Matching recommendation method and system based on large model digital talent inventory

Abstract

The invention provides a matching recommendation method and system based on large-model digital talent inventory, relates to the technical field of artificial intelligence and human resource management, integrates large-model analysis, employee portrait modeling, intelligent matching recommendation and visual display of comparative analysis, has the characteristics of automation, high precision and interpretability, and is suitable for popularization and application. The problems of difficult man-post matching, inaccurate talent evaluation, opaque selection process and the like are solved, and digital talent selection and intelligent decision making are realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and human resource management technology, and in particular to a matching recommendation method and system based on large-scale model digital talent inventory. Background Art

[0002] When implementing internal talent selection mechanisms, companies currently face the following systemic challenges:

[0003] First, job descriptions are often expressed in natural language, lacking structured and standardized processing methods, resulting in inefficient information extraction and matching. Second, the diverse human resources data within companies is decentralized, and the lack of a unified standardization system across various data sources makes effective integration and collaborative application difficult.

[0004] At the selection and decision-making level, the existing mechanism overly relies on the subjective judgment of the HR department or supervisors, which not only results in significant subjective biases but also leaves the overall selection efficiency in a critical need for improvement. Furthermore, the system lacks intelligent recommendation capabilities based on scientific algorithms, failing to provide explainable criteria for talent matching and ensuring scientific and transparent decision-making. Furthermore, regarding data security management, sensitive operations such as the export of key data lack comprehensive security control mechanisms, posing potential information security risks.

[0005] While some human resource management systems already possess basic data integration capabilities, significant deficiencies remain at the overall solution level. Specifically, they lack a comprehensive semantic understanding system, systematic talent profiling capabilities, and an effective closed-loop between intelligent matching algorithms and visualization recommendation mechanisms. These factors collectively hinder the systematic and intelligent development of corporate talent selection. Summary of the Invention

[0006] The purpose of the present invention is to provide a matching recommendation method and system based on large-scale digital talent inventory to solve at least one of the above problems.

[0007] In a first aspect, the present invention provides a matching recommendation method based on a large-scale digital talent inventory, comprising:

[0008] Obtaining a job requirement text and multi-dimensional data of at least one candidate employee; wherein the job requirement text is described in natural language, and the multi-dimensional data includes basic information data, behavioral resume data, performance evaluation data, potential inventory data, and ability assessment data;

[0009] Use a large language model to perform semantic analysis on the job requirement text and generate a job profile vector through normalization. The job profile vector is composed of the normalized weights corresponding to the structured label set.

[0010] Perform label mapping and standardization on each field of each candidate employee's multi-dimensional data to obtain an employee portrait vector, which is composed of scores corresponding to a structured label set.

[0011] Based on the matching degree between the employee profile vector and the job profile vector of each candidate employee and combined with the multi-dimensional data of each candidate employee, a first recommendation list and an explainable first recommendation reason are generated;

[0012] When a comparative analysis request is received for at least one first employee in the first recommendation list, a visual display of the comparative analysis is performed based on the employee portrait vector and job portrait vector of each first employee. The visual display of the comparative analysis includes one or more of a prominent display of label hit points, a visual display of label score comparison, a visual display of the matching degree of label dimensions, and a visual display of the contribution of labels to the total matching degree.

[0013] In an optional embodiment, the structured tag set is divided into multiple first-class tags; based on the matching degree between the employee profile vector and the job profile vector of each candidate employee and combined with the multi-dimensional data of each candidate employee, a first recommendation list and an explainable first recommendation reason are generated, including:

[0014] Based on the employee profile vector and job profile vector of each candidate employee, calculate the label hit rate, total matching degree, and matching degree of each first-class label corresponding to each candidate employee;

[0015] The comprehensive score of each candidate employee is calculated based on the matching degree and weight coefficient of each category of labels corresponding to each candidate employee;

[0016] Generate a first recommendation list based on the tag hit rate, total matching degree and comprehensive score corresponding to each candidate employee;

[0017] A first recommendation reason for the second employee is generated based on one or more of the tag hit rate, total matching degree, comprehensive score, and multi-dimensional data corresponding to the second employee; wherein the second employee is a preset number of candidate employees ranked highest in the first recommendation list.

[0018] In an optional embodiment, each first-class tag includes at least one second-class tag; the tag hit rate is calculated by the following formula:

[0019]

[0020] Among them, Hit% represents the label hit rate, v i represents the score corresponding to the i-th second-category label in the employee portrait vector, θ represents the preset score threshold, and n represents the number of second-category labels.

[0021] In an optional embodiment, the structured tag set is divided into multiple first-class tags, each first-class tag includes at least one second-class tag; and a visual display of comparative analysis is performed based on the employee portrait vector and the position portrait vector of each first employee, including:

[0022] Determine the second-category labels in the employee portrait vector of the first employee whose scores are greater than a preset score threshold as label hit points of the first employee, and highlight the label hit points of the first employee; and / or,

[0023] When there are multiple first employees, a label matrix is ​​constructed based on the employee portrait vectors of each first employee, where each row of the label matrix corresponds to the employee portrait vector of one first employee; based on the label matrix, a relative difference value of each first employee on each second-category label is calculated, where the relative difference value is used to represent the degree of deviation from the average value; the relative difference value corresponding to each first employee under each second-category label is displayed using a radar chart or a bar chart; and / or,

[0024] Calculate the matching degree of each first-category label corresponding to the first employee based on the employee portrait vector and the job portrait vector of the first employee; display the matching degree of each first-category label corresponding to the first employee through a bar chart; and / or,

[0025] Based on the employee portrait vector and job portrait vector of the first employee, the contribution of each second-category label corresponding to the first employee to the total matching degree is calculated; and the contribution of each second-category label corresponding to the first employee to the total matching degree is displayed using a pie chart.

[0026] In an optional embodiment, the above method further includes:

[0027] Aggregate and count the employee portrait vectors of all members of the target team to obtain the team portrait vector of the target team;

[0028] Based on the organizational fit between each candidate employee's employee profile vector and the target team's team profile vector, combined with each candidate employee's multi-dimensional data, a second recommendation list and explainable second recommendation reasons are generated; wherein the organizational fit includes one or more of the personality complementarity index, Euclidean distance, and cosine similarity;

[0029] When a comparative analysis request is received for at least one third employee in the second recommendation list, a capability complementarity map and a personality diversity index are visualized based on the employee portrait vector of each third employee and the team portrait of the target team. The capability complementarity map indicates the score difference between each third employee and the target team under different label dimensions, and the personality diversity index indicates the degree of personality difference in the target team after the third employee joins the target team.

[0030] In an optional embodiment, the employee portrait vectors of all members of the target team are aggregated and statistically analyzed to obtain the team portrait vector of the target team, including:

[0031] According to the preset aggregation methods corresponding to each label dimension, the label scores of the employee portrait vectors of all members of the target team are aggregated to obtain the team portrait vector of the target team; among them, the aggregation methods corresponding to the ability label include mean, variance, maximum or minimum value; the aggregation methods corresponding to the personality label include the five-dimensional personality average and / or diversity index, the aggregation methods corresponding to the performance data and inventory data include weighted mean, and the aggregation methods corresponding to the label density include label hit rate and / or shared label ratio.

[0032] In an optional embodiment, the above method further includes:

[0033] Regularly generate new dynamic tags based on each employee's behavioral history data, and mark their source type as dynamic;

[0034] Using the preset rule engine and the prompt words of the large language model, AI speculation labels are generated and their source type is marked as AI speculation;

[0035] Displays newly added dynamic tags and AI-inferred tags so that users can update the structured tag set.

[0036] In an optional embodiment, the above method further includes:

[0037] When a data export operation for target data is detected, a secondary authentication of the operation authority is performed; wherein the target data includes one or more of the first recommendation list, the first recommendation reason, and a visual display of comparative analysis;

[0038] After the secondary authentication is passed, a download link for the target data is provided for the user to download, and the relevant operation information is written into the audit log.

[0039] In a second aspect, the present invention provides a matching recommendation system based on a large-scale digital talent inventory, comprising:

[0040] A data acquisition module is used to obtain the job requirement text and multi-dimensional data of at least one candidate employee; the job requirement text is described in natural language, and the multi-dimensional data includes basic information data, behavioral resume data, performance evaluation data, potential inventory data, and ability assessment data;

[0041] The first generation module is used to perform semantic analysis on the job requirement text using a large language model and generate a job profile vector through normalization. The job profile vector is composed of the normalized weights corresponding to the structured label set;

[0042] The second generation module is used to perform label mapping and standardization on each field of the multi-dimensional data of each candidate employee to obtain an employee portrait vector. The employee portrait vector is composed of the scores corresponding to the structured label set;

[0043] The first recommendation module is used to generate a first recommendation list and an explainable first recommendation reason based on the matching degree between the employee profile vector and the job profile vector of each candidate employee and the multi-dimensional data of each candidate employee;

[0044] The first display module is used to perform a visual display of the comparative analysis based on the employee portrait vector and job portrait vector of each first employee when a comparative analysis request for at least one first employee in the first recommendation list is received. The visual display of the comparative analysis includes one or more of the following: highlighting the label hit points, visualizing the label score comparison, visualizing the matching degree of the label dimension, and visualizing the contribution of the label to the total matching degree.

[0045] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method of any one of the aforementioned embodiments is implemented.

[0046] The matching recommendation method and system based on large-scale digital talent inventory provided by the present invention obtains the job requirement text and multi-dimensional data of at least one candidate employee when recommending candidate employees for a position; wherein the job requirement text is described in natural language, and the multi-dimensional data includes basic information data, behavioral resume data, performance evaluation data, potential inventory data and ability assessment data; the job requirement text is semantically parsed using a large language model, and a job portrait vector is generated through normalization processing. The job portrait vector is composed of normalized weights corresponding to a structured label set; label mapping and normalization processing are performed on each field data of the multi-dimensional data of each candidate employee to obtain an employee portrait vector , the employee portrait vector is composed of scores corresponding to a set of structured labels; based on the matching degree between the employee portrait vector and the job portrait vector of each candidate employee, combined with the multi-dimensional data of each candidate employee, a first recommendation list and an explainable first recommendation reason are generated; when a comparative analysis request for at least one first employee in the first recommendation list is received, a visual display of the comparative analysis is performed based on the employee portrait vector and the job portrait vector of each first employee. The visual display of the comparative analysis includes one or more of the following: highlighting the label hit points, visual display of the label score comparison, visual display of the matching degree of the label dimension, and visual display of the label's contribution to the total matching degree. This integrates large-scale model analysis, employee portrait modeling, intelligent matching recommendation, and visual display of comparative analysis. It has the characteristics of automation, high precision, and explainability. It solves the problems of difficult job matching, inaccurate talent evaluation, and opaque selection process, and realizes digital talent selection and intelligent decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A flowchart of a matching recommendation method based on a large-scale digital talent inventory provided by an embodiment of the present invention;

[0049] Figure 2 A flowchart of another matching recommendation method based on large-scale digital talent inventory provided by an embodiment of the present invention;

[0050] Figure 3 A schematic diagram of the structure of a matching recommendation system based on a large-scale digital talent inventory provided by an embodiment of the present invention;

[0051] Figure 4A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] The embodiment of the present invention provides a matching recommendation method and system based on large-model digital talent inventory, and proposes an "AI selection" solution based on natural language processing (NLP) and multi-source talent data fusion. It is used in scenarios such as internal job selection, cadre evaluation, and succession planning within the enterprise. Through job requirement analysis, employee portrait modeling, and intelligent matching recommendations, digital talent selection and intelligent decision-making are achieved.

[0054] To facilitate understanding of this embodiment, a matching recommendation method based on a large-model digital talent inventory disclosed in an embodiment of the present invention is first introduced in detail.

[0055] The embodiment of the present invention provides a matching recommendation method based on a large-scale digital talent inventory. The method can be executed by an electronic device with data processing capabilities and can be implemented in an enterprise's existing human resources management platform or a separately deployed "digital talent system" environment, supporting integration with other enterprise applications. Figure 1 The flowchart of a matching recommendation method based on a large-scale digital talent inventory is shown, and the method mainly includes the following steps S110 to S150:

[0056] Step S110, obtaining a job requirement text and multi-dimensional data of at least one candidate employee; wherein the job requirement text is described in natural language, and the multi-dimensional data includes basic information data, behavioral resume data, performance evaluation data, potential inventory data, and ability assessment data.

[0057] The above-mentioned input job requirement text can be input by the user through a browser or mobile terminal, and the input job requirement text may include job-related information and employment description information, etc. The multi-dimensional data of candidate employees may come from a data warehouse, and the multi-dimensional data may include basic information data, behavioral resume data, performance evaluation data, potential inventory data and ability assessment data, among which basic information data may include education, length of service and rank, etc.; behavioral resume data may include project experience, rotation records and award certificates, etc.; performance evaluation data may include performance level (such as the last 3 times + the most recent before calibration), performance score and target achievement rate, etc.; potential inventory data may include nine-square grid score, inventory year and potential level, etc. The nine-square grid refers to the talent inventory nine-square grid (also known as the talent matrix or performance potential nine-square grid); ability assessment data may include competency and cognitive assessment, etc.

[0058] A possible example of the above multi-dimensional data is shown in Table 1 below.

[0059] Table 1

[0060]

[0061]

[0062] In one possible example, the multi-dimensional data of each candidate employee can be pushed in batches to the digital talent system through an API (Application Programming Interface), or the user can use the system's built-in import function to import the multi-dimensional data of each candidate employee into the digital talent system.

[0063] In step S120, a large language model is used to perform semantic analysis on the job requirement text, and a job portrait vector is generated through normalization processing. The job portrait vector is composed of normalized weights corresponding to a structured tag set.

[0064] LLMs (Large Language Models) such as GPT and Ernie can be used to perform natural language processing and semantic analysis on job requirements described in natural language, outputting a standard structured talent model, i.e., structured labels + weights, for example, "strategic planning ability: 8; overseas project experience: 7; team management: 9". The weights are then normalized to the [0, 1] interval to generate the job profile vector T:

[0065] T={(t1,w1),(t2,w2),...,(t n ,w n )};

[0066] Among them, t i represents the i-th label, wi It represents the normalized weight corresponding to the i-th label, and the value of i is 1, 2,···, n.

[0067] The LLM can be, but is not limited to, GPT-4, ChatGLM, or Wenxinyiyan. Specifically, a natural language description of a job requirement can be sent via an API to an authorized large language model service (such as GPT-4 or Ernie), along with a prompt. For example, consider breaking down the following job requirement into several key talent tags and assigning importance weights (1-10) to each tag: "With more than 5 years of team management experience..." The large language model returns the structured tags and weights in JSON format. The "1-10" values ​​are then converted to the range [0, 1] to obtain the final job profile vector. Users can add or remove tags and fine-tune weights in the user interface, and can also save the job profile vector to "My Talent Model" for future reuse.

[0068] The above-mentioned structured tag set is a system-built-in, uniformly managed talent tag set, which can include static tags, dynamic tags, and AI tags, etc. Among them, static tags are tags that do not change over time and can be generated based on basic information data; dynamic tags are tags that change over time and can be generated based on behavioral resume data, performance evaluation data, potential inventory data, and capability assessment data, etc.; AI tags are tags that are automatically assigned by AI based on employees' multi-dimensional data, and may be repeated with static tags / dynamic tags. AI tags are optional tags, and whether to use them is determined manually. The structured tag set can be divided into multiple first-class tags, and the first-class tags can be further divided into multiple second-class tags. The above-mentioned t i Can belong to the second category of labels.

[0069] The above weight normalization formula can be but is not limited to the following formula:

[0070]

[0071] Among them, s i Indicates the weight corresponding to the i-th label, s max Represents the maximum value of the weights corresponding to all labels, s min Indicates the minimum value among the weights corresponding to all labels.

[0072] Step S130 , label mapping and standardization processing are performed on each field data of the multi-dimensional data of each candidate employee to obtain an employee portrait vector, which is composed of scores corresponding to the structured label set.

[0073] The field data of different field types and different value distributions can be converted into a unified scale label score v i∈[0, 1], so as to be consistent with the normalized weight w of the label in the job portrait vector i Consistent alignment and participation in subsequent matching calculations. The system can automatically label employees based on pre-set dynamic labeling rules (e.g., automatically generating "National Award: ×××" based on "Award Field" + "Award Level"); if an employee's "Overseas Projects" field = "More than 2", the label "Multiple Overseas Project Experiences" is automatically generated; various labels are uniformly mapped to internal label IDs for ease of subsequent algorithm processing; structured scores (e.g., performance = 95) are normalized or Z-scored so that all scores are in the [0,1] range (or -1 to +1). After this stage is completed, each employee has one or more "label-score" pairs or "label-Boolean value" pairs within the system. A fixed-dimensional employee portrait vector can then be formed, and all candidate employees are mapped to the same-dimensional vector (if a label is missing, the score = 0).

[0074] The employee portrait vector is processed as follows: Based on the field type of each field in the multi-dimensional data, the corresponding label and normalization method are determined; each field data is normalized according to the corresponding normalization method to obtain the score corresponding to the corresponding label; and the employee portrait vector is obtained based on the score components corresponding to each label. Normalization methods can include one-hot encoding, multiple-select TF-IDF (Term Frequency-Inverse Document Frequency), numerical field normalization, etc. The employee portrait vector can be expressed as:

[0075] C={(t1,v1),....,(t n , v n )};

[0076] Among them, t i represents the i-th label, v i It represents the score corresponding to the i-th label, and the value of i is 1, 2,···, n.

[0077] Possible examples of normalization methods for different field types include: using Min-Max normalization or Z-score normalization for numeric fields; mapping categorical fields to preset graded scores; converting Boolean fields to 0 / 1 binary values; and using TF-IDF weighting to generate sparse vectors for multiple-choice enumeration fields. This is shown in Table 2 below.

[0078] Table 2

[0079]

[0080] For ease of understanding, some of the above-mentioned standardization methods are described in detail below.

[0081] (1) Min-Max normalization (often used for performance and evaluation):

[0082]

[0083] Among them, X i is the value of the i-th label, X min and X max are the minimum and maximum values ​​in the dataset to which the i-th label belongs, respectively.

[0084] This is applicable to scoring fields with a clear interval and a limited value range. For example, if the performance score is 92 and the overall score range is [60, 100], the normalized score is 0.8.

[0085] (2) Z-score standardization + Sigmoid compression (suitable for evaluating skewed distribution):

[0086]

[0087] Among them, X i is the original value of the i-th label, μ is the mean value of the dataset to which the i-th label belongs, σ is the standard deviation of the dataset to which the i-th label belongs, and Z i is the score of the i-th label after Z-Score standardization.

[0088] This method uses standard deviation normalization and compression to the [0,1] range, making it suitable for normal or long-tailed distributed data. It can be applied to cognitive scores, leadership indices, and more.

[0089] (3) Classification field mapping table:

[0090] An example of the mapping score for an educational level is shown in Table 3 below.

[0091] Table 3

[0092] Education level <![CDATA[Mapping score value v i > PhD 1 master 0.8 undergraduate 0.6 Junior college and below 0.3

[0093] You can also use one-hot encoding and then superimpose the value into the employee portrait vector (for example, an employee has both a "Master's" degree and "overseas project experience").

[0094] (4)Multiple selection field processing:

[0095] TF-IDF + dynamic tag generation: For employees with multiple award, project, or role tags, each tag is modeled separately as a candidate tag dimension. If the system already has a large number of tags, TF-IDF weighting can be used:

[0096]

[0097] Among them, TF 员工，子标签 Indicates the frequency of employees having this subtag, DF 子标签 Indicates the number of people who appear in this subtag, and N is the total number of people.

[0098] The generated labels and their values ​​are automatically included in the employee portrait vector.

[0099] Finally, the processed label-score pairs are combined into a complete portrait vector. The label order is defined by the job portrait vector. If a label does not exist for the employee (for example, no team management experience), the default score is 0.

[0100] For example, the construction method of an employee portrait vector of a certain employee is shown in Table 4 below.

[0101] Table 4

[0102]

[0103]

[0104] Step S140 , based on the matching degree between the employee portrait vector and the job portrait vector of each candidate employee and combined with the multi-dimensional data of each candidate employee, generates a first recommendation list and an explainable first recommendation reason.

[0105] In some possible embodiments, step S140 may include: calculating the total match between the employee profile vector and the job profile vector of each candidate employee, and sorting the candidate employees in descending order of total match, obtaining and displaying a first recommendation list, wherein the first recommendation list may include, among other things, the candidate employee's name and total match; generating and displaying a first recommendation reason for the second employee based on the total match and multi-dimensional data corresponding to the second employee; wherein the second employee is a preset number of candidate employees ranked highest in the first recommendation list. The second employee is the employee for whom a recommendation reason is to be generated, and may be one or more.

[0106] Optionally, the above total matching degree can be calculated using a weighted cosine similarity algorithm, as shown in the following formula:

[0107]

[0108] Among them, Match(T, C) represents the similarity between the job portrait vector T and the employee portrait vector C (i.e., the total matching degree).

[0109] Recommendation reasoning text can be generated based on the LLM or a pre-set recommendation reasoning template to assist in decision-making. For example, the overall match and multi-dimensional data for the second employee are input into the LLM, which then generates natural language output. An example of a first recommendation reason might be: "A fully meets the requirements in terms of team management experience (6 years) and international project experience, and also has a high score for strategic planning (0.78), ranking first in overall match."

[0110] In other possible embodiments, the above-mentioned structured tag set can be divided into multiple first-class tags; the above-mentioned step S140 may include: calculating the tag hit rate, total matching degree and matching degree of each first-class tag corresponding to each candidate employee based on the employee portrait vector and job portrait vector of each candidate employee; calculating the comprehensive score of each candidate employee based on the matching degree and weight coefficient of each first-class tag corresponding to each candidate employee; generating a first recommendation list based on the tag hit rate, total matching degree and comprehensive score corresponding to each candidate employee; generating a first recommendation reason for the second employee based on the tag hit rate, total matching degree, comprehensive score and multi-dimensional data corresponding to the second employee; wherein the second employee is a preset number of candidate employees ranked highest in the first recommendation list.

[0111] Optionally, each first-class label includes at least one second-class label; the above label hit rate can be calculated by the following formula:

[0112]

[0113] Among them, Hit% represents the label hit rate, v i represents the score corresponding to the i-th second-category label in the employee portrait vector, θ represents the preset score threshold, and n represents the number of second-category labels.

[0114] Optionally, the matching degree of each type of label can also be calculated using a weighted cosine similarity algorithm. It is only necessary to calculate the weighted cosine similarity between the vector corresponding to the type of label in the employee portrait vector of the candidate employee and the vector corresponding to the corresponding type of label in the job portrait vector to obtain the matching degree of the type of label corresponding to the candidate employee.

[0115] The above comprehensive score can be calculated using the following formula:

[0116]

[0117] Among them, the total score is the comprehensive score, which represents the final score of the position and employee matching (normalized to the interval [0, 1]), which is used for sorting and recommendation; d represents the index (i.e., serial number) of a class of tags, such as "basic information", "behavioral resume", "performance evaluation", "potential inventory", "ability assessment", etc.; D represents the number of tags in a class, which can be 3 to 6, for example, 5, depending on the system settings; α d Represents the weight coefficient of the d-th first-class label, which is used to regulate the influence of this dimension on the total score, satisfying ∑α d =1;T d C represents the vector representation of the job portrait vector in the dth first-class label (label-weight pair); d Represents the vector representation of the employee portrait vector in the dth first-class label (label-score pair); Sim d (T d ,C d ) represents the matching degree between the position and the employee under the d-th first-class label, which can be scored by cosine similarity or cross entropy.

[0118] The configuration of a class of labels and weight coefficients for an example is shown in Table 5 below.

[0119] Table 5

[0120] The dimension d of a class label Second-class label <![CDATA[Weight coefficient α d > Basic information dimension Education, length of service, and rank 0.2 Behavioral resume dimension Project experience, awards, overseas experience 0.2 Performance evaluation dimensions Performance level, goal completion 0.25 Potential Inventory Dimension Potential score, grid position 0.2 Ability assessment dimensions Competency score, cognitive level 0.15

[0121] This embodiment provides two methods for generating the first recommendation list, which are as follows:

[0122] Generation method 1: First, select candidates from all candidate employees whose label hit rate is greater than the preset hit rate threshold and whose total matching degree is greater than the preset matching degree threshold. Then, sort the candidates in descending order of comprehensive scores to obtain the first recommendation list;

[0123] Generation method 2: First, select from all candidate employees those whose label hit rate is greater than a preset hit rate threshold and whose comprehensive score is greater than a preset comprehensive score threshold. Then, sort the candidate employees in descending order of total matching degree to obtain the first recommendation list.

[0124] The corresponding generation method can be selected according to the user's preference or specific application scenario. If the company pays more attention to the comprehensive quality match of the candidate employees, method one can be selected; if more emphasis is placed on the high degree of fit with the job requirements and the overall evaluation, method two can be selected. It should be noted that the generation method of the first recommendation list is not limited to the above two methods. In other embodiments, it is also possible to first screen out the candidates whose total matching degree is greater than the preset matching degree threshold, and then sort the candidates in order from high to low according to the comprehensive score and label hit rate to obtain the first recommendation list, where the total matching degree has a higher priority than the label hit rate. Of course, other generation methods can also be used, which are not listed here one by one.

[0125] The LLM can input the second employee's tag hit rate, total match, comprehensive score, and multi-dimensional data into the first recommendation reason, generating a natural language output. An example of a first recommendation reason is as follows: "A fully meets the requirements in terms of team management experience (6 years) and international project experience. He also has a high score for strategic planning (0.78), ranking first in the overall score."

[0126] This embodiment implements large-scale model interpretability enhancement (XAI), which improves the trust of HR and management by enhancing the transparency and human understandability of recommendation reasons.

[0127] Step S150: When a comparative analysis request is received for at least one first employee in the first recommendation list, a visual display of the comparative analysis is performed based on the employee portrait vector and the job portrait vector of each first employee. The visual display of the comparative analysis includes one or more of a prominent display of the label hit points, a visual display of the label score comparison, a visual display of the matching degree of the label dimension, and a visual display of the contribution of the label to the total matching degree.

[0128] Users can view the first recommended list and select some employees from it for comparative analysis. These selected employees are called first employees. A label matrix M can be constructed to perform dimension comparisons, display key points, and display performance and performance maps and radar charts.

[0129] In some possible embodiments, the above-mentioned structured tag set is divided into multiple first-class tags, each first-class tag may include at least one second-class tag; the above-mentioned highlighting of the tag hit points may include: determining the second-class tags in the employee portrait vector of the first employee with a score greater than a preset score threshold as the tag hit points of the first employee, and highlighting the tag hit points of the first employee.

[0130] The visual display of the above-mentioned label score comparison may include: when there are multiple first employees, according to the employee portrait vector of each first employee, by aggregating the label scores of all first employees into the same dimensional matrix (that is, the label dimensions are aligned), constructing a label matrix, and each row of the label matrix corresponds to the employee portrait vector of a first employee; according to the label matrix, calculating the relative difference of each first employee on each second-category label, and the relative difference is used to represent the degree of deviation from the average value; and displaying the relative difference corresponding to each first employee under each second-category label through a radar chart or a bar chart.

[0131] In addition, you can also display the label matrix, which can be:

[0132]

[0133] Among them, v ij It represents the score of the i-th employee on the j-th second-category label, k is the number of employees, and n is the number of second-category labels.

[0134] The above relative difference can be calculated by the following formula:

[0135]

[0136] Where Δv ij represents the relative difference of the i-th employee on the j-th second-category label; It represents the average score of all employees on the jth second-category label, which is used to measure the group benchmark.

[0137] The visual display of the matching degree of the above-mentioned label dimension may include: calculating the matching degree of each type of label corresponding to the first employee based on the employee portrait vector and position portrait vector of the first employee; and displaying the matching degree of each type of label corresponding to the first employee through a bar chart.

[0138] The visual display of the contribution of the above-mentioned labels to the total matching degree may include: calculating the contribution of each second-category label corresponding to the first employee to the total matching degree based on the employee portrait vector and position portrait vector of the first employee; and displaying the contribution of each second-category label corresponding to the first employee to the total matching degree through a pie chart.

[0139] The contribution of the two types of labels can be calculated by the following formula:

[0140]

[0141] Among them, Contrib ij Indicates the contribution of the jth second-class label to the total matching degree of the i-th employee, satisfying ∑ j Contrib ij=1, used to represent the proportion of the total matching degree of the i-th employee that comes from the contribution of the j-th second-class label; v ij 、v ik represents the scores corresponding to the jth and kth second-class labels of the i-th employee, ω j 、ω k It represents the normalized weights corresponding to the jth and kth two-category labels in the job profile vector, and n represents the number of two-category labels.

[0142] For ease of understanding, assume that the labels and weights of job profiles are as follows: strategic planning -0.9, team management -0.8, and cross-border project experience -0.7; the labels and scores of an employee profile are: strategic planning -0.6, team management -0.9, and cross-border project experience -0.5; then, the contribution of strategic planning is (0.9×0.6) / (0.9×0.6+0.8×0.9+0.7×0.5)≈33.5%. Similarly, the contribution of team management is 44.7%, and the contribution of cross-border project experience is 21.7%.

[0143] Furthermore, the above method also includes: when a data export operation for target data is monitored, performing secondary authentication of operation permissions; wherein the target data includes one or more of a first recommendation list, a first recommendation reason, and a visual display of comparative analysis; after the secondary authentication is passed, providing a download link of the target data for the user to download, and writing relevant operation information into the audit log.

[0144] When the user wants to export target data (PDF export is supported), if the user needs to share it with management, he or she can click "Export Comparison PDF". The system will trigger a secondary authentication (such as sending a text message verification code and / or sending an email verification code). After successful authentication, the export is allowed, such as generating a report file and providing a download link; in addition, relevant operation records are written into the audit log, including user ID, export time, content type, IP and device ID, etc., to ensure the security, compliance and traceability of the system.

[0145] For ease of understanding, refer to Figure 2 The specific implementation process of the above method is introduced. Figure 2As shown, after the job requirement text is input, the large language model LLM is called to perform semantic label extraction and weight generation; then, through label standardization and vector normalization, the job portrait vector T is constructed; after the employee portrait vector C is constructed, the matching degree is calculated by the matching engine, and then the recommendation results are sorted and displayed; it is determined whether to initiate a talent comparison analysis; if yes (i.e., initiated), a visual display of the comparison analysis is performed, and then the results are exported / continued to be screened; if no (i.e., not initiated), the results are exported / continued to be screened; it is determined whether to export a PDF; if yes (i.e., export is required), the secondary authentication process is triggered, and a report is generated and a log is recorded after the authentication is passed; if no (export is not required), the process ends.

[0146] The embodiment of the present invention closely combines functional modules such as talent files, talent tags, inventory grids, custom exports, and secondary authentication control to form an end-to-end intelligent recommendation and selection system to meet the company's intelligent employment needs in cadre selection, succession planning, and project configuration; by integrating large model analysis, employee tag portrait modeling, intelligent matching recommendation and secure export, it solves the problems of difficult job matching, inaccurate talent evaluation, and opaque selection process, and realizes the intelligence, standardization and traceability of corporate talent selection decisions.

[0147] Furthermore, the job requirements text input supports multiple languages. Specifically, by accessing multilingual translation services (such as DeepL / Google Translate), job requirements texts in non-standard languages ​​can be automatically translated into a preset standard language. Dynamic switching of multilingual prompt templates supports recognition of different language structures.

[0148] Furthermore, in order to improve the accuracy of job matching, historical job positions, their recommended candidates, and actual hiring / performance data can be collected, and contrastive learning or Siamese Network can be used to fine-tune the employee portrait vector. Enterprise data can be used to train the matching model to achieve "semantic preference" adaptation and improve the accuracy of the matching results between the job portrait vector and the employee portrait vector.

[0149] It is also possible to dynamically update the job profile vector and the employee profile vector in both directions, so that the job profile vector and the employee profile vector have "learning capabilities" and are continuously optimized based on the actual selection results. Job profile vector feedback mechanism: records the difference between the portrait vector of each successfully recommended "job candidate" and the job profile vector, and uses rolling average or online learning to update the job profile vector. Employee portrait vector update logic: automatically corrects the portrait dimensions based on the projects in which the employee has participated, the certificates obtained, the latest inventory and evaluation; introduces an event-driven mechanism: such as "performance improvement once" or "receiving a certain type of award" triggers the portrait refresh logic; can be updated in real time or on a scheduled basis (such as daily / weekly) to ensure that the employee portrait vector reflects the latest status in a timely manner.

[0150] Furthermore, this embodiment also provides an automatic tag derivation mechanism. Based on this, the above method also includes: regularly generating new dynamic tags based on each employee's behavioral history data (such as internal training records, project clock-in records, and reward and punishment records), and marking their source type as dynamic; using a preset rule engine (such as Drools) and prompt words from a large language model to generate AI-inferred tags, and marking their source type as AI-inferred; displaying the newly added dynamic tags and AI-inferred tags for users to update the structured tag set.

[0151] Furthermore, this embodiment also provides a multi-round job requirement clarification dialogue function: if the job requirement text is incomplete or unclear, clarification questions will be automatically triggered, such as "Is management experience required?", "Does this position require frequent business trips?", etc.; a chatbot can be used to ask supplementary questions to the user.

[0152] Furthermore, this embodiment provides a team matching function, supporting not only single-point "position-employee" matching but also "employee-team" matching, helping to build highly collaborative and diverse teams. This function can be applied to scenarios such as recommending candidates for project team formation, assessing the fit of reserve managers within existing leadership teams, and simulating organizational structure optimization paths that may result from "replacement plans." Based on this, the above method also includes: aggregating and statistically analyzing the employee portrait vectors of all members of the target team to obtain the team portrait vector of the target team; generating a second recommendation list and explainable second recommendation reasons based on the organizational fit between the employee portrait vector of each candidate employee and the team portrait vector of the target team, combined with the multi-dimensional data of each candidate employee; wherein the organizational fit includes one or more of the personality complementarity index, Euclidean distance and cosine similarity; when a comparative analysis request for at least one third employee in the second recommendation list is received, a capability complementarity map and a personality diversity index are visualized based on the employee portrait vector of each third employee and the team portrait of the target team; wherein the capability complementarity map is used to indicate the score difference between each third employee and the target team under different label dimensions, and the personality diversity index is used to indicate the degree of personality difference in the target team after the third employee joins the target team.

[0153] The team portrait vector refers to the aggregated statistical results of the current target team members in various label dimensions. It can extract aggregate indicators of the target team members in dimensions such as ability, personality, performance, inventory, etc., such as mean, standard deviation, characteristic distribution (biased / balanced), etc.

[0154] The set of employee portrait vectors for all members of the target team is:

[0155] c={C1,C2,...,C m};

[0156] Among them, C i represents the employee portrait vector of the i-th member, C i ={(t j , v ij )},t j represents the jth label, v ij represents the score (0-1) of the i-th member on the j-th label; m represents the number of members.

[0157] This embodiment provides extraction methods for different label dimensions, as shown in Table 6 below.

[0158] Table 6

[0159]

[0160] Based on this, the employee portrait vectors of all members of the target team are aggregated and counted to obtain the team portrait vector of the target team, which may include: aggregating the label scores of the employee portrait vectors of all members of the target team according to the preset aggregation methods corresponding to each label dimension to obtain the team portrait vector of the target team; wherein, the aggregation methods corresponding to the ability label include mean, variance, maximum or minimum value; the aggregation methods corresponding to the personality label include the five-dimensional personality average and / or diversity index, the aggregation methods corresponding to the performance data and inventory data include weighted mean, and the aggregation methods corresponding to the label density include label hit rate and / or shared label ratio.

[0161] The organizational fit between a candidate and the target team can be calculated using one or more of Euclidean distance, cosine similarity, and personality complementarity index. A team profile vector can contain more labels than an employee profile vector. For example, a team profile vector may also include team-specific metrics such as "personality variance," "stability index," and "label distribution dispersion." When calculating Euclidean distance and cosine similarity, label dimension alignment is required. Label misses are marked as 0 (or the default value). In actual calculations, "intersection labels" can be selected based on the labels in the position profile vector for matching:

[0162] T={t1,t2,...,t n} = Position portrait label set ∩ Team portrait label set ∩ Employee portrait label set.

[0163] The organizational fit calculation formula includes:

[0164] (a) Weighted cosine similarity (default):

[0165]

[0166] Among them, G represents the team portrait vector, C represents the employee portrait vector, g j represents the score of the jth label in the team portrait vector, c j represents the score of the jth label in the employee portrait vector, and n represents the number of labels.

[0167] (b) Euclidean distance (applicable to difference calculation):

[0168]

[0169] (c) Personality Complementarity Index (for Big5 personality):

[0170] First calculate the five-dimensional personality distance:

[0171]

[0172] Then converted into a complementary score:

[0173]

[0174] The higher the complementarity, the greater the difference in personality composition, which is suitable for team scenarios that encourage "personality diversity".

[0175] The ability complementarity map and personality diversity index of candidate employees and target teams can be presented visually. The x-axis of the ability complementarity map is the ability label required for the position (such as leadership, technology, communication, etc.), and the y-axis is the difference between the candidate and the team average. The ability complementarity map can be used to identify whether the employee fills the team's shortcomings in a certain key dimension (+ value represents "contribution"). The Personality Diversity Index (PDI) is defined as:

[0176]

[0177] Among them, trait j represents a personality label (such as openness or conscientiousness), j represents the label's index, n represents the number of personality labels, and Var represents the variance of the distribution of that personality within the team. The personality diversity index can be used to avoid personality homogeneity (e.g., team members are all high in extroversion or neuroticism).

[0178] The team matching feature generates the following output: organizational fit score (team vs. candidate), tag-level radar chart (completion of competency dimensions), diversity score (whether personality diversity is present), and automatic recommendation reasoning (Prompt + LLM). An example scenario outputs: "Candidate A's communication skills are above the current team average (+12%) and have overseas project experience, which can fill the gap in the team's international technical experience. Furthermore, their introversion score is significantly below average (-15%), which will help improve the team's personality balance. The recommendation index is 0.92."

[0179] The embodiment of the present invention not only covers the process of job analysis → talent portrait → recommendation matching → comparison export, but also has strong scalability. It can also introduce capabilities such as organizational matching, job evolution, reverse recommendation, intelligent inventory simulation and AI service to create a truly intelligent and growable "enterprise-level human resource decision-making middle platform". Based on semantic analysis, employee portrait structuring, vector matching and recommendation algorithms, a complete intelligent personnel selection engine is constructed, which has the characteristics of automation, high precision, explainability, security and compliance. Especially at the algorithm level, combined with mathematical methods such as weighted cosine similarity, label hit rate, dimensional weighted scoring, dynamic label generation, and comparison matrix difference, a set of recommendation mechanisms with technological innovation and algorithmic originality is formed. It can not only be expanded to the AI ​​decision-making infrastructure within the enterprise, but also has a strong market application prospect. It is suitable for core scenarios such as enterprise cadre selection, internal recruitment, succession planning, job portrait management, etc., and has the ability of continuous model iteration and business expansion flexibility.

[0180] Furthermore, this embodiment also provides a reverse job recommendation function (i.e., employees view jobs), which supports employees to recommend suitable jobs based on employee portrait vectors, assisting internal mobility and talent incentives. It can batch calculate the matching degree between employee portrait vectors and job portrait vector sets; generate job recommendation lists and capability gap analysis; and output development suggestions in conjunction with the training path planning module. Specifically, the employee portrait vector C and the job portrait vector set T i Batch similarity calculations are available; a ranked list is provided, displaying recommended positions by job match, along with reasons for the recommendation and a prompt indicating the degree of achievement. This feature can be integrated with the "Capability Gap" function to identify gaps in current job matches and guide employee development paths. This can be used for pre-recommendations before internal competitive positions are opened, and by allowing employees to review "suitable positions for me," employee retention can be improved.

[0181] Furthermore, this embodiment also provides AI inventory simulation and virtual succession drill functions, which can quickly predict potential candidates and filling plans when there are vacancies / potential succession needs in positions. Simulated position input: HR inputs the hypothetical position requirement text → the system generates a position portrait vector → screens matching candidates within the organization. Generates a "succession competency map" for candidate employees: displays the fitness curve and development potential estimation; provides "time path simulation": estimates the training / rotation time required for candidate employees to grow from their current status to fully competent positions. Application scenarios include: preparing succession echelons for executive positions in advance, simulating replacement candidates for key project talent flows, and analyzing candidates before talent inventory meetings.

[0182] Furthermore, the recruitment capabilities can be opened up as microservices (AI-as-a-Service), allowing the core modules of AI recruitment to be embedded in more systems, such as recruitment, succession, and performance management systems. Capabilities such as job semantic parsing, person-job matching, recommendation lists, and comparative analysis can be encapsulated as RESTful API services for external systems such as recruitment, succession, and performance management systems to call. These external systems input job requirement data and multi-dimensional employee data to obtain recommendation results, including recommendation lists and reasons, in JSON format. Permission verification and log auditing can also be integrated into a unified authentication system.

[0183] Corresponding to the above-mentioned matching recommendation method based on large-scale digital talent inventory, the embodiment of the present invention also provides a matching recommendation system based on large-scale digital talent inventory. Figure 3 The structure diagram of a matching recommendation system based on a large-scale digital talent inventory is shown, and the system includes:

[0184] Data acquisition module 301 is used to obtain a job requirement text and multi-dimensional data of at least one candidate employee; wherein the job requirement text is described in natural language, and the multi-dimensional data includes basic information data, behavioral history data, performance evaluation data, potential inventory data, and ability assessment data;

[0185] The first generation module 302 is used to perform semantic analysis on the job requirement text using a large language model and generate a job profile vector through normalization processing. The job profile vector is composed of normalized weights corresponding to the structured label set;

[0186] The second generation module 303 is used to perform label mapping and normalization processing on each field of the multi-dimensional data of each candidate employee to obtain an employee portrait vector. The employee portrait vector is composed of scores corresponding to the structured label set;

[0187] A first recommendation module 304 is configured to generate a first recommendation list and an explainable first recommendation reason based on the matching degree between the employee profile vector and the job profile vector of each candidate employee and in combination with the multi-dimensional data of each candidate employee;

[0188] The first display module 305 is used to perform a visual display of the comparative analysis based on the employee portrait vector and job portrait vector of each first employee when a comparative analysis request for at least one first employee in the first recommendation list is received. The visual display of the comparative analysis includes one or more of a prominent display of label hit points, a visual display of label score comparison, a visual display of the matching degree of label dimensions, and a visual display of the contribution of labels to the total matching degree.

[0189] The system provided by the embodiment of the present invention integrates large-scale model analysis, employee portrait modeling, intelligent matching recommendation and visual display of comparative analysis. It has the characteristics of automation, high precision and explainability. It solves problems such as difficulty in matching people with jobs, inaccurate talent evaluation, and opaque selection process, and realizes digital talent selection and intelligent decision-making.

[0190] Furthermore, the structured tag set is divided into multiple first-class tags; the first recommendation module 304 is specifically used to: calculate the tag hit rate, total matching degree and matching degree of each first-class tag corresponding to each candidate employee based on the employee portrait vector and job portrait vector of each candidate employee; calculate the comprehensive score of each candidate employee based on the matching degree and weight coefficient of each first-class tag corresponding to each candidate employee; generate a first recommendation list based on the tag hit rate, total matching degree and comprehensive score corresponding to each candidate employee; generate a first recommendation reason for the second employee based on the tag hit rate, total matching degree, comprehensive score and multi-dimensional data corresponding to the second employee; wherein the second employee is a preset number of candidate employees ranked highest in the first recommendation list.

[0191] Furthermore, each of the above first-class labels includes at least one second-class label; the label hit rate is calculated by the following formula:

[0192]

[0193] Among them, Hit% represents the label hit rate, v i represents the score corresponding to the i-th second-category label in the employee portrait vector, θ represents the preset score threshold, and n represents the number of second-category labels.

[0194] Furthermore, the structured tag set is divided into a plurality of first-class tags, each of which includes at least one second-class tag;

[0195] Furthermore, the first display module 305 is specifically configured to: determine the second-category labels in the employee portrait vector of the first employee whose scores are greater than a preset score threshold as the label hit points of the first employee, and highlight the label hit points of the first employee; and / or,

[0196] When there are multiple first employees, a label matrix is ​​constructed based on the employee portrait vectors of each first employee, where each row of the label matrix corresponds to the employee portrait vector of one first employee; based on the label matrix, a relative difference value of each first employee on each second-category label is calculated, where the relative difference value is used to represent the degree of deviation from the average value; the relative difference value corresponding to each first employee under each second-category label is displayed using a radar chart or a bar chart; and / or,

[0197] Calculate the matching degree of each first-category label corresponding to the first employee based on the employee portrait vector and the job portrait vector of the first employee; display the matching degree of each first-category label corresponding to the first employee through a bar chart; and / or,

[0198] Based on the employee portrait vector and job portrait vector of the first employee, the contribution of each second-category label corresponding to the first employee to the total matching degree is calculated; and the contribution of each second-category label corresponding to the first employee to the total matching degree is displayed using a pie chart.

[0199] Furthermore, the above system also includes:

[0200] The third generation module is used to aggregate and count the employee portrait vectors of all members of the target team to obtain the team portrait vector of the target team;

[0201] A second recommendation module is configured to generate a second recommendation list and explainable second recommendation reasons based on the organizational fit between the employee profile vector of each candidate employee and the team profile vector of the target team, combined with the multi-dimensional data of each candidate employee; wherein the organizational fit includes one or more of the personality complementarity index, Euclidean distance, and cosine similarity;

[0202] The second display module is used to, when receiving a comparative analysis request for at least one third employee in the second recommendation list, visually display a capability complementarity map and a personality diversity index based on the employee portrait vector of each third employee and the team portrait of the target team; wherein the capability complementarity map is used to indicate the score difference between each third employee and the target team under different label dimensions, and the personality diversity index is used to indicate the degree of personality difference in the target team after the third employee joins the target team.

[0203] Furthermore, the above-mentioned third generation module is specifically used to: aggregate the label scores of the employee portrait vectors of all members of the target team according to the preset aggregation methods corresponding to each label dimension, to obtain the team portrait vector of the target team; wherein, the aggregation methods corresponding to the ability label include mean, variance, maximum or minimum value; the aggregation methods corresponding to the personality label include the five-dimensional personality average and / or diversity index, the aggregation methods corresponding to the performance data and inventory data include weighted mean, and the aggregation methods corresponding to the label density include label hit rate and / or shared label ratio.

[0204] Furthermore, the above system also includes a label update module, which is used to: regularly generate new dynamic labels based on the behavioral history data of each employee, and mark their source type as dynamic; use the preset rule engine and the prompt words of the large language model to generate AI inference labels, and mark their source type as AI inference; display new dynamic labels and AI inference labels for users to update the structured label set.

[0205] Furthermore, the above system also includes a secondary authentication module, which is used to: perform secondary authentication of operation permissions when a data export operation for target data is monitored; wherein the target data includes one or more of a first recommendation list, a first recommendation reason, and a visual display of comparative analysis; after the secondary authentication is passed, a download link of the target data is provided for the user to download, and relevant operation information is written into the audit log.

[0206] The system provided in this embodiment has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the system embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0207] like Figure 4 As shown, an electronic device 400 provided by an embodiment of the present invention includes: a processor 401, a memory 402 and a bus, the memory 402 stores a computer program that can be run on the processor 401, when the electronic device 400 is running, the processor 401 and the memory 402 communicate through the bus, and the processor 401 executes the computer program to implement the above-mentioned matching recommendation method based on large-model digital talent inventory.

[0208] Specifically, the memory 402 and processor 401 can be general-purpose memories and processors, which are not specifically limited here.

[0209] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program that, when executed by a processor, executes the matching recommendation method based on a large-scale digital talent inventory described in the preceding method embodiment. The computer-readable storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory, RAM, a magnetic disk, or an optical disk.

[0210] The term "and / or" herein simply describes an association relationship between associated objects, indicating that three relationships can exist. For example, "A and / or B" can represent the existence of three situations: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" herein refers to any combination of at least two of any one or more of a plurality of items. For example, "at least one of A, B, and C" can represent any one or more elements selected from the set consisting of A, B, and C.

[0211] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not limiting, and thus other examples of the exemplary embodiments may have different values.

[0212] The flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of code, and a part of the module, program segment or code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0213] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interface, the indirect coupling or communication connection of the system or module can be electrical, mechanical or other forms.

[0214] Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of these modules may be selected to achieve the purpose of this embodiment based on actual needs.

[0215] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0216] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A matching recommendation method based on large-scale digital talent inventory, characterized in that: include: Obtaining a job requirement text and multi-dimensional data of at least one candidate employee; wherein the job requirement text is described in natural language, and the multi-dimensional data includes basic information data, behavioral history data, performance evaluation data, potential inventory data, and ability assessment data; Using a large language model to perform semantic analysis on the job requirement text, and generating a job profile vector through normalization processing, wherein the job profile vector is composed of normalized weights corresponding to a structured tag set; Performing label mapping and standardization on each field of the multi-dimensional data of each candidate employee to obtain an employee portrait vector, wherein the employee portrait vector is composed of scores corresponding to the structured label set; generating a first recommendation list and an explainable first recommendation reason based on the matching degree between the employee portrait vector of each candidate employee and the job profile vector, combined with the multi-dimensional data of each candidate employee; When a comparative analysis request is received for at least one first employee in the first recommendation list, a visual display of the comparative analysis is performed based on the employee portrait vector and the position portrait vector of each first employee. The visual display of the comparative analysis includes one or more of a prominent display of label hit points, a visual display of label score comparison, a visual display of the matching degree of label dimensions, and a visual display of the contribution of labels to the total matching degree.

2. The method according to claim 1, characterized in that The structured tag set is divided into a plurality of first-class tags; and generating a first recommendation list and an explainable first recommendation reason based on the matching degree between the employee portrait vector of each candidate employee and the position portrait vector, combined with the multi-dimensional data of each candidate employee, includes: Calculate the label hit rate, total matching degree, and matching degree of each of the labels of the candidate employee according to the employee portrait vector and the job portrait vector of each candidate employee; Calculate a comprehensive score for each candidate employee based on the matching degree and weight coefficient of each of the first-category labels corresponding to each candidate employee; Generate a first recommendation list based on the tag hit rate, total matching degree and comprehensive score corresponding to each candidate employee; Generate a first recommendation reason for the second employee based on one or more of a tag hit rate, a total matching degree, a comprehensive score, and multi-dimensional data corresponding to the second employee; wherein the second employee is a preset number of candidate employees ranked highest in the first recommendation list.

3. The method according to claim 2, characterized in that Each of the first-class labels includes at least one second-class label; the label hit rate is calculated by the following formula: Among them, Hit% represents the label hit rate, v i represents the score corresponding to the i-th second-category label in the employee portrait vector, θ represents the preset score threshold, and n represents the number of second-category labels.

4. The method according to claim 1, wherein The structured tag set is divided into a plurality of first-class tags, each of which includes at least one second-class tag; and the visual display of comparative analysis based on the employee portrait vector and the position portrait vector of each first employee includes: Determine the second-category labels in the employee portrait vector of the first employee whose scores are greater than a preset score threshold as label hit points of the first employee, and highlight the label hit points of the first employee; and / or, When there are multiple first employees, a label matrix is ​​constructed based on the employee portrait vectors of each first employee, where each row of the label matrix corresponds to an employee portrait vector of the first employee; a relative difference value of each first employee on each of the two labels is calculated based on the label matrix, where the relative difference value is used to represent the degree of deviation from the average value; and the relative difference value corresponding to each first employee under each of the two labels is displayed using a radar chart or a bar chart; and / or, Calculating the matching degree of each of the first employee's labels based on the first employee's employee portrait vector and the job portrait vector; displaying the matching degree of each of the first employee's labels using a bar chart; and / or, Based on the employee portrait vector and the position portrait vector of the first employee, the contribution of each of the second-category labels corresponding to the first employee to the total matching degree is calculated; and the contribution of each of the second-category labels corresponding to the first employee to the total matching degree is displayed through a pie chart.

5. The method according to claim 1, characterized in that The method further comprises: Aggregate and count the employee portrait vectors of all members of the target team to obtain the team portrait vector of the target team; generating a second recommendation list and an explainable second recommendation reason based on the organizational fit between the employee profile vector of each candidate employee and the team profile vector of the target team, in combination with the multi-dimensional data of each candidate employee; wherein the organizational fit includes one or more of personality complementarity index, Euclidean distance, and cosine similarity; When a comparative analysis request is received for at least one third employee in the second recommendation list, a capability complementarity map and a personality diversity index are visualized based on the employee portrait vector of each third employee and the team portrait of the target team; wherein the capability complementarity map is used to indicate the score difference between each third employee and the target team under different label dimensions, and the personality diversity index is used to indicate the degree of personality difference in the target team after the third employee joins the target team.

6. The method according to claim 5, characterized in that The aggregating statistics of the employee portrait vectors of all members of the target team to obtain the team portrait vector of the target team includes: According to the preset aggregation method corresponding to each label dimension, the label scores of the employee portrait vectors of all members of the target team are aggregated to obtain the team portrait vector of the target team; among which, the aggregation method corresponding to the ability label includes mean, variance, maximum or minimum value; the aggregation method corresponding to the personality label includes the five-dimensional personality average and / or diversity index, the aggregation method corresponding to the performance data and inventory data includes weighted mean, and the aggregation method corresponding to the label density includes label hit rate and / or shared label ratio.

7. The method according to claim 1, characterized in that The method further comprises: Regularly generate new dynamic tags based on each employee's behavioral history data, and mark their source type as dynamic; Using the preset rule engine and the prompt words of the large language model, AI speculation labels are generated and their source type is marked as AI speculation; The newly added dynamic tags and the AI-inferred tags are displayed so that the user can update the structured tag set.

8. The method according to claim 1, characterized in that The method further comprises: When a data export operation for target data is detected, performing secondary authentication of the operation authority; wherein the target data includes one or more of the first recommendation list, the first recommendation reason, and the visual display of the comparative analysis; After the secondary authentication is passed, a download link of the target data is provided for the user to download, and relevant operation information is written into the audit log.

9. A matching recommendation system based on large-scale digital talent inventory, characterized by: include: A data acquisition module is configured to acquire a job requirement text and multi-dimensional data of at least one candidate employee; wherein the job requirement text is described in natural language, and the multi-dimensional data includes basic information data, behavioral history data, performance evaluation data, potential inventory data, and ability assessment data; A first generation module is used to perform semantic analysis on the job requirement text using a large language model and generate a job profile vector through normalization processing. The job profile vector is composed of normalized weights corresponding to a structured tag set; A second generation module is configured to perform label mapping and normalization processing on each field of the multi-dimensional data of each candidate employee to obtain an employee portrait vector, wherein the employee portrait vector is composed of scores corresponding to the structured label set; A first recommendation module is configured to generate a first recommendation list and an explainable first recommendation reason based on the matching degree between the employee portrait vector and the job profile vector of each candidate employee and in combination with the multi-dimensional data of each candidate employee; The first display module is used to perform a visual display of the comparative analysis based on the employee portrait vector and the position portrait vector of each of the first employees when a comparative analysis request for at least one first employee in the first recommendation list is received. The visual display of the comparative analysis includes one or more of a prominent display of label hit points, a visual display of label score comparison, a visual display of the matching degree of label dimensions, and a visual display of the contribution of labels to the total matching degree.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

Images