the matching degree of the personal multi-mode occupation portrait and the enterprise post demand portrait; s41.3, outputting a matching result of the personal multi-mode occupation portrait and the enterprise post requirement portrait; S41.4, synchronously sending the matching result serving as a matching result feedback signal to the personal terminal side capacity evolution simulation model and the enterprise terminal side post demand simulation model.

Description

Multimode professional image construction and intelligent post matching method based on simulation model and federal learning Technical Field The invention relates to the technical field of professional portraits construction, in particular to a multimode professional portraits construction and intelligent post matching method based on a simulation model and federal learning. Background In digital human resource management and intelligent recruitment, professional image construction and post accurate matching are the cores for improving configuration efficiency. The explosive growth of multi-mode data makes the traditional single-dimension matching difficult to meet the requirements of precision and dynamics, and the personal and enterprise data relate to privacy and business secrets, and centralized processing is easy to cause leakage risk. Therefore, integrating the advantages of multi-mode data and privacy protection technology, constructing dynamic and accurate professional images and realizing efficient matching becomes a key problem to be solved urgently in the industry. In the prior art, related patents have been explored in the field of job image generation and post matching. The invention patent CN202411330402.8 discloses a recruitment talent portrait generation method and a recruitment talent portrait generation system based on multidimensional information, which collect candidate resume information, extract a behavior mode and interest points through natural language processing after cleaning and conversion, train a model to generate a portrait with skill capability, professional potential and cultural adaptation degree, and further provide post matching recommendation and professional suggestion. As another example, the invention patent CN202411530537.9 discloses an intelligent post matching and employee growth path planning system, method and storage medium, the system constructs a comprehensive portrait by collecting information such as employee performance, educational background and the like, clustering and factor analysis, realizes post adaptation by using a recommendation algorithm, and provides personalized professional development path and training advice by combining industry trend. Although the above technical scheme has corresponding design advantages, the above technical scheme has the technical defects that firstly, the data processing cannot achieve both privacy security and dynamic adaptability, the invention patent CN202411330402.8 and the CN202411530537.9 adopt centralized data acquisition and processing modes, related technical designs of data privacy protection are not set, personal professional data and enterprise demand data are easy to face privacy leakage or business secret leakage risks in the centralized storage and model training processes, meanwhile, the CN202411330402.8 and the CN202411530537.9 construct an image and matching model based on historical static data, only the current capability state of an individual and the fixed requirements of an enterprise are captured, the evolution rule of personal skills along with time is not considered, and the post demand elastic change caused by enterprise business fluctuation is not adapted, so that the image and matching results are difficult to dynamically fit the actual scene. Secondly, the matching logic lacks a skill association support and dynamic optimization mechanism, a standardized professional skill association system is not established by both CN202411330402.8 and CN202411530537.9, portrait construction and matching calculation are only carried out around a single skill dimension or independent features, association strength, hierarchical relation and conversion logic among different skills cannot be embodied, the acceptable range of enterprises on skill gaps is not quantized, so that the matching accuracy is limited, in addition, feedback optimization paths of matching results on model parameters and feature dimensions are not designed by both CN202411330402.8 and CN202411530537.9, once the models and portraits are constructed, the models and the portraits are fixed, the optimization is difficult to adjust according to actual matching effects, and the self-adaption capability is insufficient. In view of this, we propose a multi-modal professional image construction and intelligent post matching method based on simulation model and federal learning. Disclosure of Invention The invention aims to provide a multimode professional image construction and intelligent post matching method based on a simulation model and federal learning, so as to solve the problem that the data processing proposed in the background art cannot be compatible with privacy security and dynamic adaptability and the matching logic lacks a skill association support and dynamic optimization mechanism. In order to solve the technical problems, the invention provides a multi-mode professional image construction and intelligent post matching method