A talent education experience graph construction method based on time sequence and application

Abstract

The present application provides a kind of talent education experience graph construction method and application based on timing, comprising: obtaining standardized education experience;Education experience graph schema design is carried out;Time sequence knowledge graph construction and representation.The present application is based on the knowledge of recruitment industry field, systematically proposes the standardization schema of resume education experience in recruitment field, and establishes corresponding table, completes the standardization of education experience;Solve the problems such as misalignment, alias, abbreviation, former name, information missing and incomplete in the English information of education experience in resume;Firstly, the talent education experience knowledge graph based on timing is proposed, the time of fact occurrence in education experience is considered, the problem that static knowledge graph cannot distinguish the time sequence relationship and evolution process of fact is solved;At the same time, combined with the priority of multi-education experience in recruitment field, the parameters of TDGNN model are optimized, the representation problem of time sequence education graph is solved;There are a variety of practical application scenarios.

Description

Technical Field

[0001] This invention relates to the field of information technology, specifically to a method and application for constructing a time-series-based talent education experience graph. Background Technology

[0002] Educational background (school, major, degree, etc.) is one of the most important sections of a resume, depicting a candidate's educational trajectory and level. In the recruitment field, educational background can also directly reflect the degree of match between a job seeker and the school, degree, major, skills, etc. required for the target position. Furthermore, based on relevant supplementary information such as the location of the school, it can be used to predict the candidate's "desired city" and "desired salary." However, when job seekers fill out their resumes, due to the lack of fixed templates and unified standards, the format, structure, and content of resumes vary widely, and incomplete information is also common. It is difficult to directly use the results of resume analysis as features for recruitment business.

[0003] Currently, when dealing with user information related to educational background, the mainstream approach is to directly store the original text or OCR-generated text into the database and use some rules to try to fill in the missing information. Each user's educational background is a discrete and isolated piece of information in the database, and the relationship between educational backgrounds and between talents lacks expressive ability, making it difficult to handle related reasoning tasks.

[0004] Chinese knowledge graphs are crucial for Chinese natural language understanding, and dedicated communities and organizations are promoting their popularization and application. Several common knowledge graph projects include Zhishi.me, XLore, and CN-DBpedia. Currently, knowledge graph products covering education-related knowledge include Fudan University's Knowledge Factory, CN-DBpedia. However, these knowledge graphs are based on general domains and lack domain-specific knowledge, thus making them unsuitable for knowledge representation and graph reasoning applications in recruitment. Secondly, these knowledge graphs only focus on a single feature of educational experience. For example, the school graph in Fudan's Knowledge Factory is biased towards schools, with nodes centered around schools and edges mainly extending school attributes. However, from the perspective of using educational experience information in recruitment, we not only focus on the job seeker's school but also... This involves comprehensively considering information such as the applicant's education, major, and even enrollment and graduation dates to assess whether they meet the basic requirements of the position. Therefore, there is an urgent need to create a talent education experience graph based on (recruitment) domain knowledge. Finally, and most importantly, the graphs mentioned above are static knowledge graphs. Static knowledge graphs cannot reflect the evolution of time; they are only a cross-section of each talent's education experience. However, education experience is strongly correlated with time. Multiple education experiences unfold chronologically, forming a talent's learning trajectory. This trajectory is subject to change, and the nodes and attributes on this trajectory are dynamic. For example, if a talent's last education experience at the present moment is a domestic undergraduate degree, and a year later they obtain a master's degree from an overseas university, their education experience chain should be updated, with the latest education experience added to the end. Furthermore, over time, there may be changes in school names, program names, and QS rankings. Although modeling talent education experience based on static knowledge graphs can update relevant information for a talent, static knowledge graphs cannot reflect the fact that multiple education experiences are time-dependent.

[0005] In summary, this paper proposes a time-series-based method for constructing a talent education experience graph. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and application for constructing a time-series-based talent education experience graph.

[0007] To achieve the above objectives, the specific solution of the present invention is as follows:

[0008] This invention provides a method for constructing a time-series-based talent education experience graph, comprising:

[0009] S1, to obtain a standardized educational experience;

[0010] S2, Design the educational experience mapping schema;

[0011] S3, construction and representation of time-series knowledge graphs.

[0012] Furthermore, acquiring a standardized educational experience specifically includes:

[0013] S101, standardization of multilingual resumes, converts resumes in other languages ​​into Chinese information;

[0014] S102, standardization of abbreviations, aliases, and former names, maps together entities with the same meaning but different names in school names and majors;

[0015] S103, Filling in missing information: Filling in missing information in a resume by setting completion rules or using a model.

[0016] Furthermore, the education experience graph schema design includes three types of entities: schools, degrees, and majors.

[0017] The attributes of school entities include: school name, school country, school region, whether it is a 985 university, whether it is a 211 university, whether it is a Double First-Class university, and whether it is QS500.

[0018] The attributes of an academic qualification entity include: qualification name, qualification category, and qualification level;

[0019] The attributes of professional entities include: professional name, professional category, first-level discipline, and second-level discipline.

[0020] Furthermore, the construction and representation of temporal knowledge graphs specifically include:

[0021] Incorporating time as an element into the graph, it can be represented by the following formula:

[0022] ,

[0023] in,

[0024] Represents the head node of the relation. Represents the tail node of the relation. It represents the relationship between two nodes;

[0025] In time interval Valid within or at a specific time point If it is valid, then a four-element group can be used. To indicate;

[0026] It is a collection of entities, containing Different entities,

[0027] It is a set of relations, containing Different kinds of relationships,

[0028] Time is from the initial time End time The set of time segments

[0029] It is a set of timestamps, containing Different timestamps.

[0030] Furthermore, the quadruplets in the graph contain time information. The attribute pairs are stored in a graph structure in the Neo4j database for easy querying and retrieval.

[0031] Furthermore, the TDGNN model is adopted, and a power function is used for the temporal knowledge graph, with different weights applied to adjacent nodes;

[0032] Nodes that are closer to the target node are given greater weight.

[0033] Introducing coefficients into the TDGNN model , used to represent vertices and The time-series weighting coefficients between them;

[0034] For nodes with temporal edges, different weights are assigned based on time; accordingly, Formula 1 is improved into Formula 2:

[0035] (Formula 1)

[0036] (Formula 2)

[0037] in, The specific calculation formula is Formula 3:

[0038] (Formula 3)

[0039] Furthermore, the temporal weights in TDGNN are optimized.

[0040] For different educational experiences at the same point in time, the most recent educational experience is given the highest weight.

[0041] The optimized parameters, using The calculation method is shown in Formula 4:

[0042] (Formula 4)

[0043] in,

[0044] node The Layer vector is represented as ,

[0045] for The parameters that the layer needs to train. This is the activation function.

[0046] Based on the talent education experience map, fact-finding and inference are performed;

[0047] The following formula is used to complete the educational background:

[0048] (Formula 5);

[0049] in, - The school that represents talent A.

[0050] Talent profiling and stratification; based on the talent education experience map, scoring is conducted from the dimensions of school, major, degree, and grades to form a talent education experience radar chart; talent profiling is then conducted.

[0051] Then, based on the comprehensive scores from multiple dimensions, talent is stratified.

[0052] Conduct talent seed package selection and community building; based on the talent education experience map, select core talent seed packages according to the dimensions of school, major and degree;

[0053] Based on the talent seed package, build and maintain relevant talent communities.

[0054] The technical solution of this invention has the following beneficial effects:

[0055] 1. Based on industry knowledge, a standardized schema for educational background in resumes in the recruitment field was systematically proposed, and corresponding dimension tables were established, thus completing the standardization of educational background; problems such as misalignment of English and Chinese information in educational background in resumes, aliases, abbreviations, former names, and missing or incomplete information were solved;

[0056] 2. This paper proposes a time-series-based knowledge graph of talent education experience for the first time. It considers the time when facts occur in the education experience and solves the problem that static knowledge graphs cannot distinguish the temporal relationship and evolution process of facts. At the same time, it optimizes the parameters of the TDGNN model in combination with the fact that multiple education experiences in the recruitment field have certain priorities, and solves the representation problem of time-series education graph.

[0057] 3. It has a variety of practical application scenarios. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the nodes and their attributes in this invention;

[0059] Figure 2 This is the node information diagram after adding time attributes according to the present invention;

[0060] Figure 3 This is a schematic diagram illustrating the use of the knowledge graph of the present invention for fact completion and inference;

[0061] Figure 4 This is a schematic diagram illustrating the use of the knowledge graph of the present invention for talent profiling and stratification;

[0062] Figure 5 This is the overall flowchart of the present invention. Detailed Implementation

[0063] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0064] In the description of this invention, unless otherwise explicitly specified and limited, the terms "connected," "linked," and "fixed" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0065] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0066] In the description of this embodiment, the terms "upper," "lower," "front," "rear," "left," and "right," etc., refer to the orientation or positional relationship shown in the accompanying drawings. They are used only for ease of description and simplification of operation, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used only for descriptive distinction and have no special meaning.

[0067] like Figure 1-5 As shown, this invention provides a method and application for constructing a time-series-based talent education experience graph. The education experience graph construction method includes:

[0068] S1, to obtain a standardized educational experience;

[0069] S2, Design the educational experience mapping schema;

[0070] S3, construction and representation of time-series knowledge graphs; using graph neural networks to model and vectorize educational experience graphs.

[0071] Obtaining a standardized educational experience specifically includes:

[0072] S101, standardization of multilingual resumes, converts resumes in other languages ​​into Chinese information;

[0073] S102, standardization of abbreviations, aliases, and former names, maps together entities with the same meaning but different names in school names and majors;

[0074] S103, Filling in missing information: Filling in missing information in a resume by setting completion rules or using a model.

[0075] (1) Multilingual resume issues

[0076] Because the resume database contains resumes in multiple languages, primarily Simplified Chinese, Traditional Chinese, and English, the names of many schools, degrees, and majors in the educational background are not correlated. For example, candidate A's resume is in Chinese, listing Cambridge University as their alma mater, while candidate B's resume is in English, listing "University of Cambridge." Without mapping and standardization of school information, these two resumes would be treated as two different schools. Therefore, it would be difficult to make inferences based on school information or fulfill customized recruitment needs, such as alumni referral networks. Similarly, degree names and major names face the same challenges. Furthermore, some multilingual resumes also involve the conversion between Simplified and Traditional Chinese characters.

[0077] (2) Abbreviation, alias, former name

[0078] This is a common issue in resumes. For example, even with the same name, "Peking University," some resumes will say "Peking University" while others will say "PKU." Similarly, a major like "Computer Application Technology" might be written as "Computer" in a resume. These universities and majors use different abbreviations, making it difficult to cover them all with a single rule. Likewise, some universities, majors, and degree levels have changed their names, or even merged or been abolished, due to historical, developmental, or reform-related reasons. Therefore, this information must also be standardized.

[0079] (3) Information missing

[0080] The reasons for missing information mainly stem from two aspects. First, the original resume may lack some or all of the educational experience information. Second, relevant information may have been missed during the resume parsing process, specifically during OCR (Optical Character Recognition) or entity extraction and information alignment. Missing educational experience information typically focuses on "degrees," "enrollment dates," and "graduation dates."

[0081] This solution primarily addresses the aforementioned issues by constructing dimension tables for schools, academic qualifications, and majors. The diagram illustrates the key fields of these three dimension tables. The school dimension table includes 2759 regular higher education institutions in China (1270 undergraduate institutions and 1489 vocational / junior colleges) and 254 adult higher education institutions. For each school, we refine the key fields in the diagram. Information such as the school's "Chinese name," "English name," "abbreviation," and "whether it is a 985 / 211 / Double First-Class / QS500 university" is populated using open-source general knowledge graphs such as Wikipedia and Baidu Encyclopedia. For the academic qualification dimension, this invention designs 10 academic qualification levels: "Other," "Junior High School," "Secondary Vocational School," "High School," "Junior College," "Bachelor's Degree," "Master's Degree," "MBA / EMBA," "Doctorate," and "Postdoctoral Fellow." The degree ranking increases sequentially, with "Other" being the lowest level (level 0) and "Postdoctoral Fellow" the highest level (level 10). Simultaneously, this invention also compiles the correspondence between the Chinese and English names and abbreviations of various academic qualifications to populate the relevant fields of the academic qualification dimension table. Finally, a professional dimension table was constructed using the same method, containing nearly 3,000 relevant data entries for each professional field. For each professional field, in addition to the basic "name" and other organized information, this invention also adds two professional attributes—"first-level discipline" and "second-level discipline"—that are important for filtering resumes in recruitment.

[0082] The dimension tables for school, academic qualifications, and majors have been completed. For each educational experience in each resume, the fields can be standardized based on the dimension table information. For example, in (1), "Cambridge University" and "University of Cambridge" can be uniformly converted to "Cambridge University" based on the school dimension table information, and in (2), "Peking University" and "Beijing University" can also be uniformly converted to "Beijing University". For cases where multiple universities have the same abbreviation, simple reasoning can be made using other attribute information in the dimension table. For example, the three universities "Central China Normal University", "East China Normal University", and "South China Normal University" are all abbreviated as "Huashi". If the school name on the resume is "Huashi", this invention can distinguish it by using the "address" field of the school dimension table and convert it into a standard school name. Similarly, fields such as academic qualifications and majors can also be standardized to the greatest extent possible based on dimension table information and other educational experience information. This makes entities of the same type as "school", "academic qualifications", and "major" under the same representation framework, which completes the standardization of educational experience.

[0083] The educational experience graph schema design includes three types of entities: schools, degrees, and majors.

[0084] The attributes of school entities include: school name, school country, school region, whether it is a 985 university, a 211 university, a Double First-Class university, and whether it is in the QS 500 (top 500 universities in the world).

[0085] The attributes of an academic qualification entity include: qualification name, qualification category, and qualification level;

[0086] The attributes of professional entities include: professional name, professional category, first-level discipline, and second-level discipline.

[0087] The construction and representation of time-series knowledge graphs specifically include:

[0088] Add time as an element to the graph:

[0089] In a static graph, a fact is represented by... express;

[0090] in, Represents the head node of the relation. Represents the tail node of the relation. It represents the relationship between two nodes;

[0091] If a fact also contains a time element, if in the time interval Valid within or at a specific time point If it is valid, then a four-element group can be used. To indicate;

[0092] like:

[0093] When a time series diagram involves multiple sets of facts, it is represented by the following formula:

[0094] ,

[0095] in,

[0096] It is a collection of entities, containing Different entities,

[0097] It is a set of relations, containing Different kinds of relationships,

[0098] Time is from the initial time End time The set of time segments

[0099] It is a set of timestamps, containing Different timestamps.

[0100] Quadruples with time information in the diagram The attribute pairs are stored in a graph structure in the Neo4j database for easy querying and retrieval.

[0101] In the GNN model, given a static undirected graph... ,

[0102] Obtain nodes using GCN The Layer vector is represented as ,

[0103] To preserve the network topology of the graph, the target node The feature representation requires aggregating the features of all its neighboring nodes. for The parameters that the layer needs to train. For activation functions;

[0104] node The The layer representation is based on the sum of its neighboring nodes and then non-linear processing. The weights of the neighboring nodes are consistent, so there is no temporal relationship between the nodes.

[0105] This solution uses the TDGNN model, employing a power function for the time-series knowledge graph and different weights for adjacent nodes;

[0106] Nodes that are closer to the target node are given greater weight.

[0107] Introducing coefficients into the TDGNN model , used to represent vertices and The time-series weighting coefficients between them;

[0108] For nodes with temporal edges, different weights are assigned based on time; accordingly, Formula 1 is improved into Formula 2:

[0109] (Formula 1)

[0110] (Formula 2)

[0111] in, The specific calculation formula is Formula 3:

[0112] (Formula 3)

[0113] It can be seen that, in terms of time, the earlier an edge is established, the greater its weight.

[0114] In fact, the most recent educational experience should be given greater importance than other educational experiences.

[0115] For example, when comparing high school education experience and postgraduate education experience, the school and major of the most recent postgraduate study will attract more attention; while high school education will only be used as a supplement or will be ignored.

[0116] Therefore, the temporal weights in TDGNN need to be optimized.

[0117] For different educational experiences at the same point in time, the most recent educational experience is given the highest weight.

[0118] The optimized parameters, using The calculation method is shown in Formula 4:

[0119] (Formula 4)

[0120] Application of Knowledge Graph of Talent Education Experience

[0121] After constructing and vectorizing the time-series-based knowledge graph of talent education experience, it can be combined with domain applications to develop higher-level applications based on the knowledge graph of academic qualifications. The time-series-based knowledge graph of talent education experience proposed in this invention has the following main applications.

[0122] (1) Fact completion and inference

[0123] In the recruitment field, due to the lack of a fixed and standardized format for resume content and format, and various reasons such as privacy, incomplete information in job seekers' resumes is very common. Furthermore, omissions and errors in resume parsing exacerbate the loss of information. Therefore, the incompleteness of resume information has become a major challenge in recruitment. After establishing a talent's educational background graph, the inferential capabilities of the graph can be used to supplement and infer some missing educational information. For example, given that talent A's resume contains two educational qualifications, the first qualification is complete: [Information about attending school 1 during time period t]. It indicates that you are studying major 1, using This indicates that, and obtained degree 1, using This indicates that the second part of A's educational information is missing; it is only known that A obtained the degree within time t. Studying a major These are facts. However, the resume does not include information about the schools involved in the second educational experience. In this case, one can rely on the inference capabilities of the educational experience map to predict... The reliability of this event.

[0124] like Figure 1 As shown, the left side of the figure provides an example of the education map for talent A, while the area within the dashed box on the right side represents... The process of this event prediction / inference involves first calculating node embeddings for entity nodes using the TDAgg module. In this method, TDAgg uses a TD-GNN model with optimized temporal coefficients tailored to the recruitment domain for model encoding. (The last sentence, "Acquiring talent," appears to be an unrelated fragment and is omitted from the translation.) and embedding, that is and Using the EdgeAgg model to - The edges (facts) between the nodes are encoded. In this scheme, the EdgeAgg module concatenates the embeddings of the nodes on both sides of the edge and then uses ReLU nonlinear activation to generate the edge embedding representation. In the example above... - The edges between them can be represented by the following formula 5:

[0125] (Formula 5)

[0126] Once the edge embedding is obtained, it can be used for prediction, reliability analysis, error correction, etc.

[0127] (2) Talent profiling and stratification

[0128] In the recruitment industry, especially for mid-to-high-level positions, there are strict requirements for candidates' educational backgrounds. For example, graduates are often required to have graduated from prestigious universities such as "985" or "211" institutions, or renowned domestic and international universities. There are also more customized educational background requirements. For instance, a company in the apparel and textile industry needs a "fabric design analyst." This position is highly specialized, and the company requires candidates with strong textile backgrounds. For highly specialized positions, candidates' educational backgrounds, professional backgrounds, and past work experience must be highly aligned. By combining resume data from the recruitment field, a talent education graph can be constructed, providing a multi-dimensional profile of talent education. This includes aspects such as school, major, and degree, addressing the issues of accurate talent profiling and stratification. Based on this constructed educational experience graph, this solution proposes using graph capabilities to maintain a talent education profile and scoring model. The evaluation of talent education experience focuses on multiple dimensions, including "school," "major," "degree," "grades," and "achievements," with normalized scores forming a talent education experience radar chart. Then, based on the comprehensive scores across these dimensions, talent is stratified. Its simplified diagram is as follows:

[0129] (3) Talent seed selection and community building

[0130] In the recruitment field, "internal referral" is an efficient and high-quality recruitment model. Internal referrals mainly involve recommending job openings through classmates, alumni, friends, colleagues, and other recommenders.

[0131] This model relies on the referrer's understanding of the candidate's abilities and the match between the recommended company and the position. However, often the referrer cannot directly assess the candidate's skill level, resulting in a large number of ineffective referrals. There are also situations where candidates cannot find a sponsor within their social circles to refer them.

[0132] Therefore, the talent education map proposed in this solution can select core talent pools from dimensions such as "school," "major," and "degree," and then expand them in a lookalike fashion. Customized talent communities can be built and maintained, such as "Alumni Group of [University Name]" or "Senior Engineer Group of [Industry Name]." This can improve the quality and efficiency of referrals, allowing candidates to quickly find suitable recommenders. This can also serve as a new recruitment business model.

[0133] This invention proposes a method and application for constructing a time-series-based talent education experience graph, comprising several main inventive steps, in the following order:

[0134] (1) Standardize the educational background of the resume;

[0135] (2) Educational map schema design;

[0136] (3) Construction and representation of time-series knowledge graphs;

[0137] (4) Application of time series knowledge graphs.

[0138] The following is a detailed explanation of the process for each step.

[0139] Step 1. Standardize the educational background information in the resume.

[0140] The resume education and experience information standardization module primarily addresses various issues encountered during resume parsing by standardizing the parsed resume information. Common scenarios requiring resume standardization in this solution include multilingual resumes, abbreviations, aliases, former names, and missing information. These can be resolved through the following steps:

[0141] Step 1.1 Standardize Multilingual Resumes

[0142] The multilingual resume standardization process primarily addresses situations where information in a resume is described in multiple languages, including entirely foreign language resumes or resumes where a specific field in the education section is in a foreign language. This step standardizes the resume by unifying the content from other languages ​​into Chinese.

[0143] Step 1.2 Standardization of abbreviations, aliases, former names, etc.

[0144] Abbreviations, aliases, and former names primarily target fields such as school names and majors in the resume's education experience section. This step maps multiple entities with the same reference but different names together, thus completing the standardization of education experience information.

[0145] Step 1.3 Standardization of the Information Missing Problem

[0146] Information gaps can originate from the original resume itself or result from information loss during resume parsing. This step primarily uses certain rules or lightweight models to complete the missing information.

[0147] Step 2. Educational Map Schema Design

[0148] After completing step 1, which standardizes family education experience, in order to construct a knowledge graph for the recruitment domain, and incorporating business data, step 2 of this invention provides the schema design for the graph.

[0149] Step 3: Construction and Representation Learning of the Time-Sequence Talent Education Graph

[0150] After completing the standardization of educational experience in Step 1 and the schema design in Step 2, a talent map can be established. Step 3 details the embedding method for the talent education time sequence map proposed in this scheme.

[0151] Step 4: Application of Knowledge Graph of Talent Education Experience

[0152] After completing the talent graph construction in step 3, you can fully utilize the graph's basic functions such as factual reasoning and knowledge completion based on specific downstream scenarios. Additionally, you can customize use cases according to business needs, such as talent profiling and stratification, talent seed selection, and community building. The implementation of step 4 depends on specific business requirements, including but not limited to the scenarios mentioned above.

[0153] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of the present invention.

Claims

1. A method for constructing a time-series-based talent education experience graph, characterized in that, include: S1, Obtain standardized educational background, which includes: standardization of multilingual resumes, standardization of abbreviations, aliases, former names, and completion of missing information; and in the standardization process, a unified mapping of entities is achieved based on the school dimension table, academic qualification dimension table, and major dimension table; S2, Design the education experience graph schema, which includes school entities, degree entities, and major entities; S3, construction and representation of temporal knowledge graphs, in which time is added as an element to the graph, so that educational experience facts can be represented by quadruples (h, r, t, γ) when they are valid within a time interval or at a point in time. The construction and representation of time-series knowledge graphs specifically include: Incorporating time as an element into the graph, it can be represented by the following formula: , in, Represents the head node of the relation. Represents the tail node of the relation. It represents the relationship between two nodes; In time interval Valid within or at a specific time point If it is effective, then a four-element group can be used. To indicate; It is a collection of entities, containing Different entities, It is a set of relations, containing Different kinds of relationships, Time is from the initial time End time The set of time segments It is a set of timestamps, containing Different timestamps; The TDGNN model is used to apply a power function to the time-series knowledge graph and different weights to adjacent nodes. Nodes that are closer to the target node are given greater weight. Introducing coefficients into the TDGNN model , used to represent vertices and The time-series weighting coefficients between them; For nodes with temporal edges, different weights are assigned based on time; accordingly, Formula 1 is improved into Formula 2: (Official 1) (Official 2) in, The specific calculation formula is Formula 3: (Official 3) Furthermore, the temporal weights in TDGNN are optimized. For different educational experiences at the same point in time, the most recent educational experience is given the highest weight. The optimized parameters, using The calculation method is shown in Formula 4: (Official 4) in, node The Layer vector is represented as , for The parameters that the layer needs to train. This is the activation function.

2. The method for constructing a time-series-based talent education experience graph according to claim 1, characterized in that, Obtaining a standardized educational experience specifically includes: S101, standardization of multilingual resumes, converts resumes in other languages ​​into Chinese information; S102, standardization of abbreviations, aliases, and former names, maps together entities with the same meaning but different names in school names and majors; S103, Completion of missing information: By setting completion rules or using a model, the missing information in the resume is completed. The missing information includes education level, enrollment time, and graduation time.

3. The method for constructing a time-series-based talent education experience graph according to claim 1, characterized in that, The education experience graph schema design includes three types of entities: schools, degrees, and majors. The attributes of school entities include: school name, school country, school region, whether it is a 985 university, whether it is a 211 university, whether it is a Double First-Class university, and whether it is QS500. The attributes of an academic qualification entity include: qualification name, qualification category, and qualification level; The attributes of professional entities include: professional name, professional category, first-level discipline, and second-level discipline.

4. The method for constructing a time-series-based talent education experience graph according to claim 1, characterized in that, Quadruples with time information in the diagram The attribute pairs are stored in a graph structure in the Neo4j database for easy querying and retrieval.

5. The method for constructing a time-series-based talent education experience graph according to any one of claims 1-4, characterized in that, The method is applied to fact completion and inference based on talent education experience graphs; The following formula is used to complete the educational background: (Official 5); in, - The school that represents talent A.

6. The method for constructing a time-series-based talent education experience graph according to any one of claims 1-4, characterized in that, The method is applied to talent profiling and stratification; Based on the talent education experience map, scores are generated from the dimensions of school, major, degree, and grades to form a talent education experience radar chart; and talent profiles are created. Then, based on the comprehensive scores from multiple dimensions, talent is stratified.

7. The method for constructing a time-series-based talent education experience graph according to any one of claims 1-4, characterized in that, The method is applied to talent seed selection and community building; Based on the talent education experience map, and according to the dimensions of school, major and degree, a core talent seed package is selected; Based on the talent seed package, build and maintain relevant talent communities.

Images