A talent evaluation system and method based on intelligent graphs and heterogeneous data.

Abstract

This invention discloses a multi-source heterogeneous data talent evaluation system and method based on intelligent graphs, including a multi-source heterogeneous data acquisition module, a heterogeneous knowledge graph construction module, and a personalized recommendation module. After acquiring and determining the sources of heterogeneous network data, a heterogeneous graph neural network is formed. Based on knowledge distillation technology, complex social network entity links of relationships are established, and talent data after knowledge distillation is generated by creating a person profile. Combined with skill-based talent evaluation and intelligent recommendation using the knowledge graph, this invention achieves knowledge complementarity between multiple heterogeneous knowledge graphs, effectively ensures the integrity of knowledge through a distributed data storage mechanism, and provides accurate and trustworthy person profiles. It also achieves precise semantic understanding, resulting in high accuracy in talent evaluation and intelligent recommendations tailored to different individuals.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a talent evaluation system and method based on intelligent graphs and heterogeneous data. Background Technology

[0002] Currently, knowledge graphs and their application in key technologies for talent evaluation are showing positive development trends. Knowledge graphs are considered a key technology for the sustainable development of the digital economy, and they have shown great potential, especially in promoting the integration of knowledge graphs with traditional industries.

[0003] In the field of skills talent evaluation, the current stage is still at the level of traditional single data collection and flat evaluation. There are still shortcomings in the construction of knowledge graphs for complex social network entity links of human relationships and human profiles. Inconsistencies in cross-domain entity and relationship references can easily lead to system chaos. The accuracy and recall of entity links are low, and the completeness and accuracy of talent evaluation knowledge are generally poor. At the same time, there is a lack of systematic and holistic talent evaluation and intelligent recommendation functions. To address these issues, we propose a multi-dimensional heterogeneous data talent evaluation system and its method based on intelligent graphs. Summary of the Invention

[0004] In view of the problems existing in the current talent evaluation in knowledge graphs, this invention is proposed.

[0005] Therefore, one objective of this invention is to provide a multi-dimensional heterogeneous data talent evaluation system based on intelligent graphs. This system constructs complex social network entity links and user profiles based on knowledge distillation technology, while simultaneously combining knowledge graph-based skill-based talent evaluation and intelligent recommendation. By constructing a demand intent tree through an end-to-end deep memory network, it automatically generates demand recommendations based on user demand preferences in the knowledge graph, offering advantages such as high accuracy and effectiveness in talent evaluation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] On the one hand, this invention provides a talent evaluation system based on intelligent graphs and multi-dimensional heterogeneous data, comprising:

[0008] The multi-source heterogeneous data acquisition module, based on data interfaces, data sharing, web crawlers, and public resources, acquires and determines the sources of heterogeneous network data to form a heterogeneous graph neural network. Based on the heterogeneous graph neural network, it extracts document information to generate structured data. Specifically, it incorporates evaluation individuals, skill types, and reward / punishment behavior elements based on graph attention mechanisms to construct a heterogeneous graph composed of sentence nodes, word nodes, and skill evaluation element nodes to capture the relationships between sentences and classify sentences to generate structured data.

[0009] The heterogeneous knowledge graph construction module establishes complex social network entity links and person profiles based on knowledge distillation technology to generate knowledge-distilled talent data. It then constructs a corresponding heterogeneous knowledge graph of skilled personnel based on this talent data, performs talent evaluation based on the heterogeneous knowledge graph of skilled personnel, and generates evaluation results. The heterogeneous knowledge graph construction module includes a knowledge distillation unit for disambiguation features of structured data, an analysis unit for talent evaluation analysis, and a generation unit for generating evaluation results.

[0010] The personalized recommendation module combines skill and talent evaluation and intelligent recommendation based on knowledge graphs. It constructs a demand intent tree through an end-to-end deep memory network and automatically generates demand information for personalized recommendations based on the user's demand preferences in the knowledge graph.

[0011] As a preferred embodiment of the present invention, the heterogeneous data is divided into text and structured data, and further divided into news, academic papers, monographs and patents. Different mapping mechanisms are matched for different types of data, and the mapped data is then integrated into a heterogeneous knowledge graph of skilled personnel. The heterogeneous data is converted using Jena's built-in conversion tool.

[0012] Based on the different skill-based heterogeneous knowledge graphs formed after mapping, the DBpedia and Sedris programs are used to standardize the different types of knowledge graphs to obtain multiple types of heterogeneous knowledge graphs.

[0013] In a preferred embodiment of the present invention, the construction of a heterogeneous graph consisting of sentence nodes, word nodes, and skill evaluation element nodes is for building a database of skilled talent nodes. Specifically, it is based on the person relationship matching algorithm in the person relationship matching model of machine learning technology to extract skilled talent information, person relationship information, and person technical information from public social media and competition media. Then, the skilled talent person relationship matching model is trained and verified using the learning, training, and verification process of machine learning technology. The skilled talent person profile database contains person skill information, skilled talent competition information, and skilled talent achievements. At the same time, the extracted person technical information, skilled talent competition information, and skilled talent achievement information are classified and added to build a person relationship knowledge base.

[0014] As a preferred embodiment of the present invention, the steps for constructing the character relationship matching model are as follows: the training process and testing process are the same as the offline model training and verification process; data reading and processing; extraction of character information; storing the training set into the graph character relationship matching model, inputting the training set, completing the model training through the training process, and obtaining the result character matching.

[0015] As a preferred embodiment of the present invention, the knowledge distillation unit performs word segmentation and named entity recognition on the text by calling the LTP system during the training of the graph person relationship matching model. Specifically, a word segmentation user dictionary and a part-of-speech tagging user dictionary are added to the LTP system. The process is divided into two stages. In the first stage, a ranking model is obtained by using four types of disambiguation features to rank the manually annotated training corpus. This model is used to predict the target entity of the referent in the corpus. After the prediction is completed, the process proceeds to the second stage. In the second stage, the loss function of knowledge distillation is used to optimize the model and improve the accuracy of the prediction results.

[0016] The loss function optimization model for knowledge distillation is as follows:

[0017] L=α*L soft +β*L hard

[0018]

[0019] Where L is the cross-entropy loss function, L soft For the teacher model, soft labeling using sofemax, L hard The hard labels for the student model's loss are α and β, which are the corresponding coefficients, respectively.

[0020] H represents L soft The divergence of the soft-label model (Sofemax) outputs a probability distribution after high-temperature processing, where K represents L. hard The student model's student loss is the hard-label cross-entropy, G. j Here are the parameters of the loss function model for knowledge distillation, where a is a hyperparameter and represents the ratio of the true label cross-entropy to the divergence of the distribution of the loss function model for knowledge distillation.

[0021] n represents the loss function model parameters for the j-th knowledge distillation. The smoothing value for the soft label of the teacher model (Sofemax). The probability of the soft label output by the teacher model sofemax at temperature stage T. The smoothed value of the hard label for the student model's loss is given. Output the probability of hard labels for the student model (student loss).

[0022] As a preferred embodiment of the present invention, the four types of disambiguation features are four hypothetical relationships, specifically: the term text matching any entity reference has semantic relevance to the text to be disambiguated; the entity reference and the matching entity are of the same type; there is a correlation between entities in the text to be disambiguated; and the entities in the knowledge base are incomplete.

[0023] By utilizing the corresponding hypothetical relationships of the four types of disambiguation features, a series of candidate entities, i.e., terms, in each reference in the text to be disambiguated are sorted to obtain matching terms, thus completing entity recognition after optimization of the structured data disambiguation features.

[0024] As a preferred embodiment of the present invention, the multi-source heterogeneous data acquisition module further includes a heterogeneous data comparison unit and a data processing unit. The heterogeneous data comparison unit is used to compare the social relationship graphs among skilled talents and match them with the talent database. If the graphs already exist, they are completed to form a skilled talent metadata ontology dataset. If the graphs do not already exist, they are recorded in the skilled talent metadata ontology set. Finally, the data is manually sorted and analyzed.

[0025] The data processing unit uses a metadata database to perform data fusion on the extracted metadata knowledge according to feature values ​​in the metadata database. The skill data fusion method includes: reorganizing non-standard data from different sources using relational mapping to construct a centralized and unified data set; performing standardization processing on the centralized dataset, including content structuring, standardizing data elements, cleaning, filtering to remove invalid information and data duplication, and uniformly mapping to a basic entity set.

[0026] In a preferred embodiment of the present invention, the personalized recommendation module includes a human-computer interaction model construction unit, a human-computer interaction analysis unit, and a recommendation unit.

[0027] The human-computer interaction model building unit is used to build a multi-turn human-computer interaction model to support the acquisition of requirements based on multi-turn interaction.

[0028] The multi-turn human-computer interaction model analysis unit is used to obtain and analyze the user's real-time needs in a multi-turn interaction manner during the real-time interaction needs acquisition process, based on the user's needs and preferences knowledge in the user's knowledge graph, and then generate the needs intent tree based on the intent tree needs model.

[0029] The recommendation unit is used to acquire and automatically generate user demand information based on the user's demand preference information in the user's knowledge graph, and then push the user demand information to the corresponding user terminal for the user to make personalized selections.

[0030] As a preferred embodiment of the present invention, the human-computer interaction model construction unit adopts a deep memory network to construct a language understanding layer, wherein the memory of the memory network is used to encode text sequences, and the memory network pointer can identify the user's needs through an attention mechanism;

[0031] The multi-turn human-computer interaction model analysis unit is used to initialize and encode the user's demand preferences in the user's knowledge graph to generate preference vectors, and uses the GRU model to generate corresponding interaction strategy information based on the preference vectors and interaction states. The recommendation unit generates corresponding push information content based on the interaction strategy information.

[0032] On the one hand, this invention provides a method for a multi-dimensional heterogeneous data talent evaluation system based on intelligent graphs, comprising:

[0033] Step 1: Based on data interfaces, data sharing, web crawlers, and public resources, a heterogeneous graph neural network is formed after acquiring and determining the source of heterogeneous network data. The document information is extracted and structured data is generated based on the heterogeneous graph neural network.

[0034] Step 2: Based on knowledge distillation technology, establish complex social network entity links and person profiles to generate knowledge-distilled talent data. Construct a heterogeneous knowledge graph of skilled talents based on the talent data. Perform talent evaluation based on the heterogeneous knowledge graph of skilled talents and generate evaluation results.

[0035] Step 3: Combining skill talent evaluation and intelligent recommendation with knowledge graphs, construct a demand intent tree through an end-to-end deep memory network, and automatically generate demand information for personalized recommendations based on the user's demand preferences in the knowledge graph.

[0036] The beneficial effects of this invention are as follows: This invention has the functions of linking entities in complex social networks based on relationships and creating character profiles; it achieves entity linking in complex networks by using cross-domain knowledge fusion network based on knowledge distillation technology, preventing system chaos caused by inconsistencies in cross-domain entity and relationship references, and realizing entity linking in complex social networks based on relationships; through the association representation model of preference targets and text, it introduces external knowledge to approximate the distribution between domain knowledge, realizes knowledge complementarity between multiple heterogeneous knowledge graphs, and effectively ensures the integrity of knowledge by utilizing a distributed data storage mechanism, resulting in accurate and trustworthy character profiles.

[0037] It also features skills talent evaluation and intelligent recommendation functions that combine knowledge graphs; it constructs a demand intent tree through an end-to-end deep memory network, automatically generates demand recommendations based on the user's demand preferences in the knowledge graph, and prevents noise from inactive information through a dynamic attention mechanism gating structure, achieving accurate semantic understanding, high accuracy in talent evaluation, and intelligent recommendations tailored to different individuals. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0039] Figure 1 This is a schematic diagram of the modular structure of the system of the present invention;

[0040] Figure 2 This is a schematic diagram of the knowledge distillation process in the system of the present invention;

[0041] Figure 3 This is a schematic diagram of a multi-round human-computer interaction processing scenario for the personalized recommendation module in the system of the present invention;

[0042] Figure 4 This is a flowchart illustrating the method of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0044] To comprehensively assess the true skill level of skilled personnel and improve the accuracy and efficiency of talent selection, this embodiment, based on the talent intelligent assessment system, studies technologies such as intelligent collection of multi-source heterogeneous data, complex social network entity linking and character profiling based on interpersonal relationships, and skill talent evaluation and intelligent recommendation combined with knowledge graphs. This addresses technical issues related to data collection, character profiling, and artificial intelligence, ensuring not only the fairness of the skill talent evaluation process and the authenticity of the results, but also providing reliable analysis for subsequent talent prediction and career planning, thereby improving the efficiency and quality of the entire skill talent evaluation process.

[0045] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides one aspect of the present invention: the present invention provides a multi-dimensional heterogeneous data talent evaluation system based on intelligent graphs, comprising:

[0046] The multi-source heterogeneous data acquisition module, based on data interfaces, data sharing, web crawlers, and public resources, acquires and determines the sources of heterogeneous network data to form a heterogeneous graph neural network. Based on the heterogeneous graph neural network, it extracts document information to generate structured data. Specifically, it incorporates evaluation individuals, skill types, and reward / punishment behavior elements based on graph attention mechanisms to construct a heterogeneous graph composed of sentence nodes, word nodes, and skill evaluation element nodes to capture the relationships between sentences and classify sentences to generate structured data.

[0047] The heterogeneous knowledge graph construction module uses knowledge distillation technology to establish complex social network entity links and person profiles to generate knowledge-distilled talent data. Based on the talent data, it constructs a corresponding heterogeneous knowledge graph of skilled talents. Based on the heterogeneous knowledge graph of skilled talents, it performs talent evaluation and generates evaluation results. The heterogeneous knowledge graph construction module includes a knowledge distillation unit for disambiguation features of structured data, an analysis unit for talent evaluation analysis, and a generation unit for generating evaluation results.

[0048] The personalized recommendation module combines skill and talent evaluation with intelligent recommendation based on a knowledge graph. It constructs a demand intent tree through an end-to-end deep memory network and automatically generates demand information for personalized recommendations based on user demand preferences in the user's knowledge graph. After extracting and labeling relevant dialogue samples and the user's knowledge graph, a multi-turn human-computer interaction model needs to be built to support demand acquisition based on multi-turn interactions. This model must acquire the user's real-time demands through multiple rounds of interaction, based on the user's demand and preference knowledge in the user's knowledge graph. Therefore, constructing a multi-turn human-computer interaction model based on the user's knowledge graph is the core issue in demand intent tree construction. After model training, the model can interact and identify demands in real time, and construct a demand intent tree based on the demand tree model.

[0049] Specifically, in this embodiment, heterogeneous data is divided into text and structured data, and further divided into news, academic papers, monographs and patents. Different mapping mechanisms are matched for different types of data, and the mapped data is then integrated into a heterogeneous knowledge graph of skilled personnel. The heterogeneous data is converted using Jena's built-in conversion tool.

[0050] Based on the different skill-based heterogeneous knowledge graphs formed after mapping, the DBpedia and Sedris programs are used to standardize the different types of knowledge graphs to obtain multiple types of heterogeneous knowledge graphs.

[0051] Specifically, this embodiment constructs a heterogeneous graph composed of sentence nodes, word nodes, and skill evaluation element nodes to build a database of skilled talent nodes. Specifically, it extracts skilled talent information, relationship information, and technical information from public social media and competition media using a relationship matching algorithm in a machine learning-based graph-based relationship matching model. Then, it trains and validates the constructed skilled talent relationship matching model using a machine learning training and validation process. The skilled talent profile database contains talent skill information, competition information, and achievements. Simultaneously, the extracted technical information, competition information, and achievement information are categorized and added to construct a relationship knowledge base.

[0052] Specifically, in this embodiment, the steps for constructing the person-relationship matching model are as follows: training process and testing process and offline model training and verification process; data reading and processing; extraction of person information; storing the training set into the graph person-relationship matching model, inputting the training set, completing the model training through the training process, and obtaining the result person matching.

[0053] It is important to emphasize in this embodiment that the knowledge distillation unit calls the LTP (Language Technology Platform) system to perform word segmentation and named entity recognition on the text during the training of the graph person relationship matching model. Specifically, a word segmentation user dictionary and a part-of-speech tagging user dictionary are added to the LTP system. The process is divided into two stages. In the first stage, the manually annotated training corpus is sorted and learned by using four types of disambiguation features to obtain a sorting model, which is used to predict the target entity of the referent in the corpus. After the prediction is completed, the second stage is entered. In the second stage, the loss function of knowledge distillation is used to optimize the model and improve the accuracy of the prediction results.

[0054] The loss function optimization model for knowledge distillation is as follows:

[0055] L=α*L soft +β*L hard

[0056]

[0057] Where L is the cross-entropy loss function, L soft For the teacher model, soft labeling using sofemax, L hard The hard labels for the student model's loss are α and β, which are the corresponding coefficients, respectively.

[0058] H represents L soft The divergence of the soft-label model (Sofemax) outputs a probability distribution after high-temperature processing, where K represents L. hardThe student model's student loss is the hard-label cross-entropy, G. j Here are the parameters of the loss function model for knowledge distillation, where a is a hyperparameter and represents the ratio of the true label cross-entropy to the divergence of the distribution of the loss function model for knowledge distillation.

[0059] n represents the loss function model parameters for the j-th knowledge distillation. The smoothing value for the soft label of the teacher model (Sofemax). The probability of the soft label output by the teacher model sofemax at temperature stage T. The smoothed value of the hard label for the student model's loss is given. The student loss model outputs the probability of hard labels. From the perspective of the relevance between entity extraction and entity linking tasks, a knowledge distillation framework is used to effectively improve the accuracy of person profiling in the field of vocational skills talent evaluation.

[0060] In this embodiment, the four types of disambiguation features are four hypothetical relationships: the term text that matches any entity reference has semantic relevance to the text to be disambiguated; the entity reference and the matching entity are of the same type; there is a correlation between entities in the text to be disambiguated; and the entities in the knowledge base are incomplete.

[0061] By utilizing the corresponding hypothetical relationships of four types of disambiguation features, a series of candidate entities, i.e., terms, in each reference in the text to be disambiguated are sorted to obtain matching terms, thus completing entity recognition after optimizing the disambiguation features of structured data.

[0062] Based on the above, this embodiment has the functions of linking complex social network entities based on relationships and creating character profiles; it uses knowledge distillation technology to achieve cross-domain knowledge fusion network to realize entity linking of complex networks, preventing system chaos caused by inconsistencies in cross-domain entity and relationship references, and realizing complex social network entity linking based on relationships; through the association representation model of preference targets and text, it introduces external knowledge to approximate the distribution between domain knowledge, realizes knowledge complementarity between multiple heterogeneous knowledge graphs, and effectively ensures the integrity of knowledge by utilizing a distributed data storage mechanism, resulting in high accuracy and trustworthiness of character profiles.

[0063] To address the challenges in collecting multi-source heterogeneous data, which involves diverse data sources (e.g., web crawlers, APIs, sensors, and IoT devices), and various data formats (e.g., structured, semi-structured, and unstructured data), this embodiment specifically includes a heterogeneous data acquisition module. This module addresses common data acquisition problems such as low efficiency, high error rates, difficulty in handling large datasets, difficulty in collecting complex data, and poor scalability. The heterogeneous data comparison unit compares the social relationship graphs among skilled personnel with a talent database. If existing data exists, it completes the database to form a skilled personnel metadata ontology dataset. If no existing dataset exists, it is recorded in the skilled personnel metadata ontology set. Finally, manual analysis and processing are performed.

[0064] The data processing unit utilizes the metadata database to perform data fusion on the extracted metadata knowledge according to feature values. The data fusion methods include: reorganizing non-standard data from different sources using relational mapping to construct a centralized and unified data set; performing standardization processing on the centralized dataset, including content structuring, standardizing data elements, cleaning, filtering to remove invalid information and data duplication, and uniformly mapping to the basic entity set.

[0065] Specifically, in this embodiment, the personalized recommendation module includes a human-computer interaction model construction unit, a human-computer interaction analysis unit, and a recommendation unit. The human-computer interaction model construction unit is used to construct a multi-turn human-computer interaction model to support demand acquisition based on multi-turn interactions. The multi-turn human-computer interaction model analysis unit is used to analyze the user's real-time demands obtained through multi-turn interactions based on the demand and preference knowledge in the user's knowledge graph during the real-time interaction demand acquisition process, and then generate a demand intent tree based on the intent tree demand model. The recommendation unit is used to obtain and automatically generate user demand information based on the demand and preference information in the user's knowledge graph, and then push the user demand information to the corresponding user terminal for the user to make personalized selections and determinations.

[0066] In addition, the human-computer interaction model construction unit uses a deep memory network to construct a language understanding layer, where the memory network memory is used to encode text sequences, and the memory network pointers can identify user needs through an attention mechanism; the multi-turn human-computer interaction model analysis unit is used to initialize and encode the user's needs and preferences in the user's knowledge graph to generate preference vectors, and uses a GRU model to generate corresponding interaction strategy information based on the preference vectors and interaction states; the recommendation unit generates corresponding push information content based on the interaction strategy information.

[0067] After implementation in software, this embodiment can support a knowledge graph triple of no less than 100,000; the number of users who can access the AI ​​at the same time is no less than 200; the accuracy and recall of entity links are greater than 85% and 90% respectively; and the overall system response time is less than 500ms.

[0068] Based on the above, this embodiment has the function of skill talent evaluation and intelligent recommendation combined with knowledge graph; it constructs a demand intent tree through an end-to-end deep memory network, automatically generates demand recommendations based on the user's demand preferences in the knowledge graph, and prevents noise from inactive information through a dynamic attention mechanism gating structure, thus achieving accurate semantic understanding, high accuracy in talent evaluation, and intelligent recommendation for different individuals; and finally, it realizes demand confirmation through interactive feedback with users, thereby achieving knowledge graph-based evaluation and intelligent recommendation for vocational skills talents.

[0069] Understandably, the personalized recommendation module employs a deep memory network to construct the language understanding layer. The memory network's memory is used to encode text sequences, and the memory network pointers, through an attention mechanism, can identify user needs. Vector initialization encoding of user needs and preferences in the user's knowledge graph is then performed to realize the function of the latent need inference layer. The interaction strategy is controlled through the dialogue state management layer. A GRU (Gate Recurrent Unit) model is used to generate corresponding interaction strategies based on preference vectors and interaction states. Finally, the interaction generation layer generates corresponding response content based on the interaction strategies. The multi-turn dialogue and need tree construction model is developed in an end-to-end manner, using supervised learning methods for backpropagation training. After training, the model is used for real-time online interaction to realize a multi-turn interaction process based on the user's knowledge graph, thereby completing the construction of a need intent tree based on multi-turn interaction and realizing the core technologies for skills talent evaluation and intelligent recommendation.

[0070] Furthermore, this embodiment also provides a method for a multi-dimensional heterogeneous data talent evaluation system based on intelligent graphs, as detailed below:

[0071] Step 1: Based on data interfaces, data sharing, web crawlers, and public resources, heterogeneous network data sources are acquired and identified to form a heterogeneous graph neural network. Document information is extracted and structured data is generated based on the heterogeneous graph neural network.

[0072] Step 2: Based on knowledge distillation technology, establish complex social network entity links and person profiles to generate knowledge-distilled talent data. Construct a heterogeneous knowledge graph of skilled talents based on the talent data. Then, conduct talent evaluation based on the heterogeneous knowledge graph of skilled talents to generate evaluation results.

[0073] Step 3: Combining skill talent evaluation and intelligent recommendation with knowledge graphs, construct a demand intent tree through an end-to-end deep memory network, and automatically generate demand information for personalized recommendations based on the user's demand preferences in the knowledge graph.

[0074] In summary, this invention's research on technologies such as intelligent acquisition of multi-source heterogeneous data, complex social network entity linking and character profiling based on interpersonal relationships, and skill talent evaluation and intelligent recommendation combined with knowledge graphs demonstrates that knowledge graph-based systems can analyze users' potential needs, providing strong support for personalized recommendations, intelligent question answering, and talent management.

[0075] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0076] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0077] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0078] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0079] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0080] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware, the program being stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiments.

[0081] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. This storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0082] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A talent evaluation system based on intelligent graphs and heterogeneous data, characterized in that, include: The multi-source heterogeneous data acquisition module, based on data interfaces, data sharing, web crawlers and public resources, acquires and determines the source of heterogeneous network data to form a heterogeneous graph neural network, and extracts document information to generate structured data based on the heterogeneous graph neural network; The heterogeneous knowledge graph construction module establishes complex social network entity links and person profiles based on knowledge distillation technology to generate knowledge-distilled talent data, and constructs a corresponding heterogeneous knowledge graph of skilled talents based on the talent data. After evaluating talents based on the heterogeneous knowledge graph of skilled talents, it generates evaluation results. Specifically, by incorporating evaluation individuals, skill types, and reward / punishment behavior elements based on graph attention mechanisms, a heterogeneous graph composed of sentence nodes, word nodes, and skill evaluation element nodes is constructed to capture the relationships between sentences, build a database of skill talent nodes, and classify sentences to generate structured data. The construction of a heterogeneous graph consisting of sentence nodes, word nodes, and skill evaluation element nodes includes: extracting skill talent information, relationship information, and technical information from public social media and competition media based on a person-relationship matching model, and then using machine learning technology for learning, training, and verification processes to train and verify the constructed person-relationship matching model; During the training process of the character relationship matching model, the LTP system is called to perform word segmentation and named entity recognition on the text. Specifically, based on the added user dictionary for word segmentation and user dictionary for part-of-speech tagging, the LTP system is called to perform word segmentation and named entity recognition on the text, generating a set of referential items and a set of candidate entities. Four types of disambiguation features are used to rank and learn the manually annotated training corpus to obtain a ranking model, which is used to predict the target entity of the referential items in the corpus. After the prediction is completed, based on the prediction results, the trained ranking model is called, and the model is optimized by using the knowledge distillation loss function to optimize the accuracy of the prediction results. The personalized recommendation module combines skill and talent evaluation and intelligent recommendation based on knowledge graphs. It constructs a demand intent tree through an end-to-end deep memory network and automatically generates demand information for personalized recommendations based on the user's demand preferences in the knowledge graph.

2. The talent evaluation system based on intelligent graphs and heterogeneous data as described in claim 1, characterized in that, Heterogeneous data is divided into text and structured data, and further divided into news, academic papers, monographs and patents. Different mapping mechanisms are matched for different types of data. The mapped data is then integrated into a heterogeneous knowledge graph of skilled personnel. The heterogeneous data is transformed using Jena's built-in conversion tool. Based on the different heterogeneous knowledge graphs of skilled talents formed after mapping, the DBpedia and Sedris programs are used to standardize the different types of knowledge graphs to obtain multiple types of heterogeneous knowledge graphs.

3. The talent evaluation system based on intelligent graphs and heterogeneous data as described in claim 1, characterized in that, The personalized recommendation module includes a human-computer interaction model construction unit, a human-computer interaction analysis unit, and a recommendation unit; The human-computer interaction model building unit is used to build a multi-turn human-computer interaction model to support the acquisition of requirements based on multi-turn interaction. The multi-turn human-computer interaction model analysis unit is used to obtain and analyze the user's real-time needs in a multi-turn interaction manner during the real-time interaction needs acquisition process, based on the user's needs and preferences knowledge in the user's knowledge graph, and then generate the needs intent tree based on the intent tree needs model. The recommendation unit is used to acquire and automatically generate user demand information based on the user's demand preference information in the user's knowledge graph, and then push the user demand information to the corresponding user terminal for the user to make personalized selections.

4. The talent evaluation system based on intelligent graphs and heterogeneous data as described in claim 3, characterized in that, The human-computer interaction model construction unit uses a deep memory network to construct a language understanding layer, wherein the memory of the memory network is used to encode text sequences, and the memory network pointers can identify user needs through an attention mechanism; The multi-turn human-computer interaction model analysis unit is used to initialize and encode the user's demand preferences in the user's knowledge graph to generate preference vectors, and uses the GRU model to generate corresponding interaction strategy information based on the preference vectors and interaction states. The recommendation unit generates corresponding push information content based on the interaction strategy information.

5. The talent evaluation system based on intelligent graphs and heterogeneous data as described in claim 1, characterized in that, The loss function optimization model for knowledge distillation is as follows: Where L is the cross-entropy loss function, L soft For the teacher model, soft labeling using Softmax, L hard The student loss model has hard labels, α and β are the corresponding coefficients; H represents L soft The divergence of the soft-label model (Sofemax) outputs a probability distribution after high-temperature processing, where K represents L. hard The student model's student loss is the hard-label cross-entropy, G. j Here, represents the loss function model parameters for knowledge distillation, 'a' is a hyperparameter, and represents the ratio of the true label cross-entropy to the divergence of the knowledge distillation loss function model's output distribution; 'n' represents the loss function model parameters for the j-th knowledge distillation. The smoothing value for the soft label of the teacher's model (sofemax). The probability of the soft label output by the teacher model sofemax at temperature stage T. The smoothed value of the hard label for the student model's loss is given. Output the probability of hard labels for the student model (student loss).

6. The talent evaluation system based on intelligent graphs and heterogeneous data as described in claim 1, characterized in that, The multi-source heterogeneous data acquisition module also includes a heterogeneous data comparison unit and a data processing unit. The heterogeneous data comparison unit is used to compare the social relationship graphs among skilled talents and match them with the talent database. If the graphs already exist, they are completed to form a skilled talent metadata ontology dataset. If the graphs do not already exist, they are recorded in the skilled talent metadata ontology set. Finally, the data is manually sorted and analyzed.

7. The talent evaluation system based on intelligent graphs and heterogeneous data as described in claim 6, characterized in that, The data processing unit uses a metadata database to perform data fusion on the extracted metadata knowledge according to feature values. The data fusion method includes: reorganizing non-standard data from different sources using relational mapping to construct a centralized and unified data set; performing standardization processing on the centralized dataset, including content structuring, standardizing data elements, cleaning, filtering to remove invalid information and data duplication, and uniformly mapping to a basic entity set.

8. The talent evaluation system based on intelligent graphs and heterogeneous data as described in claim 1, characterized in that, By utilizing the corresponding hypothetical relationships of the four types of disambiguation features, a series of candidate entities, i.e., terms, in each reference in the text to be disambiguated are sorted to obtain matching terms, thus completing entity recognition after optimization of the structured data disambiguation features.

9. A method for a multi-dimensional heterogeneous data talent evaluation system based on intelligent graphs as described in claim 1, characterized in that, include: Step 1: Based on data interfaces, data sharing, web crawlers, and public resources, a heterogeneous graph neural network is formed after acquiring and determining the source of heterogeneous network data. The document information is extracted and structured data is generated based on the heterogeneous graph neural network. Step 2: Based on knowledge distillation technology, establish complex social network entity links and person profiles to generate knowledge-distilled talent data. Construct a heterogeneous knowledge graph of skilled talents based on the talent data. Perform talent evaluation based on the heterogeneous knowledge graph of skilled talents and generate evaluation results. Step 3: Combining skill talent evaluation and intelligent recommendation with knowledge graphs, construct a demand intent tree through an end-to-end deep memory network, and automatically generate demand information for personalized recommendations based on the user's demand preferences in the knowledge graph.

Images