A cross-domain talent mobility recommendation method based on skill graph decoupling from large language models

Abstract

This invention belongs to the field of human resource management technology and provides a cross-domain talent mobility recommendation method based on skill graph decoupling using a large language model. The method includes: decoupling a candidate's skills using a large language model to obtain a set of explicit atomic capabilities and their proficiency; obtaining the atomic capability requirements and weights of the target position, calculating the initial job capability matching degree, and generating an initial recommendation score; determining the focus depth influence coefficient based on the number and proficiency of explicit atomic capabilities, and adjusting it to obtain a comprehensive recommendation score; further, decoupling the candidate's experience data using a large language model to extract a set of potential atomic capabilities and their proficiency, and calculating potential atomic capability recommendation scores; and weightedly fusing the comprehensive recommendation score and potential capability recommendation scores to obtain a final recommendation score, which is then used for talent mobility recommendations. This invention achieves refined skill decoupling and potential capability mining, improving the accuracy and applicability of cross-domain talent recommendations.

Description

Technical Field

[0001] This invention belongs to the field of human resource management technology, and in particular relates to a cross-domain talent mobility recommendation method based on decoupling of skill graphs from large language models. Background Technology

[0002] Cross-disciplinary talent mobility is a core element in optimizing human resource allocation and promoting industrial integration and development. Its effectiveness directly depends on the accurate identification of the match between candidates' abilities and those of target positions.

[0003] Current talent recommendation systems generally adopt traditional methods based on resume keyword matching, skill tag classification, or collaborative filtering of historical behavior. Some solutions introduce knowledge graph technology to abstract skills into nodes and calculate matching relationships through path similarity, or use deep learning models to encode features of work experience text to generate appropriate scores.

[0004] However, these existing technologies have fundamental limitations. First, the methods rely excessively on manually pre-defined discrete skill tagging systems or structured resume information explicitly provided by candidates, resulting in coarse-grained expression of abilities and an inability to depict continuous differences in skill proficiency. For example, simply labeling programming ability as "familiar" without quantifying specific levels is insufficient. Second, the rich semantic information contained in free-text candidate descriptions (such as interview statements or project documents) is not effectively utilized, making it difficult to identify and assess potential abilities not explicitly stated in the resume. More importantly, existing technical frameworks fail to leverage the deep semantic understanding and decoupling advantages of large language models: skill modeling remains at the level of composite labels, lacking a mechanism to decompose complex skills into atomic ability units and accurately quantify proficiency, resulting in a lack of refined support for matching degree calculation; at the same time, the potential atomic abilities implicit in candidates' experience data cannot be systematically extracted, and the processes of explicit ability matching and potential ability mining are disconnected, ignoring the focus depth characteristics reflected by the quantitative distribution of highly proficient explicit atomic abilities, and lacking a unified fusion framework to comprehensively evaluate candidates' immediate adaptability and long-term growth potential, ultimately restricting the accuracy and comprehensiveness of cross-domain talent mobility recommendations. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a cross-domain talent mobility recommendation method based on decoupling of a large language model skill graph, thus solving the aforementioned problems.

[0006] To achieve the above objectives, this invention provides the following technical solution: a cross-domain talent mobility recommendation method based on decoupling of a large language model skill graph, which specifically includes:

[0007] Based on a large language model, the candidate's skills are decoupled to determine the candidate's set of explicit atomic abilities and the proficiency of each explicit atomic ability.

[0008] Obtain the required atomic skill proficiency for the target position and the importance weight coefficient of the atomic skill to the target position. Based on the candidate's explicit atomic skill proficiency, the required atomic skill proficiency for the target position, and the importance weight coefficient of the atomic skill to the target position, determine the candidate's initial job skill matching degree.

[0009] Based on the candidate's initial job competency match, determine the candidate's initial recommendation score;

[0010] Based on the number of dominant atomic abilities in the candidate's dominant atomic ability set and the proficiency of the dominant atomic abilities, the candidate's initial recommendation score is adjusted, and the candidate's comprehensive recommendation score is output.

[0011] Based on big data, we collect candidates’ experience data, and then decouple the candidates’ experience data through a big language model to determine the candidates’ potential atomic ability set and the proficiency of each potential atomic ability.

[0012] The recommended score for a candidate's potential atomic ability is determined based on the candidate's set of potential atomic abilities and the proficiency level of each potential atomic ability.

[0013] The final recommendation score for each candidate is determined based on their potential atomic ability recommendation score and their overall recommendation score.

[0014] Cross-disciplinary talent mobility recommendations will be made based on the candidates' final recommendation scores.

[0015] Based on the above technical solutions, the present invention also provides the following optional technical solutions:

[0016] Further technical solution: The determination of the preliminary job competency matching degree of the candidate specifically includes:

[0017] Through the formula: ;

[0018] Determine the initial match between the candidate's job skills and the position. ;

[0019] In the formula, This refers to the proficiency of candidate i in the kth explicit atomic ability of the same type as the atomic ability required for the target position. This refers to the proficiency requirement of the k-th atomic ability for the target position. , refers to the importance weight coefficient of the k-th atomic ability for the target position, and N refers to the total number of atomic ability proficiency requirements for the target position; where, if candidate i does not have an explicit atomic ability of the same type as the atomic ability required for the target position, then the candidate's explicit atomic ability proficiency is 0.

[0020] Further technical solution: The determination of the preliminary recommendation score for candidates specifically includes:

[0021] Through the formula: ;

[0022] Determine the preliminary recommendation score for candidates ;

[0023] In the formula, This refers to the initial match between candidate i's job skills and the candidate's qualifications. This refers to the matching sensitivity adjustment coefficient.

[0024] Further technical solutions: The specific method for outputting the candidate's comprehensive recommendation score includes:

[0025] The focus depth influence coefficient of a candidate is determined based on the number of dominant atomic abilities in the candidate's set of dominant atomic abilities and the proficiency of the dominant atomic abilities.

[0026] Based on the candidate's focus depth influence coefficient, the initial recommendation score for the candidate is adjusted, and the overall recommendation score for the candidate is output.

[0027] Further technical solution: The determination of the candidate's focus depth influence coefficient based on the number of dominant atomic abilities in the candidate's dominant atomic ability set and the proficiency of those dominant atomic abilities specifically includes:

[0028] Through the formula: ;

[0029] Determine the candidate's focus depth impact coefficient ;

[0030] In the formula, This refers to the proficiency of the j-th dominant atom in candidate i. This refers to the threshold for atomic ability proficiency. This refers to the number of dominant atoms in candidate i. For indicator functions, when the condition When it was established The value is 1 if it is not 1, otherwise the value is 0.

[0031] Further technical solution: The preliminary recommendation score for the candidate is adjusted based on the candidate's focus depth influence coefficient, and a comprehensive recommendation score for the candidate is output, specifically including:

[0032] Through the formula: ;

[0033] Output the overall recommendation score of the candidates ;

[0034] In the formula, This refers to the focus depth influence coefficient of candidate i. This refers to the initial recommendation score for candidate i. This refers to the scaling factor that affects the depth of focus; p is the adjustment factor. The function of the item is to... The value range is adjusted to ±1.

[0035] Further technical solution: The determination of the candidate's potential atomic capability recommendation score specifically includes:

[0036] Through the formula: ;

[0037] Determine the recommended score for the candidate's potential atomic ability ;

[0038] In the formula, This refers to the proficiency of the r-th potential atomic ability of candidate i. This refers to the importance weight coefficient of candidate i's r-th potential atomic ability for the target position. This refers to the r-th atomic ability proficiency requirement for the target position, and N refers to the total number of atomic ability proficiency requirements for the target position. If candidate i does not have any potential atomic ability of the same type as the atomic ability required for the target position, then the candidate's explicit atomic ability proficiency is 0.

[0039] Further technical solution: The step of determining the final recommendation score of a candidate based on the candidate's potential atomic capability recommendation score and the candidate's comprehensive recommendation score specifically includes:

[0040] Through the formula: ;

[0041] Determine the final recommendation score for each candidate. ;

[0042] In the formula, This refers to the recommended score for the potential atomic ability of candidate i. This refers to the overall recommendation score for candidate i. This refers to the weighting coefficient of training resources based on the target position.

[0043] Further technical solution: The cross-domain talent mobility recommendation based on the candidate's final recommendation score specifically includes:

[0044] Sort all candidates by their final recommendation scores and output the sequence of final recommendation scores for each candidate.

[0045] Cross-disciplinary talent mobility recommendations are made based on the final recommendation score sequence of candidates.

[0046] Further technical solution: The step of sorting the final recommendation scores of all candidates and outputting the final recommendation score sequence of candidates specifically includes:

[0047] When sorting all candidates' final recommendation scores, arrange them in descending order of their final recommendation scores and output the sequence of final recommendation scores.

[0048] The number of recommended candidates is preset to Z. The candidates corresponding to the top Z final recommendation scores in the candidate final recommendation score sequence are selected for cross-domain talent mobility recommendation.

[0049] This invention provides a cross-domain talent mobility recommendation method based on decoupling of skill graphs from large language models, which has the following advantages compared with existing technologies:

[0050] This invention uses a large language model to decouple a candidate's skills and experience into atomic capabilities, quantifies the proficiency of explicit and potential abilities, and incorporates a focus depth influence coefficient to correct the initial matching score. Finally, it combines potential ability recommendation scores to generate a comprehensive recommendation result. This method overcomes the limitations of traditional coarse modeling of skill tags, achieving refined quantification of abilities and in-depth mining of potential abilities, while taking into account both the candidate's specialization and development potential, significantly improving the accuracy and comprehensiveness of cross-domain talent recommendations. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating the cross-domain talent mobility recommendation method based on decoupling from a large language model skill graph provided by the present invention.

[0052] Figure 2 This is a schematic diagram of the S40 process provided by the present invention.

[0053] Figure 3 This is a schematic diagram of the S80 process provided by the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0056] Please see Figure 1 The present invention provides a method for cross-domain talent mobility recommendation based on decoupling of a large language model skill graph, comprising the following steps:

[0057] S10: Based on a large language model, decouple the candidate's skills to determine the candidate's set of explicit atomic abilities and the proficiency of each explicit atomic ability;

[0058] S20: Obtain the atomic skill proficiency requirements for the target position and the importance weight coefficient of the atomic skills to the target position. Based on the candidate's explicit atomic skill proficiency, the atomic skill proficiency requirements for the target position, and the importance weight coefficient of the atomic skills to the target position, determine the candidate's preliminary job skill matching degree.

[0059] S30: Determine the initial recommendation score for the candidate based on the candidate's preliminary job competency match.

[0060] S40: Adjust the candidate's initial recommendation score based on the number of dominant atomic abilities in the candidate's dominant atomic ability set and the proficiency of the dominant atomic abilities, and output the candidate's comprehensive recommendation score;

[0061] S50: Based on big data, collect candidates' experience data, and then decouple the candidates' experience data through a large language model to determine the candidate's potential atomic ability set and the proficiency of each potential atomic ability;

[0062] S60: Determine the recommended score for a candidate's potential atomic abilities based on the candidate's set of potential atomic abilities and the proficiency level of each potential atomic ability;

[0063] S70: Determine the final recommendation score for the candidate based on the candidate's potential atomic ability recommendation score and the candidate's comprehensive recommendation score;

[0064] S80: Recommend cross-disciplinary talent mobility based on the candidate's final recommendation score;

[0065] Among them, the Large Language Model (LLM) is a deep learning-based artificial intelligence model that has gained powerful natural language understanding, generation, and reasoning capabilities by being trained on massive amounts of text data. In this method, the Large Language Model is used to perform semantic analysis, information extraction, and knowledge decoupling on unstructured text data, thereby achieving refined identification and quantification of candidates' abilities and experiences.

[0066] Skill map decoupling refers to breaking down complex, multi-faceted skills into more basic, atomic ability units and constructing the relationships between these atomic abilities. Through decoupling, the actual ability composition of candidates can be assessed more accurately, avoiding the coarse-grained problem of traditional skill labels and providing a more refined matching basis for cross-domain talent mobility.

[0067] The set of explicit atomic capabilities refers to the set of basic capability units that can be directly observed and evaluated after analyzing and decoupling the candidate's current skills;

[0068] Dominant atomic ability proficiency refers to the degree of mastery or expertise a candidate has in a particular dominant atomic ability, and is usually expressed in a quantitative way.

[0069] The importance weighting coefficient of atomic capabilities for the target position refers to the criticality of different atomic capabilities for the successful performance of the position when assessing the match between the candidate and the target position. It is quantified by the weighting coefficient.

[0070] Experience data refers to a candidate's educational background, work experience, project participation, training records, and other historical information related to their abilities.

[0071] The potential atomic capability set refers to the set of basic capability units that may not yet be reflected in explicit skills but have development potential, identified after in-depth mining and decoupling of candidate experience data through large language models;

[0072] Potential atomic ability proficiency refers to a candidate's mastery or development potential of a particular potential atomic ability, and is usually expressed in a quantitative way.

[0073] Specifically, in S10, based on a large language model, the candidate's skills are decoupled to determine the candidate's set of explicit atomic abilities and the proficiency level of each explicit atomic ability. In one implementation, a rule-based text matching system can be used, with preset keywords and phrases to extract skill information from the candidate's text description and to roughly estimate proficiency based on the frequency of keyword occurrences or contextual information. For example, if the system detects "proficient in Python," it will identify "Python programming" as an explicit atomic ability and assign it a high proficiency level.

[0074] In another implementation, human experts can review candidates' resumes and self-descriptions and manually annotate and assess their proficiency based on a pre-defined list of atomic capabilities. For example, for a software engineer, an expert might identify explicit atomic capabilities such as "Java programming" and "database management" and judge their proficiency based on their project experience.

[0075] Furthermore, in S20, determining the initial match between the candidate's job skills and the requirements can be done manually by the hiring manager or domain expert, who can set the required atomic skills and their proficiency levels for the target position and subjectively assess the importance weight of each atomic skill for the position. For example, a "Senior Data Analyst" position might require a high level of proficiency in "Statistics Knowledge" and a medium level of proficiency in "SQL Query," with "Statistics Knowledge" considered to have a higher importance weight than "SQL Query." Subsequently, the candidate's explicit atomic skill proficiency is compared with the requirements of the target position using a simple weighted average or linear summation method to obtain the initial match.

[0076] Based on this, in S30, the initial recommendation score for the candidate is determined according to the candidate's preliminary job competency match. One implementation is to use a linear mapping function to directly convert the preliminary job competency match into a recommendation score proportionally; for example, a match of 0.8 would result in a recommendation score of 80. Another implementation is to set multiple match ranges, each corresponding to a fixed recommendation score; for example, a match between 0.6 and 0.7 would uniformly receive 60 points.

[0077] Subsequently, in S40, the candidate's overall recommendation score is output. This can be achieved by simply calculating the total number of explicit atomic abilities the candidate possesses and setting a fixed coefficient. This total is then multiplied by the coefficient and added to the initial recommendation score. For example, if a candidate possesses 10 explicit atomic abilities but has a low proficiency in each ability, it may indicate insufficient focus. In this case, 1 point is deducted from each ability, resulting in a reduction of 10 points from the initial recommendation score. Alternatively, if the candidate possesses 10 explicit atomic abilities, each contributing 1 point, then 10 points are added to the initial recommendation score.

[0078] Furthermore, in S50, the set of potential atomic capabilities of the candidate and the proficiency of each potential atomic capability are determined. In one implementation, a large oracle model can be used to extract keywords related to the preset potential capabilities from experience data, and the proficiency of the potential atomic capabilities can be roughly estimated based on the frequency of keyword occurrence. In another implementation, unstructured texts such as the candidate's project reports and work summaries can be manually reviewed, and the potential capabilities contained therein can be judged based on experience. For example, the potential capability of "risk management" can be identified from project management experience, and its proficiency can be subjectively assessed.

[0079] Furthermore, in S60, the recommended score for a candidate's potential atomic ability can be determined by simply summing the candidate's proficiency in potential atomic abilities and multiplying it by a fixed weighting coefficient.

[0080] Based on this, in S70, the final recommendation score for the candidate is determined in two ways: one is to simply take the arithmetic average of the potential atomic ability recommendation score and the overall recommendation score to obtain the final recommendation score; another is to set a fixed ratio, for example, the potential atomic ability recommendation score accounts for 30% and the overall recommendation score accounts for 70%, and then perform a weighted summation; in addition, the proportion of the potential atomic ability recommendation score can also be determined according to the training resources of the target position, so as to determine the candidate's final recommendation score by weighted summation.

[0081] Finally, in S80, cross-domain talent mobility recommendations are made based on the candidates' final recommendation scores. The final recommendation scores of all candidates can be listed, and the top-ranked candidates can be manually selected from the list for recommendation.

[0082] Through the above technical solution, this invention decouples the skills and experiences of candidates using a large language model to quantify the proficiency of explicit and potential abilities, and integrates the focus depth influence coefficient to correct the initial matching score. Finally, it combines the potential ability recommendation score to generate a comprehensive recommendation result. This method breaks through the limitations of the coarse modeling of traditional skill tags, realizes the fine quantification of abilities and the in-depth mining of potential abilities, and takes into account the candidate's specialization and development potential, significantly improving the accuracy and comprehensiveness of cross-domain talent recommendation.

[0083] Preferably, the present invention further proposes the determination of the preliminary job competency matching degree of the candidate, specifically including:

[0084] Through the formula: ;

[0085] Determine the initial match between the candidate's job skills and the position. ;

[0086] In the formula, This refers to the proficiency of candidate i in the kth explicit atomic ability of the same type as the atomic ability required for the target position. This refers to the proficiency requirement of the k-th atomic ability for the target position. This refers to the importance weight coefficient of the k-th atomic ability for the target position, and N refers to the total number of atomic ability proficiency requirements for the target position; where, if candidate i does not have an explicit atomic ability of the same type as the atomic ability required for the target position, then the candidate's explicit atomic ability proficiency is 0.

[0087] Among them, the kth atomic ability proficiency requirement of the target position represents the standard level of the target position's requirement for that atomic ability, which is usually set by the job description, industry standards or expert experience.

[0088] The importance weight coefficient of the kth atomic capability to the target position reflects the criticality of the atomic capability in the target position. This weight coefficient can be set by methods such as expert scoring, analytic hierarchy process (AHP) or machine learning model training.

[0089] Specifically, this invention quantifies and compares candidate i's proficiency in various explicit atomic skills with the requirements of the target position, and then performs a weighted summation based on the importance weights of each skill. This calculation method can accurately assess the degree of matching between candidate i and the target position in terms of explicit atomic skills, ensuring the objectivity and accuracy of the initial matching assessment. By incorporating the refined skill proficiency decoupled from the large language model into this quantitative model, this scheme can provide a more scientific and reliable foundation for subsequent recommendation score calculation, thereby improving the effectiveness of the entire cross-domain talent mobility recommendation method.

[0090] Through the above technical solution, the present invention significantly improves the accuracy and reliability of the initial job competency matching degree, laying a solid foundation for subsequent recommendation score calculation, thereby improving the effectiveness and reliability of the entire cross-domain talent mobility recommendation method.

[0091] Preferably, the present invention further proposes the determination of the preliminary recommendation score for candidates, specifically including:

[0092] Through the formula: ;

[0093] Determine the preliminary recommendation score for candidates ;

[0094] In the formula, This refers to the initial match between candidate i's job skills and the candidate's qualifications. This refers to the matching sensitivity adjustment coefficient;

[0095] The above formula is used to non-linearly map the initial match between a candidate's job competence and their performance to an initial recommendation score. This non-linear mapping better simulates the complex relationship between the match and the recommendation score in real-world scenarios. For example, when the match is low, a small improvement may lead to a large increase in the recommendation score, while when the match is high, further improvements may result in a slower increase in the recommendation score. Besides using an exponential function, other non-linear functions, such as the Sigmoid function, the Tanh function, or a piecewise linear function, can be used to achieve different mapping characteristics and sensitivity curves.

[0096] The initial job competency matching degree of candidate i is calculated based on candidate i's explicit atomic competency proficiency, the atomic competency proficiency requirements of the target position, and the importance weight coefficient of the atomic competencies to the target position. It represents the initial degree of fit between candidate i and the target position in terms of explicit competencies. Its value is usually between 0 and 1, with a higher value indicating a higher degree of matching.

[0097] The matching sensitivity adjustment coefficient refers to the coefficient used to adjust how sensitive the initial recommendation score is to changes in the initial match between job skills and qualifications. When the matching sensitivity adjustment coefficient is large, the initial recommendation score is more sensitive to changes in the match, meaning that small changes in the match can lead to significant fluctuations in the recommendation score. Conversely, when the matching sensitivity adjustment coefficient is small, the recommendation score is relatively insensitive to changes in the match. This coefficient can be optimized through historical data analysis, expert experience, or machine learning algorithms to adapt to the recommendation needs of different industries and job positions. For example, for highly specialized positions, a larger matching sensitivity adjustment coefficient can be set to highlight subtle differences in the match; for more general positions, a smaller matching sensitivity adjustment coefficient can be set to accommodate a wider range of candidates.

[0098] Specifically, after obtaining the initial job competency matching degree of candidate i, the system utilizes the characteristics of an exponential function to allow the initial recommendation score to change non-linearly with the change in the initial job competency matching degree of candidate i. For example, when the initial job competency matching degree of candidate i is low, even a small increase in the initial job competency matching degree may result in a relatively significant increase in the initial recommendation score, thus encouraging consideration of candidates with potential low matching degrees but room for improvement. Conversely, when the initial job competency matching degree of candidate i is already high, further improvements may contribute less to the initial recommendation score, avoiding the problem of over-saturation of the recommendation score after the matching degree reaches a certain level. Furthermore, by adjusting the matching degree sensitivity adjustment coefficient, the sensitivity of this non-linear transformation can be flexibly controlled, allowing the system to adjust the influence weight of the matching degree on the recommendation score according to the characteristics of the specific job or recommendation strategy. For example, for positions with strict requirements, the sensitivity adjustment coefficient for matching degree can be increased, so that even small differences in matching degree can be reflected in a larger difference in the recommendation score; for positions with higher tolerance for error, the sensitivity adjustment coefficient for matching degree can be decreased, so that when the matching degree fluctuates within a certain range, the change in recommendation score is less drastic. This mechanism ensures that the initial recommendation score can more accurately and reasonably reflect the fit between the candidate and the target position.

[0099] Through the above technical solution, the present invention helps to improve the differentiation and accuracy of the preliminary recommendation score, so that the recommendation results can more accurately reflect the actual suitability of the candidate, thereby laying a more solid foundation for the subsequent comprehensive recommendation score calculation, and thus improving the effectiveness of cross-domain talent mobility recommendation.

[0100] For preferred options, please refer to [link / reference]. Figure 2 The present invention further proposes a specific method for outputting the comprehensive recommendation score of the candidate, including:

[0101] S41: Determine the focus depth influence coefficient of the candidate based on the number of dominant atomic abilities in the candidate's set of dominant atomic abilities and the proficiency of the dominant atomic abilities.

[0102] S42: Adjust the candidate's initial recommendation score based on the candidate's focus depth influence coefficient, and output the candidate's comprehensive recommendation score;

[0103] The focus depth influence coefficient for determining a candidate aims to quantify the degree of specialization or skill concentration of a candidate within their set of explicit atomic abilities. The concept is to assess whether a candidate possesses significant strengths or expertise in a specific skill area, rather than simply having a broad range of skills. This coefficient can be determined in various ways. For example, it can be based on counting the number of atomic abilities among the candidate's explicit atomic abilities that reach a preset high proficiency threshold and comparing them with the total number of atomic abilities. Alternatively, it can be based on analyzing the distribution of proficiency in the candidate's various explicit atomic abilities, such as calculating the variance or concentration index of the proficiency distribution to reflect the degree of concentration of their skills.

[0104] Adjusting the initial recommendation score for candidates to output a comprehensive recommendation score aims to incorporate the candidate's professional depth information into the recommendation score, resulting in a more comprehensive and accurate overall recommendation score. There are various ways to adjust the initial recommendation score. For example, a weighted average method can be used, incorporating the focus depth influence coefficient as one of the weights, combined with the initial recommendation score. Alternatively, a non-linear function can be designed to gain or lose weight on the initial recommendation score based on the magnitude of the focus depth influence coefficient, thus highlighting or weakening the impact of professional depth on the recommendation result.

[0105] Specifically, after the initial recommendation score has been calculated by matching the candidate's explicit atomic competency proficiency with the atomic competency requirements of the target position, this scheme further analyzes the quantity and proficiency of explicit atomic competencies in the candidate's set of explicit atomic competencies to quantify their level of specialization in specific skill areas, generating a focus depth impact coefficient. This focus depth impact coefficient is then used to revise the initial recommendation score. This revision mechanism allows candidates with deep expertise in key skills to receive higher overall recommendation scores, rather than simply those with broad but insufficient skill coverage. In this way, this scheme can more accurately identify candidates with unique advantages and potential in cross-domain mobility, ensuring that the recommendation results consider not only the breadth of skill matching but also the depth of skill matching, thereby improving the effectiveness and accuracy of talent recommendation.

[0106] Through the above technical solution, this application introduces a focus depth influence coefficient, which enables the recommendation results to more comprehensively reflect the candidate's skill structure. This allows for more effective identification of candidates with unique professional advantages who can quickly adapt to new fields and play a key role, thereby improving the accuracy and practicality of talent recommendation and avoiding the loss of outstanding talents due to neglecting professional depth.

[0107] Preferably, the present invention further proposes determining the focus depth influence coefficient of a candidate based on the number of dominant atomic abilities in the candidate's dominant atomic ability set and the proficiency of the dominant atomic abilities, specifically including:

[0108] Through the formula: ;

[0109] Determine the candidate's focus depth impact coefficient ;

[0110] In the formula, This refers to the proficiency of the j-th dominant atom in candidate i. This refers to the threshold for atomic ability proficiency. This refers to the number of dominant atoms in candidate i. For indicator functions, when the condition When it was established The value is 1 if it is set to 1, otherwise the value is 0.

[0111] The above formula is used to quantify the focus depth influence coefficient of candidate i. Its function is to reflect the candidate's professional depth in the skill area by statistically analyzing the number of explicit atomic abilities that candidate i possesses that reach a specific proficiency threshold and comparing them with the total number of explicit atomic abilities.

[0112] The atomic ability proficiency threshold refers to a preset value used to determine whether a candidate's explicit atomic ability has reached the level of "mastery" or "proficiency". Only atomic abilities with proficiency at or above this threshold will be considered in the depth of focus assessment. This threshold can be adjusted based on industry standards, job requirements, or experience data. For example, it can be set to 0.7 (when the proficiency range is 0-1), meaning that only atomic abilities with a proficiency of 70% or higher are considered to effectively contribute to the depth of focus. Of course, different proficiency thresholds can also be set for different atomic abilities based on their importance or scarcity.

[0113] The number of explicit atomic capabilities of candidate i refers to the total number of explicit atomic capabilities identified after candidate i is decoupled through the large language model. This number reflects one aspect of the candidate's skill breadth. For example, if candidate i is decoupled into three explicit atomic capabilities: "Python programming", "data analysis", and "machine learning", then the number of explicit atomic capabilities of candidate i is 3.

[0114] An indicator function is a mathematical function used to convert the result of a conditional judgment into a numerical value. The indicator function takes the value 1 when the internal condition is true; otherwise, it takes the value 0. Its purpose is to count the atomic abilities that meet the proficiency threshold, while ignoring those that do not.

[0115] In this application, to more precisely assess the candidate's professional depth and thus reasonably adjust the initial recommendation score, this method introduces a focus depth influence coefficient. The determination of this coefficient first relies on evaluating the proficiency of each explicit atomic ability of candidate i. By setting an atomic ability proficiency threshold, this method uses an indicator function to judge each explicit atomic ability: if its proficiency reaches or exceeds the threshold, the ability is considered to contribute to the candidate's focus depth, and the indicator function outputs 1; otherwise, it outputs 0. Subsequently, the results of processing all explicit atomic abilities through the indicator function are summed to obtain the total number of explicit atomic abilities possessed by candidate i that reach the proficiency threshold. Finally, this number is divided by the total number of explicit atomic abilities possessed by candidate i, i.e., the total number of explicit atomic abilities possessed by candidate i, thereby calculating the focus depth influence coefficient. This calculation method cleverly combines the "quality" and "quantity" of the candidate's skills. It not only considers how many explicit atomic abilities the candidate has mastered, but more importantly, it quantifies how many of these abilities have truly reached the level of "mastery" or "proficiency." In this way, this method can distinguish between candidates with broad but shallow skills and those with deep expertise in a specific field. For example, a candidate with 10 explicit atomic skills but only 2 at a high proficiency level will have a lower focus depth impact coefficient than a candidate with 5 explicit atomic skills but 4 at a high proficiency level. This quantification of focus depth allows subsequent adjustments to the initial recommendation score to more accurately reflect the candidate's actual professional ability, avoiding the bias of judging solely based on the number of skills or average proficiency, thus providing a more insightful basis for cross-disciplinary talent mobility recommendations.

[0116] Through the above technical solution, this application can effectively distinguish whether a candidate's skills are broad but not specialized or deeply ingrained, thus avoiding the limitations of judging solely based on the number of skills or average proficiency. This allows for a more accurate reflection of a candidate's professional ability and potential when adjusting the initial recommendation score, especially in cross-disciplinary talent mobility recommendation scenarios. It can more effectively identify candidates with core competitiveness in specific fields, thereby improving the accuracy and effectiveness of talent recommendation.

[0117] Preferably, the present invention further proposes adjusting the preliminary recommendation score of a candidate based on the candidate's focus depth influence coefficient, and outputting a comprehensive recommendation score for the candidate, specifically including:

[0118] Through the formula: ;

[0119] Output the overall recommendation score of the candidates ;

[0120] In the formula, This refers to the focus depth influence coefficient of candidate i. This refers to the initial recommendation score for candidate i. This refers to the scaling factor that affects the depth of focus; p is the adjustment factor. The function of the item is to... The value range is adjusted to ±1;

[0121] The above formula is used to combine the initial recommendation score of candidate i with its focus depth influence coefficient to calculate the comprehensive recommendation score of candidate i. The formula introduces an adjustment term so that the initial recommendation score can be amplified or reduced according to the focus depth.

[0122] Among them, the transformation term The goal is to map the original focus depth influence coefficient to a range of -1 to 1. This transformation allows the focus depth influence coefficient to have a two-way impact on the initial recommendation score: when the focus depth influence coefficient is high, the adjustment item is positive, increasing the initial recommendation score; when the focus depth influence coefficient is low, the adjustment item is negative, decreasing the initial recommendation score.

[0123] The adjustment coefficient p can be a positive integer, used to adjust the degree of nonlinearity of the conversion.

[0124] Specifically, firstly, the system obtains a preliminary recommendation score for candidate i, which initially reflects the match between the candidate and the target position's explicit atomic capabilities. Simultaneously, based on the number of explicit atomic capabilities in the candidate's set and their proficiency, the system calculates a focus depth influence coefficient for candidate i. This coefficient quantifies the professional depth of the candidate's skills. To allow the focus depth influence coefficient to flexibly adjust the preliminary recommendation score, the system utilizes a transformation term... Will The value range of is adjusted from [0,1] to [-1,1]. Then, the weight of this adjustment term is adjusted by using the scaling factor that affects focus depth, multiplied by the initial recommendation score, and summed to obtain the comprehensive recommendation score. This mechanism allows the recommendation system to consider not only the breadth of the candidate's skills but also their depth and focus, effectively solving the problem that the initial recommendation score fails to fully reflect the candidate's skill focus depth, resulting in a more comprehensive and accurate final recommendation.

[0125] Through the aforementioned technical solution, this application introduces a focus depth influence coefficient and employs a specific adjustment formula. This allows the recommendation system to consider not only the match between candidates' explicit atomic abilities and the target position, but also the depth of their professional skills and level of focus when evaluating candidates. This enables the comprehensive recommendation score to more accurately reflect the candidate's actual abilities and potential, avoiding evaluation biases caused by skills that are broad but not specialized or specialized but not broad. Furthermore, by introducing a focus depth influence scaling coefficient, the influence weight of focus depth on the final recommendation result can be flexibly adjusted according to the characteristics of different positions or recommendation strategies, further improving the flexibility and accuracy of the recommendation. This provides a more reliable and refined recommendation basis for cross-disciplinary talent mobility.

[0126] Preferably, the present invention further proposes the following method for determining the recommended score for the potential atomic capabilities of candidates:

[0127] Through the formula: ;

[0128] Determine the recommended score for the candidate's potential atomic ability ;

[0129] In the formula, This refers to the proficiency of the r-th potential atomic ability of candidate i. This refers to the importance weight coefficient of candidate i's r-th potential atomic ability for the target position. This refers to the r-th atomic ability proficiency requirement for the target position, and N refers to the total number of atomic ability proficiency requirements for the target position. If candidate i does not have any potential atomic ability of the same type as the atomic ability required for the target position, then the candidate's explicit atomic ability proficiency is 0.

[0130] The candidate's potential atomic competence recommendation score aims to quantify the degree of match between the candidate's potential atomic competences and the requirements of the target position, and to transform this into a score that can be used for recommendation decisions. Its purpose is to compensate for the shortcomings of assessments that may arise from considering only explicit atomic competences, and to more comprehensively reflect the candidate's overall value. Furthermore, this score can be used as an independent evaluation indicator, or it can be weighted and combined with other scores to form the final recommendation basis. For example, it can serve as an important measure of a candidate's potential to adapt to a new field or a new position, or as a predictive indicator of a candidate's future development potential.

[0131] The above formula is used to calculate the recommended score of the potential atomic ability of candidate i. The formula adopts the form of an exponential function, which can non-linearly map the potential ability matching degree to the recommended score. This makes the score increase more gradual as the matching degree is higher, thereby avoiding the excessive influence of extreme values ​​on the overall evaluation.

[0132] The r-th potential atomic ability proficiency of candidate i in the formula represents the degree of mastery of candidate i in a specific potential atomic ability. Potential atomic ability proficiency can be obtained in various ways. For example, by using a large language model to conduct in-depth analysis and reasoning on the candidate's project experience, learning experience, hobbies and other indirect skill descriptions, to identify atomic abilities that the candidate may possess but have not explicitly stated, and to assess their proficiency; or by using an expert system combined with an industry knowledge graph to mine the candidate's background information and infer their potential abilities.

[0133] The importance weight coefficient of candidate i's r-th potential atomic ability to the target position reflects the degree of demand and importance the target position places on a specific potential atomic ability. Different potential atomic abilities have different values ​​for different target positions. For example, for a position that requires rapid learning of new technologies, the weight of the potential atomic ability of learning ability will be higher; for a position that requires innovative thinking, the weight of the potential atomic ability of innovation ability will be higher. This weight coefficient can be set by job experts based on job descriptions and industry trends, or it can be obtained by analyzing and learning from historical data through machine learning models.

[0134] The total number of atomic competency proficiency requirements for the target position represents the total categories or items of atomic competencies that need to be assessed for the target position. In addition, if candidate i does not have any potential atomic competencies of the same type as the atomic competencies required for the target position, the candidate's corresponding explicit atomic competency proficiency is 0. This calculation rule ensures that when a candidate lacks a specific potential atomic competency, that competency contributes zero to the total score, thereby avoiding evaluation bias due to missing data.

[0135] Specifically, the core of the above formula lies in constructing a weighted summation term. This term comprehensively considers the ratio of each potential atomic skill proficiency of candidate i to the atomic skill proficiency requirement of the target position, multiplied by the importance weight coefficient of that potential atomic skill to the target position. This design allows potential skills that are highly matched to the requirements of the target position and are important to the position to contribute more to the summation term. When candidate i has no potential atomic skills of the same type as the atomic skills required by the target position, its corresponding proficiency is set to 0, ensuring the accuracy of the calculation. Subsequently, this weighted summation term undergoes a non-linear transformation through an exponential function, where a sensitivity adjustment coefficient can adjust the influence curve of the potential skill matching degree on the final score. This non-linear transformation causes the score to increase rapidly when the matching degree is low, while the increase tends to be slower when the matching degree is high, thus better reflecting the marginal contribution of potential skills to the recommended score. In this way, this application can transform the candidate's hidden, unexpressed potential skills into quantified recommended scores, thereby more comprehensively assessing the candidate's adaptability and development potential in cross-domain mobility. This effectively complements schemes that rely solely on explicit atomic capabilities for evaluation, enabling recommendation results to focus not only on what candidates "can do now" but also on "what they might do in the future," greatly enhancing the depth and breadth of cross-disciplinary talent mobility recommendations.

[0136] Through the aforementioned technical solution, this application can quantify and assess a candidate's potential, unexpressed, atomic abilities and convert them into recommendation scores. This allows for the consideration of not only a candidate's existing explicit skills but also their potential abilities in learning, adaptation, and innovation when making recommendations for cross-disciplinary talent mobility. This comprehensive assessment method effectively avoids the "regret of overlooking gems" that may result from focusing solely on explicit abilities, enabling the identification and recommendation of candidates with strong learning abilities and adaptability, even those whose current explicit skills do not perfectly match the target position. Therefore, this application can significantly improve the accuracy and comprehensiveness of cross-disciplinary talent mobility recommendations, providing strong support for enterprises to discover and cultivate multi-skilled talents with development potential, thereby promoting the optimal allocation and efficient flow of human resources.

[0137] Preferably, the present invention further proposes determining the final recommendation score of a candidate based on the candidate's potential atomic capability recommendation score and the candidate's comprehensive recommendation score, specifically including:

[0138] Through the formula: ;

[0139] Determine the final recommendation score for each candidate. ;

[0140] In the formula, This refers to the recommended score for the potential atomic ability of candidate i. This refers to the overall recommendation score for candidate i. This refers to the weighting coefficient based on training resources for the target position;

[0141] The weighting coefficient based on training resources for the target position is used to adjust the relative importance of the potential atomic ability recommendation score and the comprehensive recommendation score in the final recommendation score. It reflects the target position's willingness and ability to invest in talent development, as well as its emphasis on the candidate's future development potential. This coefficient can be set according to the amount of training resources that the target position can provide. For example, the richer the training resources, the larger the coefficient can be set to increase the proportion of the potential atomic ability recommendation score in the final score. In addition, this coefficient can also be dynamically adjusted according to factors such as the strategic needs of the target position, talent reserves, or industry development trends. For example, for positions that urgently need innovative talents, this coefficient can be appropriately increased to give priority to candidates with high potential.

[0142] Specifically, the potential atomic ability recommendation score for candidate i measures the candidate i's potential to be competent for the target position after training or development, while the comprehensive recommendation score for candidate i reflects the level of explicit abilities that candidate i currently possesses and can be directly applied to the target position. By weighting and summing these two scores and adjusting them using weighting coefficients, the final recommendation score can flexibly reflect the different emphases of the target position on the candidate's potential and explicit abilities. When the target position has abundant training resources, the weighting coefficient can be appropriately increased. The value of the potential atomic ability recommendation score is increased, thereby increasing its weight in the final recommendation score, making the system more inclined to recommend candidates with higher development potential. Conversely, when the target position requires high on-the-job skills and training resources are limited, the weighting coefficient can be reduced. The value of the recommendation score is adjusted to increase its weight in the overall recommendation score, prioritizing candidates whose current abilities are a better match. This dynamic adjustment mechanism allows talent recommendation results to better adapt to the specific needs of different target positions, thereby improving the accuracy and effectiveness of cross-disciplinary talent mobility recommendations.

[0143] Through the aforementioned technical solution, this application can flexibly adjust the weights of the candidate's potential atomic ability recommendation score and comprehensive recommendation score in the final recommendation score based on the training resources available for the target position. This allows the final recommendation result to more accurately match the actual needs of the target position. This effectively solves the problem of inaccurate recommendation results in cross-domain talent mobility recommendations due to insufficient consideration of the differences in training resources for target positions. By introducing a weighting coefficient based on the training resources of the target position, this application can better balance the candidate's current ability and future potential. This enables the recommendation system to not only identify candidates with matching current abilities but also those with high development potential. Especially when the target position is willing to invest resources in talent development, this greatly enhances the flexibility and adaptability of talent recommendations, thereby improving talent mobility efficiency and job matching.

[0144] For preferred options, please refer to [link / reference]. Figure 3 The present invention further proposes the cross-domain talent mobility recommendation based on the candidate's final recommendation score, specifically including:

[0145] S81: Sort all candidates by their final recommendation scores and output the sequence of final recommendation scores for each candidate.

[0146] S82: Recommend cross-disciplinary talent mobility based on the final recommendation score sequence of candidates;

[0147] The process involves sorting the final recommendation scores of all candidates. This operation aims to establish a clear priority or ranking based on the final recommendation score obtained by each candidate. Its purpose is to integrate the scattered individual evaluation results into an ordered whole, thereby intuitively showing the relative strengths and weaknesses of different candidates. The implementation methods may include, but are not limited to: using standard data sorting algorithms, such as quicksort, mergesort, or heapsort, to arrange the final recommendation scores of all candidates in descending or ascending order; or using the sorting function provided by the database management system to query the stored candidate score data and return it in a specified order.

[0148] Outputting the final recommendation score sequence of candidates involves presenting the sorted candidates and their corresponding final recommendation scores in a structured format. This provides a clear and ordered data list that can be directly used for subsequent recommendation decisions. Implementation methods can include, but are not limited to: generating a list, array, or table data structure from the sorted results, containing candidate identification information (such as ID or name) and their corresponding final recommendation scores; or visually presenting the sequence on a user interface for manual review or further processing, or storing it in a file or database for use by automated systems.

[0149] Cross-domain talent mobility recommendations are made based on the final recommendation score sequence of candidates. This operation aims to perform specific cross-domain talent mobility recommendations based on the ranked final recommendation score sequence of candidates. Its purpose is to transform the aforementioned evaluation and ranking results into actual talent mobility suggestions or decisions. The implementation methods may include, but are not limited to: setting a recommendation quantity threshold (e.g., recommending the top Z candidates) and directly selecting the top-ranked candidates in the sequence for recommendation; or setting a recommendation score threshold and including all candidates with a final recommendation score higher than the threshold in the recommendation scope; or providing the sequence to the recruiter or talent management department as a reference for talent screening and matching.

[0150] After calculating the final recommendation score for each candidate, this application's solution first ranks all candidates' final recommendation scores to translate these quantified evaluation results into actual recommendation actions. This ranking process places all candidates within a unified evaluation system, clarifying their relative competitiveness for the target positions. Subsequently, the system outputs a sequence of final recommendation scores for candidates, which directly reflects the ranking results and provides a clear and structured data foundation for subsequent recommendation decisions. Finally, based on this ordered sequence, the system can conduct targeted cross-domain talent mobility recommendations. For example, it can select the most suitable candidates from the sequence based on preset recommendation numbers or score criteria. In this way, this application's solution decouples complex skill decoupling and matching calculation results, transforming them into an intuitive and actionable recommendation list through ranking and sequence output. This effectively solves the problem of how to translate abstract scores into concrete recommendation actions, ensuring the objectivity and efficiency of the talent recommendation process.

[0151] Through the above technical solutions, this application not only improves the transparency and efficiency of the recommendation process, but also ensures that outstanding talents based on comprehensive evaluation can be recommended first, thereby optimizing the decision-making process for cross-domain talent mobility and making talent matching more accurate and efficient.

[0152] Preferably, the present invention further proposes to sort all candidates' final recommendation scores and output a sequence of candidates' final recommendation scores, specifically including:

[0153] When sorting all candidates' final recommendation scores, arrange them in descending order of their final recommendation scores and output the sequence of final recommendation scores.

[0154] With a preset number of recommended candidates Z, the candidates corresponding to the top Z final recommendation scores in the candidate final recommendation score sequence are selected for cross-domain talent mobility recommendations.

[0155] The preset number of recommendations Z can be a fixed value, such as 5 or 10 people depending on the recruitment plan; or it can be a dynamic value, such as being adjusted in real time based on the urgency of the target position, the size of the candidate pool, or the historical recommendation results.

[0156] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cross-domain talent mobility recommendation method based on decoupling from a large language model skill graph, characterized in that, The method specifically includes: Based on a large language model, the candidate's skills are decoupled to determine the candidate's set of explicit atomic abilities and the proficiency of each explicit atomic ability. Obtain the required atomic skill proficiency for the target position and the importance weight coefficient of the atomic skill to the target position. Based on the candidate's explicit atomic skill proficiency, the required atomic skill proficiency for the target position, and the importance weight coefficient of the atomic skill to the target position, determine the candidate's initial job skill matching degree. Based on the candidate's initial job competency match, determine the candidate's initial recommendation score; Based on the number of dominant atomic abilities in the candidate's dominant atomic ability set and the proficiency of the dominant atomic abilities, the candidate's initial recommendation score is adjusted, and the candidate's comprehensive recommendation score is output. Based on big data, we collect candidates’ experience data, and then decouple the candidates’ experience data through a big language model to determine the candidates’ potential atomic ability set and the proficiency of each potential atomic ability. The recommended score for a candidate's potential atomic ability is determined based on the candidate's set of potential atomic abilities and the proficiency level of each potential atomic ability. The final recommendation score for each candidate is determined based on their potential atomic ability recommendation score and their overall recommendation score. Cross-disciplinary talent mobility recommendations will be made based on the candidates' final recommendation scores.

2. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 1, characterized in that, The determination of the preliminary job competency match of the candidate specifically includes: Through the formula: ; Determine the initial match between the candidate's job skills and the position. ; In the formula, This refers to the proficiency of candidate i in the kth explicit atomic ability of the same type as the atomic ability required for the target position. This refers to the proficiency requirement of the k-th atomic ability for the target position. , refers to the importance weight coefficient of the k-th atomic ability for the target position, and N refers to the total number of atomic ability proficiency requirements for the target position; where, if candidate i does not have an explicit atomic ability of the same type as the atomic ability required for the target position, then the candidate's explicit atomic ability proficiency is 0.

3. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 1, characterized in that, The determination of the preliminary recommendation score for candidates specifically includes: Through the formula: ; Determine the preliminary recommendation score for candidates ; In the formula, This refers to the initial match between candidate i's job skills and the candidate's qualifications. This refers to the matching sensitivity adjustment coefficient.

4. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 1, characterized in that, The specific methods for outputting the candidate's comprehensive recommendation score include: The focus depth influence coefficient of a candidate is determined based on the number of dominant atomic abilities in the candidate's set of dominant atomic abilities and the proficiency of the dominant atomic abilities. Based on the candidate's focus depth influence coefficient, the initial recommendation score for the candidate is adjusted, and the overall recommendation score for the candidate is output.

5. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 4, characterized in that, The determination of the candidate's focus depth influence coefficient based on the number of dominant atomic abilities in the candidate's dominant atomic ability set and the proficiency of those dominant atomic abilities specifically includes: Through the formula: ; Determine the candidate's focus depth impact coefficient ; In the formula, This refers to the proficiency of the j-th dominant atom in candidate i. This refers to the threshold for atomic ability proficiency. This refers to the number of dominant atoms in candidate i. For indicator functions, when the condition When it was established The value is 1 if it is not 1, otherwise the value is 0.

6. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 4, characterized in that, The process of adjusting the initial recommendation score of a candidate based on the candidate's focus depth influence coefficient and outputting the candidate's comprehensive recommendation score specifically includes: Through the formula: ; Output the overall recommendation score of the candidates ; In the formula, This refers to the focus depth influence coefficient of candidate i. This refers to the initial recommendation score for candidate i. This refers to the scaling factor that affects the depth of focus; p is the adjustment factor. The function of the item is to... The value range is adjusted to ±1.

7. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 1, characterized in that, The determination of the candidate's potential atomic capability recommendation score specifically includes: Through the formula: ; Determine the recommended score for the candidate's potential atomic ability ; In the formula, This refers to the proficiency of the r-th potential atomic ability of candidate i. This refers to the importance weight coefficient of candidate i's r-th potential atomic ability for the target position. This refers to the r-th atomic ability proficiency requirement for the target position, and N refers to the total number of atomic ability proficiency requirements for the target position. If candidate i does not have any potential atomic ability of the same type as the atomic ability required for the target position, then the candidate's explicit atomic ability proficiency is 0.

8. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 1, characterized in that, The process of determining the final recommendation score for a candidate based on their potential atomic capability recommendation score and overall recommendation score includes: Through the formula: ; Determine the final recommendation score for each candidate. ; In the formula, This refers to the recommended score for the potential atomic ability of candidate i. This refers to the overall recommendation score for candidate i. This refers to the weighting coefficient of training resources based on the target position.

9. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 1, characterized in that, The cross-disciplinary talent mobility recommendation based on the candidate's final recommendation score specifically includes: Sort all candidates by their final recommendation scores and output the sequence of final recommendation scores for each candidate. Cross-disciplinary talent mobility recommendations are made based on the final recommendation score sequence of candidates.

10. The cross-domain talent mobility recommendation method based on large language model skill graph decoupling according to claim 9, characterized in that, The process of sorting all candidates' final recommendation scores and outputting a sequence of final recommendation scores specifically includes: When sorting all candidates' final recommendation scores, arrange them in descending order of their final recommendation scores and output the sequence of final recommendation scores. The number of recommended candidates is preset to Z. The candidates corresponding to the top Z final recommendation scores in the candidate final recommendation score sequence are selected for cross-domain talent mobility recommendation.

Images