Multi-modal data fused talent recommendation knowledge graph construction method and system

Abstract

The invention relates to the technical field of data processing, in particular to a talent recommendation knowledge graph construction method and system fusing multi-modal data. Comprising the steps of mining a relationship between historical talents and posts through multi-source data, extracting entities and associations, and constructing a knowledge graph for talent recommendation; the method comprises the following steps: extracting change events from multi-source data, constructing an initial talent change set, searching nodes related to industry field preference and post matching indexes in a knowledge graph, and generating an affected node list; the nodes are processed through priority ranking, an updating sequence is determined, and semantic association rules and relation edge weights are adjusted, so that the graph structure is optimized; and a notification mechanism is triggered, an update view is pushed to the decision support module, confirmation is obtained, and the real-time performance and accuracy of the atlas are ensured. According to the method, the problems of lagging data updating, inaccurate post matching and the like in a traditional talent recommendation method are solved, dynamic updating of the knowledge graph is realized, and the accuracy and efficiency of talent recommendation and post matching are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for constructing a talent recommendation knowledge graph that integrates multimodal data. Background Technology

[0002] With the continuous development of internet technology and the arrival of the big data era, talent recommendation systems have gradually become an important tool in enterprise human resource management. Traditional talent recommendation methods typically rely on a single data source and static matching rules, easily overlooking the dynamic changes and complex relationships between talent and job requirements. This often makes it difficult for talent recommendation systems to perform timely and accurate job matching and talent recommendation when facing rapidly changing market demands, job changes, and talent mobility.

[0003] In recent years, with the widespread application of multimodal data, how to effectively integrate information from multiple data sources to build a more comprehensive, dynamic, and accurate talent recommendation system has become an urgent problem to be solved in the industry. Multimodal data includes, but is not limited to, historical data of talents, skill tags, industry preferences, job requirements, and talent change records. This data provides rich background information, which helps to more accurately assess the matching degree between talents and positions.

[0004] This invention provides a method and system for constructing a talent recommendation knowledge graph that integrates multimodal data, aiming to solve the problems of lagging data updates and inaccurate job matching in existing technologies. By mining and integrating change events from multiple data sources, and combining dynamic updates of industry preferences and job matching indices, a real-time updated and accurately matched talent recommendation knowledge graph is constructed. This method can flexibly adjust the matching relationship between talent and jobs based on historical data, talent change information, and changes in industry demand, providing enterprises with accurate talent recommendation services and improving recruitment efficiency and decision-making quality. Summary of the Invention

[0005] This invention provides a method and system for constructing a talent recommendation knowledge graph that integrates multimodal data. By integrating and dynamically updating multi-source data, it achieves accurate talent recommendation and job matching, optimizes the accuracy and real-time performance of traditional talent recommendation systems, effectively solves the adaptability problem of talent recommendation systems in the face of rapid market and industry changes, and can track changes in talent and positions in real time, thereby improving the efficiency of recruitment and talent management.

[0006] In a first aspect, the present invention provides a method for constructing a talent recommendation knowledge graph that integrates multimodal data, the method comprising: Step S1: By mining multi-source relationship data of historical talents and positions, entities and associations are automatically extracted to construct a knowledge graph for talent recommendation; Step S2: Extract change events from multi-source data input to construct an initial talent change set; based on the initial talent change set, search for nodes in the knowledge graph that are associated with industry domain preferences and job matching index, and generate a list of affected nodes; Step S3: Sort the nodes in the affected node list according to priority to generate a sorted sequence to be processed; extract node data from the sorted sequence to be processed and determine the update order of the nodes in the relation edge weights. Step S4: Adjust the semantic association rules and relation edge weights according to the update order; perform storage system update using the adjusted relation edge weights and semantic association rules, and synchronize entity node definitions and job matching indexes; Step S5: Trigger a notification mechanism using the updated graph to push the updated view to the decision support module and obtain a response confirmation.

[0007] As a preferred embodiment of the present invention, step S2, constructing an initial talent change set, includes: By monitoring changes in data input from multiple sources, the system extracts affected talent skill tags and professional experience records from a pre-defined change log. Using a data source integration approach, the extracted talent skill tags and professional experience records are correlated to generate preliminary entity node definitions. Based on the initial entity node definition, an initial talent change set is generated. If there is missing data in the initial talent change set, complete data is obtained through supplementary queries, and the reliability of the data source is verified to form the final talent change set.

[0008] As a preferred embodiment of the present invention, step S2, determining the list of affected nodes, includes: Search for directly related nodes in the knowledge graph; analyze the semantic relevance of the searched nodes to industry preferences; filter nodes highly correlated with job matching index based on the semantic relevance; if the number of filtered nodes exceeds a preset threshold, adjust the search scope to narrow the node set; generate a preliminary list of affected nodes based on the filtered nodes; verify the accuracy of the matching between the preliminary list of affected nodes and industry preferences; adjust the preliminary list of affected nodes based on the verification results to generate the final node list.

[0009] As a preferred embodiment of the present invention, step S3, generating the sorted sequence to be processed, includes: If the number of nodes in the affected node list exceeds a preset threshold, nodes with high team collaboration scores and social network influence are processed first. The priority weight of each node is determined by analyzing its team collaboration score. The update frequency of the knowledge graph is adjusted based on the priority weight and social network influence. The affected node list is then sorted using the adjusted update frequency. A sorted sequence to be processed is generated, and the rationality of the sorting logic is verified. Based on the verification results, the node order in the sequence to be processed is adjusted to obtain the adjusted sequence to be processed.

[0010] As a preferred embodiment of the present invention, step S3, determining the update order of nodes in the relation edge weights, includes: Detailed data of the first node is extracted from the sorted sequence of nodes to be processed. Learning curves and geographic distribution characteristics related to the first node are preloaded using a node attribute expansion mechanism. The association strength of the first node in the knowledge graph is analyzed based on the preloaded information. The update order of the first node in the relation edge weights is determined based on the association strength. If the update order does not meet a preset standard, the processing priority of the node is adjusted. A confirmation record of the update order is generated for the adjusted priority. The position of the node in subsequent processing is determined based on the confirmation record.

[0011] As a preferred embodiment of the present invention, step S4, adjusting the semantic association rules and relation edge weights according to the update order, includes: Based on the update order, obtain a graph snapshot from the historical version tracing record; analyze the differences between the professional experience record and industry domain preference of each node using the graph snapshot; determine the semantic association rules that need to be adjusted based on the differences; calculate the adjustment range of the relation edge weights for the semantic association rules; if the adjustment range exceeds a preset range, recalculate the relation edge weights; generate the adjusted relation edge weights based on the recalculation results; update the associated data in the knowledge graph based on the adjusted relation edge weights.

[0012] As a preferred embodiment of the present invention, step S4, which synchronizes the entity node definition and job matching index, includes: Using adjusted relational edge weights and semantic association rules, batch write operations are performed in a high-concurrency environment; through the batch write operations, updated data is written to the main storage system; for the written data, entity node definitions are synchronously updated; and based on the synchronously updated entity node definitions, the job matching index is adjusted. If the adjustment result of the job matching index does not meet the preset standard, the index is recalculated; the final job matching index is generated based on the recalculated result; and the update status of the knowledge graph is determined based on the final job matching index.

[0013] As a preferred embodiment of the present invention, step S5, obtaining response confirmation, includes: generating an instant synchronization identifier based on the updated map; triggering an automated notification mechanism based on the instant synchronization identifier; pushing the updated view to the decision support module through the automated notification mechanism; if the decision support module does not respond in a timely manner, resending the updated view; obtaining the final response confirmation through the resent updated view; and recording the execution status of the notification mechanism based on the response confirmation.

[0014] Secondly, the present invention also provides a talent recommendation knowledge graph construction system that integrates multimodal data, for implementing the above-mentioned method, the system comprising: The graph construction unit is used to automatically extract entities and associations based on mining multi-source relationship data of historical talents and positions, and to build a knowledge graph for talent recommendation. The data extraction unit is used to extract change events from multi-source data input and construct an initial talent change set; based on the initial talent change set, it searches the knowledge graph for nodes associated with industry preferences and job matching indices to determine the list of affected nodes. A node sorting unit is used to generate a sorted sequence of nodes to be processed based on the list of affected nodes; extract node data from the sorted sequence of nodes to be processed; and determine the update order of nodes in the relation edge weights. The rule adjustment unit is used to adjust the semantic association rules and relation edge weights according to the update order; and to perform storage system updates using the adjusted relation edge weights and semantic association rules, synchronizing entity node definitions and job matching indices. The notification push unit is used to trigger a notification mechanism through the updated graph, push the updated view to the decision support module, and obtain a response confirmation.

[0015] Thirdly, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0016] The beneficial effects of this invention are as follows: This invention mines multi-source relationship data of historical talent and positions, automatically extracts entities and associations between talent and positions, and constructs a complete knowledge graph. This provides foundational data support for subsequent recommendation and matching, ensuring that the graph reflects real market demand and talent capabilities throughout the recommendation process. Furthermore, it extracts talent change events from multi-source data, constructs an initial talent change set, and searches for associated nodes in the knowledge graph based on industry preferences and job matching indices, generating a list of affected nodes. This ensures the graph can be dynamically updated, reflecting changes between talent and positions in real time, and ensuring that talent recommendations are based on the latest data and meet industry needs and job requirements. By prioritizing the list of affected nodes, it ensures that during the graph update process, priority is given to... The system processes nodes that significantly impact job matching accuracy, identifying the most important nodes and determining their update order based on their relational edge weights in the talent graph. This avoids over-processing low-priority nodes and improving efficiency. Semantic association rules and relational edge weights are adjusted according to the update order. Specifically, by analyzing node attributes and changes, association rules related to industry preferences and job requirements are adjusted to ensure more accurate job-talent matching. The adjusted weights and rules are applied to the storage system to ensure real-time graph synchronization after updates, enabling the decision support module to promptly obtain the latest recommendation information. Updating the graph view provides real-time feedback to the decision support module, ensuring recruiters receive accurate job matching results in the shortest possible time. This step demonstrates the efficiency and accuracy of the technical solution, playing a crucial role in the talent recommendation system. Through the synergy of these technical solutions, dynamically tracking talent changes, adjusting job matching rules in real time, and updating graph data promptly, the accuracy of job matching is greatly improved. This solves the problems of data lag and inaccurate matching in traditional recommendation systems, providing enterprises with a more intelligent and efficient recruitment and talent management solution. Attached Figure Description

[0017] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for constructing a talent recommendation knowledge graph that integrates multimodal data, as shown in this embodiment. Figure 2 This is a structural diagram of a talent recommendation knowledge graph construction system that integrates multimodal data, as shown in the embodiment. Figure 3 This is a flowchart of the method for determining the list of affected nodes in the embodiment. Detailed Implementation

[0019] This invention provides a method and system for constructing a talent recommendation knowledge graph that integrates multimodal data. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data used can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] For ease of understanding, the specific process of the embodiments of the present invention will be described below, such as... Figure 1 As shown in the figure, an embodiment of the present invention provides a method for constructing a talent recommendation knowledge graph that integrates multimodal data, comprising: Step S1: By mining multi-source relationship data of historical talents and positions, entities and associations are automatically extracted to construct a knowledge graph for talent recommendation; Specifically, by mining multi-source relationship data of historical talent and positions, the system automatically extracts relevant entity information, i.e., node information, such as talent skill tags, work experience, industry preferences, and positions, as well as the relationships between them, such as the matching relationship between skills and positions, and the relationship between talent and industry needs. Through the integration and analysis of the above data, a knowledge graph containing multi-node information including talent nodes (personnel nodes), skill tag nodes, industry nodes, industry preference nodes, work experience nodes, and position nodes is constructed. The above graph represents the matching situation between talent and positions through the relationship between nodes and edges, and how various skills and experiences affect job requirements, thereby providing rich structured data support for the subsequent recommendation process.

[0021] Step S2: Extract change events from multi-source data input to generate a talent change set; search for nodes in the knowledge graph that are associated with industry preferences and job matching index based on the talent change set to determine the list of affected nodes; In step S2, a talent change set is generated, including: By monitoring changes in data input from multiple sources, the system extracts affected talent skill tags and professional experience records from a pre-defined change log. Using a data source integration approach, the extracted talent skill tags and professional experience records are correlated to generate preliminary entity node definitions. Based on the initial entity node definition, an initial talent change set is generated. If there is missing data in the initial talent change set, complete data is obtained through supplementary queries, and the reliability of the data source is verified to form the final talent change set.

[0022] Specifically, in this embodiment, by monitoring change events from different data sources, such as recruitment platforms, social media, and internal enterprise systems, key information related to talent changes is extracted. These change events typically include job changes, skill updates, and increases in work experience. Corresponding talent skill tags and professional experience records are obtained from a pre-set change log. These change events allow for the capture of dynamic changes in talent, providing necessary data support for subsequent processing. The skill tags and professional experience records extracted from different data sources are then correlated to generate preliminary entity node definitions. Specifically, talent skill tags and professional experience are mapped to a unified... In the data model, the aforementioned data model defines the association rules between skill tags such as programming languages ​​and leadership skills and professional experiences such as years of work experience and project experience. In order to quantify the degree of association between the above data, matching algorithms such as cosine similarity are used to calculate the similarity between skill tags and professional experiences, and based on this, preliminary entity node definitions are generated. That is, various information about talents obtained from different data sources, such as their skills, experience, and past job positions, are organized into a "data package". This not only effectively processes diverse data from different data sources, but also avoids the existence of data silos, and provides accurate data support for subsequent node searches in the knowledge graph.

[0023] After obtaining the initial entity node definitions, an initial talent change set is constructed based on these definitions. This set will be used for subsequent node searches in the knowledge graph. The entity nodes are aggregated to form a list, facilitating graph traversal and subsequent processing. If data gaps are found in the initial talent change set during this process, a supplementary query is automatically triggered to retrieve relevant data via API, ensuring data integrity and accuracy. For each entity node in the initial talent change set, the reliability of its data source is verified. Specifically, cross-validation is used to compare the consistency of data from different data sources, such as comparing the name, position, and date of the same person on recruitment platforms and social media. Information is processed and a consistency score is calculated. If the data consistency score reaches a preset standard, such as above 80%, the node is marked as reliable; otherwise, an alarm is triggered and the node is excluded. This prevents low-quality data from affecting the subsequent construction of the knowledge graph and generates the final set of talent changes. The above verification process is particularly important in business scenarios with high reliability requirements. For example, in the corporate recruitment process, cross-validation can effectively avoid fake resumes and ensure the authenticity and reliability of the data. Through the above technical solution, not only can different data from multiple sources be fully integrated, but also the accuracy and timeliness of talent change information can be ensured through precise data processing and verification, thereby effectively supporting the construction of the knowledge graph and the realization of intelligent talent recommendation.

[0024] Further, in step S2, a list of affected nodes is determined, such as... Figure 3 As shown, it includes: Search for directly related nodes in the knowledge graph; analyze the semantic relevance of the searched nodes to industry preferences; filter nodes highly correlated with job matching index based on the semantic relevance; if the number of filtered nodes exceeds a preset threshold, adjust the search scope to narrow the node set; generate a preliminary list of affected nodes based on the filtered nodes; verify the accuracy of the matching between the preliminary list of affected nodes and industry preferences; adjust the preliminary list of affected nodes based on the verification results to generate the final node list.

[0025] Specifically, based on the final talent change set, nodes associated with industry preferences and job matching indices are searched in the knowledge graph to determine the list of affected nodes. This involves first searching the knowledge graph for directly related nodes using a path query efficiency-optimized traversal method. Specifically, the entity node corresponding to each change event in the talent change set is extracted as the starting point, and a breadth-first search algorithm is used to traverse the knowledge graph. This means starting from the starting node, visiting adjacent nodes layer by layer, and prioritizing queries for directly connected relationship edges, such as connections from skill tags to industry domains, to ensure efficient location of directly related nodes. For example, in a talent management scenario, if the talent change set contains a talent's skill update event, the traversal starts from that talent node and searches for directly related job nodes, avoiding inefficiencies caused by excessively deep recursive queries. This approach provides a faster response time, especially reducing computational overhead in large knowledge graphs. The optimized traversal method avoids excessively deep recursive queries, thus significantly improving query efficiency, especially in large knowledge graphs, effectively reducing computational overhead and increasing efficiency. The system boasts rapid response speed. After acquiring initial associated nodes, it further analyzes the semantic relevance between these nodes and industry preferences. Specifically, the attributes of each node, such as skill tags, are converted into word vectors. A pre-trained word embedding model, such as Word2Vec, is used to generate standard vectors. The cosine similarity between the node and the industry preference vector is then calculated to quantify the similarity between the node and the preference. The closer the value is to 1, the higher the relevance between the node and the industry preference, thus accurately quantifying semantic relationships and improving matching accuracy. The industry preference vector is usually constructed based on the characteristics and needs of the industry. It can be formed by quantifying widely used skills and experiences in the industry to create a high-dimensional vector. For example, the preferences of a certain industry may include different skills such as programming languages ​​and management experience, along with their importance weights. During the analysis, if the skill tag represented by a node has a high similarity to the preference vector of the target industry, then the matching degree between the node and the industry preference is relatively strong, and it may become a candidate node that meets the job requirements.

[0026] Next, nodes highly correlated with the job matching index are selected based on semantic relevance. If the number of highly correlated nodes exceeds a preset threshold, the search range is adjusted to narrow down the node set. This adjustment includes limiting the traversal depth or adding additional filtering conditions, such as reducing the traversal range to within two hops or excluding relation edges with lower than a preset weight. This reduces the number of nodes and processing burden, ensuring that too many irrelevant nodes are not generated during processing, thereby effectively improving computational efficiency and accuracy. After node selection, a preliminary list of affected nodes is generated. Cross-validation is used for verification. This method is a statistical evaluation technique that divides the data into training and test sets, trains the model, and evaluates the accuracy on the test set. Specifically, the node attributes in the preliminary list are divided into subsets. A subset is used to train the matching model, and then the matching accuracy is verified on the remaining subset. If the accuracy is below 90%, it is marked as needing adjustment. For example, in talent skill matching... In the matching scenario, the skill tag and domain preference of a node in the list are predicted to match using a model. The verification results show an accuracy of 95%, confirming its effectiveness and improving the reliability of the list, which is beneficial for subsequent decision-making. The nodes in the initial list are divided into a training subset and a test subset in a ratio of 7:3. A matching model, such as a logistic regression model, is built using the training subset. This model predicts probabilities by fitting the relationship between features and tags. Based on the above matching model, the accuracy of the matching degree is evaluated on the test set. If the accuracy of the matching model is lower than the predetermined standard, the node list is adjusted, removing nodes that do not meet the conditions, and generating a more accurate final node list. The above technical solution obtains node attributes in the final node list, such as team collaboration scores and job matching indices, to provide a basis for subsequent decision support and dynamically updates the knowledge graph, ensuring that the generated talent recommendation scheme is efficient and accurate, thus ensuring the efficiency and accuracy of the talent recommendation system.

[0027] Step S3: Sort the nodes in the affected node list according to priority to generate a sorted sequence to be processed; extract node data from the sorted sequence to be processed and determine the update order of the nodes in the relation edge weights. In step S3, generating the sorted sequence to be processed includes: If the number of nodes in the affected node list exceeds a preset threshold, nodes with high team collaboration scores and social network influence are processed first. The priority weight of each node is determined by analyzing its team collaboration score. The update frequency of the knowledge graph is adjusted based on the priority weight and social network influence. The affected node list is then sorted using the adjusted update frequency. A sorted sequence to be processed is generated, and the rationality of the sorting logic is verified. Based on the verification results, the node order in the sequence to be processed is adjusted to obtain the adjusted sequence to be processed.

[0028] Specifically, when the number of nodes in the affected node list exceeds a preset threshold, nodes with high team collaboration scores and social network influence, such as personnel nodes, are prioritized for processing. The priority weight is determined by analyzing the team collaboration score of each node. The team collaboration score is calculated based on the node's contribution and interaction frequency in historical collaborative projects. The contribution and interaction frequency are multiplied to generate a base score, which is then normalized to convert it into a weight value. This weight value is used for subsequent node ranking to highlight nodes with strong collaboration, ensuring that these nodes receive priority responses during processing. Furthermore, the update frequency of the knowledge graph is adjusted by analyzing the social network influence of each node, combined with the team collaboration score. Social network influence is calculated based on the node's connection count and propagation depth. The influence score is obtained by multiplying the connection count by a depth factor, where the depth factor is derived based on the graph traversal level. The update frequency of the graph is adjusted based on these scores to optimize the processing speed of high-influence nodes. This ensures that changes to these key nodes are prioritized in the graph during high-concurrency scenarios, reducing latency and improving the accuracy of the job matching index. For example, the update frequency adjustment formula is... The initial frequency, such as once per minute, may be adjusted to once every 30 seconds to dynamically optimize the processing speed of high-impact nodes. For example, in a talent change set, if a node represents a key talent with high influence, adjusting its update frequency can reduce latency, improve the accuracy of the job matching index, and thus lead to the rapid generation of real-time synchronization markers, which is beneficial to the response efficiency of the decision support module. In this process, the adjusted update frequency is used to sort the list of affected nodes. Based on the update frequency, the nodes in the list are arranged in descending order of priority, generating a sorted sequence of nodes to be processed. This sorting process ensures that high-priority nodes are processed first when resources are limited. The sorted sequence of nodes to be processed is then verified to check whether its sorting logic is reasonable, ensuring that the priority weights of adjacent nodes are arranged in descending order, and calculating the overall sorting stability index, such as Kendall's algorithm. The TAU distance is used to evaluate the rationality of the ranking. When an unreasonable ranking is found, such as high-influence nodes being placed later, the above sequence is marked as needing adjustment, and the order of nodes in the sequence to be processed is adjusted according to the verification results to optimize the final processing order. This provides more accurate input data for the subsequent decision support module. The above technical solution effectively ensures the efficiency and accuracy of the talent recommendation knowledge graph construction, and ensures the reasonable allocation of node priorities and the timely updating of high-influence nodes.

[0029] Further, in step S3, determining the update order of nodes in the relation edge weights includes: Detailed data of the first node is extracted from the sorted sequence of nodes to be processed. Learning curves and geographic distribution characteristics related to the first node are preloaded using a node attribute expansion mechanism. The association strength of the first node in the knowledge graph is analyzed based on the preloaded information. The update order of the first node in the relation edge weights is determined based on the association strength. If the update order does not meet a preset standard, the processing priority of the node is adjusted. A confirmation record of the update order is generated for the adjusted priority. The position of the node in subsequent processing is determined based on the confirmation record.

[0030] Specifically, detailed data of the first node is extracted from the sorted sequence to be processed. This is achieved by directly accessing the head of the sequence, ensuring that the extracted node possesses complete basic attributes and associated records. This node is the talent node, providing a reliable foundation for subsequent data processing. Combined with a node attribute expansion mechanism—a dynamic loading method based on graph queries—this mechanism expands the node's attribute set to include additional dimensions of data. It preloads the node's learning and growth curve and regional distribution characteristics. The learning and growth curve reflects the trajectory of talent skills over time, such as the skill progression path from basic to advanced levels. The regional distribution characteristics are further expanded by statistically analyzing the distribution ratio of different cities or regions involved in the node's career experience. This preloaded information enriches the node's feature data, helping to more comprehensively analyze the relationship between the node and other nodes. Especially when the experience background and skill development path of the personnel represented by the node in different regions may influence their future job matching, this data provides strong support for subsequent decision-making.

[0031] After acquiring and preloading relevant node information, the association strength of nodes in the knowledge graph is analyzed based on learning growth curves and geographical distribution characteristics. This involves calculating the connection strength between nodes and other nodes to assess the correlation between nodes and target industry preferences and job matching indices. Specifically, a weighted association strength value is calculated by quantifying the matching degree between the node's growth trajectory and skill tags, and the overlap between geographical distribution and industry preferences. The association strength can be obtained by weighted summation, where the contribution weight of the learning growth curve is 0.6 and the weight of geographical distribution is 0.4, ensuring the analysis results reflect the comprehensive impact. This association strength reflects the node's potential influence, and the update order of nodes in the relation edge weights is determined based on the strength value, ensuring that nodes with high influence are processed first. If the calculation results show that the node's update order does not meet the preset standard, the processing priority of that node is adjusted to ensure that high-influence nodes are processed in a timely manner, avoiding duplicate processing. Delays in node updates can impact subsequent decisions. After adjusting node priorities, a confirmation record of the update order is generated, containing the new priority and processing order of the nodes. This confirmation record helps the system locate the new position of nodes in the sequence during subsequent processing. Based on this position data, the adjusted sequence to be processed serves as input for subsequent semantic rule adjustments, ensuring that the node relationships in the knowledge graph can be updated in a timely and accurate manner, thereby optimizing the structure and semantic accuracy of the knowledge graph. For example, for a technical talent node, the pre-loaded learning growth curve shows a 5-year progress trajectory from beginner to expert, and the geographical distribution characteristics show that the node is mainly active in Beijing and Shanghai. When analyzing the association strength, the slope of the growth curve is calculated to be 0.8, which has a high matching degree with the skill tag. Combined with the geographical overlap of 0.7, the strength value is 0.75. If the strength is higher than the threshold of 0.7, it is judged as a high update order, which is beneficial for quickly synchronizing talent data and avoiding delays in job matching.

[0032] The aforementioned technical solution ensures that the update order and priority of nodes meet actual needs, making it particularly suitable for processing large-scale talent data in high-concurrency scenarios. It effectively reduces processing delays or errors caused by improper node sorting, improves the real-time performance and accuracy of the knowledge graph updates, and ultimately provides efficient data input for the decision support module, enhancing the response speed and reliability of the talent recommendation system. Through learning curve analysis, regional distribution characteristics, and node strength analysis, the knowledge graph for talent recommendation can be dynamically adjusted and optimized, flexibly adapting to the needs of different scenarios in practical applications and improving the overall processing capacity and adaptability of the system.

[0033] Step S4: Adjust the semantic association rules and relation edge weights according to the update order; perform storage system update using the adjusted relation edge weights and semantic association rules, and synchronize entity node definitions and job matching indexes; In step S4, adjusting the semantic association rules and relation edge weights according to the update order includes: Based on the update order, obtain a graph snapshot from the historical version tracing record; analyze the differences between the professional experience record and industry domain preference of each node using the graph snapshot; determine the semantic association rules that need to be adjusted based on the differences; calculate the adjustment range of the relation edge weights for the semantic association rules; if the adjustment range exceeds a preset range, recalculate the relation edge weights; generate the adjusted relation edge weights based on the recalculation results; update the associated data in the knowledge graph based on the adjusted relation edge weights.

[0034] Specifically, based on the update order, a knowledge graph snapshot is obtained from the historical version tracing records. This snapshot represents a snapshot of the knowledge graph's state at a specific historical point in time or a series of points in time. The snapshot provides complete foundational data for subsequent analysis, ensuring data accuracy and consistency during the update process. By analyzing the snapshot, the differences between the professional experience records of the talent entity at that node and industry preferences are compared to identify mismatches. Specifically, relevant talent professional experience information, such as job changes and skill tags, is extracted from the snapshot and compared with current industry preferences, such as programming skills and management experience. The difference values ​​are calculated to form a difference list, highlighting the discrepancies between industry preferences and talent experience. For example, in a talent change scenario, assuming an engineer transitions from manufacturing to software development, the knowledge graph snapshot shows that their experience records include mechanical design skills, while industry preferences emphasize code optimization. The difference analysis highlights the skill mismatches, facilitating precise adjustment of association rules and ensuring the knowledge graph is more accurate. The new system is more accurate and avoids invalid associations. Based on the discrepancy list, it identifies the semantic association rules that need adjustment and identifies those association values ​​that need to be modified. For example, if there is a discrepancy between a talent's skills and job requirements, the association between the job and skills is adjusted according to this discrepancy. This process is achieved by screening rules with high adjustment priority, prioritizing those rules that have a significant impact on job matching index and talent recommendation results. For example, for talents in the technology industry, if the discrepancy shows a lack of data analysis experience, an adjustment rule is determined to change the association between the skill and the job from weak to strong, which is beneficial to improving matching efficiency. After the adjustment is completed, the above rules are refined by calculating the adjustment range of the relation edge weights. The adjustment range is calculated using the cosine similarity algorithm, which quantifies the change in relation edge weights by comparing the similarity between professional experience and industry preferences. If the calculated adjustment range exceeds the preset range, it is backtracked and recalculated using cosine similarity to ensure the stability and accuracy of the edge weights.

[0035] Through the above technical solution, new and adjusted relation edge weights are generated and updated to the associated data in the knowledge graph. The updated graph will be synchronized in real time to ensure that the data in the graph reflects the latest changes and optimizations. This allows the knowledge graph to adjust association rules and weights in a timely manner based on the latest industry needs and talent changes, thereby improving the accuracy of talent recommendation and the system's response speed. It also optimizes the data processing efficiency and matching degree of the talent recommendation system, providing strong support for efficient and accurate talent allocation.

[0036] Furthermore, in step S4, the entity node definition and job matching index are synchronized, including: Using adjusted relational edge weights and semantic association rules, batch write operations are performed in a high-concurrency environment; through the batch write operations, updated data is written to the main storage system; for the written data, entity node definitions are synchronously updated; and based on the synchronously updated entity node definitions, the job matching index is adjusted. If the adjustment result of the job matching index does not meet the preset standard, the index is recalculated; the final job matching index is generated based on the recalculated result; and the update status of the knowledge graph is determined based on the final job matching index.

[0037] Specifically, the storage system is updated using the adjusted relation edge weights and semantic association rules, synchronizing entity node definitions and job matching indices. The core of performing batch write operations in a high-concurrency environment is to use the adjusted relation edge weights and semantic association rules to process data in the knowledge graph in batches and write it to the main storage system. First, the adjusted relation edge weights are extracted from the change log and combined with predefined semantic association rules to ensure that each written data item conforms to the graph structure requirements. To cope with the load brought by the high-concurrency environment, a distributed transaction management mechanism is used, employing technologies such as two-phase commit protocols to ensure data consistency and avoid data conflicts or loss due to concurrent operations.

[0038] Next, the entity node definitions are updated synchronously to ensure that the attributes of each node, such as skill tags and professional experience, are consistent with the latest changes. This involves not only synchronizing the node data itself but also adjusting the relevant job matching index based on the latest node attributes. The adjustment of the job matching index is based on the latest definition of the entity node. Through methods such as weighted summation, multiple factors such as skill tags and professional experience are comprehensively considered to calculate the matching degree of each job. If the calculation result does not meet the preset standard, such as the matching degree being lower than a certain threshold, the job matching index is recalculated based on the differences between the node and the industry sector to ensure that the final index is more accurate and meets actual needs. During the recalculation of the job matching index, the differences between professional experience and industry preferences are analyzed, and algorithms such as cosine similarity are used to adjust the weights to ensure that the matching degree of each job reflects the latest industry needs and talent skills. The adjusted index will be used to generate the final job matching index, which will serve as a key parameter for updating the knowledge graph and measure the degree of matching between talent and job.

[0039] Ultimately, the knowledge graph's update status is determined based on the adjusted job matching index. The system marks the graph as updated and triggers a subsequent notification mechanism. This mechanism ensures timely delivery of updated graph information to the decision support module, enabling automated notification and feedback processing, thereby improving system response speed and decision-making efficiency. This technical solution ensures stability and efficiency under high concurrency, especially during large-scale data changes. Batch write operations and synchronous updates of node attributes significantly improve system throughput and reduce latency. Furthermore, the dynamic adjustment of the job matching index and the graph update mechanism guarantee accurate knowledge graph updates and talent recommendations. The entire system can respond more accurately to talent changes, optimize job recommendations, and improve overall talent management efficiency.

[0040] Step S5: Trigger a notification mechanism using the updated map to push the updated view to the decision support module and obtain a response confirmation; specifically including: The updated graph is used to generate an instant synchronization identifier; an automated notification mechanism is triggered based on the instant synchronization identifier; the updated view is pushed to the decision support module through the automated notification mechanism; if the decision support module does not respond in time, the updated view is resent; the final response confirmation is obtained through the resent updated view; and the execution status of the notification mechanism is recorded based on the response confirmation.

[0041] Specifically, the updated graph triggers a notification mechanism to push the updated view to the decision support module and obtain a response confirmation. First, an instant synchronization identifier is generated from the updated graph. This identifier, typically composed of the current timestamp and graph version number, is generated after the graph completes a batch write operation. This identifier marks the graph's current status, ensuring the accuracy of subsequent data processing. Based on this generated instant synchronization identifier, an automated notification mechanism is triggered. This mechanism is activated by a preset script, checks the identifier's validity, and initiates the notification process, pushing the updated graph view to the decision support module. This updated view contains key data such as talent skill tags and team collaboration scores. This data is packaged via a message queue and sent to the decision support module's interface. If the decision support module does not respond promptly, the updated view is resent according to a preset timeout threshold, which is updated based on the graph. The scale is dynamically adjusted. If the number of updated nodes exceeds a certain limit, the threshold will be extended accordingly to adapt to high-load environments. During retransmission, the retransmission interval is calculated based on the lack of response. Strategies such as linear interpolation or exponential backoff algorithms are used to increase the retry time to ensure that notifications can be reliably sent under different network conditions. Taking exponential backoff as an example, the interval is increased by multiplying it by a factor such as 1.5, for example, 10 seconds for the first time, 15 seconds for the second time, and 22.5 seconds for the third time. This algorithm is based on the congestion avoidance principle in computer networks and can effectively reduce repeated conflicts during network fluctuations, thereby ensuring the reliability of pushed updated views in talent management scenarios. When the number of retransmissions exceeds the maximum limit, the request is marked as a failure. After each retransmission, the feedback from the decision support module is monitored. If a confirmation signal is received, retrying is stopped and confirmation details are extracted. If all retryings are unresponsive, a default confirmation record is generated, and the process ends.

[0042] The system records the execution status of the response confirmation notification mechanism and writes confirmation details to a log file for subsequent optimization and adjustment. This log data serves as a reference, helping the system optimize threshold settings for subsequent notifications and improve the success rate of the notification mechanism. This technical solution ensures that updated graph notifications are delivered to the decision support module promptly and accurately, and provides a resend mechanism when necessary to prevent data loss or processing errors due to response delays or network fluctuations, thereby improving the reliability and real-time performance of the entire graph synchronization process. This technical solution plays a crucial role in the construction and optimization of the talent recommendation knowledge graph, especially in scenarios with frequent changes in talent skills or large-scale data updates. It effectively improves the response speed of the decision support module to graph updates and ensures the accuracy of decision-making. Under high concurrency or large-scale data updates, the system optimizes the graph synchronization process through automated notification and retry mechanisms, making talent recommendation and job matching more accurate and efficient.

[0043] This invention also provides a talent recommendation knowledge graph construction system that integrates multimodal data, used to implement the above-mentioned methods, such as... Figure 2 As shown, the system includes: The graph construction unit is used to automatically extract entities and associations based on mining multi-source relationship data of historical talents and positions, and to build a knowledge graph for talent recommendation. The data extraction unit is used to extract change events from multi-source data input and construct an initial talent change set; based on the initial talent change set, it searches the knowledge graph for nodes associated with industry preferences and job matching indices to determine the list of affected nodes. A node sorting unit is used to generate a sorted sequence of nodes to be processed based on the list of affected nodes; extract node data from the sorted sequence of nodes to be processed; and determine the update order of nodes in the relation edge weights. The rule adjustment unit is used to adjust the semantic association rules and relation edge weights according to the update order; and to perform storage system updates using the adjusted relation edge weights and semantic association rules, synchronizing entity node definitions and job matching indices. The notification push unit is used to trigger a notification mechanism through the updated graph, push the updated view to the decision support module, and obtain a response confirmation.

[0044] The present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0045] In summary, this invention mines multi-source relationship data of historical talent and positions, automatically extracts entities and associations between talent and positions, and constructs a complete knowledge graph. This provides foundational data support for subsequent recommendation and matching, ensuring that the graph reflects real market demand and talent capabilities throughout the recommendation process. Furthermore, it extracts talent change events from multi-source data, constructs an initial talent change set, and searches for associated nodes in the knowledge graph based on industry preferences and job matching indices, generating a list of affected nodes. This ensures the graph can be dynamically updated, reflecting changes between talent and positions in real time, and ensuring that talent recommendations are based on the latest data and meet industry needs and job requirements. By prioritizing the list of affected nodes, it ensures that during the graph update process, optimal talent is prioritized. First, nodes that significantly impact job matching are processed to identify the most important nodes. The update order is determined based on the weights of their relationships within the talent graph, avoiding over-processing of low-priority nodes and improving efficiency. Semantic association rules and relationship weights are adjusted according to the update order. Specifically, by analyzing node attributes and changes, association rules related to industry preferences and job requirements are adjusted to ensure more accurate job-talent matching. The adjusted weights and rules are applied to the storage system to ensure real-time graph synchronization after updates, enabling the decision support module to promptly obtain the latest recommendation information. Updating the graph view provides real-time feedback to the decision support module, ensuring recruiters receive accurate job matching results in the shortest possible time. This step demonstrates the efficiency and accuracy of the technical solution, playing a crucial role in the talent recommendation system. Through the synergy of these technical solutions, dynamically tracking talent changes, adjusting job matching rules in real-time, and updating graph data promptly, the accuracy of job matching is greatly improved. This solves the problems of data lag and inaccurate matching in traditional recommendation systems, providing enterprises with a more intelligent and efficient recruitment and talent management solution.

[0046] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0047] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0048] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a talent recommendation knowledge graph that integrates multimodal data, characterized in that, The method includes: Step S1: By mining multi-source relationship data of historical talents and positions, entities and associations are automatically extracted to construct a knowledge graph for talent recommendation; Step S2: Extract change events from multi-source data input to construct an initial talent change set; based on the initial talent change set, search for nodes in the knowledge graph that are associated with industry domain preferences and job matching index, and generate a list of affected nodes; Step S3: Sort the nodes in the affected node list according to priority to generate a sorted sequence to be processed; extract node data from the sorted sequence to be processed and determine the update order of the nodes in the relation edge weights. Step S4: Adjust the semantic association rules and relation edge weights according to the update order; perform storage system update using the adjusted relation edge weights and semantic association rules, and synchronize entity node definitions and job matching index; Step S5: Trigger a notification mechanism using the updated graph to push the updated view to the decision support module and obtain a response confirmation.

2. The method as described in claim 1, characterized in that, In step S2, an initial set of talent changes is constructed, including: By monitoring changes in data input from multiple sources, the system extracts affected talent skill tags and professional experience records from a pre-defined change log. Using a data source integration approach, the extracted talent skill tags and professional experience records are correlated to generate preliminary entity node definitions. Based on the initial entity node definition, an initial talent change set is generated. If there is missing data in the initial talent change set, complete data is obtained through supplementary queries, and the reliability of the data source is verified to form the final talent change set.

3. The method as described in claim 2, characterized in that, In step S2, the list of affected nodes is determined, including: Search for directly related nodes in the knowledge graph; analyze the semantic relevance of the searched nodes to industry preferences; filter nodes highly correlated with job matching index based on the semantic relevance; if the number of filtered nodes exceeds a preset threshold, adjust the search scope to narrow the node set; generate a preliminary list of affected nodes based on the filtered nodes; verify the accuracy of the matching between the preliminary list of affected nodes and industry preferences; adjust the preliminary list of affected nodes based on the verification results to generate the final node list.

4. The method as described in claim 1, characterized in that, In step S3, the sorted sequence to be processed is generated, including: If the number of nodes in the affected node list exceeds a preset threshold, nodes with high team collaboration scores and social network influence are processed first. The priority weight of these nodes is determined by analyzing their team collaboration scores. The update frequency of the knowledge graph is adjusted based on the priority weight and social network influence. The affected node list is then sorted using the adjusted update frequency. A sorted sequence of nodes to be processed is generated, and the rationality of the sorting logic is verified. Based on the verification results, the order of nodes in the sequence of nodes to be processed is adjusted to obtain the adjusted sequence of nodes to be processed.

5. The method as described in claim 4, characterized in that, In step S3, determining the update order of nodes in the relation edge weights includes: Detailed data of the first node is extracted from the sorted sequence of nodes to be processed. Learning curves and geographic distribution characteristics related to the first node are preloaded using a node attribute expansion mechanism. The association strength of the first node in the knowledge graph is analyzed based on the preloaded information. The update order of the first node in the relation edge weights is determined based on the association strength. If the update order does not meet a preset standard, the processing priority of the node is adjusted. A confirmation record of the update order is generated for the adjusted priority. The position of the node in subsequent processing is determined based on the confirmation record.

6. The method as described in claim 1, characterized in that, In step S4, the semantic association rules and relation edge weights are adjusted according to the update order, including: Based on the update order, obtain a graph snapshot from the historical version tracing record; analyze the differences between the professional experience record and industry domain preference of each node using the graph snapshot; determine the semantic association rules that need to be adjusted based on the differences; calculate the adjustment range of the relation edge weights for the semantic association rules; if the adjustment range exceeds a preset range, recalculate the relation edge weights; generate the adjusted relation edge weights based on the recalculation results; update the associated data in the knowledge graph based on the adjusted relation edge weights.

7. The method as described in claim 6, characterized in that, In step S4, the entity node definition and job matching index are synchronized, including: Using adjusted relational edge weights and semantic association rules, batch write operations are performed in a high-concurrency environment; through the batch write operations, updated data is written to the main storage system; for the written data, the entity node definition is updated synchronously; based on the synchronously updated entity node definition, the job matching index is adjusted. If the adjustment result of the job matching index does not meet the preset standard, the index is recalculated; the final job matching index is generated based on the recalculated result; and the update status of the knowledge graph is determined based on the final job matching index.

8. The method as described in claim 1, characterized in that, In step S5, obtaining response confirmation includes: generating an instant synchronization identifier based on the updated map; triggering an automated notification mechanism based on the instant synchronization identifier; pushing the updated view to the decision support module through the automated notification mechanism; if the decision support module does not respond in time, resending the updated view; obtaining the final response confirmation through the resent updated view; and recording the execution status of the notification mechanism based on the response confirmation.

9. A talent recommendation knowledge graph construction system integrating multimodal data, used to implement the method as described in any one of claims 1-8, characterized in that, The system includes: The graph construction unit is used to automatically extract entities and associations based on mining multi-source relationship data of historical talents and positions, and to build a knowledge graph for talent recommendation. The data extraction unit is used to extract change events from multi-source data input and construct an initial talent change set; based on the initial talent change set, it searches the knowledge graph for nodes associated with industry preferences and job matching indices to determine the list of affected nodes. The node sorting unit is used to generate a sorted sequence of nodes to be processed based on the list of affected nodes; extract node data from the sorted sequence of nodes to be processed; and determine the update order of nodes in the relation edge weights. The rule adjustment unit is used to adjust the semantic association rules and relation edge weights according to the update order; and to perform storage system updates using the adjusted relation edge weights and semantic association rules, synchronizing entity node definitions and job matching indices. The notification push unit is used to trigger a notification mechanism through the updated graph, push the updated view to the decision support module, and obtain a response confirmation.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method as described in any one of claims 1-8.

Images