Recruitment information standardization processing method and device

Abstract

This application relates to the field of information processing technology and discloses a method and apparatus for standardizing recruitment information processing. The method includes: acquiring a job posting data object containing a job identifier, field version number, and structured fields; matching the current field semantic version node based on the field version number; calculating the optimal evolution path to the target node based on a field evolution graph containing semantic version nodes and edge-weighted evolution paths; performing version conversion on the structured fields according to the path to obtain a unified field object; generating a standard structure object conforming to the standard field structure; parsing free text to obtain a semantic entity set; constructing a semantic constraint graph containing field nodes, text entity nodes, and edges based on a constraint rule base; and detecting conflicts and correcting the standard structure object. This method achieves cross-version semantic unification of recruitment information, improves the accuracy and efficiency of data processing, and ensures the reliability of data statistics.

Description

Technical Field

[0001] This application relates to the field of information processing technology, specifically to a method and apparatus for standardizing recruitment information processing. Background Technology

[0002] In the practical application of recruitment information management systems, the standardized processing of recruitment information is a key link in ensuring the accuracy of data statistical analysis and improving system operating efficiency. However, existing technologies still have many problems to be solved in the recruitment information processing process. On the one hand, implicit semantic conflicts, mutually exclusive benefit expressions, and inconsistent time conditions are prone to occur between structured fields and free text descriptions within the same job posting version. Existing systems can only complete the field mandatory field validation and cannot achieve semantic logic closed-loop detection between structured fields and free text, making it difficult to identify and correct various semantic-level contradictions.

[0003] On the other hand, with system upgrades and policy adjustments, the semantics of recruitment information fields continuously evolve. The expression and definition standards of fields such as salary, age, and benefits change, and historical data cannot maintain consistency with the new semantic standards. This leads to data statistical distortion, invalidation of conflict detection rules, and loss of reference value for cross-year data comparisons of the same position. Furthermore, existing technologies execute field version evolution processing and semantic conflict detection independently, lacking a mechanism for coordinated action. This fails to guarantee the internal logical consistency of data after field evolution and transformation, easily generating new semantic conflicts. It also makes it difficult to construct a dynamically evolving and logically stable standardized processing framework for recruitment information, thus hindering the intelligent development of recruitment information management systems and the full realization of data value. Summary of the Invention

[0004] The purpose of this application is to provide a standardized processing method and apparatus for recruitment information to solve the problems mentioned in the background art.

[0005] According to a first aspect of this application, a method for standardizing recruitment information processing is provided, comprising the following steps: obtaining a job posting data object, wherein the job posting data object includes a job identifier, a field version number, and a structured field; querying a pre-stored field version mapping table using the field version number as a search keyword to match and obtain the current field semantic version node corresponding to the structured field; calculating the optimal evolution path from the current field semantic version node to a predefined target field semantic version node based on a field evolution graph; wherein the field evolution graph includes multiple field semantic version nodes and evolution paths connecting adjacent field semantic version nodes, each evolution path having a corresponding edge weight; and sequentially performing version conversion processing on the structured field according to the traversal order of the field semantic version nodes in the optimal evolution path to obtain a unified field. The process involves: 1) Performing version conversion processing on the corresponding conversion rules based on the currently processed adjacent field semantic version nodes; 2) Generating a standard structure object that conforms to the standard field structure defined by the target field semantic version node based on the unified field object; 3) Semantically parsing the free text in the job posting data object to obtain a semantic entity set; 4) Constructing a semantic constraint graph based on the standard structure object and the semantic entity set; 5) Nodes of the semantic constraint graph include field nodes converted from standard fields in the standard structure object and text entity nodes converted from semantic entities in the semantic entity set, and the edges of the semantic constraint graph are constructed based on a semantic constraint rule library; 6) Detecting conflicts in the semantic constraint graph to obtain a conflict list, and performing correction processing on the standard structure object based on the conflict list.

[0006] Preferably, the process involves obtaining the first semantic definition text corresponding to the first node and the second semantic definition text corresponding to the second node in the adjacent field semantic version node pair; inputting the first semantic definition text and the second semantic definition text into a pre-trained Sentence-BERT model to obtain a first semantic vector and a second semantic vector; calculating the semantic deviation based on the first semantic vector and the second semantic vector; obtaining the number of first constraint rules corresponding to the first node and the number of second constraint rules corresponding to the second node, and calculating the rule difference based on the number of first constraint rules and the number of second constraint rules; and determining the edge weight of the evolution path between the adjacent field semantic version node pairs based on the weighted sum of the semantic deviation and the rule difference.

[0007] Preferably, calculating the optimal evolution path from the current field semantic version node to the target field semantic version node includes: initializing a path weight table, setting the cumulative path weight of the current field semantic version node to 0, and setting the cumulative path weights of other field semantic version nodes in the field evolution graph to infinity; starting from the current field semantic version node, traversing its adjacent field semantic version nodes, and updating the cumulative path weights of the adjacent field semantic version nodes in the path weight table based on the sum of the cumulative path weights of the current field semantic version node and the edge weights of the corresponding evolution paths; selecting the node with the smallest cumulative path weight among the unvisited nodes from the path weight table as the new current node, and repeating the traversal and update steps until the new current node is the target field semantic version node; backtracking to obtain the traversal order of the node with the smallest cumulative path weight from the current field semantic version node to the target field semantic version node, which is used as the optimal evolution path.

[0008] Preferably, semantic parsing is performed on the free text in the job posting data object to obtain a semantic entity set, including: inputting the free text into a RoBERTa-CRF named entity recognition model to obtain an entity sequence, wherein the entity sequence contains semantic entities and their corresponding entity types; inputting the entity sequence into a Biaffine relation extraction model to obtain a list of triples, wherein each triple in the triple list contains a first semantic entity, a second semantic entity, and a semantic relationship between the first semantic entity and the second semantic entity; and integrating the entity sequence and the triple list to obtain the semantic entity set.

[0009] Preferably, constructing the semantic constraint graph includes: converting each standard field in the standard structure object into a field node, the node attributes of which include field name, field value, and field source; converting each semantic entity in the semantic entity set into a text entity node, the node attributes of which include entity name, entity type, and entity value; traversing the field nodes and the text entity nodes, constructing corresponding edges for node pairs with corresponding constraint rules based on the semantic constraint rule library, the edge types including numerical constraint edges, proportional constraint edges, mutually exclusive edges, or time-dependent edges; and integrating the field nodes, text entity nodes, and constructed edges in the form of an adjacency list to generate the semantic constraint graph.

[0010] Preferably, conflict detection is performed on the semantic constraint graph to obtain a conflict list, including: performing a depth-first search on the mutually exclusive edges in the semantic constraint graph to detect closed-loop paths formed by the connection of mutually exclusive edges, and recording the closed-loop paths as mutually exclusive edge closed-loop conflicts; traversing the numerical constraint edges or proportional constraint edges in the semantic constraint graph to obtain the corresponding first node values ​​and second node values, calculating theoretical values ​​according to the constraint rules of the numerical constraint edges or proportional constraint edges, and calculating the deviation value between the theoretical value and the actual calculated value. If the deviation value is greater than the rule threshold of the numerical constraint edges or proportional constraint edges, it is recorded as a numerical interval conflict; integrating the mutually exclusive edge closed-loop conflicts and the numerical interval conflicts to generate the conflict list.

[0011] Preferably, the step of correcting the standard structure object based on the conflict list includes: for each conflict in the conflict list, determining the field node and text entity node involved; calculating the field confidence of the field node and the text confidence of the text entity node; if the text confidence is greater than the field confidence, modifying the field value of the corresponding field node in the standard structure object according to the entity value of the text entity node; if the field confidence is greater than the text confidence, recording a text conflict tag for the corresponding text entity node in the semantic entity set.

[0012] Preferably, the method for determining the credibility of the field includes: obtaining the ratio of the conflict-free runtime to the total runtime of the field semantic version corresponding to the field node within a preset time range, as the version stability; obtaining the historical values ​​of the field node in the historical recruitment information of the enterprise that published the job posting data object, calculating the average similarity between the historical values ​​and the current field values, as the enterprise's historical consistency rate; and determining the credibility of the field based on the weighted sum of the version stability and the enterprise's historical consistency rate.

[0013] The text credibility is determined by: obtaining the average of the entity confidence score output by the named entity recognition model and the relation confidence score output by the relation extraction model when extracting the text entity node from the free text, as the model confidence score; obtaining the ratio of the context text length of the text entity node in the free text to the preset text length, as the context completeness score; and determining the text credibility score based on the product of the model confidence score and the context completeness score.

[0014] A second aspect of this application also provides a standardized processing device for recruitment information, comprising: a data acquisition module for acquiring job posting data objects, the job posting data objects including job identifiers, field version numbers, and structured fields; a version matching module for querying a pre-stored field version mapping table using the field version number as a search keyword to match and obtain the current field semantic version node corresponding to the structured field; a path calculation module for calculating the optimal evolution path from the current field semantic version node to a predefined target field semantic version node based on a field evolution graph; wherein the field evolution graph includes multiple field semantic version nodes and evolution paths connecting adjacent field semantic version nodes, each evolution path having a corresponding edge weight; and a version conversion module for sequentially performing version conversion processing on the structured fields according to the traversal order of the field semantic version nodes in the optimal evolution path to obtain a unified field object; wherein... The system includes: a version conversion module for the corresponding conversion rules based on the currently processed adjacent field semantic version nodes; a standard generation module for generating a standard structure object that conforms to the standard field structure defined by the target field semantic version node based on the unified field object; a semantic parsing module for performing semantic parsing on the free text in the job posting data object to obtain a semantic entity set; a constraint graph construction module for constructing a semantic constraint graph based on the standard structure object and the semantic entity set; wherein the nodes of the semantic constraint graph include field nodes converted from standard fields in the standard structure object and text entity nodes converted from semantic entities in the semantic entity set, and the edges of the semantic constraint graph are constructed based on a semantic constraint rule library; and a conflict detection and correction module for performing conflict detection on the semantic constraint graph to obtain a conflict list, and performing correction processing on the standard structure object based on the conflict list.

[0015] This application effectively solves problems such as data inconsistency and semantic conflicts between structured fields and free text caused by cross-version evolution of recruitment information fields by constructing a dual-graph collaborative mechanism of field evolution graph and semantic constraint graph. The field evolution graph, combined with the optimal path algorithm, achieves minimum loss transformation of fields across different versions, ensuring consistency between historical data and new semantic standards and avoiding statistical distortion. The semantic constraint graph models the association between structured fields and text entities, accurately identifying various semantic conflicts through loop closure detection. The dual-graph linkage and reliable quantification decision-making mechanism ensure the logical consistency of data after evolution and achieve automated conflict correction. This solution significantly improves the accuracy and efficiency of standardized processing of recruitment information, reduces the workload of manual review, ensures the reliability of cross-version data comparison and statistical analysis, and significantly enhances the intelligence level and data value of the recruitment information management system. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0018] Figure 1 This is a schematic diagram of a standardized recruitment information processing method provided in an embodiment of this application;

[0019] Figure 2 This is a schematic diagram of the calculation of the optimal evolution path provided in the embodiments of this application;

[0020] Figure 3 This is a schematic diagram of the semantic constraint graph construction process provided in the embodiments of this application;

[0021] Figure 4 This is a schematic diagram of the modified standard structure object flow provided in the embodiments of this application;

[0022] Figure 5 This is a schematic diagram of a recruitment information standardization processing device provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be noted that all user information (including but not limited to user device information, user personal information, object information corresponding to device usage data, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, device usage data, etc.) involved in all embodiments of this application are information and data authorized by the user or fully authorized by all parties.

[0025] The following detailed description of the implementation process of the recruitment information standardization processing method described in this application, with reference to specific embodiments, is provided in detail. It should be noted that this embodiment is only used to explain this application and is not intended to limit the scope of protection of this application. Any conventional adjustments or substitutions made by those skilled in the art to each step without departing from the concept of this application should be included within the scope of protection of this application.

[0026] This method is applicable to recruitment information management platforms operating on a server architecture. Such platforms can encompass data input units such as mini-program publishing modules, PC backend publishing modules, third-party interface import modules, and data analysis service nodes. The standardization processing is executed by a server-side standardization processing engine, positioned between the business logic layer and the data layer, serving as a pre-processing layer for job posting data. The server's database pre-stores field version mapping tables, field evolution graph structures, semantic constraint rule bases, and pre-trained model parameter files. The standardization processing engine can receive job posting data transmitted from each input unit in real time, independently completing the standardization processing flow for each data entry. The processed standardized data is then written to the database, adapting to the semantic evolution requirements of fields brought about by system upgrades and policy changes. This method resolves semantic conflicts between structured fields and free text in recruitment information, achieving semantic uniformity across recruitment data versions.

[0027] like Figure 1 As shown in the figure, this application discloses a schematic diagram of a method for standardizing recruitment information, including the following steps: S1, obtaining a job posting data object, the job posting data object including a job identifier, a field version number, and a structured field; S2, using the field version number as a search keyword, querying a pre-stored field version mapping table to match and obtain the current field semantic version node corresponding to the structured field; S3, based on the field evolution graph, calculating the optimal evolution path from the current field semantic version node to a predefined target field semantic version node; wherein, the field evolution graph includes multiple field semantic version nodes and evolution paths connecting adjacent field semantic version nodes, each evolution path having a corresponding edge weight; S4, according to the traversal order of the field semantic version nodes in the optimal evolution path, sequentially performing version conversion processing on the structured field to obtain a unified... S5. Based on the unified field object, a standard structure object conforming to the standard field structure defined by the target field semantic version node is generated; S6. Semantic parsing is performed on the free text in the job posting data object to obtain a semantic entity set; S7. Based on the standard structure object and the semantic entity set, a semantic constraint graph is constructed; wherein, the nodes of the semantic constraint graph include field nodes converted from standard fields in the standard structure object and text entity nodes converted from semantic entities in the semantic entity set, and the edges of the semantic constraint graph are constructed based on a semantic constraint rule library; S8. Conflict detection is performed on the semantic constraint graph to obtain a conflict list, and the standard structure object is corrected based on the conflict list.

[0028] In some embodiments, for step S1, the standardization processing engine is in a continuous standby state and can receive job posting data objects transmitted by each data input unit of the recruitment information management platform in real time. The data object includes job identifier, field version number, structured fields, and also includes two types of free text data: free text description and free text job requirements. The job identifier is a unique identifier of string type used to distinguish different job posting data; the field version number is the identification information representing the version of the field semantic definition used by the data; the structured fields are structured data in the form of key-value pair sets, covering the core content of recruitment information such as salary range, education requirements, age range, and welfare tags.

[0029] For example, structured fields may contain key-value pairs corresponding to salary ranges and key-value pairs corresponding to bachelor's degree or above requirements. The key of each key-value pair is the field name, and the value is the specific numerical value or attribute of the field. After receiving the data object, the standardization processing engine performs data integrity verification, retaining only valid data objects with no missing data fields. For invalid data objects missing core fields, they can be directly marked as failing the review and feedback can be sent to the corresponding data input unit.

[0030] In some embodiments, for step S2, the field version number is used as the search keyword to query the pre-stored field version mapping table and match the current field semantic version node corresponding to the structured field.

[0031] For example, a field semantic version node is a collection representing the semantic definitions, constraint rules, and value ranges of recruitment information fields under different system versions, and is a basic building block of the field evolution graph. The standardization processing engine extracts the field version number from the valid job posting data object, and uses this field version number as the search keyword to perform a precise matching query in the field version mapping table pre-stored in the database. This field version mapping table stores the association relationship between all online field version numbers and their corresponding field semantic version nodes. Through this association relationship, the current field semantic version node corresponding to the structured field can be directly matched and obtained.

[0032] In actual implementation, if no matching relationship is found in the field version mapping table, the field semantic version node corresponding to the structured field is marked as the initial version node by default. This initial version node is the original baseline node for field semantic evolution, containing the most basic field semantic definition, constraint rules, and value range settings. The standardization processing engine records the matched or marked nodes as the current field semantic version node.

[0033] In some embodiments, for step S3, edge weights represent the evolution cost between nodes with different semantic versions of the field. The node traversal order with the lowest evolution cost is found by accumulating path weights, thereby achieving a unified conversion of the semantics of different versions of the field and ensuring the semantic integrity of the converted data.

[0034] Specifically, please refer to Figure 2 , Figure 2 This is a schematic diagram illustrating the process of calculating the optimal evolution path provided in an embodiment of this application. In S201, a field evolution graph is constructed. In this embodiment, the field evolution graph is a directed weighted graph, consisting of multiple field semantic version nodes and evolution paths connecting adjacent field semantic version nodes. Each evolution path corresponds to a unique edge weight, which represents the transformation cost from one field semantic version node to another adjacent node, including cost considerations for both semantic deviation and rule differences.

[0035] The nodes in the field evolution graph are semantic version nodes of the field under different system versions, covering the initial version node and the iterative version nodes formed after each system upgrade. The attributes of the nodes include the semantic definition text of the field, the set of constraint rules, and the range of values. The evolution path between nodes is a directed edge. The evolution path is only constructed between adjacent nodes that have a direct semantic evolution relationship. The direction of the evolution path is consistent with the actual evolution direction of the field semantics. For example, the salary field evolves from the version node of the interval definition to the version node of the interval plus the probationary period ratio definition, that is, a directed evolution path is constructed from the interval definition node to the interval plus the probationary period ratio node.

[0036] In S202, the edge weights of the evolution path are determined. The edge weights of the evolution path are the transformation cost quantification values ​​between adjacent field semantic version node pairs. They are determined by the weighted sum of semantic deviation and rule difference. The specific implementation process is as follows: First, the first node and the second node in the adjacent field semantic version node pair are obtained. The first semantic definition text corresponding to the first node and the second semantic definition text corresponding to the second node are extracted. This semantic definition text is a complete description of the semantics of the node field, including the core information such as the connotation, extension, calculation method, and association relationship of the field. At the same time, the number of first constraint rules corresponding to the first node and the number of second constraint rules corresponding to the second node are extracted. The number of constraint rules is the number of rules contained in the constraint rule set in the node attribute.

[0037] The first semantic definition text and the second semantic definition text are respectively input into the pre-trained Sentence-BERT model. After the model encodes the text, it outputs the first semantic vector and the second semantic vector respectively. The semantic vector is a fixed-dimensional numerical vector that can accurately represent the semantic features of the semantic definition text. The pre-training data of the Sentence-BERT model consists of field semantic description texts in the field of recruitment information. After pre-training, it has the ability to accurately encode the semantics of recruitment field.

[0038] The semantic deviation is calculated based on the first and second semantic vectors. The formula for calculating the semantic deviation is as follows: ,in This is the first semantic vector. For the second semantic vector, The cosine similarity calculation function has a value range of [-1, 1]. The higher the result value, the higher the similarity between the two semantic vectors, and the lower the corresponding semantic deviation. That is, the smaller the semantic difference between the semantic version nodes of the two fields. The semantic deviation value range is [0, 2].

[0039] The rule difference degree is calculated based on the number of first constraint rules and the number of second constraint rules. The formula for calculating the rule difference degree is as follows: ,in The number of first constraint rules. This represents the number of the second constraint rules. The maximum value calculation function has a rule difference range of [0,1]. The higher the result value, the greater the difference in constraint rules between the semantic version nodes of the two fields.

[0040] Edge weights are calculated using a weighted sum of semantic deviation and rule difference. The formula for calculating edge weights is as follows: ,in The edge weights of the evolution path. This is the weighting coefficient for semantic bias. This is the weighting coefficient for the rule difference. and The preset weighting coefficients were obtained by the system developers based on actual data statistics derived from a large number of recruitment information fields, and they satisfy the following conditions: The constraints, The value range is [0.5, 0.9]. The value range is [0.1, 0.5]. For example, it can be configured as follows: , This value configuration ensures that semantic deviation is the core consideration for evolution cost while also taking into account the impact of constraint rule differences on the evolution process, making the edge weights more closely match the actual field evolution needs. The edge weights range from [0, 1.9]. Higher values ​​indicate higher evolution costs between adjacent nodes, and greater semantic and rule losses after evolution.

[0041] In S203, the optimal evolution path is calculated. The optimal evolution path is the node traversal order with the minimum cumulative edge weight from the current field semantic version node to the predefined target field semantic version node, i.e., the evolution path with the minimum sum of semantic loss and rule loss. The target field semantic version node is the latest standard field semantic version node adopted by the current recruitment information management system, which is predefined by the system developers based on business needs and technical upgrade requirements. In this embodiment, the optimal evolution path is calculated using an improved Dijkstra algorithm, specifically including: initializing a path weight table containing all field semantic version nodes in the field evolution graph; configuring a cumulative path weight attribute for each node; setting the cumulative path weight of the current field semantic version node to 0; setting the cumulative path weight of all other nodes in the field evolution graph to infinity; and configuring an access status attribute for each node, initially setting all nodes to unvisited.

[0042] Starting with the current field semantic version node, mark its access status as visited. Traverse all adjacent field semantic version nodes of this node. For each adjacent node, calculate its temporary cumulative path weight using the following formula: ,in For temporary cumulative path weights, The cumulative path weight of the starting node. This is the edge weight of the evolution path between the starting node and its neighboring node. The calculated temporary cumulative path weight is compared with the cumulative path weight already configured for the neighboring node in the path weight table. If the temporary cumulative path weight is smaller, the cumulative path weight of the neighboring node is updated to the temporary cumulative path weight; otherwise, the original value remains unchanged.

[0043] Select all nodes whose access status is unvisited from the path weight table, filter out the node with the smallest cumulative path weight, and take it as the new current node. Mark the access status of the new current node as visited. Repeat the above steps of traversing adjacent nodes, calculating temporary cumulative path weight, and updating cumulative path weight until the new current node is the target field semantic version node.

[0044] Using the target field semantic version node as the endpoint, the node traversal order is backtracked based on the cumulative path weight of each node in the path weight table. That is, the current field semantic version node is traced back from the endpoint node. All nodes and evolution paths in this backtracking process are recorded. The nodes obtained by backtracking are arranged in ascending order to obtain the node traversal order with the minimum cumulative path weight from the current field semantic version node to the target field semantic version node. This order is the optimal evolution path.

[0045] In actual implementation, if the current field semantic version node and the target field semantic version node are the same node, there is no need to calculate the evolution path. The path with this node as the only node is recorded as the optimal evolution path, and subsequent steps are executed directly based on this node.

[0046] In some embodiments, for step S4, the standardization processing engine performs version conversion processing on the structured fields sequentially according to the traversal order of the field semantic version nodes in the optimal evolution path. Each conversion is based on the corresponding conversion rule of the adjacent field semantic version node being processed. The conversion rule is pre-stored in the semantic constraint rule library of the database and is pre-defined by the system developers based on the field semantic evolution characteristics of the adjacent nodes. It includes field recombination rules, field splitting rules, field addition rules, field assignment rules, etc. Different adjacent nodes correspond to unique conversion rules.

[0047] Specifically, starting from the starting node of the optimal evolution path, i.e., the current field semantic version node, each pair of adjacent nodes is processed sequentially according to the traversal order. For the currently processed pair of adjacent nodes, the corresponding transformation rule is first read from the semantic constraint rule library. Then, the structured field is reorganized according to the transformation rule, and the field structure of the current node is re-integrated according to the field semantic definition of the next node. For the newly added field in the next node, if there is a mapping rule for the field in the transformation rules, the relevant data is extracted from the original field of the current node and assigned according to the mapping rule. If there is no mapping rule, a default value is assigned to the newly added field. This default value is pre-set by the system developers based on the general data characteristics of the recruitment information domain. For the old field that needs to be split in the current node, it is split into multiple fields of the next node according to the field splitting rule in the transformation rules, and the value of the old field is distributed to each of the split fields according to the splitting ratio. After completing the version conversion of a set of adjacent node pairs, the content of the structured field is updated, and the updated structured field is used as the input for the next adjacent node pair conversion process, until the conversion process of all adjacent node pairs in the optimal evolution path is completed. The structured field obtained at this time is the unified field object, and the field semantics of the object are completely consistent with the definition of the target field semantic version node.

[0048] In some embodiments, for step S5, in specific implementation, the target field semantic version node predefines a standard field structure, which is the standard recruitment information field structure of the current system, including core standard fields such as basic salary, performance salary, subsidy salary, education requirements, lower age limit, upper age limit, benefits list, and working hour type.

[0049] The standardization processing engine reorganizes the unified field object according to the standard field structure. First, it extracts all fields from the unified field object and matches them one by one with the standard fields in the standard field structure. For fields that match successfully, their values ​​and attributes are directly retained in the corresponding standard fields. For fields that exist in the standard field structure but are missing from the unified field object, the standardization processing engine queries the default assignment rule corresponding to the missing field from the database's default rule base. Based on this rule, a default value is assigned to the missing field, and a source attribute is marked for the field. This attribute represents how the field's value was obtained; for example, the source attribute of a missing field is marked as "transformed and generated," while the source attribute of a successfully matched field is marked as "original data." After completing the matching, assignment, and source marking of all standard fields, a standard structure object containing all standard fields is generated. The field structure and semantic definition of this object conform to the requirements of the target field's semantic version node.

[0050] In some embodiments, for step S6, entity recognition and relation extraction are performed on the free text in the job posting data object using a pre-trained deep learning model to extract semantic entities and semantic relationships between entities. The deep learning model includes a RoBERTa-CRF named entity recognition model and a bilinear Biaffine relation extraction model.

[0051] In one embodiment, the RoBERTa-CRF named entity recognition model is an end-to-end deep learning model. Its structure includes an input layer, a Transformer encoding layer, a Bidirectional Long Short-Term Memory (BiLSTM) network layer, and a Conditional Random Field (CRF) layer. The layers are connected in a serial manner, with the output of the previous layer serving as the input of the next layer.

[0052] The model's input is a concatenated text of free text descriptions and job requirements from job posting data. The input layer is a tokenizer module, which first segments the concatenated text into discrete word sequences and assigns a pre-trained word embedding vector to each word. Next, the word sequences are normalized by padding with zeros or truncation to a fixed length, outputting a fixed-length word vector sequence to the Transformer encoding layer. The Transformer encoding layer is based on the RoBERTa model, employing a multi-layer bidirectional Transformer encoder structure. It uses a self-attention mechanism to capture the contextual semantic relationships of the word vector sequences. The system performs deep semantic encoding on the data, outputting a sequence of semantically encoded vectors containing contextual semantic information to a BiLSTM layer. The BiLSTM layer performs bidirectional temporal feature extraction on the semantically encoded vector sequence, capturing forward and backward temporal semantic features respectively. The two types of feature vectors are concatenated to output a sequence of feature vectors containing bidirectional temporal semantic information to a CRF layer. The CRF layer is a sequence labeling layer that performs sequence labeling on the feature vector sequence by learning the transition probability matrix of entity labeling. Each word is assigned an entity label containing entity type and entity boundary. The optimal entity labeling sequence is then obtained by decoding using the Viterbi algorithm. Finally, semantic entities and their corresponding entity types are extracted from this sequence to generate an entity sequence. Entity types include amount, proportion, time, welfare, age, education, etc.

[0053] In one embodiment, the training process of the model is as follows: An entity recognition annotation dataset in the recruitment information domain is constructed, and the free text is annotated with entities using the BIO annotation method. The dataset is divided into a training set, a validation set, and a test set. Based on the training set, the model is trained using the stochastic gradient descent optimization algorithm, with the cross-entropy loss of entity annotation as the loss function. The model parameters are updated through forward and backward propagation. During training, the training effect is verified in real time using the validation set. Training stops when the recognition accuracy of the validation set reaches a preset threshold. Finally, the model performance is evaluated using the test set, and the evaluation metrics include precision, recall, and F1 score.

[0054] In one embodiment, the Biaffine relation extraction model is a deep learning model based on bilinear scoring. Its structure includes a bidirectional LSTM encoding layer and a bilinear scoring layer, with each layer connected sequentially. The input to the model is the aforementioned entity sequence. The bidirectional LSTM encoding layer first performs vector encoding on each semantic entity in the entity sequence, converting the entity text into a fixed-dimensional entity feature vector. Then, it performs bidirectional temporal encoding on all entity feature vectors to capture the contextual association features between entities, and outputs a sequence of entity encoded vectors containing association information to the bilinear scoring layer.

[0055] The bilinear scoring layer calculates the relationship score between any two entities through bilinear transformation. This score represents the probability that a certain semantic relationship exists between the entities. The relationship label with the highest probability is assigned to each entity pair. If the probability of all relationship labels is lower than a preset threshold, it is determined that the entity pair has no significant semantic relationship. Finally, the entity pairs with significant semantic relationships and their corresponding relationship labels are combined into triples to generate a triple list. Each triple contains a first semantic entity, a second semantic entity, and the semantic relationship between them.

[0056] In one embodiment, the model is trained as follows: a relation extraction and annotation dataset for the recruitment information domain is constructed, entity pairs and their corresponding semantic relations are labeled, and the dataset is divided into a training set, a validation set, and a test set; based on the training set, an adaptive moment estimation optimization algorithm is used to train the model, with the cross-entropy loss of relation prediction as the loss function, and the model parameters are updated through forward propagation and backpropagation; during the training process, the training effect is verified in real time through the validation set, and training is stopped when the prediction accuracy of the validation set reaches a preset threshold; finally, the model performance is evaluated through the test set.

[0057] The standardization processing engine integrates the above entity sequence with the triple list, and associates and stores the semantic entities and entity types in the entity sequence with the semantic relationships between entities in the triple list, generating a semantic entity set. This set contains all semantic entity features extracted from the free text and the relationships between entities.

[0058] In some embodiments, for step S7, by constructing a semantic constraint graph containing field nodes and text entity nodes, structured data and text semantic data are modeled in a unified manner, and the semantic association and constraint relationship between the two is visualized. The technical principle is based on the business logic and semantic association features of the recruitment information domain, constructing corresponding edges for node pairs with constraint relationships, and representing the numerical, proportional, mutual exclusion, time and other constraint relationships between nodes through graph structure.

[0059] Specifically, please refer to Figure 3 , Figure 3 This is a schematic diagram of the semantic constraint graph construction process provided in an embodiment of this application.

[0060] In S301, the nodes of the semantic constraint graph are constructed. The nodes of the semantic constraint graph include two types: field nodes and text entity nodes. All nodes are configured with a unique node identifier to facilitate subsequent edge construction and graph traversal. Field nodes are converted from standard fields in the standard structure object. Each standard field corresponds to an independent field node. The attributes of the field node include field name, field value, and field source, which are completely consistent with the information of the corresponding standard field in the standard structure object. Text entity nodes are converted from semantic entities in the semantic entity set. Each semantic entity corresponds to an independent text entity node. The attributes of the text entity node include entity name, entity type, and entity value, which are completely consistent with the information of the corresponding semantic entity in the semantic entity set. The standardization processing engine first extracts all standard fields from the standard structure object and converts them into a list of field nodes. Then, it extracts all semantic entities from the semantic entity set and converts them into a list of text entity nodes. Finally, the two lists are integrated to obtain the node set of the semantic constraint graph, completing the node construction.

[0061] In S302, edges are constructed based on a pre-stored semantic constraint rule library in the database. The edges of the semantic constraint graph are constructed based on the pre-stored semantic constraint rule library in the database. This rule library is predefined by the system developers based on the business logic, semantic association features, and data constraint relationships in the recruitment information domain. It contains constraint rules between various types of nodes. The edge types include numerical constraint edges, proportional constraint edges, mutually exclusive edges, and time-dependent edges. All edges are undirected, representing bidirectional constraint relationships between nodes. Each edge is configured with a unique edge identifier and rule threshold. The rule threshold is the deviation tolerance range of the constraint relationship between nodes.

[0062] Specifically, numerical constraint edges represent the numerical operation constraint relationships between nodes. For example, a numerical constraint edge is constructed between the base salary field node, the performance salary field node, and the total salary text entity node, with the constraint rule that the sum of the base salary and the performance salary equals the total salary. Proportional constraint edges represent the proportional constraint relationships between nodes. For example, a proportional constraint edge is constructed between the probationary period salary field node and the regular salary field node, with the constraint rule that the ratio of the probationary period salary to the regular salary is a preset ratio. Mutually exclusive edges represent the mutually exclusive relationships between nodes, that is, the values ​​or attributes of two nodes cannot be true at the same time. For example, a mutually exclusive edge is constructed between the five social insurances and one housing fund welfare text entity node and the no social insurance welfare text entity node. Time dependency edges represent the temporal sequence or time range constraint relationships between nodes. For example, a time dependency edge is constructed between the work experience years text entity node and the onboarding time text entity node.

[0063] The standardization processing engine traverses the node set of the semantic constraint graph, extracts the attribute information of all nodes, and uses the node attributes as the retrieval basis to query the constraint rules corresponding to each node pair in the semantic constraint rule library. For node pairs with corresponding constraint rules, corresponding edges are constructed according to the rule type, and edge identifiers, constraint rules, and rule thresholds are configured for the edges. For node pairs without corresponding constraint rules, no edges are constructed, and finally, the edge set of the semantic constraint graph is generated.

[0064] In S303, a semantic constraint graph is generated by integrating the aforementioned node set and edge set into an adjacency list. The standardization processing engine integrates the node set and edge set to generate the semantic constraint graph in the form of an adjacency list. Each row of the adjacency list corresponds to a node, storing the node's identifier, node attributes, and the edge identifiers, attributes, and node identifiers of all edges connected to that node. This adjacency list-based semantic constraint graph enables efficient storage and fast traversal of nodes and edges, facilitating the execution of subsequent conflict detection algorithms. It also clearly represents the semantic associations and constraints between nodes, achieving unified modeling of standardized structured data and free text semantic data.

[0065] In some embodiments, for step S8, global semantic conflict detection is achieved by graph traversal of the semantic constraint graph, and automatic correction of the conflict is achieved by combining credibility calculation. The technical principle is to detect closed-loop conflicts and numerical deviation conflicts in the constraint graph through graph search algorithm, and to calculate the credibility of fields and text through quantification.

[0066] Specifically, please refer to Figure 4 , Figure 4 This is a schematic diagram of the modified standard structure object flow provided in the embodiments of this application. In S401, conflicts in the semantic constraint graph are detected. The core of conflict detection is to perform graph traversal on the semantic constraint graph, combine the depth-first search algorithm with numerical deviation calculation, detect mutually exclusive edge closed-loop conflicts and numerical interval conflicts, and integrate all conflict information to generate a conflict list. Specifically, it may include: mutually exclusive edge closed-loop conflicts refer to closed-loop paths in the semantic constraint graph formed by connecting several mutually exclusive edges. That is, starting from a certain node, passing through several nodes connected by mutually exclusive edges, and finally returning to the starting node, the mutually exclusive edge constraint rules in the closed-loop path are contradictory, resulting in semantic logic inconsistency. The standardization processing engine first extracts all mutually exclusive edges from the edge set of the semantic constraint graph, constructs a mutually exclusive edge subgraph containing only the original node set and mutually exclusive edges, and then uses a depth-first search algorithm to traverse the subgraph. Starting from each node, it traverses all possible paths and determines whether there is a closed-loop path that starts from the starting node and returns to the starting node. If so, it is determined to be a mutually exclusive edge closed-loop conflict and records relevant information about the conflict, including the conflicting node identifier, the mutually exclusive edge identifier, the closed-loop path, and the conflict type.

[0067] Numerical interval conflict refers to a situation in the semantic constraint graph where there are node pairs with numerical or proportional constraint edges, and the actual numerical calculation results deviate from the requirements of the constraint rules, with the deviation exceeding the rule threshold of the edge, resulting in inconsistent numerical logic. The standardization processing engine extracts all numerical and proportional constraint edges from the edge set of the semantic constraint graph and processes each edge individually. First, it extracts the values ​​of the two nodes corresponding to the edge, calculates the theoretical value according to the constraint rules of the edge, then performs actual calculations on the values ​​of the two nodes to obtain the actual calculated value, and finally calculates the deviation between the theoretical value and the actual calculated value.

[0068] For numerically constrained edges, the formula for calculating the deviation value is: ,in This is the deviation value. This is the actual calculated value. This is the theoretical value; for proportionally constrained edges, the formula for calculating the deviation value is: ,in This is the actual ratio value. The calculated deviation values ​​are compared with the corresponding rule thresholds for each edge. If either deviation value exceeds the rule threshold, a numerical range conflict is identified, and relevant information is recorded, including the conflict node identifier, edge identifier, constraint rule, theoretical value, actual calculated value, deviation value, and conflict type. The rule threshold is a preset deviation tolerance range, obtained by system developers based on numerical characteristics of the recruitment information domain. For example, for salary-related numerical constraint edges, the rule threshold can be configured to a reasonable monetary value.

[0069] The standardization engine integrates information on all detected mutual exclusion edge closed-loop conflicts and numerical range conflicts, assigns a unique conflict identifier to each conflict, and categorizes and stores conflict information according to conflict type, generating a conflict list. This list contains complete information on all conflicts, providing a basis for subsequent corrections to the standard structure object. In actual implementation, if no conflicts are detected, an empty conflict list is generated, and subsequent steps directly use the standard structure object as the standardization result.

[0070] In S402, standard structure objects are corrected based on the conflict list. The correction of standard structure objects is performed based on conflict information in the conflict list. The core is to provide a quantitative basis for conflict correction by calculating the field credibility of field nodes and the text credibility of text entity nodes. The specific implementation process is as follows: First, the field credibility is calculated. Field credibility is a quantitative value representing the credibility of the field node's numerical value, ranging from [0,1]. The higher the value, the more reliable the field node's numerical value. It is determined by a weighted sum of version stability and enterprise historical consistency rate. The calculation formula is: ,in For field credibility, For version stability, For the company's historical consistency rate, This is a weighting factor for version stability. This is the weighting coefficient for the company's historical consistency rate. and The preset weighting coefficients can be obtained by system developers based on data analysis of a large amount of recruitment information, and must meet the following requirements: The constraints, The value range is [0.5, 0.7]. The value range is [0.3, 0.5]. For example, it can be configured as follows: , .

[0071] Version stability is the ratio of conflict-free runtime to total runtime of the semantic version of the field corresponding to the field node within a preset time range. The preset time range is set by the system developers according to actual business needs; for example, it can be configured to the past six months, and the value range of version stability is [0,1]. Enterprise historical consistency rate is the average similarity between the historical value of the field node and the current value of the field in the historical recruitment information of the company that posted the job posting data. The similarity is calculated using the cosine similarity algorithm, and the value range of enterprise historical consistency rate is [0,1]. The standardization processing engine extracts the runtime data of the semantic version of the field and the company's historical recruitment data from the database, calculates the version stability and enterprise historical consistency rate respectively, and substitutes them into the formula to obtain the field credibility.

[0072] Simultaneously, text credibility is calculated. Text credibility is a quantified value representing the credibility of text entity node values, ranging from [0,1]. Higher values ​​indicate greater credibility of text entity node values. It is determined by the product of model confidence and context completeness, and the calculation formula is as follows: ,in For text credibility, For model confidence, For contextual completeness.

[0073] Here, model confidence is the average of the entity confidence output by the RoBERTa-CRF model and the relationship confidence output by the Biaffine model when extracting the text entity node from free text. The value ranges from [0,1], with higher values ​​indicating greater accuracy in extracting the text entity. Context completeness is the ratio of the length of the context text of the text entity node in the free text to a preset text length. If the context text length exceeds the preset text length, the context completeness value is 1. The preset text length can be obtained by system developers based on the average length of free text in recruitment information. The context completeness value ranges from [0,1], with higher values ​​indicating more complete contextual semantics for the text entity. The standardization processing engine extracts model confidence from the semantic entity set, performs context analysis on the free text to calculate context completeness, and substitutes this into the formula to obtain text credibility.

[0074] For each conflict in the conflict list, the standardization processing engine first identifies the field nodes and text entity nodes involved, extracts the field confidence and text confidence of these two nodes, and performs a conflict correction decision based on the numerical comparison of the two:

[0075] Specifically, if the text credibility is greater than the field credibility, indicating that the text entity node value extracted from the free text is more reliable, then the field value of the corresponding field node in the standard structure object is modified based on the entity value of the text entity node. Simultaneously, the modification record for this field is marked in the standard structure object, including the value before modification, the value after modification, and the basis for modification. If the field credibility is greater than the text credibility, indicating that the field node value in the standard structure object is more reliable, then the field node value is retained. At the same time, a text conflict label is recorded for the corresponding text entity node in the semantic entity set. This label indicates that there is a conflict between the text entity value and the field node value, and that the field node value is more valuable. After all conflict correction decisions are executed, the content of the standard structure object is updated, resulting in the corrected standard structure object.

[0076] In some embodiments, the method also includes reselecting and reprocessing the evolution path for version-induced conflicts. In practice, if the conflict list contains version-induced conflicts, that is, conflicts introduced by the semantic version evolution process of the field, it indicates that the currently selected optimal evolution path has generated new semantic conflicts during the version conversion process. In this case, the evolution path needs to be reselected, and subsequent processing steps need to be re-executed based on the new optimal evolution path.

[0077] Specifically, the evolutionary paths related to version-induced conflicts are first extracted from the optimal evolutionary paths. The edge weights of these paths are then weighted and increased by multiplying the original edge weights by a weighting coefficient greater than 1. This coefficient is calculated by the system developers based on the severity of the conflict and ranges from [1.2, 2.0]. For example, for severe version-induced conflicts, the weighting coefficient can be configured to 1.8. After the edge weights are increased, the conversion cost of the evolutionary path increases accordingly, making it more difficult to select as the optimal path in subsequent path calculations.

[0078] Based on the improved edge weights, the calculation steps for the optimal evolution path in this method are re-executed. The improved Dijkstra algorithm is used to recalculate the node traversal order with the minimum cumulative path weight from the current field semantic version node to the target field semantic version node, and this order is taken as the new optimal evolution path. For details of the specific execution process, please refer to the aforementioned embodiment, which will not be repeated in this embodiment.

[0079] The steps of version conversion processing, standard structure object generation, free text semantic parsing, semantic constraint graph construction, conflict detection, and standard structure object correction are re-executed according to the new optimal evolution path until there are no version-induced conflicts in the conflict list, or it is determined that there is no better evolution path. For details of the specific execution process, please refer to the aforementioned embodiment, which will not be repeated in this embodiment.

[0080] Optionally, after completing all the above processing steps, the standardization engine uses the corrected standard structure object as core data, integrates information such as the conflict list, optimal evolution path, and field and text credibility, and generates a standardized job posting data object. This object is then formatted according to database storage specifications and written to the database. Simultaneously, the core information of the standardization result is written to the Redis cache for subsequent fast queries and historical consistency metric statistics. At this point, the standardization process for a single job posting data entry is complete, and the standardization engine returns to the pending state to receive the next job posting data object and execute the above processing steps.

[0081] Therefore, this method achieves unified semantic transformation of recruitment information fields across different versions by constructing a field evolution graph, solving the problem of data statistical distortion caused by cross-version field evolution; it achieves unified modeling of structured fields and free text semantics by constructing a semantic constraint graph, and combines graph traversal algorithms to achieve global semantic logic closed-loop detection, solving the problem of implicit semantic conflicts within fields; through the collaborative linkage of the two graphs, the conflict detection results are fed back to the evolution path calculation, realizing dynamic adjustment of the evolution path, solving the problem of inconsistent data logic after evolution caused by the independence of field evolution and conflict detection. This method can independently complete the standardized processing of single job posting data, adapt to various data input units of recruitment information management platforms, and the processed recruitment data achieves version unification, semantic closed-loop verification, and conflict correction, ensuring semantic and logical consistency of cross-version recruitment data, significantly improving the quality of recruitment data and the accuracy of statistical analysis, while reducing the workload of manual review and improving the automation and intelligence level of recruitment information processing.

[0082] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0083] Please see Figure 5 , Figure 5This application provides a structural block diagram of a recruitment information standardization processing device. The device specifically includes: a data acquisition module 501, used to acquire job posting data objects, the job posting data objects including job identifiers, field version numbers, and structured fields; a version matching module 502, used to query a pre-stored field version mapping table using the field version number as a search keyword to match the current field semantic version node corresponding to the structured field; a path calculation module 503, used to calculate the optimal evolution path from the current field semantic version node to a predefined target field semantic version node based on a field evolution graph; wherein the field evolution graph includes multiple field semantic version nodes and evolution paths connecting adjacent field semantic version nodes, each evolution path having a corresponding edge weight; and a version conversion module 504, used to sequentially perform version conversion processing on the structured fields according to the traversal order of the field semantic version nodes in the optimal evolution path to obtain a unified field object; wherein, based on the currently processed adjacent... The field semantic version node performs version conversion processing on the corresponding conversion rules; the standard generation module 505 is used to generate a standard structure object that conforms to the standard field structure defined by the target field semantic version node based on the unified field object; the semantic parsing module 506 is used to perform semantic parsing on the free text in the job posting data object to obtain a semantic entity set; the constraint graph construction module 507 is used to construct a semantic constraint graph based on the standard structure object and the semantic entity set; wherein, the nodes of the semantic constraint graph include field nodes converted from standard fields in the standard structure object and text entity nodes converted from semantic entities in the semantic entity set, and the edges of the semantic constraint graph are constructed based on the semantic constraint rule library; the conflict detection and correction module 508 is used to perform conflict detection on the semantic constraint graph to obtain a conflict list, and perform correction processing on the standard structure object based on the conflict list.

[0084] It should be noted that the working process of each module in the recruitment information standardization processing device described in this embodiment can refer to the working process of the recruitment information standardization processing method described in the above embodiments, and the technical effect achieved is the same as that of the recruitment information standardization processing method described in the above embodiments, so it will not be repeated here.

[0085] The above description represents the preferred embodiments of the present invention. It should be noted that, for those skilled in the art, various improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for standardizing recruitment information processing, characterized in that, The process includes the following steps: obtaining a job posting data object, which includes a job identifier, a field version number, and a structured field; querying a pre-stored field version mapping table using the field version number as the search keyword to match and obtain the current field semantic version node corresponding to the structured field; calculating the optimal evolution path from the current field semantic version node to a predefined target field semantic version node based on the field evolution graph; wherein, the field evolution graph includes multiple field semantic version nodes and evolution paths connecting adjacent field semantic version nodes, and each evolution path has a corresponding edge weight; performing version conversion processing on the structured field sequentially according to the traversal order of the field semantic version nodes in the optimal evolution path to obtain a unified field object; wherein, based on the current processing... Adjacent field semantic version nodes perform version conversion processing on the corresponding conversion rules; based on the unified field object, a standard structure object conforming to the standard field structure defined by the target field semantic version node is generated; semantic parsing is performed on the free text in the job posting data object to obtain a semantic entity set; based on the standard structure object and the semantic entity set, a semantic constraint graph is constructed; wherein, the nodes of the semantic constraint graph include field nodes converted from standard fields in the standard structure object and text entity nodes converted from semantic entities in the semantic entity set, and the edges of the semantic constraint graph are constructed based on a semantic constraint rule library; conflict detection is performed on the semantic constraint graph to obtain a conflict list, and the standard structure object is corrected based on the conflict list.

2. The standardized processing method for recruitment information according to claim 1, characterized in that, Obtain the first semantic definition text corresponding to the first node and the second semantic definition text corresponding to the second node in the adjacent field semantic version node pair; input the first semantic definition text and the second semantic definition text into the pre-trained Sentence-BERT model respectively to obtain the first semantic vector and the second semantic vector; calculate the semantic deviation based on the first semantic vector and the second semantic vector; obtain the number of first constraint rules corresponding to the first node and the number of second constraint rules corresponding to the second node, and calculate the rule difference based on the number of first constraint rules and the number of second constraint rules; The edge weights of the evolution paths between adjacent field semantic version node pairs are determined based on the weighted sum of the semantic deviation and the rule difference.

3. The standardized processing method for recruitment information according to claim 1, characterized in that, Calculating the optimal evolution path from the current field semantic version node to the target field semantic version node includes: initializing a path weight table, setting the cumulative path weight of the current field semantic version node to 0, and setting the cumulative path weights of other field semantic version nodes in the field evolution graph to infinity; starting from the current field semantic version node, traversing its adjacent field semantic version nodes, and updating the cumulative path weights of the adjacent field semantic version nodes in the path weight table based on the sum of the cumulative path weights of the current field semantic version node and the edge weights of the corresponding evolution paths; selecting the node with the smallest cumulative path weight among the unvisited nodes in the path weight table as the new current node, and repeating the traversal and update steps until the new current node is the target field semantic version node; backtracking to obtain the traversal order of the node with the smallest cumulative path weight from the current field semantic version node to the target field semantic version node, which is used as the optimal evolution path.

4. The standardized processing method for recruitment information according to claim 1, characterized in that, Semantic parsing is performed on the free text in the job posting data object to obtain a semantic entity set, including: inputting the free text into a RoBERTa-CRF named entity recognition model to obtain an entity sequence, wherein the entity sequence contains semantic entities and their corresponding entity types; inputting the entity sequence into a Biaffine relation extraction model to obtain a list of triples, wherein each triple in the triple list contains a first semantic entity, a second semantic entity, and a semantic relationship between the first semantic entity and the second semantic entity; and integrating the entity sequence and the triple list to obtain the semantic entity set.

5. The standardized processing method for recruitment information according to claim 1, characterized in that, The construction of the semantic constraint graph includes: converting each standard field in the standard structure object into a field node, the node attributes of which include field name, field value, and field source; converting each semantic entity in the semantic entity set into a text entity node, the node attributes of which include entity name, entity type, and entity value; traversing the field nodes and the text entity nodes, constructing corresponding edges for node pairs with corresponding constraint rules based on the semantic constraint rule library, the edge types including numerical constraint edges, proportional constraint edges, mutually exclusive edges, or time-dependent edges; and integrating the field nodes, text entity nodes, and constructed edges in the form of an adjacency list to generate the semantic constraint graph.

6. The standardized processing method for recruitment information according to claim 5, characterized in that, The semantic constraint graph is subjected to conflict detection to obtain a conflict list, including: performing a depth-first search on the mutually exclusive edges in the semantic constraint graph to detect closed-loop paths formed by the mutual exclusive edges, and recording the closed-loop paths as mutually exclusive edge closed-loop conflicts; traversing the numerical constraint edges or proportional constraint edges in the semantic constraint graph to obtain the corresponding first node values ​​and second node values, calculating theoretical values ​​according to the constraint rules of the numerical constraint edges or proportional constraint edges, and calculating the deviation value between the theoretical value and the actual calculated value. If the deviation value is greater than the rule threshold of the numerical constraint edges or proportional constraint edges, it is recorded as a numerical interval conflict; integrating the mutually exclusive edge closed-loop conflicts and the numerical interval conflicts to generate the conflict list.

7. The standardized processing method for recruitment information according to claim 1, characterized in that, The step of correcting the standard structure object based on the conflict list includes: for each conflict in the conflict list, determining the field node and text entity node involved; calculating the field confidence of the field node and the text confidence of the text entity node; if the text confidence is greater than the field confidence, modifying the field value of the corresponding field node in the standard structure object according to the entity value of the text entity node; if the field confidence is greater than the text confidence, recording a text conflict tag for the corresponding text entity node in the semantic entity set.

8. The method for standardizing recruitment information processing according to claim 7, characterized in that, The method for determining the credibility of a field includes: obtaining the ratio of the conflict-free runtime to the total runtime of the field semantic version corresponding to the field node within a preset time range, as the version stability; obtaining the historical values ​​of the field node in the historical recruitment information of the enterprise that published the job posting data object, calculating the average similarity between the historical values ​​and the current field values, as the enterprise's historical consistency rate; and determining the credibility of the field based on the weighted sum of the version stability and the enterprise's historical consistency rate.

9. A method for standardizing recruitment information processing according to claim 7, characterized in that, The text credibility is determined by: obtaining the average of the entity confidence score output by the named entity recognition model and the relation confidence score output by the relation extraction model when extracting the text entity node from the free text, as the model confidence score; obtaining the ratio of the context text length of the text entity node in the free text to the preset text length, as the context completeness score; and determining the text credibility score based on the product of the model confidence score and the context completeness score.

10. A standardized processing device for recruitment information, characterized in that, include: The data acquisition module is used to acquire job posting data objects, which include job identifier, field version number, and structured fields; The version matching module is used to query a pre-stored field version mapping table using the field version number as the search keyword to match the current field semantic version node corresponding to the structured field; the path calculation module is used to calculate the optimal evolution path from the current field semantic version node to a predefined target field semantic version node based on the field evolution graph; wherein, the field evolution graph includes multiple field semantic version nodes and evolution paths connecting adjacent field semantic version nodes, and each evolution path has a corresponding edge weight; the version conversion module is used to perform version conversion processing on the structured field sequentially according to the traversal order of the field semantic version nodes in the optimal evolution path to obtain a unified field object; wherein, the version conversion processing is performed on the corresponding conversion rules based on the currently processed adjacent field semantic version nodes; The standard generation module is used to generate a standard structure object that conforms to the standard field structure defined by the semantic version node of the target field based on the unified field object; the semantic parsing module is used to perform semantic parsing on the free text in the job posting data object to obtain a semantic entity set; the constraint graph construction module is used to construct a semantic constraint graph based on the standard structure object and the semantic entity set; wherein, the nodes of the semantic constraint graph include field nodes converted from standard fields in the standard structure object and text entity nodes converted from semantic entities in the semantic entity set, and the edges of the semantic constraint graph are constructed based on a semantic constraint rule base; the conflict detection and correction module is used to perform conflict detection on the semantic constraint graph to obtain a conflict list, and to perform correction processing on the standard structure object based on the conflict list.

Images