CN-122022752-A - Intelligent scheduling and dynamic configuration optimizing system and method for enterprise human resources

CN122022752ACN 122022752 ACN122022752 ACN 122022752ACN-122022752-A

Abstract

The application discloses an intelligent dispatching and dynamic configuration optimizing system and method for human resources of enterprises, which relate to the technical field of talent resource information processing, and the scheme comprises that a data processing module is used for respectively analyzing posts issued by each enterprise and resume uploaded by each job seeker to obtain corresponding post portrait vector and talent portrait vector; the talent-post matching module is used for mapping the post portrait vector and the talent portrait vector to the same semantic vector space by utilizing the pre-training language model, acquiring a candidate resume set of a to-be-recruited post based on the similarity of the post portrait vector and the talent portrait vector, and the multi-task recommended resource scheduling module is used for executing an intelligent resource scheduling mechanism based on the reinforcement learning model so as to optimally allocate resources for the candidate resume set of the to-be-recruited post and updating parameters of the pre-training language model based on the result output by the reinforcement learning model. The technology solves the problem of unreasonable scheduling of human resources of enterprises.

Inventors

Lin Lvwei
LIN WANLING
LIN WANLING

Assignees

睿德(广东)人力资源服务有限公司

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (10)

1. An enterprise human resource intelligent scheduling and dynamic configuration optimizing system, which is characterized by comprising: The data processing module is used for monitoring post release requests of all enterprises and resume uploading requests of job seekers through preset interfaces, and respectively analyzing posts released by all enterprises and resume uploaded by all job seekers to obtain corresponding post portrait vectors and talent portrait vectors; The talent-post matching module is used for mapping the post portrait vector and the talent portrait vector to the same semantic vector space by utilizing the pre-training language model, and acquiring a candidate resume set of a to-be-recruited post based on the similarity of the post portrait vector and the talent portrait vector; The multi-task recommended resource scheduling module is used for triggering an intelligent resource scheduling mechanism based on the reinforcement learning model to optimally allocate resources for the candidate resume set of the recruitment post if multi-task concurrence and recommended resource constraint exists, updating the parameters of the pre-training language model based on the result output by the reinforcement learning model, and otherwise outputting talent-post recommendation results according to the candidate resume set of the recruitment post.
2. The intelligent scheduling and dynamic configuration optimizing system for human resources of enterprises according to claim 1, wherein the intelligent resource scheduling mechanism based on the reinforcement learning model specifically comprises: Predicting a recommended success rate and an expected conversion period of the to-be-recruited post based on the current state characteristics by utilizing the reinforcement learning model; the state characteristics comprise post emergency degree, talent scarcity degree and resume processing efficiency; Defining the product of the recommended success rate of the post to be recruited and the importance weight of the post as the comprehensive recommended score of the post to be recruited, and synchronously acquiring a corresponding emergency grade, wherein the emergency grade is used for quantifying the urgency degree of the post recruitment requirement; Distributing all the posts to be recruited to corresponding priority scheduling pools according to the emergency level, and carrying out descending arrangement in each priority scheduling pool according to the comprehensive recommendation score to generate a scheduling queue; And traversing the scheduling queue in sequence from high priority to low priority, dynamically distributing resources to the positions to be recruited, calculating the residual allocable limit of the positions in the current recruitment period for the positions traversed currently, and dynamically distributing the residual allocable limit according to the allocation limit of the current round.
3. The intelligent scheduling and dynamic configuration optimizing system for human resources of enterprises according to claim 2, wherein the specific obtaining method for the emergency level is as follows: if the post emergency degree of the post to be recruited belongs to the first post emergency interval or the talent scarcity degree belongs to the first talent scarcity interval, marking the emergency grade as a first emergency grade; in addition, if the post emergency belongs to the second post emergency interval or the talent scarcity belongs to the second talent scarcity interval, the emergency grade is marked as a second emergency grade, Otherwise, recording as a third emergency level; The lower numerical limit of the first post emergency interval is higher than the lower numerical limit of the second post emergency interval; the lower numerical limit of the first talent scarcity region is higher than the lower numerical limit of the second talent scarcity region; And the emergency degrees corresponding to the first emergency level, the second emergency level and the third emergency level are gradually reduced.
4. The intelligent scheduling and dynamic configuration optimization system for human resources of an enterprise according to claim 2, wherein the intelligent resource scheduling mechanism based on the reinforcement learning model further comprises: generating a global optimal scheduling scheme based on a preset fitness function in a genetic algorithm, wherein the global optimal scheduling scheme comprises talent-post recommendation results and priority ordering; Monitoring state characteristics in real time, if an abnormal event is detected, locking a task subset influenced by the abnormal event, performing plug-in heuristic search on the task subset, searching the latest available time window in an original scheduling queue, and outputting a global optimal scheduling scheme after reassignment; If no abnormal event is detected, outputting talent-post recommendation results according to the candidate resume set of the post to be recruited; The abnormal event represents an event which is caused by unexpected state change and causes the failure of the global optimal scheduling scheme in the execution process of the global optimal scheduling scheme.
5. The intelligent scheduling and dynamic configuration optimizing system for human resources of enterprises according to claim 1, wherein the pre-training language model parameters are updated based on the result output by the reinforcement learning model, and the specific flow is as follows: Outputting a predicted value based on the current state characteristics by using the reinforcement learning model, wherein the predicted value comprises an expected recommended success rate, an expected conversion period and state value data; counting the actual recording quantity, the quantity of currently allocated resume and the time length from resume recommendation to successful recording; Recording the ratio of the actual recording quantity to the currently allocated resume quantity as a conversion rate index; calculating recruitment efficiency indexes according to the time from resume recommendation to successful recording; performing weighted coupling processing on the conversion rate index and the recruitment efficiency index to obtain an efficiency evaluation value; acquiring state value data of the next state, obtaining a time sequence difference error according to the efficiency evaluation value, the state value data of the current state and the state value data of the next state, and inputting the time sequence difference error as a gradient signal into the reinforcement learning model to update model parameters.
6. The intelligent scheduling and dynamic configuration optimization system for human resources of an enterprise of claim 5, wherein updating the pre-trained language model parameters based on the results output by the reinforcement learning model further comprises: Aiming at the resume which is not checked in the preset time period, obtaining the history checking rate corresponding to the resume based on the statistical data of posts delivered by the resume; If the historical viewing rate is not less than the lower limit of the viewing rate, setting the sample weight of the corresponding resume as the reciprocal of the historical viewing rate, and judging the resume processing efficiency of the monitored recruitment post; if the historical viewing rate is smaller than the lower limit of the viewing rate, setting the sample weight of the corresponding resume as the maximum weight; And introducing the sample weight, carrying out weighting processing on a loss function used for mapping the post portrait vector and the talent portrait vector into the same semantic vector space in the pre-training language model, and training the pre-training language model by using the weighted loss function.
7. The intelligent scheduling and dynamic configuration optimizing system for human resources of enterprises according to claim 6, wherein the judging of the resume processing efficiency of the monitored recruitment post comprises the following specific procedures: The resume processing efficiency comprises resume average processing time length and resume backlog rate; If the detected resume average processing time length is larger than the resume average processing time length limit value or the resume backlog rate is larger than the resume backlog rate limit value, judging that the post is in an inefficient processing state, and triggering a sequencing attenuation function; if the conditions are not met, judging that the post is not in an inefficient processing state, and not processing; The triggering ordering attenuation function specifically comprises the following steps: Taking the product of the integrated recommendation score of the post in the dispatching queue and a preset attenuation factor as the updated integrated recommendation score of the post, and re-determining the sequencing priority of the post in the dispatching queue according to the updated integrated recommendation value score; In a preset sorting recovery period, if the average processing duration of the resume corresponding to the post in the low-efficiency processing state is not greater than the average processing duration limit value of the resume and the resume backlog rate is not greater than the resume backlog rate limit value, gradually increasing the comprehensive recommendation score of the post until the post is recovered to the original comprehensive recommendation score of the post in the scheduling queue, otherwise, not increasing the comprehensive recommendation score of the post.
8. The intelligent scheduling and dynamic configuration optimizing system for human resources of enterprises according to claim 1, wherein the analyzing of posts issued by each enterprise and resume uploaded by each job seeker comprises the following specific procedures: Performing named entity recognition and dependency syntactic analysis on the issued post description text by utilizing a pre-training language model to extract post portrait entities, and mapping the extracted post portrait entities into structured post portrait vectors; Analyzing and optimizing the resume, and acquiring coordinate information of each text block in the resume image by utilizing OCR to recognize the resume image corresponding to the resume; Introducing a layout analysis model to typeset the resume image, and identifying the text region type in the resume; and extracting talent portrait entities from texts corresponding to the logical reading sequence restored based on the coordinate information of each text block, and mapping the extracted talent portrait entities into structured talent portrait vectors.
9. The intelligent scheduling and dynamic configuration optimizing system for human resources of enterprises according to claim 1, wherein the talent-post recommendation result is output according to the candidate resume set of the post to be recruited, and the specific flow is as follows: Obtaining a matching degree score between each candidate resume and a post to be recruited, wherein the matching degree score is used for reflecting the similarity degree between talent portrait vectors of the candidate resume and the post portrait vectors of the post to be recruited; According to the matching degree score, all candidate resume in the candidate resume set are sorted in a descending order; And selecting the candidate resume with the target number which is ranked at the front from the ranked candidate resume set as a final recommendation result to be output according to the maximum receiving resume number of the recruitment post, wherein the target number does not exceed the maximum receiving resume number.
10. An enterprise human resource intelligent scheduling and dynamic configuration optimizing method is characterized by comprising the following steps: s1, monitoring post release requests of enterprises and resume uploading requests of job seekers through preset interfaces, and respectively analyzing posts released by the enterprises and resume uploaded by the job seekers to obtain corresponding post portrait vectors and talent portrait vectors; S2, mapping the post portrait vector and the talent portrait vector to the same semantic vector space by utilizing a pre-training language model, and acquiring a candidate resume set of a job to be recruited based on the similarity of the post portrait vector and the talent portrait vector; and S3, triggering an intelligent resource scheduling mechanism based on the reinforcement learning model to optimally allocate resources for the candidate resume set of the recruitment post if the multi-task concurrence and the recommended resource constraint exist, updating the parameters of the pre-training language model based on the result output by the reinforcement learning model, and otherwise, outputting talent-post recommendation results according to the candidate resume set of the recruitment post.

Description

Intelligent scheduling and dynamic configuration optimizing system and method for enterprise human resources Technical Field The invention relates to the technical field of talent resource information processing, in particular to an intelligent scheduling and dynamic configuration optimizing system and method for human resources of enterprises. Background With the expansion of enterprise scale and the acceleration of the digitization process of the human resource market, recruitment businesses face the double challenges of massive job requirements and massive resume supply. In order to improve the efficiency of post matching, existing recruitment management systems typically integrate natural language processing (Natural Language Processing, NLP), optical character recognition (Optical Character Recognition, OCR) and recommendation algorithms, aimed at assisting in screening talents by automated means. In the prior art, the conventional implementation process mainly comprises the following steps of firstly, carrying out text extraction on a resume document by utilizing an OCR technology in a data analysis layer, matching with a keyword matching algorithm based on rules or statistical machine learning, comparing skills, students and other entities in the resume with job descriptions, secondly, in a post matching layer, generating a recommendation list by a system through calculating similarity scores of candidates and job positions and pushing the recommendation list to recruiters, thirdly, carrying out resource allocation by the system by adopting static rules when facing a plurality of concurrent posts in a resource scheduling layer, for example, presetting a fixed recommendation number according to a post release time sequence or by manpower, and finally, training the model offline in a model iteration layer by collecting historical click, communication and recording data of the recruiters. The intelligent talent resource information matching system and method based on big data, for example, disclosed by Chinese patent publication No. CN116433201B, comprises the steps of firstly predicting recruitment requirements of recruitment enterprises at each delay time point, secondly screening talent information in talent libraries based on acquired related professional names and recruitment professional names, thirdly screening talent information secondarily, thirdly intelligently matching talents according to constructed talent maps and sending the matched talent information to enterprise recruitment ends. An intelligent recognition method based on talent number intelligent brain system is disclosed in Chinese patent application with publication number of CN120494775A, and comprises the steps of firstly constructing a talent database based on the talent number intelligent brain system, then collecting enterprise recruitment requirement text, historical recruitment data and industry trend information, extracting key skills, experience requirements, cultural compliance and other core indexes, combining enterprise historical data with industry dynamics, recognizing potential requirements which are not explicitly expressed by an enterprise through a correlation rule mining algorithm, integrating the dominant requirements with the potential requirements to form a structured enterprise recruitment requirement set, intelligently screening candidates based on the talent number intelligent brain system and combining the enterprise recruitment requirement set, and generating an individualized recommendation report by the system after screening is completed. The above technology is found to have at least the following technical problems: In the prior art, a static allocation strategy which is obtained first from first to first or depends on manual presetting is generally adopted. The lack of dynamic awareness and global optimization capabilities of post deadlines and talent scarcity results in a system that is unable to adaptively adjust resource allocation. Therefore, the emergency post is difficult to acquire enough recommended resources, and the recommended resources are unreasonable to be allocated. In addition, the existing matching model has a remarkable data deviation problem in the training process. The hot post or top-ranked resume produces a large amount of click data due to the predominance of exposure, resulting in the model over fitting these high-heat samples during training. In contrast, long-tail premium resume, which is not viewed because of poor presentation location, lacks an effective feedback sample. The deviation of the data layer directly reacts to the matching model, so that potential talents are difficult to excavate through learning, and along with model iteration, the preference of the popular sample is continuously strengthened, so that the system finally loses the excavation capability of long tails and special talents, and the problem of unreasonable scheduling of human resources of enterprise