Semantic analysis driven resume and post intelligent matching method and system

Abstract

The invention relates to the technical field of semantic analysis, and discloses a semantic analysis-driven resume and post intelligent matching method and system, and the method comprises the steps: carrying out the fine-grained semantic analysis of to-be-matched resume data and post data, and obtaining a first semantic feature tuple set and a second semantic feature tuple set; constructing a double-layer semantic index map for semantic matching, and analyzing the compatibility of feature entities in the first semantic feature tuple set to a topological structure of the double-layer semantic index map; in the double-layer semantic index atlas, feature matching nodes and semantic association nodes corresponding to the feature entities are retrieved according to the preliminary matching atlas, a local association atlas representing posts is constructed, the local association atlas is used for conducting directional propagation and energy attenuation on the feature entities, and energy distribution features are obtained; and according to the energy distribution characteristics, correlation scores for the resume data and the post data are generated, and sorting and matching are performed based on the correlation scores. According to the invention, the accuracy during intelligent matching can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semantic analysis, and particularly relates to a resume and post intelligent matching method and system driven by semantic analysis. BACKGROUND

[0002] In the field of human resource management and online recruitment, how to efficiently and accurately match a large number of resumes with recruitment posts is a long-standing and crucial problem. Existing automatic matching methods mainly rely on keyword matching, vector space model based on bag-of-words model or shallow machine learning model.

[0003] The existing technology is usually one-way and one-time calculation, and lacks the ability to mine implicit and networked associations in the matching process. For example, a resume may not directly meet all the requirements of the target post, but its skill combination is highly compatible with multiple posts associated with the company. This potential association value is difficult to be discovered and quantified under the existing technical framework. Therefore, the industry urgently needs an intelligent matching method that can perform deep semantic understanding, model complex association networks and quantify implicit matching relationships. SUMMARY

[0004] The present application provides a resume and post intelligent matching method and system driven by semantic analysis, which mainly aims to solve the problem of low accuracy in intelligent matching.

[0005] To achieve the above purpose, the present application provides a resume and post intelligent matching method driven by semantic analysis, comprising: performing fine-grained semantic analysis on the pre-acquired resume data and post data to be matched respectively to obtain a first semantic feature tuple set and a second semantic feature tuple set; constructing a double-layer semantic index graph for semantic matching based on the second semantic feature tuple set, analyzing the compatibility of the feature entities in the first semantic feature tuple set with the topological structure of the double-layer semantic index graph, and determining a preliminary matching graph related to the resume data to be matched according to the compatibility; In the double-layer semantic index graph, retrieving the feature matching nodes and semantic association nodes corresponding to the feature entities according to the preliminary matching graph; constructing a local association graph representing the post according to the feature matching nodes and the semantic association nodes, and performing directional propagation and energy attenuation on the feature entities using the local association graph to obtain energy distribution features for the propagation of feature entities; generating a relevance score for the resume data and post data according to the energy distribution features, and sorting and matching based on the relevance score.

[0006] To solve the above problems, the application further provides a resume and post intelligent matching system driven by semantic analysis, which comprises: A fine-grained semantic analysis module is configured to perform fine-grained semantic analysis on the pre-acquired resume data and post data to be matched respectively, and obtain a first semantic feature tuple set and a second semantic feature tuple set; A preliminary matching graph construction module is configured to construct a double-layer semantic index graph for semantic matching based on the second semantic feature tuple set, analyze the compatibility of feature entities in the first semantic feature tuple set with the topological structure of the double-layer semantic index graph, and determine a preliminary matching graph related to the resume data to be matched according to the compatibility; A node retrieval module is configured to retrieve, in the double-layer semantic index graph, a feature matching node and a semantic association node corresponding to the feature entities according to the preliminary matching graph; An energy distribution feature analysis module is configured to construct a local association graph representing a post based on the feature matching node and the semantic association node, perform directional propagation and energy attenuation on the feature entities by using the local association graph, and obtain an energy distribution feature for the propagation of the feature entities; A data matching module is configured to generate a correlation score for the resume data and the post data according to the energy distribution feature, and perform sorting and matching based on the correlation score.

[0007] The embodiment of the application converts unstructured text into structured semantic representation through fine-grained semantic analysis and semantic feature tuple set construction, thereby solving the semantic gap problem existing in keyword matching and realizing accurate semantic understanding of the content; the double-layer semantic index graph is constructed and the topological compatibility is analyzed, the complex association between posts and skills is modeled as a calculable network structure, thereby realizing deep and interpretable preliminary matching based on network topology; the energy directional propagation and attenuation model based on the local association graph simulates the diffusion process of skill influence in the post network, thereby solving the problem of being unable to mine implicit and networked associations; the correlation score is generated by fusing structure matching and dynamic propagation energy, thereby effectively improving the accuracy and rationality of resume and post matching. Therefore, the resume and post intelligent matching method and system driven by semantic analysis provided by the application can solve the problem of low accuracy in intelligent matching. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 A flowchart of the semantic analysis driven resume and post intelligent matching method provided by an embodiment of the application is shown in the figure; Figure 2 A functional module diagram of the semantic analysis driven resume and post intelligent matching system provided by an embodiment of the application is shown in the figure.

[0009] The objectives, functional characteristics and advantages of the present application will be further illustrated in conjunction with the embodiments with reference to the accompanying drawings. DETAILED DESCRIPTION

[0010] It should be understood that the specific embodiments described herein merely exemplify the present application and are not intended to limit the present application.

[0011] Embodiments of the present application provide a resume and post intelligent matching method driven by semantic analysis. The execution subject of the resume and post intelligent matching method driven by semantic analysis includes but is not limited to at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiments of the present application. In other words, the resume and post intelligent matching method driven by semantic analysis can be executed by software or hardware installed in a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to a single server, a server cluster, a cloud server, or a cloud server cluster, etc. The server can be a stand-alone server, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and big data and artificial intelligence platforms, etc. basic cloud computing services.

[0012] Reference Figure 1 Fig. 1 is a flowchart of a resume and post intelligent matching method driven by semantic analysis provided by an embodiment of the present application. In the embodiment, the resume and post intelligent matching method driven by semantic analysis includes: S1, performing fine-grained semantic analysis on pre-acquired resume data and post data to be matched respectively to obtain a first semantic feature tuple set and a second semantic feature tuple set.

[0013] In the embodiments of the present application, the resume data to be matched refers to text information provided by a job seeker to describe his / her educational background, work experience, and skill expertise; the post data refers to text information published by an employer to describe the job responsibilities and requirements of the position. The first semantic feature tuple set refers to a structured semantic feature set parsed from the resume data. The second semantic feature tuple set refers to a structured semantic feature set parsed from the post data.

[0014] In the embodiments of the present application, the fine-grained semantic analysis on the pre-acquired resume data and post data to be matched respectively to obtain the first semantic feature tuple set and the second semantic feature tuple set includes: performing multi-granularity word segmentation and part-of-speech tagging on the resume data and post data to be matched based on a pre-set domain ontology dictionary, and identifying named entities and core verb phrases in the resume data and post data; extracting a triple structure for skills according to the grammatical relationship between the named entity and the core verb phrase; According to the hierarchical and synonymous relationship in the domain ontology dictionary, the object in the triple structure is semantically normalized, and different words with the same semantics are mapped to the same standard concept; According to the preset data source, the triple structure after semantic normalization is aggregated into a first semantic feature tuple set and a second semantic feature tuple set respectively.

[0015] In detail, the domain ontology dictionary is a predefined knowledge base containing concepts, attributes, relationships and axioms of a specific field (such as computer software). For example, it is defined that Java is a programming language, Spring is a development framework, and master, proficient and familiar are verbs representing skill levels. Load the domain ontology dictionary as a custom dictionary using natural language processing tools (such as HanLP, spaCy) to perform word segmentation and part-of-speech tagging on the input text. Multi-granularity segmentation refers to combining maximum forward matching and statistical model-based segmentation methods to segment the text; part-of-speech tagging is to label the part-of-speech of each segmented result, such as Java as a noun and proficient as a verb. According to the labeling results and the dictionary, named entities (such as Java, MySQL) representing skills, tools, projects, etc. and core verb phrases (such as proficient, responsible for design) representing actions or states are identified. For example, from the resume sentence "proficient in Java and Spring framework", the named entities Java, Spring framework and the core verb phrase proficient are identified.

[0016] Specifically, the triple structure is a semantic unit representing the subject-predicate-object relationship. The dependency syntax analysis is performed on the segmented and annotated sentence to obtain the syntax dependency relationship between words. The core verb phrase is taken as an action, and the named entity dominated by the core verb phrase is taken as a subject or object to construct a triple of the form (subject, action, object). For example, from "I use Spring Boot to develop microservices", two triples (subject: developer, action: use, object: Spring Boot) and (subject: developer, action: develop, object: microservice) are extracted. For the focus skill, the subject can be defaulted to the job seeker or the job requirement, thereby simplifying it to an (action, object) pair, such as (master, Java). The domain ontology dictionary is queried, and if the object word in the triple has a synonym or is considered as a more general concept of a lower-level word, the object word is replaced by the standard concept. For example, JDK8 and Java8 are all normalized to the standard concept Java, and MySQL database and MySql are normalized to MySQL. The action word can also be normalized, that is, different words with the same semantics are mapped to the same standard concept, such as mastering and being proficient in being normalized to mastering, so that different expressions can be unified and the interference caused by lexical differences is eliminated.

[0017] Further, the source of each triple is distinguished between a resume text or a job description text. All triples derived from the same resume are aggregated together to form a first semantic feature tuple set for the resume. Similarly, all triples derived from the same job description are aggregated to form a second semantic feature tuple set for the job. Each tuple can include a normalized action, object, and possibly a context modifier (such as project duration). The unstructured text data is converted into a structured semantic feature set, laying a foundation for subsequent network analysis.

[0018] S2, constructing a double-layer semantic index graph for semantic matching based on the second semantic feature tuple set, analyzing the compatibility of feature entities in the first semantic feature tuple set to the topological structure of the double-layer semantic index graph, and determining a preliminary matching graph related to the to-be-matched resume data according to the compatibility.

[0019] In the embodiment of the application, the double-layer semantic index graph is a graph data structure for representing and retrieving complex association relationships between jobs and skills, and is composed of a job main network and a concept sub-network that are coupled to each other.

[0020] In the embodiment of the application, the double-layer semantic index graph for semantic matching is constructed based on the second semantic feature tuple set, comprising: Each job is taken as a job node, a first connection edge is established between the job nodes according to the hierarchical relationship between the jobs, and a job main network representing the aggregation relationship of the jobs is constructed according to the job nodes and the first connection edges. Taking the normative concept in each tuple in the second semantic feature tuple set as a concept node, establishing a second connection edge between concept nodes with existence logic relationship, and constructing a concept sub-network representing skill association relationship according to the concept nodes and the second connection edge; Identifying a first index number of each post node in the post main network and a second index number of each concept node in the concept sub-network; According to the first index number and the second index number, cross-layer linking each post node in the post main network to all the post corresponding concept nodes in the concept sub-network, to obtain a double-layer semantic index graph.

[0021] In detail, the post main network is a graph, in which nodes represent different posts. The membership, sequence or functional similarity relationship between posts is extracted from post data. For example, there may be a promotion sequence relationship between Java junior engineer, Java senior engineer and technical manager, and a connection edge is established between their corresponding nodes, and the weight of the edge can reflect the closeness of the relationship. For example, by analyzing the similarity of post description text, a connection can be established for functionally similar posts (such as back-end development and server development). The concept sub-network is a graph, in which nodes represent normalized skill concepts extracted from all post requirements (such as Java, distributed system). The logical relationship between skill concepts can be defined based on domain knowledge (such as Spring depends on Java), co-occurrence frequency (frequently appearing in the same post requirement) or semantic similarity. For example, the co-occurrence frequency of skill pairs in all post data is counted, and if it exceeds a threshold, a connection edge is established between the corresponding concept nodes, and the weight is positively correlated with the co-occurrence strength, thereby constructing a network reflecting the inherent association between skills.

[0022] Specifically, the first index number is a unique identification number allocated to each post node in the post main network, which is used to realize the cross-layer association of the post node and the concept node. The second index number is a unique identification number allocated to each concept node in the concept sub-network, which is used to realize the cross-layer association of the concept node and the post node. A unique identifier is allocated to each node to facilitate efficient linking and retrieval. A self-incrementing integer numbering method is used to allocate a unique first index number to each post node in the post main network, and the numbering starts from 1 and increases sequentially, for example, the first index number of a junior data analyst is 1, the first index number of a data analyst is 2, and the first index number of a senior data analyst is 3. During the allocation process, the nodes are numbered in the order of low to high post level, which facilitates quick identification of the level attributes of the post through the index number. Similarly, a self-incrementing integer numbering method is used to allocate a unique second index number to each concept node in the concept sub-network, and the numbering starts from 100 and increases sequentially, for example, the second index number of Python programming is 101, the second index number of SQL query is 102, the second index number of machine learning is 103, and the second index number of data visualization is 104. During the allocation process, the nodes are sorted according to the popularity and association breadth of the skills, and the skills with high popularity and wide association are allocated to the index numbers at the front, which facilitates subsequent quick retrieval. To ensure the uniqueness and stability of the index number, a mapping table of index number and node information is established and stored in the metadata of the graph database. When a node is added or deleted, the mapping table is updated in time to ensure the one-to-one correspondence between the index number and the node.

[0023] Further, for each post node (such as Java backend engineer) in the post main network, according to the corresponding second semantic feature tuple set, all skill concept nodes (such as Java, Spring, MySQL) required by the post in the concept sub-network are queried, and cross-layer links are established between them, thereby forming a double-layer network: the upper layer is a post association network, and the lower layer is a skill association network, and the two layers are tightly coupled through post-skill links. As a whole, the graph constitutes an index structure for semantic matching.

[0024] In the embodiment of the application, the compatibility of the topological structure is an index for quantifying the matching degree of the structure of the resume skill combination and the post skill network.

[0025] In the embodiment of the application, the analysis of the compatibility of the topological structure of the double-layer semantic index graph by the feature entities in the first semantic feature tuple set comprises: mapping each feature entity in the first semantic feature tuple set to the corresponding node in the concept sub-network of the double-layer semantic index graph to obtain a mapping node; constructing a mapping sub-graph according to the mapping node, and taking the ratio of the structure cohesion of the mapping sub-graph to the global average cohesion of the concept sub-network as a first topological index; The coincidence degree between each node in the mapping subgraph and the directly associated post node in the double-layer semantic index graph is taken as a second topology index; The first topology index and the second topology index are weighted and fused to obtain the compatibility degree of the topology structure of the double-layer semantic index graph.

[0026] In detail, the mapping node refers to a node in the concept subnetwork corresponding to the resume characteristic entity (skill), and the characteristic entity refers to a normalized concept in the tuple. Each normalized skill concept in the first semantic feature tuple set obtained by parsing the resume is found in the concept subnetwork to find a concept node with the same name. For example, the resume skill Java is mapped to the node named Java in the concept subnetwork. The structural cohesion degree is an index for measuring the close degree of connection of nodes in a subgraph in graph theory, and is represented by an average clustering coefficient or a subgraph density. A mapping subgraph is formed by all mapping nodes and the connection edges between them in the concept subnetwork. The density (ratio of actual edge number to maximum possible edge number) of the mapping subgraph is calculated as its structural cohesion degree. At the same time, the average density of the entire concept subnetwork is calculated as a global benchmark, and the ratio (mapping subgraph density / global average density) of the two is taken as the first topology index. The index reflects the cohesion of the skill combination possessed by the resume in the skill network. If the ratio is greater than 1, it means that the skills are closely related and may form a professional direction; if the ratio is close to or less than 1, it means that the skills are relatively scattered.

[0027] Specifically, the coincidence degree is used to measure the consistency of the set of posts associated with the resume skills. For each skill node in the mapping subgraph, all post nodes directly associated with it are found through cross-layer linking to obtain multiple post sets. The ratio (Jaccard similarity) of the intersection size to the union size of these post sets is calculated as the second topology index. The index reflects the concentration degree of the post direction pointed to by the resume skills. A high ratio means that the skill combination points to a few specific posts; a low ratio means that the skill combination is relatively generalized and can be associated with multiple posts. The first topology index (cohesion) and the second topology index (directivity) are assigned a preset weight (such as 0.6 and 0.4, respectively), and a weighted sum is performed to obtain the final topology compatibility score. The score comprehensively evaluates the matching quality of the resume skill combination and the entire post skill network in structure.

[0028] Further, a compatibility threshold is set. All post nodes in the post main network are traversed, and the topology compatibility of the subgraph composed of the post node and its directly associated skill nodes with the resume characteristic entity is calculated. All post nodes with a compatibility score higher than the threshold and their directly associated skill nodes are extracted from the double-layer semantic index graph to form the preliminary matching graph. The graph narrows down the search range and focuses on potential matching posts with high structural compatibility with the resume and their related skill networks.

[0029] Further, the constructed double-layer semantic index graph and the determined preliminary matching graph are core ranges for searching the feature matching nodes and the semantic association nodes. The preliminary matching graph filters out the post nodes and the skill nodes related to the resume data, thereby avoiding the low efficiency problem caused by searching in the entire double-layer semantic index graph. Meanwhile, the topological structure of the double-layer semantic index graph provides key association paths and data support for searching the direct matching nodes, the potential association nodes and the semantic association nodes. Only based on the constructed structured graph, the node searching can be accurately and efficiently completed.

[0030] S3, searching the feature matching nodes and the semantic association nodes corresponding to the feature entity in the double-layer semantic index graph according to the preliminary matching graph.

[0031] In the embodiment of the application, the feature matching node refers to a post node having a direct and explicit corresponding relationship with the skill in the resume in the preliminary matching graph. The semantic association node refers to a post node indirectly related through post network association in the preliminary matching graph.

[0032] In the embodiment of the application, searching the feature matching nodes and the semantic association nodes corresponding to the feature entity in the double-layer semantic index graph according to the preliminary matching graph includes: filtering out, from the post nodes in the preliminary matching graph, a post node having a direct cross-layer link with the skill node mapped by the feature entity, marking the filtered post node as a feature matching node set, and determining the node in the feature matching node set as a feature matching node; taking the feature matching node set as a starting point, querying a target post node having a strong association with the post node in the post main network in the double-layer semantic index graph within a preset hop range according to the connection edge corresponding to the post node in the feature matching node set, and marking the target post node as a potential association node set; querying, for each node in the potential association node set, a skill node in the double-layer semantic index graph, and calculating a semantic similarity between the skill node and the feature entity; taking a node having a semantic similarity higher than a preset association threshold as a semantic association node in a semantic association node set.

[0033] In detail, in the preliminary matching graph, for each feature entity (skill node) of the resume, find all job nodes directly connected to the skill node through cross-layer links. For example, the resume skill Java is directly linked to the job Java backend engineer and Android development engineer, then these two job nodes are selected into the feature matching node set, which represents the job that the resume skill meets the requirements. From each job node in the feature matching node set, perform a breadth-first search in the job main network to explore the neighbor job nodes within 1 to 2 hops. Only keep the neighbor nodes connected to the starting node with an edge weight higher than a certain strength threshold, which are considered as strong correlation jobs (such as other positions in the department, positions on the promotion path). All these discovered neighbor nodes constitute the potential correlation node set.

[0034] Specifically, for each job node in the potential correlation node set, find all skill nodes required by the job through cross-layer links. Calculate the semantic similarity of these skill nodes with the resume feature entity (i.e. the original skill node). The semantic similarity can be calculated by cosine similarity based on word vectors, or by using concept distance in the domain ontology dictionary. Set a correlation threshold, if the total average semantic similarity of the required skills of a potential correlation job node with the resume skills is higher than the correlation threshold, it is considered that there is a meaningful semantic correlation between the job and the resume, and it is included in the semantic correlation node set. For example, although the resume does not directly mention cloud computing, its Java and distributed system skills are highly similar to the skills required by the cloud computing development engineer job, so the job may be determined as a semantic correlation node. These nodes represent jobs that the resume may be indirectly suitable for or have potential for.

[0035] Further, the retrieved feature matching nodes and semantic correlation nodes are the core node basis for constructing the local correlation graph. The local correlation graph needs to be constructed around these nodes and their associated skill nodes, the feature matching nodes provide the core matching jobs, the semantic correlation nodes provide the correlation expansion jobs, and the two together constitute the job node core of the local correlation graph; at the same time, the skill nodes corresponding to these nodes and their association provide key data support for calculating the comprehensive correlation weight, energy propagation and attenuation, only based on these node data, the local correlation graph that can accurately quantify the correlation strength between the job and the resume can be constructed.

[0036] S4, constructing a local correlation graph representing a job according to the feature matching node and the semantic correlation node, using the local correlation graph to perform directed propagation and energy attenuation on the feature entity, to obtain an energy distribution feature propagated for the feature entity.

[0037] In the embodiment of the present application, the local correlation graph is a closely connected subgraph extracted from the double-layer semantic index graph, containing feature matching nodes, semantic correlation nodes and their directly correlated skills.

[0038] In the embodiment of the present application, the local correlation graph representing the post is constructed according to the feature matching nodes and the semantic correlation nodes, comprising: extracting skill nodes having a direct correlation relationship with the feature matching nodes and the semantic correlation nodes in the concept subnetwork; performing subgraph search on the double-layer semantic index graph according to the feature matching nodes, the semantic correlation nodes and the skill nodes to obtain a target closed subgraph; identifying the original connection strength of the feature matching nodes and the semantic correlation nodes in the post main network, and identifying the link weight of the skill nodes; calculating the comprehensive correlation weight between all nodes in the target closed subgraph according to the original connection strength and the link weight; constructing a local correlation graph representing the post according to the comprehensive correlation weight and the target closed subgraph.

[0039] In detail, the direct correlation relationship refers to the correlation established through cross-layer links, and all skill concept nodes directly connected to the feature matching nodes or the semantic correlation nodes through cross-layer links are queried. In the extraction process, the corresponding second index number is collected by traversing the cross-layer links of each feature matching node and semantic correlation node, and then the corresponding skill nodes are extracted from the concept subnetwork according to the second index number, so as to ensure that no directly correlated skill node is missed. Taking all the above nodes as a seed set, in the double-layer semantic index graph, a maximum connected subgraph containing the seed nodes and all edges (including the edges in the post main network, the edges in the concept subnetwork and the cross-layer links) connecting these nodes is extracted, that is, the subgraph search adopts a depth-first search algorithm, taking the feature matching nodes, the semantic correlation nodes and the skill nodes as core nodes, searching all correlation edges between these nodes, including the first connection edges in the post main network, the second connection edges in the concept subnetwork and the cross-layer links. This subgraph is the target closed subgraph, which ensures that all correlation relationships between the nodes in the subgraph are completely retained without missing key correlation paths, and focuses on the local post-skill network highly related to the current resume.

[0040] Specifically, the comprehensive correlation weight is used to quantify the correlation strength between any two connected nodes in the local correlation graph. The original connection strength between the feature matching node and the semantic correlation node in the post main network, that is, the weight of the first connection edge between these post nodes; the link weight of the skill node includes two parts: one is the weight of the second connection edge between the skill nodes, and the other is the weight of the cross-layer link between the skill nodes and the post nodes. By querying the weight attribute of each correlation edge in the target closed subgraph, the corresponding weight value is directly extracted. For the node pair that does not exist the correlation edge, the original connection strength or the link weight is set to 0.

[0041] Further, the calculation of the comprehensive correlation weight adopts different calculation methods according to the different node types. The comprehensive correlation weight between the post nodes is the original connection strength (the weight of the first connection edge), that is, for the edge inside the post main network or the concept subnetwork, the original weight is adopted; the comprehensive correlation weight between the skill nodes is the link weight (the weight of the second connection edge), that is, for the cross-layer link (post-skill), the weight can be set based on the requirement degree (such as necessary or priority) of the post to the skill; the comprehensive correlation weight between the post node and the skill node is the link weight (the weight of the cross-layer link), that is, for the node pair that has no direct edge but is indirectly correlated through the shared neighbor node, the link weight can be calculated. Finally, the comprehensive correlation weight value is given to each edge in the local correlation graph, and the local correlation graph is constructed to form a structured graph based on the node set of the target closed subgraph and the weight of the correlation edge. The graph includes the post nodes, the skill nodes, and the correlation edges with the comprehensive correlation weight between the nodes. The local correlation graph is stored and visualized by using the graph database, which is convenient for subsequent energy propagation calculation.

[0042] In the embodiment of the application, the energy distribution feature is a quantitative value accumulated on each post node after the influence of the resume skill on the local post network is propagated and attenuated.

[0043] In the embodiment of the application, the energy distribution feature is a quantitative value accumulated on each post node after the influence of the resume skill on the local post network is propagated and attenuated. Each of the feature entities is initialized as an energy source, and the initial energy value corresponding to the energy source is uniformly distributed to the skill nodes corresponding to the feature entities in the local correlation graph; According to the comprehensive correlation weight, energy is propagated from the skill nodes to the post nodes and between the post nodes; In the process of energy propagation along the connection edge, the initial energy value is attenuated according to the exponential function of the comprehensive correlation weight for each edge, and when the attenuated energy value is lower than a preset propagation threshold, the energy propagation in the propagation direction corresponding to the current connection edge is terminated; When the energy propagation of all connection edges is stable or terminated in the energy propagation process, record the accumulated energy value of each post node, and aggregate the energy values of all post nodes into the energy distribution feature.

[0044] In detail, assuming that the resume has N characteristic entities (skills), initialize N energy sources, and the total energy of each energy source is 1. The energy of each energy source is evenly distributed to the skill nodes in the local association graph to which it is mapped. For example, if the resume has Java and Spring skills, the Java energy source has 0.5 energy assigned to the Java node in the graph, and the Spring energy source has 0.5 energy assigned to the Spring node in the graph. For skill nodes that do not have a direct corresponding characteristic entity, the initial energy value is 0, and an energy storage unit is established for each skill node to record the initial energy value, ensuring the accuracy and uniformity of energy distribution. Set the directionality of energy propagation, and energy first flows from skill nodes to post nodes directly connected through cross-layer links (simulating the direct influence of skills on posts). Between post nodes, energy propagates along the edges of the post main network (simulating the associated influence between posts), and energy cannot flow back from post nodes to skill nodes. The propagation process uses an iterative propagation method, and in each iteration, energy is distributed according to the comprehensive association weight, i.e., the energy of a node is distributed to each adjacent node according to the proportion of its comprehensive association weight to the total comprehensive association weight of all adjacent nodes. After each iteration, update the energy value of each node, and the number of iterations is preset to 5 times, or until the energy propagation is stable (i.e., the energy value change of all nodes in the adjacent two iterations is less than the preset stability threshold 0.01).

[0045] Specifically, when energy propagates from a node u along an edge to node v, the energy value received by node v is: the current energy value of node u multiplied by a decay factor, which is an exponential function of the edge comprehensive association weight, for example, set as where is the normalized comprehensive association weight (between 0 and 1), is the decay coefficient, and the higher the weight of the edge, the smaller the decay. At the same time, a propagation threshold (such as 0.01) is set, and when the energy value to be propagated is lower than this threshold, the propagation along the edge is stopped to improve the calculation efficiency and simulate the natural decay of influence. The propagation process is iteratively executed until the energy change on all paths is less than a small value or the propagation has terminated. At this time, the accumulated energy value on each post node represents the total influence of the resume skills on the post through direct and indirect association. The vector composed of the energy values of all post nodes is the energy distribution feature, which dynamically quantifies the association strength between the resume and each post.

[0046] Further, the energy value of each post node in the energy distribution feature directly reflects the association strength between the post and the resume skill, providing a data basis for calculating the direct energy association degree; at the same time, the distribution of the energy distribution feature also provides support for calculating the energy structure coefficient. Only the accurate energy distribution feature is generated, can the objective and accurate correlation score be generated, and then the sorting and matching are completed.

[0047] S5, generating a correlation score for the resume data and the post data according to the energy distribution feature, and sorting and matching based on the correlation score.

[0048] In the embodiment of the application, the correlation score is a comprehensive quantitative value for measuring the matching degree of the resume and the post.

[0049] In the embodiment of the application, the correlation score for the resume data and the post data according to the energy distribution feature comprises: extracting the energy value of each post node in the energy distribution feature, and taking the energy value as the direct energy association degree between the post and the resume; taking the ratio of the average energy value of the feature matching node set to the average energy value of the semantic association node set as the energy structure coefficient for measuring the matching concentration degree; performing a product operation on the direct energy association degree and the energy structure coefficient to obtain a preliminary correlation score of each post; correcting the preliminary correlation score based on the timeliness factor of the post publishing time length and the urgency degree to obtain the correlation score.

[0050] In detail, the cumulative energy value on each post node calculated is obtained The value range of the direct energy association degree is consistent with the energy value in the energy distribution feature, and the higher the value is, the greater the direct association strength between the post and the resume skill is. By traversing the energy distribution feature vector, the energy value of each post node is directly mapped to the corresponding direct energy association degree, ensuring the accuracy of data extraction. The energy structure coefficient is used to distinguish the concentration of the influence of the resume skill, and the energy average value of all feature matching nodes (direct matching posts) and the energy average value of all semantic association nodes (indirectly associated posts) are calculated respectively. If the C value is large, it means that the influence of the resume is mainly concentrated on the directly matched post, and the matching direction is clear; if the C value is close to 1 or less than 1, it means that the influence of the resume is relatively dispersed, or the indirectly associated post also has potential. In the calculation process, if the average energy value of the semantic association node set is 0, the energy structure coefficient is set to a preset maximum value 5.0 to avoid the case that the divisor is 0.

[0051] Specifically, for each post its preliminary relevance score i.e. the post with high direct energy relevance will get a high score, meanwhile, if the resume is overall well-matched and the direction is clear (C value is large), the score of the direct matching post will be further strengthened. Define the timeliness factor for example, where is the number of days since the post was published, is the decay constant. The newer the post, the closer to 1; the older the post, the smaller. At the same time, a bonus factor can be given in combination with the urgency degree (such as urgent recruitment) marked by the post. The final relevance score ensures that the matching result considers not only the depth of semantic and network relevance, but also the timeliness requirement of recruitment.

[0052] In the embodiment of the application, sorting and matching are the process of prioritizing posts according to the relevance score and outputting the matching result.

[0053] In the embodiment of the application, the sorting and matching based on the relevance score include: arranging all posts in descending order according to the relevance score to generate a preliminary recommendation sequence; in the preliminary recommendation sequence, for the post group with adjacent relevance scores and a difference less than a preset threshold, performing secondary sorting according to the node centrality of the post node in the local correlation graph to obtain an optimized recommendation sequence; determining a preset number of posts in the optimized recommendation sequence as the post matching set of the resume data to be matched.

[0054] In detail, all the positions with the relevance scores are ranked from high to low to obtain a preliminary recommendation sequence. In the ranking process, a quick sorting algorithm is used to arrange the relevance scores of the positions in descending order to ensure the efficiency and accuracy of the ranking. If the relevance scores of two positions are exactly the same, the positions are arranged in ascending order according to the sequence numbers of the position nodes. A small threshold (such as 0.01) is set. In the preliminary recommendation sequence, if the scores of several positions in succession are less than the threshold, it is considered that the matching degrees of the positions and the resume are difficult to distinguish in the quantitative scores. At this time, a network structure index is introduced for secondary sorting. The node centrality (such as degree centrality or eigenvector centrality) of the position nodes in the local correlation graph is calculated. The node centrality is a weighted average of the degree centrality, the closeness centrality and the betweenness centrality, and the weight is set to 1 / 3. The degree centrality is the ratio of the number of associated edges of the node to the total number of nodes in the local correlation graph. The closeness centrality is the inverse of the average of the shortest path lengths from the node to all other nodes. The betweenness centrality is the proportion of the number of times that the node is an intermediate node in the shortest paths between other nodes to the total number of shortest paths. The positions with higher centrality are in more core or more critical positions in the local network, and may represent more mainstream or more important career development paths.

[0055] Specifically, the positions in the same segment are arranged in descending order according to the centrality, so as to optimize the logic and interpretability of the recommendation sequence. The top K positions with the highest ranking are output as the final matching result, and the whole intelligent matching process is completed.

[0056] As shown in Figure 2 FIG. 1 is a functional module diagram of a semantic analysis driven resume and position intelligent matching system provided by an embodiment of the present application.

[0057] The semantic analysis driven resume and position intelligent matching system 100 can be installed in an electronic device. According to the functions to be implemented, the semantic analysis driven resume and position intelligent matching system 100 can include a fine-grained semantic analysis module 101, a preliminary matching graph construction module 102, a node retrieval module 103, an energy distribution feature analysis module 104 and a data matching module 105. The modules of the present application can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete a fixed function, and are stored in the memory of the electronic device.

[0058] In the present embodiment, the functions of each module / unit are as follows: The fine-grained semantic analysis module 101 is configured to perform fine-grained semantic analysis on the pre-acquired resume data and position data to be matched respectively to obtain a first semantic feature tuple set and a second semantic feature tuple set. The preliminary matching graph construction module 102 is configured to construct a double-layer semantic index graph for semantic matching based on the second semantic feature tuple set, analyze the compatibility of the feature entity in the first semantic feature tuple set with the topology structure of the double-layer semantic index graph, and determine a preliminary matching graph related to the to-be-matched resume data according to the compatibility. The node retrieval module 103 is configured to retrieve, in the double-layer semantic index graph, a feature matching node corresponding to the feature entity and a semantic association node according to the preliminary matching graph. The energy distribution feature analysis module 104 is configured to construct a local association graph representing a post according to the feature matching node and the semantic association node, perform directional propagation and energy attenuation on the feature entity by using the local association graph, and obtain an energy distribution feature for the feature entity propagation. The data matching module 105 is configured to generate a relevance score for the resume data and the post data according to the energy distribution feature, and perform sorting and matching based on the relevance score.

[0059] In detail, the modules in the semantic analysis driven resume and post intelligent matching system 100 in the embodiments of the present application use the same technical means as the semantic analysis driven resume and post intelligent matching method in the above Figure 1 , and can produce the same technical effects, which will not be described here.

[0060] In the several embodiments of the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there can be another division manner.

[0061] The modules described as separate components can or can not be physically separate, and the components displayed as modules can or can not be physical units, i.e., they can be located in one place, or distributed on multiple network units. According to actual needs, some or all of the modules can be selected to achieve the purpose of the present embodiment.

[0062] In addition, each functional module in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware, or in the form of hardware plus software function modules.

[0063] It will be obvious to a person skilled in the art that the application is not limited to the details of the above-described exemplary embodiments but can be implemented in other embodiments without departing from the scope of the application.

[0064] The embodiments should therefore be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the description given above, which is intended as explanatory only and therefore the application is intended to encompass all technical equivalents which fall within the scope of the claims.

[0065] Embodiments of the present application can acquire and process related data based on artificial intelligence technology. Wherein, artificial intelligence (AI) is to use digital computers or digital computer controlled machine to simulate, extend and expand human intelligence, perceive environment, obtain knowledge and use knowledge to obtain the best results.

[0066] Furthermore, it is clear that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. A single processor or other unit can fulfill the functions of several units or systems recited in the claims. The terms "first", "second" and the like do not denote any ordinal, but do not exclude other units or steps. The suffix "s" is used as an English plural but does not show the consistency or the unconsistency.

[0067] Finally, it should be noted that the above-mentioned embodiments are merely intended to illustrate the technical solutions of the present application, rather than limit the present application. Even though the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application.

Claims

1. A semantic analysis-driven intelligent resume and job matching method, characterized in that, The method includes: Fine-grained semantic parsing is performed on the pre-acquired resume data and job data to be matched to obtain the first semantic feature tuple set and the second semantic feature tuple set; A two-layer semantic index graph for semantic matching is constructed based on the second semantic feature tuple set. The compatibility of feature entities in the first semantic feature tuple set with the topological structure of the two-layer semantic index graph is analyzed. Based on the compatibility, a preliminary matching graph related to the resume data to be matched is determined. In the two-layer semantic index graph, feature matching nodes and semantic association nodes corresponding to the feature entities are retrieved based on the preliminary matching graph; A local association graph representing the job position is constructed based on the feature matching nodes and the semantic association nodes. The local association graph is then used to perform directional propagation and energy attenuation on the feature entities to obtain the energy distribution characteristics of the propagation of the feature entities. Based on the energy distribution characteristics, a relevance score is generated for the resume data and job data, and the data is sorted and matched based on the relevance score.

2. The semantic analysis-driven intelligent resume and job matching method as described in claim 1, characterized in that, The process involves performing fine-grained semantic parsing on the pre-acquired resume data and job posting data to be matched, respectively, to obtain a first semantic feature tuple set and a second semantic feature tuple set, including: Based on a pre-defined domain ontology dictionary, the resume data and job data to be matched are segmented and labeled with part of speech at multiple granularities to identify named entities and core verb phrases in the resume data and job data. Extract the skill-specific triple structure based on the grammatical relationship between the named entity and the core verb phrase; Based on the hierarchical and synonym relationships in the domain ontology dictionary, semantic normalization is performed on the objects in the triplet structure to map different words with the same semantics to the same normative concept. According to the preset data source, the semantically normalized triple structure is aggregated into the first semantic feature tuple set and the second semantic feature tuple set.

3. The semantic analysis-driven intelligent resume and job matching method as described in claim 1, characterized in that, The construction of a two-layer semantic index graph for semantic matching based on the second semantic feature tuple set includes: Each job position is treated as a job position node. A first connection edge is established between the job position nodes according to the hierarchical relationship between the job positions. A job position main network representing the job position aggregation relationship is constructed based on the job position nodes and the first connection edge. Take the canonical concepts in each tuple of the second semantic feature tuple set as concept nodes, establish a second connection edge between concept nodes that have logical relationships, and construct a concept sub-network representing skill association relationships based on the concept nodes and the second connection edge. Identify the first index number of each job node in the main job network and the second index number of each concept node in the sub-concept network; Based on the first index number and the second index number, each job node in the main job network is linked across layers to all concept nodes corresponding to the job in the concept sub-network, resulting in a two-layer semantic index graph.

4. The semantic analysis-driven intelligent resume and job matching method as described in claim 1, characterized in that, The analysis of the compatibility of feature entities in the first semantic feature tuple set with the topological structure of the two-layer semantic index graph includes: Each feature entity in the first semantic feature tuple set is mapped to the corresponding node in the concept sub-network of the two-layer semantic index graph to obtain the mapping node; A mapping subgraph is constructed based on the mapping nodes, and the ratio of the structural cohesion of the mapping subgraph to the global average cohesion of the conceptual subnetwork is used as the first topological index. The overlap between each node in the mapping subgraph and the directly associated job nodes in the two-layer semantic index graph is used as the second topological index. The first topological index and the second topological index are weighted and fused to obtain the compatibility of the topological structure of the two-layer semantic index graph.

5. The semantic analysis-driven intelligent resume and job matching method as described in claim 1, characterized in that, The step of retrieving feature matching nodes and semantic association nodes corresponding to the feature entities in the two-layer semantic index graph based on the preliminary matching graph includes: In the preliminary matching graph, job nodes that have direct cross-layer links with the skill nodes mapped to the feature entities are selected. The selected job nodes are marked as a feature matching node set, and the nodes in the feature matching node set are determined as feature matching nodes. Starting from the set of feature matching nodes, within the main network of job positions in the two-layer semantic index graph, according to the connection edges corresponding to the job positions in the set of feature matching nodes, within a preset number of hops, target job positions that are strongly associated with the job positions are queried, and the target job positions are marked as a set of potential associated nodes. For each node in the potential associated node set, backtrack to query the skill node in the two-layer semantic index graph, and calculate the semantic similarity between the skill node and the feature entity; Nodes with semantic similarity higher than a preset association threshold are included as semantically associated nodes in the set of semantically associated nodes.

6. The semantic analysis-driven intelligent resume and job matching method as described in claim 3, characterized in that, The step of constructing a local association graph representing job positions based on the feature matching nodes and the semantic association nodes includes: Skill nodes that have a direct association with the feature matching nodes and the semantic association nodes are extracted from the conceptual sub-network; Based on the feature matching nodes, the semantic association nodes, and the skill nodes, a subgraph search is performed on the two-layer semantic index graph to obtain the target closed subgraph. Identify the original connection strength between the feature matching node and the semantic association node in the main job network, and identify the link weight of the skill node; Calculate the comprehensive association weight between all nodes within the target closed subgraph based on the original connection strength and the link weight; Based on the comprehensive association weights and the target closed subgraph, a local association graph representing the job positions is constructed.

7. The semantic analysis-driven intelligent resume and job matching method as described in claim 6, characterized in that, The step of using the local correlation map to perform directional propagation and energy attenuation on the feature entity to obtain the energy distribution characteristics of the propagation of the feature entity includes: Each of the feature entities is initialized as an energy source, and the initial energy value corresponding to the energy source is evenly distributed to the skill nodes in the local association graph that correspond to the feature entities. Energy is propagated from the skill node to the job node, and between job nodes, according to the comprehensive association weight. During the propagation of energy along the connecting edge, the initial energy value is attenuated for each edge according to the exponential function of the comprehensive association weight. When the attenuated energy value is lower than the preset propagation threshold, the energy propagation in the propagation direction corresponding to the current connecting edge is terminated. During the energy propagation process, when the energy propagation of all connected edges is stable or terminated, the accumulated energy value of each job node is recorded, and the energy values ​​of all job nodes are aggregated into energy distribution characteristics.

8. The semantic analysis-driven intelligent resume and job matching method as described in claim 5, characterized in that, The step of generating a relevance score for resume data and job data based on the energy distribution characteristics includes: Extract the energy value of each job node from the energy distribution features, and use the energy value as the direct energy correlation between the job and the resume; The ratio of the average energy value of the feature matching node set to the average energy value of the semantically related node set is used as the energy structure coefficient for measuring the degree of matching concentration. The direct energy correlation degree is multiplied by the energy structure coefficient to obtain a preliminary correlation score for each position; The preliminary relevance score is corrected based on the timeliness factors of job posting duration and urgency to obtain a relevance score.

9. The semantic analysis-driven intelligent resume and job matching method as described in claim 1, characterized in that, The sorting and matching based on the relevance score includes: All job positions are sorted in descending order according to the aforementioned relevance scores to generate a preliminary recommendation sequence; In the initial recommendation sequence, job groups with adjacent relevance scores and a difference less than a preset discrimination threshold are sorted a second time according to the node centrality of the job nodes in the local association graph to obtain the optimized recommendation sequence. A preset number of positions in the optimized recommendation sequence are determined as the position matching set with the resume data to be matched.

10. A semantic analysis-driven intelligent resume and job matching system, characterized in that, The system is used to perform the semantic analysis-driven intelligent resume and job matching method as described in any one of claims 1-9, the system comprising: The fine-grained semantic parsing module is used to perform fine-grained semantic parsing on the pre-acquired resume data and job data to be matched, respectively, to obtain the first semantic feature tuple set and the second semantic feature tuple set; The preliminary matching graph construction module is used to construct a two-layer semantic index graph for semantic matching based on the second semantic feature tuple set, analyze the compatibility of feature entities in the first semantic feature tuple set with the topological structure of the two-layer semantic index graph, and determine the preliminary matching graph related to the resume data to be matched based on the compatibility. The node retrieval module is used to retrieve the feature matching nodes and semantic association nodes corresponding to the feature entities in the two-layer semantic index graph based on the preliminary matching graph. The energy distribution feature analysis module is used to construct a local association graph representing the job position based on the feature matching node and the semantic association node, and to use the local association graph to perform directional propagation and energy attenuation on the feature entity to obtain the energy distribution feature for the propagation of the feature entity; The data matching module is used to generate a relevance score for resume data and job data based on the energy distribution characteristics, and to sort and match the data based on the relevance score.

Images