CN-122022749-A - Manpower processing method and system based on generation of countermeasure network and large language model

CN122022749ACN 122022749 ACN122022749 ACN 122022749ACN-122022749-A

Abstract

The invention provides a human processing method and a human processing system based on an antagonism network generation and a large language model, which relate to the technical field of artificial intelligence, wherein the method comprises the steps of collecting multi-industry talent data and cross-border post demand data; the method comprises the steps of mapping multi-industry talent data into three-dimensional grids in a high-dimensional semantic embedding space, wherein nodes defined in the three-dimensional grids represent specific skill points and project experience points under different industry backgrounds, mapping cross-border post demand data into spheres in the same semantic embedding space, wherein point sets defined on the surfaces and the interiors of the spheres represent core capability indexes and condition ranges required by posts, and determining feature intersection point sets of the three-dimensional grids and the spheres in the semantic embedding space to construct an initial training data set. The invention can realize the deep adaptation of cross-domain talents and cross-boundary posts and the dynamic optimization of the whole process of manpower management.

Inventors

GUO XIAOYU
LI Dansi
WANG SEN

Assignees

山东鼎驰人力资源有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. A human processing method based on generating a countermeasure network and a large language model, the method comprising: The method comprises the steps of collecting multi-industry talent data and cross-border post demand data, mapping the multi-industry talent data into a three-dimensional grid in a high-dimensional semantic embedding space, wherein nodes defined in the three-dimensional grid represent specific skill points and project experience points under different industrial contexts; Generating virtual talent capability data with cross-domain capability causal relationship by generating an antagonism network model based on the improvement of causal structure constraint according to the initial training data set; fusing the initial training data set and the virtual talent ability data, and performing fine adjustment on the pre-training large language model to obtain a fine-adjusted large language model with enhanced semantic understanding ability; Analyzing the preset target post requirements and talent capacity based on the fine-tuned large language model by combining DIKWP capacity assessment frames to obtain a capacity conversion feasibility assessment result; According to the capability conversion feasibility evaluation result, calculating the dynamic energy level matching degree of talents and posts, and generating a capability gap visual report and a cross-culture adaptation suggestion; and analyzing the feedback effect of the dynamic energy level matching degree and the capability gap visual report in the practical application, dynamically adjusting and improving to generate causal structure constraint in the antagonism network model, and synchronously updating the fine tuning parameters of the fine-tuned large language model.
2. The human processing method based on generating an antagonism network and a large language model according to claim 1, wherein the multi-industry talent data is mapped into a three-dimensional grid in a high-dimensional semantic embedding space, wherein nodes defined in the three-dimensional grid represent specific skill points and project experience points in different industry contexts, the cross-border post demand data is mapped into spheres in the same semantic embedding space, wherein the surface and the internal defined point sets of the spheres represent core capability indexes and condition ranges required by posts, and the feature intersection point sets of the three-dimensional grid and the spheres in the semantic embedding space are determined to construct an initial training data set, comprising: Acquiring multi-industry talent data and cross-border post demand data, wherein the multi-industry talent data comprises skill points and project experience points, and the cross-border post demand data comprises core capability indexes and condition ranges; Mapping skill points and project experience points in multi-industry talent data to a high-dimensional semantic embedding space to construct a three-dimensional grid, wherein nodes defined in the three-dimensional grid correspond to a specific skill point or project experience point respectively; mapping core capability indexes and condition ranges in the cross-border post demand data to the same high-dimensional semantic embedded space to construct a sphere, wherein the point sets defined on the surface and the inside of the sphere correspond to one core capability index or condition range respectively; calculating a characteristic intersection point set of the three-dimensional grid and the sphere in the high-dimensional semantic embedding space, wherein the characteristic intersection point set consists of points, which are intersected or contained in the space position, of nodes of the three-dimensional grid and the point set of the sphere; An initial training data set is generated based on the feature intersection point set.
3. The human processing method based on generating a countermeasure network and large language model according to claim 2, wherein generating virtual talent capability data with cross-domain capability causal relationships by generating a countermeasure network model with an improvement based on causal structure constraints from an initial training dataset, comprises: Based on an initial training data set, constructing a cross-domain capability causal graph by analyzing dependence and conversion relations between different skills and experience points; according to the cross-domain capability causal graph, abstracting and generating causal structure rules for constraint model training; integrating the causal structure rules into a training process for generating an antagonism network to construct an improved generation antagonism network model based on causal structure constraints; Training the improved generation countermeasure network model through an initial training data set to optimize parameters of the improved generation countermeasure network model, and obtaining a trained improved generation countermeasure network model; an countermeasure network model is generated based on the improvement of the training completion, and virtual talent capability data with cross-domain capability causal relationship is generated.
4. The human processing method based on generating an countermeasure network and a large language model according to claim 3, wherein the pre-training large language model is fine-tuned by fusing an initial training data set and virtual talent ability data to obtain a fine-tuned large language model with enhanced semantic understanding ability, comprising: fusing the initial training data set and the virtual talent ability data to form a mixed training data set; carrying out formatting pretreatment on the mixed training data set to generate model fine tuning input data with uniform structure; performing supervised fine tuning on the pre-trained large language model based on the model fine tuning input data with uniform structure so as to update model parameters of the pre-trained large language model and obtain a primarily fine-tuned large language model; analyzing the performance of the primarily fine-tuned large language model on the mixed training data set to evaluate the semantic understanding capability of the primarily fine-tuned large language model and generating a semantic understanding capability evaluation result; And when the semantic understanding capability evaluation result of the optimized large language model meets the preset condition, confirming the fine-tuned large language model with enhanced semantic understanding capability.
5. The human processing method based on the generation of the countermeasure network and the large language model according to claim 4, wherein the analyzing the preset target post requirements and talent ability based on the trimmed large language model by combining DIKWP with the ability evaluation framework to obtain the ability transformation feasibility evaluation result comprises the following steps: Acquiring preset target post demand description and talent capability data to be evaluated as analysis input data; Based on the fine-tuned large language model, carrying out deep semantic analysis on the analysis input data to extract post demand elements and talent capacity elements; Combining DIKWP capability assessment frameworks, mapping the post requirement elements and talent capability elements to five dimensions of data, information, knowledge, wisdom and purposes respectively, and generating a structured capability element set; Based on the structured capability element set, calculating the matching degree and conversion path feasibility between the talent capability element and the post requirement element according to each dimension in the five dimensions, and generating a preliminary feasibility analysis result of a plurality of dimensions; And integrating the preliminary feasibility analysis results of each dimension through a preset decision rule to form a final capability transformation feasibility assessment result.
6. The human processing method based on generating the countermeasure network and the large language model according to claim 5, wherein calculating dynamic energy level matching degree of talents and posts according to the capability transformation feasibility evaluation result, generating a capability gap visual report and a cross-cultural adaptation suggestion comprises: based on the final capability transformation feasibility assessment result, extracting the degree of matching of each dimension and the transformation path feasibility quantification index; according to a preset energy level weight coefficient, carrying out weighted calculation on the quantization index to generate dynamic energy level matching degree of talents and posts; Comparing the dynamic energy level matching degree with a preset post level threshold value to identify and locate key capability gaps and form a capability gap visual report; and combining the capability gap visual report with cross-cultural background information contained in talent data, and generating cross-cultural adaptation suggestions by performing rule and case reasoning.
7. The human processing method based on generating an countermeasure network and a large language model according to claim 6, wherein analyzing the feedback effect of the dynamic energy level matching degree and capability gap visual report in the practical application to dynamically adjust and improve the causal structure constraint in the generating countermeasure network model, synchronously updating the fine tuning parameters of the fine tuned large language model, comprises: collecting effect feedback data of a dynamic energy level matching degree and capability gap visual report in actual recruitment decision or talent development application; analyzing the effect feedback data to evaluate the accuracy of the current matching result and the effectiveness of the capacity gap identification, and generating a model performance deviation evaluation result; Identifying, based on the model performance bias evaluation results, causal structural rule defects that improve generation of the resulting bias in the antagonism network model; According to the identified causal structure rule defect, dynamically adjusting and improving generation of causal structure constraint in the antagonism network model, and generating adjusted causal structure constraint; And synchronously acting the adjusted causal structure constraint on the parameter updating process of the trimmed large language model to realize iterative updating of the trimmed parameters of the trimmed large language model, and finally completing the circular optimization of the round of model.
8. A human processing system based on generating a countermeasure network and a large language model, characterized in that the system performs the method according to any one of claims 1 to 7, comprising: The mapping characterization module is used for collecting multi-industry talent data and cross-border post demand data; mapping multi-industry talent data into three-dimensional grids in a high-dimensional semantic embedded space, wherein nodes defined in the three-dimensional grids represent specific skill points and project experience points under different industry backgrounds; mapping the cross-border post demand data into spheres in the same semantic embedding space, wherein the defined point sets on the surfaces and the interiors of the spheres represent core capability indexes and condition ranges required by posts; the model construction and coordination module is used for generating an countermeasure network model by adopting the improvement based on causal structure constraint according to the initial training data set and generating virtual talent capacity data with a cross-domain capacity causal relation; Fine tuning the pre-trained large language model by fusing the initial training data set and the virtual talent capability data to obtain a fine-tuned large language model with enhanced semantic understanding capability; The analysis evaluation module is used for analyzing the preset target post requirements and talent capabilities based on the trimmed large language model and combining DIKWP capability evaluation frames to obtain a capability conversion feasibility evaluation result; The matching degree calculation module calculates the dynamic energy level matching degree of talents and posts according to the capability transformation feasibility evaluation result, and generates a capability gap visual report and a cross-culture adaptation suggestion; and the iteration adjustment module is used for analyzing the feedback effect of the dynamic energy level matching degree and capability gap visual report in the practical application so as to dynamically adjust and improve the generation of causal structure constraint in the antagonism network model and synchronously update the fine adjustment parameters of the large language model after fine adjustment.
9. A computing device comprising a memory and a processor; wherein one or more computer programs are stored in the memory, the one or more computer programs comprising instructions, which when executed by the processor, cause the computing device to perform the method of any of claims 1-7.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium is for storing a computer program, the computer program is for performing the method of any of claims 1 to 7.

Description

Manpower processing method and system based on generation of countermeasure network and large language model Technical Field The invention relates to the technical field of artificial intelligence, in particular to a human processing method and system based on generation of an antagonism network and a large language model. Background Along with the deep fusion of artificial intelligence and various industries, ai+x cross-border posts continue to emerge, structural contradictions of talent supply and demand are increasingly prominent, and efficient manpower processing technology is needed to achieve accurate matching of talents and posts. Currently, human processing schemes based on generation of a countermeasure network (GAN) and a Large Language Model (LLM) have been gradually applied, but the prior art generally has the technical defect that data modeling and model training are only performed based on statistical correlation, and effective constraint is not performed on causal dependency between inter-domain human capability and post requirements. The limitation directly leads to the following association results that the matching result of cross-domain talents and posts only stays at the coincidence of surface features, internal conversion logic between skills and experiences is difficult to reflect, matching accuracy is insufficient, deep requirements for capability suitability under AI+ scenes cannot be met, logic contradiction easily occurs in virtual talent data generated by a generation model lacking causal constraints, a large language model finely tuned based on the data is weak in generalization capability in the face of a new cross-boundary scene, and efficient iterative optimization of model parameters cannot be realized through practical application feedback. Disclosure of Invention The invention aims to solve the technical problem of providing a human processing method based on a generation countermeasure network and a large language model, which can solve the problems of insufficient matching precision and weak generalization capability of the existing scheme and realize the deep adaptation of cross-domain talents and cross-boundary posts and the dynamic optimization of the whole process of human management. In order to solve the technical problems, the technical scheme of the invention is as follows: in a first aspect, a human processing method based on generating a countermeasure network and a large language model, the method comprising: The method comprises the steps of collecting multi-industry talent data and cross-border post demand data, mapping the multi-industry talent data into a three-dimensional grid in a high-dimensional semantic embedding space, wherein nodes defined in the three-dimensional grid represent specific skill points and project experience points under different industrial contexts; Generating virtual talent capability data with cross-domain capability causal relationship by generating an antagonism network model based on the improvement of causal structure constraint according to the initial training data set; Fine tuning the pre-trained large language model by fusing the initial training data set and the virtual talent capability data to obtain a fine-tuned large language model with enhanced semantic understanding capability; Analyzing the preset target post requirements and talent capacity based on the fine-tuned large language model by combining DIKWP capacity assessment frames to obtain a capacity conversion feasibility assessment result; According to the capability conversion feasibility evaluation result, calculating the dynamic energy level matching degree of talents and posts, and generating a capability gap visual report and a cross-culture adaptation suggestion; and analyzing the feedback effect of the dynamic energy level matching degree and the capability gap visual report in the practical application, so as to dynamically adjust and improve the generation of causal structure constraint in the antagonism network model, and synchronously updating the fine tuning parameters of the large language model after fine tuning. It should be noted that, in the method, the generation of the countermeasure network model based on the improvement of the causal structure constraint refers to the integration of causal structure rules characterizing causal dependency relationships between capabilities into a model formed in the process of generating the countermeasure network training. In a second aspect, a human processing system based on generating a countermeasure network and a large language model, the system performing the method, comprising: The mapping characterization module is used for collecting multi-industry talent data and cross-border post demand data; mapping multi-industry talent data into three-dimensional grids in a high-dimensional semantic embedded space, wherein nodes defined in the three-dimensional grids represent specific skill points and project e