CN-122022751-A - Resume recommendation method of label model based on business organization tree

CN122022751ACN 122022751 ACN122022751 ACN 122022751ACN-122022751-A

Abstract

The invention provides a resume recommendation method of a label model based on a service organization tree, which comprises the steps of constructing the service organization tree label model, generating a full-dimensional service vector by logically fusing post inherent characteristics and father node service attributes through label inheritance, carrying out structural analysis on resume data, carrying out time sequence weighted calculation on skill entities according to current time to generate resume feature vectors, calculating service capacity semantic matching results and service scene matching results between the full-dimensional service vectors and the resume feature vectors, and carrying out weighted fusion on the matching results according to weight parameters to generate a recommendation list. The invention also relates to a dynamic weight optimization mechanism based on reinforcement learning. The invention can effectively complement post business context, accurately evaluate timeliness of candidate skills, realize self-adaptive evolution of recommendation strategies and improve accuracy of post matching.

Inventors

DU GUANGWEI
ZOU HONGTAO
LI MENG
ZHANG JIAMING

Assignees

中煤科工集团信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (10)

1. A resume recommendation method of a label model based on a business organization tree is characterized by comprising the following steps: Constructing a service organization tree label model, and generating a full-dimensional service vector of a job to be recruited by fusing the inherent characteristics of the job node and the service attribute of the node on the father path of the job node; Analyzing the collected resume data, and carrying out time sequence weighted calculation on the identified skill entity according to the current system time to generate resume feature vectors containing skill strength value information; Performing adaptation calculation to obtain a service capability semantic matching result between the full-dimensional service vector and the resume feature vector and a service scene matching result between a service scene entity in the resume and a scene label set of a service node to which a post belongs; and carrying out weighted fusion on the business capability semantic matching result and the business scene fitness matching result according to the weight parameter configuration to obtain a comprehensive matching score and generate a resume recommendation list.
2. The resume recommendation method of a business organization tree-based tag model according to claim 1, wherein the step of constructing the business organization tree tag model comprises: Defining a hierarchy of a tree structure, wherein the hierarchy at least comprises a root node, a service unit node, a project node and a post node; extracting dominant keywords in a post instruction book as inherent feature vectors aiming at the post nodes; Performing vector synthesis operation, traversing upwards along the tree structure, and obtaining attribute tag vectors of all father nodes on the father path to which the post node belongs; Synthesizing the inherent feature vector and the attribute label vector of each father node by using a weighted summation algorithm to generate the full-dimensional service vector; Wherein, the closer to the post node level parent node is given a greater level weight coefficient.
3. The resume recommendation method of a tag model based on a service organization tree according to claim 1, wherein the step of structurally parsing the collected resume data comprises: denoising and normalizing the original resume data to obtain a standard text sequence; Inputting the standard text sequence into a pre-trained deep learning model, wherein the deep learning model adopts a BERT-BiLSTM-CRF network architecture; And outputting the serialized entity tags through the deep learning model, wherein the types of the entity tags at least comprise skill entities, post entities, business scene entities and time entities.
4. A resume recommendation method based on a business organizational tree tag model according to claim 3, wherein said step of performing a time-series weighted calculation on the identified skill entities based on the current system time comprises: Determining, for each skill entity identified, a project end time for the last occurrence of the skill entity according to the associated time entity; Calculating a time difference value between the project end time and the current system time; Determining a basic score of the skill entity, processing the basic score and the time difference value by using an exponential decay function, and calculating to obtain a skill strength value of the skill entity; wherein the exponential decay function has a time decay coefficient that controls the decay rate.
5. The resume recommendation method of a business organizational tree-based tag model according to claim 1, wherein said performing adaptation calculation further comprises a hard condition filtering step: before calculating the matching result of the full-dimensional service vector and the resume feature vector, extracting a structural field and a rigid constraint condition of a post in resume data; Constructing a hard index indication function which comprises a series of Boolean logic judgments; And judging whether the structured field meets the hard constraint condition by using the hard index indication function, if not, terminating subsequent calculation and setting the comprehensive matching score to be zero.
6. The resume recommendation method of a business organization tree-based tag model of claim 1, wherein the step of calculating a business capability semantic matching result between the full-dimensional business vector and the resume feature vector comprises: Mapping the full-dimensional service vector and the resume feature vector to the same vector space; calculating the cosine value of an included angle of two vectors in the vector space by adopting a cosine similarity algorithm; And taking the calculated cosine value of the included angle as the semantic matching result of the business capability to represent the semantic proximity degree of the skill distribution of the candidate and the post business technical requirement.
7. The resume recommendation method based on the tag model of the service organization tree according to claim 1, wherein the step of calculating the service scene matching result between the service scene entity in the resume and the scene tag set of the service node to which the post belongs comprises the following steps: Extracting service scene labels associated with service unit nodes and project nodes of the full-dimensional service vector source to form a post scene set; extracting business scene entities contained in the resume feature vectors to form a resume scene set; and calculating the ratio of the number of intersection elements to the number of union elements of the post scene set and the resume scene set by adopting a Jaccard similarity coefficient algorithm, and taking the ratio as a matching result of the business scene fitness.
8. The resume recommendation method of a tag model based on a service organization tree according to claim 1, wherein the step of performing weighted fusion on the service capability semantic matching result and the service scene fitness matching result according to preset weight parameter configuration comprises: Acquiring current weight parameter configuration, which comprises a first weight parameter corresponding to a business capability semantic matching result and a second weight parameter corresponding to a business scene fitness matching result; And multiplying the business capability semantic matching result by a first weight parameter, multiplying the business scene fitness matching result by a second weight parameter, and adding the two to obtain the comprehensive matching score.
9. The resume recommendation method of a business organizational tree-based tag model of claim 8, further comprising a reinforcement learning-based dynamic weight optimization step: Defining a state space, wherein the state space at least comprises current post type characteristics and current weight parameter configuration; Defining a reward function, wherein the reward function maps the operation behaviors of a user on a resume recommendation list into a numeric reward value, and the operation behaviors at least comprise browsing, downloading, interviewing and elimination; and constructing a state action cost function for evaluating long-term expected benefits brought by adjusting the weight parameter configuration in a preset state.
10. The resume recommendation method of a business organizational tree-based tag model of claim 9, wherein the reinforcement learning-based dynamic weight optimization step further comprises: Capturing operation behaviors of a user on a recommendation result in real time, and calculating a cumulative rewarding value according to the rewarding function; Updating the state action cost function according to the jackpot value using a Q-learning algorithm or a deep Q network strategy; And determining the optimal weight adjustment action at the next moment according to the updated state action cost function, generating new weight parameter configuration, and applying the new weight parameter configuration to the subsequent weighting fusion step.

Description

Resume recommendation method of label model based on business organization tree Technical Field The disclosure relates to the technical field of internet data processing and artificial intelligence, in particular to a resume recommendation method of a label model based on a business organization tree. Background With the rapid development of internet technology and the deep digital transformation of enterprises, massive resume data are accumulated in an online recruitment platform and an enterprise internal talent library. How to accurately and efficiently discover candidates meeting specific post requirements from huge data resources has become a core problem to be solved in the field of intelligent recruitment. The current mainstream solution generally utilizes natural language processing technology to realize person post matching by extracting keywords or semantic feature vectors in post specifications and resume texts and calculating the similarity between the post specifications and the resume texts. However, the existing resume recommendation technology still has obvious limitations in practical application. Most of the traditional matching algorithms consider the post specifications as independent text basis, and the business context association of posts in the complex organization architecture of the enterprise is ignored. Because the post instruction often has the conditions of short description, serious templatization or late update, the special technical stack preference of the business unit to which the post belongs or the special stage requirement of the current project cannot be completely reflected by relying on the literal content, and the recommendation result is matched with the actual requirement of a team on the deep business gene although the recommendation result is matched with the keyword layer. Meanwhile, the conventional resume analysis technology is usually focused on static entity identification, and timing dimension consideration on skill mastery level is lacked. When extracting skill labels of candidates, the prior art often cannot effectively distinguish whether the candidates use the skills recently and frequently or only in historical projects many years ago, ignoring forgetting curves and time decay characteristics of technical capabilities. The static feature extraction mode easily causes that the system recommends candidates with sparse skills or outdated technical stacks to enterprises, and the effectiveness of person post matching is reduced. In addition, the existing sentry matching model mostly adopts preset fixed weights or static rules to carry out multi-dimensional score fusion, and is difficult to adapt to dynamic change recruitment strategies. In an actual recruitment scene, different business stages or different interviewees have differences on the capability dimension emphasis of candidates, and a model with fixed weight cannot effectively utilize recruiters to carry out self-correction on real-time feedback behaviors such as browsing, interviewing or elimination of recommendation results, so that recommendation strategies are stiff, and adaptive optimization and continuous evolution cannot be realized along with the use process. Disclosure of Invention The method aims at solving at least one of the technical problems that the existing resume recommendation technology lacks depth consideration on post business context and skill timeliness to a certain extent, and the recommendation strategy cannot be adaptively adjusted according to user feedback, so that the post matching accuracy is low and business compliance is poor. Therefore, the disclosure provides a resume recommendation method of a label model based on a business organization tree. The invention provides a resume recommendation method of a label model based on a business organization tree, which comprises the following steps: constructing a service organization tree label model, mapping an enterprise organization architecture into a multi-level tree structure, configuring label inheritance logic, and generating a full-dimensional service vector of a job to be recruited by fusing inherent characteristics of the job node and service attributes of nodes on a parent path of the job node; carrying out structural analysis on the collected resume data, identifying entities and relations in the resume text, carrying out time sequence weighted calculation on the identified skill entities according to the current system time, and generating resume feature vectors containing skill strength value information; Executing multidimensional adaptation calculation, at least calculating a service capability semantic matching result between a full-dimensional service vector and a resume feature vector, and calculating a service scene matching result between a service scene entity in the resume and a scene label set of a service node to which a post belongs; And carrying out weighted fusion on the business capability seman