A job data matching method and system based on large model technology

Abstract

The present application discloses a job data matching method and system based on large model technology. By responding to the talent value index determination instruction triggered by the user on the first interface, the generative pre-trained language model is called to extract nodes from job recruitment data, and the entity nodes are combined and processed to obtain a job combination data set. The generative pre-trained language model is called to extract entities from the input personal resume data to obtain user-preferred job tags and user job combination data; the user-preferred job tags are input into the optimal job data push model to output a target job combination data set; the target job combination data set and the user job combination data are input into the data matching model to output multiple target job combination data; and based on the multiple target job combination data, the talent value index of the user is determined and displayed through the first interface; compared with the prior art, the present application can improve the real-time performance and acquisition efficiency of the talent value index.

Description

Technical Field

[0001] This application relates to the technical field of large model technology processing, and particularly to a method and system for matching job data based on large model technology. Background Art

[0002] In modern recruitment scenarios, whether it is enterprise recruitment or individual job application, determining the personal value index of applicants has become an important and effective evaluation method; the personal value index is a comprehensive indicator that quantitatively evaluates multiple dimensions of applicants, such as skills, experience, and cultural adaptability, aiming to improve recruitment efficiency and matching quality.

[0003] When the prior art determines the personal value index, it often uses a quantitative scoring method for evaluation and processing. However, the needs of enterprises and positions are dynamically changing, and the traditional quantitative scoring method cannot adapt to the constantly changing working environment and needs, resulting in the evaluation results may not reflect the current actual abilities of applicants, affecting the scientificity and objectivity of the evaluation results.

[0004] At the same time, when determining a personal value index with reference value, it is often necessary to refer to diverse data sources. If data analysis and processing are based on manual work, it may take a relatively long time, resulting in low efficiency in obtaining the personal value index. Therefore, how to integrate large model technology and utilize the existing large model technology to quickly process and analyze large-scale and diverse data sources to improve the evaluation efficiency of the personal value index is also a technical problem that urgently needs to be solved currently. Summary of the Invention

[0005] The technical problem to be solved by this application is: to provide a method and system for matching job data based on large model technology, which can improve the real-time performance and acquisition efficiency of the talent value index.

[0006] To solve the above technical problem, this application provides a method for matching job data based on large model technology, including:

[0007] Responding to a talent value index determination instruction triggered by a user on a first interface, calling a generative pre-trained language model to perform entity extraction on the job recruitment data corresponding to a target job obtained in real time, and obtaining entity nodes, where the entity nodes include professional skills, job salary, and educational background;

[0008] Performing combination processing on the professional skills, the job salary, and the educational background to obtain multiple job combination data, setting job labels for the multiple job combination data according to the target job, and integrating the job label combination data corresponding to all job labels to obtain a job combination data set;

[0009] Based on the job combination data set, train the job data push model to obtain the optimal job data push model;

[0010] Obtain the personal resume data input by the user on the first interface, call the generative pre-trained language model to perform entity extraction on the personal resume data, determine the user-preferred job tags, user professional skills, user expected salary, and user educational background corresponding to the personal resume data, and generate user job combination data based on the user professional skills, the user expected salary, and the user educational background;

[0011] Input the user-preferred job tags into the optimal job data push model, so that the optimal job data push model performs similarity matching between the user-preferred job tags and the job combination data set, and outputs a target job combination data set;

[0012] Input the target job combination data set and the user job combination data into the pre-trained data matching model, so that the data matching model outputs multiple target job combination data corresponding to the user job combination data;

[0013] Based on the multiple target job combination data, determine the talent value index corresponding to the personal resume data, and display the talent value index to the user based on the first interface.

[0014] In a possible implementation, call the generative pre-trained language model to perform entity extraction on the job recruitment data corresponding to the target job obtained in real time, and obtain entity nodes, where the entity nodes include professional skills, working years, job salary, and educational background, specifically including:

[0015] Crawl the job recruitment image data corresponding to the target job from multiple data sources, and determine the data type corresponding to the job recruitment data, where the data type includes job recruitment images and job recruitment texts;

[0016] When the data type corresponding to the job recruitment data is the job recruitment image, perform image preprocessing on the job recruitment image to obtain a preprocessed job recruitment image;

[0017] Based on OCR technology, perform text recognition processing on the preprocessed job recruitment image to obtain the first job recruitment text data, and perform standardization processing on the job recruitment text data to obtain standard job recruitment text data;

[0018] Call the generative pre-trained language model to perform feature extraction on the first standard job recruitment text data to obtain the entity nodes of the target job;

[0019] When the data type corresponding to the job recruitment data is the job recruitment text, perform standardization processing on the job recruitment text to obtain the second standard job recruitment text data;

[0020] Call a generative pre-trained language model to extract features from the second standard job recruitment text data to obtain the entity nodes of the target job;

[0021] Among them, the entity nodes include professional skills, working years, job salary, and educational background.

[0022] In a possible implementation, perform combination processing on the professional skills, the job salary, and the educational background to obtain multiple job combination data, specifically including:

[0023] Obtain the first professional skills, the first job salary, and the first educational background corresponding to each job recruitment image in the target job;

[0024] Perform encoding processing on the first professional skills, the first job salary, and the first educational background respectively to obtain the first professional skills encoding data, the first job salary encoding data, and the first educational background encoding data;

[0025] Perform splicing processing on the first professional skills encoding data, the first job salary encoding data, and the first educational background encoding data to obtain the job combination data corresponding to each job recruitment image in the target job;

[0026] Integrate the job combination data corresponding to all job recruitment images in the target job to obtain multiple job combination data.

[0027] In a possible implementation, based on the job combination data set, perform model training on the job data push model to obtain the optimal job data push model, specifically including:

[0028] Perform division processing on the job combination data set according to a preset ratio to obtain a job combination data training set;

[0029] Based on the model structure of single input - multiple outputs, construct an initial job data push model, where the job label corresponding to each job combination data in the job combination data training set is used as the single input of the model, and all job combination data corresponding to the job label in the job combination data training set is used as the multiple outputs of the model;

[0030] Based on the cross-entropy loss function and the gradient descent algorithm, perform iterative training on the initial job data push model until the model converges or reaches the preset number of iterations to determine the optimal job data push model.

[0031] In a possible implementation, the target job combination dataset and the user job combination data are input into a pre-trained data matching model, so that the data matching model outputs multiple target job combination data corresponding to the user job combination data. Specifically, it includes:

[0032] Input the job combination dataset and the user job combination data into the pre-trained data matching model;

[0033] Based on the data matching model, vectorize the user job combination data and each target job combination data in the target job combination dataset to obtain a user job combination data vector and a set of target job combination data vectors;

[0034] Map the user job combination data vector and each target job combination data vector in the set of target job combination data vectors to the corresponding word vector space;

[0035] Calculate the cosine similarity between the user job combination data vector and each target job combination data vector, and based on the cosine similarity, determine the first distance between the user job combination data vector and each target job combination data vector;

[0036] Sort the first distances in ascending order to determine a first distance sequence, obtain the first target number of target first distances in the first distance sequence, and use the target job combination data corresponding to the first target number of target first distances as the multiple target job combination data corresponding to the user job combination data, and output the multiple target job combination numbers.

[0037] In a possible implementation, based on the multiple target job combination data, determine the talent value index corresponding to the user. Specifically, it includes:

[0038] Decode the multiple target job combination data respectively to obtain first decoded data corresponding to each target job combination data, where the first decoded data includes the decoded target job salary;

[0039] Based on the decoded target job salary, determine the target job value range, adjust the target job value range to obtain an adjusted target job value range, and based on the adjusted target job value range, determine the talent value index corresponding to the user.

[0040] In a possible implementation, adjust the target job value range to obtain an adjusted target job value range. Specifically, it includes:

[0041] Based on the target position value range, determine the upper limit and the lower limit of the first target position value;

[0042] Obtain the decoded target position salary corresponding to each target position combination data in the multiple target position combination data, and calculate the average value of the decoded target position salary based on the decoded target position salary;

[0043] When it is detected that the first difference between the average value of the decoded target position salary and the lower limit of the first target position value is greater than the second difference between the average value of the decoded target position salary and the upper limit of the first target position value, calculate the first proportion of the first difference within the second difference, and based on the first proportion, adjust the lower limit of the first target position value to obtain the adjusted lower limit of the first target position value;

[0044] Based on the adjusted lower limit of the first target position value and the upper limit of the first target position value, determine the adjusted target position value range;

[0045] When it is detected that the first difference between the average value of the decoded target position salary and the lower limit of the first target position value is less than the second difference between the average value of the decoded target position salary and the upper limit of the first target position value, calculate the second proportion of the first difference within the second difference, and based on the second proportion, adjust the upper limit of the first target position value to obtain the adjusted upper limit of the first target position value;

[0046] Based on the lower limit of the first target position value and the adjusted upper limit of the first target position value, determine the adjusted target position value range.

[0047] This application also provides a position data matching system based on large model technology, including: a position recruitment image acquisition module, a data combination module, a position data push model training module, a user resume data acquisition module, a target position combination data set push module, a target position combination data matching module, and a talent value index determination module;

[0048] The position recruitment image acquisition module is used to respond to the talent value index determination instruction triggered by the user on the first interface, and call the generative pre-trained language model to perform entity extraction on the position recruitment data corresponding to the real-time obtained target position, and obtain entity nodes, where the entity nodes include professional skills, position salary, and educational background;

[0049] The data combination module is used to perform combination processing on the professional skills, the position salary, and the educational background to obtain multiple position combination data, set position labels for the multiple position combination data according to the target position, and integrate the position label combination data corresponding to all position labels to obtain a position combination data set;

[0050] The job data push model training module is used to train the job data push model based on the job combination data set to obtain the optimal job data push model;

[0051] The user resume data acquisition module acquires the personal resume data input by the user on the first interface, calls the generative pre-trained language model to perform entity extraction on the personal resume data to determine the user-preferred job tags, user professional skills, user expected salary, and user educational background corresponding to the personal resume data, and generates user job combination data based on the user professional skills, the user expected salary, and the user educational background;

[0052] The target job combination data set push module is used to input the user-preferred job tags into the optimal job data push model, so that the optimal job data push model performs similarity matching between the user-preferred job tags and the job combination data set and outputs a target job combination data set;

[0053] The target job combination data matching module is used to input the target job combination data set and the user job combination data into a pre-trained data matching model, so that the data matching model outputs multiple target job combination data corresponding to the user job combination data;

[0054] The talent value index determination module is used to determine the talent value index corresponding to the personal resume data based on the multiple target job combination data and display the talent value index to the user on the first interface.

[0055] In a possible implementation manner, the job recruitment image acquisition module is used to perform entity extraction on the job recruitment data corresponding to the target job obtained in real time by using a generative pre-trained language model to obtain entity nodes, where the entity nodes include professional skills, working years, job salary, and educational background, specifically including:

[0056] Crawl the job recruitment data corresponding to the target job from multiple data sources and determine the data type corresponding to the job recruitment data, where the data type includes job recruitment images and job recruitment texts;

[0057] When the data type corresponding to the job recruitment data is the job recruitment image, perform image preprocessing on the job recruitment image to obtain a preprocessed job recruitment image;

[0058] Perform text recognition processing on the preprocessed job recruitment image based on OCR technology to obtain job recruitment text data, and perform standardization processing on the job recruitment text data to obtain first-standard job recruitment text data;

[0059] Call the generative pre-trained language model to extract features from the first standard job recruitment text data to obtain the entity nodes of the target job;

[0060] When the data type corresponding to the job recruitment data is the job recruitment text, perform standardization processing on the job recruitment text to obtain the second standard job recruitment text data;

[0061] Call the generative pre-trained language model to extract features from the second standard job recruitment text data to obtain the entity nodes of the target job;

[0062] Among them, the entity nodes include professional skills, working years, job salary, and educational background.

[0063] In a possible implementation manner, the data combination module is used to perform combination processing on the professional skills, the job salary, and the educational background to obtain multiple job combination data, specifically including:

[0064] Obtain the first professional skills, the first job salary, and the first educational background corresponding to each job recruitment image in the target job;

[0065] Respectively perform encoding processing on the first professional skills, the first job salary, and the first educational background to obtain the first professional skills encoding data, the first job salary encoding data, and the first educational background encoding data;

[0066] Perform splicing processing on the first professional skills encoding data, the first job salary encoding data, and the first educational background encoding data to obtain the job combination data corresponding to each job recruitment image in the target job;

[0067] Integrate the job combination data corresponding to all job recruitment images in the target job to obtain multiple job combination data.

[0068] In a possible implementation manner, the job data push model training module is used to train the job data push model based on the job combination data set to obtain the optimal job data push model, specifically including:

[0069] Divide the job combination data set according to a preset ratio to obtain a job combination data training set;

[0070] Based on a single-input and multi-output model structure, an initial job data push model is constructed. Among them, the job label corresponding to each job combination data in the job combination data training set is used as the single input of the model, and all job combination data corresponding to the job label in the job combination data training set is used as the multi-output of the model;

[0071] The initial job data push model is iteratively trained based on the cross-entropy loss function and the gradient descent algorithm until the model converges or reaches a preset number of iterations to determine the optimal job data push model.

[0072] In a possible implementation manner, the target job combination data matching module is used to input the target job combination data set and the user job combination data into a pre-trained data matching model, so that the data matching model outputs multiple target job combination data corresponding to the user job combination data, specifically including:

[0073] Input the job combination data set and the user job combination data into the pre-trained data matching model;

[0074] Based on the data matching model, vectorize the user job combination data and each target job combination data in the target job combination data set to obtain a user job combination data vector and a target job combination data vector set;

[0075] Map the user job combination data vector and each target job combination data vector in the target job combination data vector set to the corresponding word vector space;

[0076] Calculate the cosine similarity between the user job combination data vector and each target job combination data vector, and based on the cosine similarity, determine the first distance between the user job combination data vector and each target job combination data vector;

[0077] Sort the first distances in ascending order to determine a first distance sequence, obtain the first target number of target first distances in the first distance sequence, and use the target job combination data corresponding to the first target number of target first distances as the multiple target job combination data corresponding to the user job combination data, and output the multiple target job combination numbers.

[0078] In a possible implementation manner, the talent value index determination module is used to determine the talent value index corresponding to the user based on the multiple target job combination data, specifically including:

[0079] Decode the multiple target position combination data respectively to obtain first decoded data corresponding to each target position combination data, where the first decoded data includes decoded target position salaries;

[0080] Based on the decoded target position salaries, determine a target position value range, adjust the target position value range to obtain an adjusted target position value range, and based on the adjusted target position value range, determine a talent value index corresponding to the user.

[0081] In a possible implementation, the talent value index determination module is used to adjust the target position value range to obtain an adjusted target position value range, specifically including:

[0082] Based on the target position value range, determine a first target position value upper limit and a first target position value lower limit;

[0083] Obtain the decoded target position salaries corresponding to each target position combination data in the multiple target position combination data, and calculate an average decoded target position salary based on the decoded target position salaries;

[0084] When it is detected that a first difference between the average decoded target position salary and the first target position value lower limit is greater than a second difference between the average decoded target position salary and the first target position value upper limit, calculate a first ratio of the first difference within the second difference, and based on the first ratio, adjust the first target position value lower limit to obtain an adjusted first target position value lower limit;

[0085] Based on the adjusted first target position value lower limit and the first target position value upper limit, determine an adjusted target position value range;

[0086] When it is detected that the first difference between the average decoded target position salary and the first target position value lower limit is less than the second difference between the average decoded target position salary and the first target position value upper limit, calculate a second ratio of the first difference within the second difference, and based on the second ratio, adjust the first target position value upper limit to obtain an adjusted first target position value upper limit;

[0087] Based on the first target position value lower limit and the adjusted first target position value upper limit, determine an adjusted target position value range.

[0088] This application also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the position data matching method based on large model technology as described in any one of the above.

[0089] The present application also provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the job data matching method based on the large model technology as described in any one of the above.

[0090] A job data matching method and system based on the large model technology in an embodiment of the present application have the following beneficial effects compared with the prior art:

[0091] Based on the talent value index determination instruction triggered by the user on the first interface, entity node combination, job label setting, etc. are performed on the entity nodes in the job recruitment data corresponding to the target job obtained by invoking the generative pre-trained language model to obtain a job combination data set; ensure the real-time nature of the obtained job combination data set, and subsequently perform model training on the job data push model based on the job combination data set with real-time nature to obtain the optimal job data push model, which can enable the optimal job data push model to learn the current latest job combination data information and avoid the information difference problem; when the user-preferred job label is input into the optimal job data push model subsequently, it can correspondingly output a target job combination data set with real-time nature; and finally, the target job combination data set and the user job combination data are input into the data matching model, so that the data matching model can match multiple target job combination data with a relatively high degree of matching with the personal resume data from the target job combination data set; based on the multiple target job combination data, determine the talent value index corresponding to the user, making the evaluation of the talent value index more scientific and objective; when facing a rapidly changing working environment and diverse job requirements, it shows stronger adaptability, flexibility and acquisition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 is a schematic flowchart of an embodiment of a job data matching method based on the large model technology provided by the present application;

[0093] Figure 2 is a schematic structural diagram of an embodiment of a job data matching system based on the large model technology provided by the present application;

[0094] Figure 3 is a schematic structural diagram of a terminal device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0095] The following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0096] Embodiment 1, refer to Figure 1 , Figure 1 which is a schematic flowchart of an embodiment of a job data matching method based on large model technology provided by the present application. As Figure 1 shown, the method includes steps 101 - 106, specifically as follows:

[0097] Step 101: In response to the talent value index determination instruction triggered by the user on the first interface, call the generative pre - trained language model to perform entity extraction on the job recruitment data corresponding to the target job obtained in real - time, and obtain entity nodes, where the entity nodes include professional skills, job salary, and educational background.

[0098] In one embodiment, the job data matching method based on large model technology can be applied to intelligent terminal devices, which include but are not limited to smart phones, laptop computers, tablet computers, desktop computers, and physical servers and cloud servers connected with display units, etc.

[0099] In one embodiment, by detecting the target area on the first interface, when a triggering operation is detected in the target area, it is determined that the user triggers the talent value index determination instruction on the first interface.

[0100] Specifically, the triggering operation includes but is not limited to operations such as mouse clicking or gesture touching.

[0101] In one embodiment, crawl the job recruitment data corresponding to each target job from multiple data sources, and determine the data type corresponding to the job recruitment data, where the data type includes job recruitment images and job recruitment texts.

[0102] Specifically, the multiple data sources include but are not limited to recruitment websites, enterprise official websites, social media, and industry reports, etc.

[0103] Specifically, the number of job recruitment data corresponding to each target job can be one or more.

[0104] Specifically, the job recruitment data corresponding to the target job is the job recruitment data published within the target time range, where the target time range is the time range from the current moment to the preset time before.

[0105] Preferably, the job recruitment data can be all job recruitment data within the previous month before the response moment of the talent value index determination instruction. By restricting the job recruitment data queried by time, problems such as excessive data load, slow data response efficiency caused by too large query data can be avoided. Moreover, the collected job recruitment data is recent data, which can better reflect the current actual situation and meet the dynamic changes of enterprise and job requirements.

[0106] Specifically, the job recruitment data corresponding to the target job is crawled from the multiple data sources based on web crawler technology.

[0107] In one embodiment, when the data type of the job recruitment data is the job recruitment image, image preprocessing is performed on the job recruitment image to obtain a preprocessed job recruitment image.

[0108] In one embodiment, the image preprocessing includes but is not limited to image scaling processing, image noise processing, and image enhancement processing.

[0109] In one embodiment, the job recruitment image is subjected to image scaling processing according to a preset image size to obtain a job recruitment image, the job recruitment image is subjected to filtering processing based on an image filtering algorithm to obtain a filtered job recruitment image, and then the filtered job recruitment image is subjected to image enhancement processing based on a contrast enhancement algorithm to obtain a preprocessed job recruitment image.

[0110] Preferably, the image filtering algorithm includes but is not limited to a mean filtering algorithm, a median filtering algorithm, and a Gaussian filtering algorithm; the contrast enhancement algorithm includes but is not limited to a histogram equalization algorithm and an adaptive histogram equalization algorithm.

[0111] By performing image preprocessing on the job recruitment image, the size of the job recruitment image can be standardized, facilitating subsequent operations. At the same time, based on image noise processing, the influence of noise on the job recruitment image can be reduced, improving the accuracy and efficiency of subsequent image processing. Finally, based on image contrast enhancement processing, the problem that the text in the job recruitment image may be blurred due to image quality can also be avoided, improving the readability of the image.

[0112] In one embodiment, text recognition processing is performed on the preprocessed job recruitment image based on OCR technology to obtain job recruitment text data, and the job recruitment text data is standardized to obtain first standard job recruitment text data.

[0113] Specifically, based on an edge detection algorithm, text localization is performed on the preprocessed job recruitment image to determine the text region image in the preprocessed job recruitment image. Character segmentation processing is performed on the text region image to obtain multiple character region images. Based on OCR technology, character recognition is performed on the multiple character region images to obtain the target character corresponding to each character region image, and the job recruitment text data is determined by integrating the target characters.

[0114] Specifically, when performing standardization processing on the job recruitment text data, it is detected whether there are incorrect characters in the job recruitment text data. If so, data filtering processing is performed on the incorrect characters, and it is detected whether there are invalid characters in the job recruitment text data. If so, data filtering processing is performed on the invalid characters to obtain the standard job recruitment text data. If it is detected that there are no such incorrect characters in the job recruitment text data and there are no such invalid characters in the job recruitment text data, the job recruitment text data is directly used as the first standard job recruitment text data.

[0115] In one embodiment, a generative pre-trained language model is called to perform feature extraction on the first standard job recruitment text data to obtain the entity nodes of the target job, where the entity nodes include professional skills, working years, job salary, and educational background.

[0116] Specifically, a generative pre-trained language model is called to perform word segmentation processing on the first standard job recruitment text data to obtain multiple word groups, and feature label annotation is performed on each word group respectively. The preset job recruitment feature word groups and the feature labels corresponding to the job recruitment feature word groups are used as extraction features, and regular expressions are used to extract feature information from the first standard job recruitment text data based on the extraction features to obtain the entity nodes of the target job.

[0117] Preferably, the generative pre-trained language model includes but is not limited to GPT (Generative Pre-trained Transformer). Since the model has learned a large number of language rules, context information, and word usage frequencies, this enables the model to effectively understand the structure and semantics of language, including the relationships between words and the meanings of context. This context understanding ability enables the model to process complex text structures and contexts and accurately identify word boundaries.

[0118] Specifically, a word segmenter is set in the generative pre-trained language model. The generative pre-trained language model is called to perform word segmentation processing on the input first standard job recruitment text data based on the word segmenter in the generative pre-trained language model to obtain multiple word groups.

[0119] Preferably, the feature tags include, but are not limited to, professional skill tags, work experience tags, position salary tags, and educational background tags.

[0120] In one embodiment, when the data type of the position recruitment data is the position recruitment text, the position recruitment text is normalized to obtain second standard position recruitment text data; a generative pre-trained language model is called to extract features from the second standard position recruitment text data to obtain entity nodes of the target position; wherein the entity nodes include professional skills, work experience, position salary, and educational background.

[0121] Preferably, when the data type of the position recruitment data is the position recruitment text, compared with the position recruitment image, the process of preprocessing and recognizing the image is reduced, and the remaining steps are the same, which will not be described in detail here.

[0122] In one embodiment, the professional skill is the main skill or professional requirement corresponding to the target position, the position salary is the salary data of the target position, and the educational background is the educational background required for the target position.

[0123] Step 102: Combine the professional skills, the position salary, and the educational background to obtain multiple position combination data, set position tags for the multiple position combination data according to the target position, and integrate the position tag combination data corresponding to all the position tags to obtain a position combination data set;

[0124] In one embodiment, the first professional skill, the first position salary, and the first educational background corresponding to each position recruitment image in the target position are obtained, and the first professional skill, the first position salary, and the first educational background are respectively encoded to obtain first professional skill encoded data, first position salary encoded data, and first educational background encoded data.

[0125] Specifically, when encoding the first professional skill, a pre-constructed professional skill encoding system is obtained, wherein the professional skill encoding system contains professional skill encoded data uniquely corresponding to each professional skill. Based on the professional skill encoding system, the first professional skill is mapped to obtain the first professional skill encoded data corresponding to the first professional skill.

[0126] Illustrative example, the professional skill encoding system includes: data analysis = 001; Python programming = 002; machine learning = 003; database management = 004; big data technology = 005.

[0127] When the first professional skills required for the target position are data analysis, Python programming, and machine learning, these first professional skills can be mapped to corresponding codes. For example, data analysis = 001, Python programming = 002, and machine learning = 003. Finally, the first professional skill code data corresponding to the target position is a code sequence 001-002-003.

[0128] Specifically, when encoding the first position salary, obtain the position salary value corresponding to the first position salary, and use the position salary value as the first position salary code data.

[0129] Specifically, when encoding the first educational background, obtain a pre-constructed educational background coding system. Among them, the educational background coding system contains the educational background code data uniquely corresponding to each educational background. Based on the educational background system, map the first educational background to obtain the first educational background code data corresponding to the first educational background.

[0130] Illustrative example, the educational background coding system includes: junior high school = A; high school = B; technical secondary school = C; junior college = D; undergraduate = E; postgraduate = F, etc.

[0131] When the first educational background required for the target position is undergraduate, these first educational backgrounds can be mapped to corresponding codes. For example, undergraduate = E. Finally, the first educational background code data corresponding to the target position is E.

[0132] In one embodiment, splice the first professional skill code data, the first position salary code data, and the first educational background code data to obtain the position combination data corresponding to each position recruitment image in the target position.

[0133] Specifically, use the first professional skill code data as the sentence start code, use the first educational background code data as the sentence middle code, and use the first position salary code data as the sentence end code. Splice the first professional skill code data, the first position salary code data, and the first educational background code data in the order of sentence start code - sentence middle code - sentence end code to obtain the position combination data corresponding to each position recruitment image in the target position.

[0134] In one embodiment, integrate the position combination data corresponding to all position recruitment images in the target position to obtain multiple position combination data.

[0135] In one embodiment, when setting job tags for the multiple job combination data according to the target job, obtain the job tags corresponding to the target job, and splice the job tags with the multiple job combination data respectively to implement tag setting for the job combination data.

[0136] Preferably, when splicing the job tags with the multiple job combination data respectively, use the job tags as the first key names, and use the job combination data as the first key values, and generate job tag combination data based on the form of the first key name - first key value.

[0137] Step 103: Based on the job combination data set, train a job data push model to obtain an optimal job data push model.

[0138] In one embodiment, divide the job combination data set according to a preset ratio to obtain a job combination data training set.

[0139] Specifically, the preset ratio is 80%.

[0140] Specifically, randomly extract 80% of the data from the job combination data set as the job combination data training set, and train the model based on the job combination data training set.

[0141] Preferably, based on the preset ratio, also randomly extract 20% of the data from the job combination data set as the job combination data test set, and detect the model performance based on the job combination data test set.

[0142] In one embodiment, construct an initial job data push model based on a single-input multi-output model structure, where the job tags corresponding to each job combination data in the job combination data training set are used as a single input, and all job combination data corresponding to the job tags in the job combination data training set are used as multi-outputs.

[0143] Specifically, use each job combination data in the job combination data training set as the model input, and use all job combination data corresponding to the job tags in the job combination data training set as the model output, and train the initial job data push model.

[0144] In one embodiment, perform iterative training on the initial job data push model based on the cross-entropy loss function and the gradient descent algorithm until the model converges or reaches a preset number of iterations to determine the optimal job data push model.

[0145] Specifically, set initial model parameters for the initial job data push model, select a first random job label from the job combination data training set, input the first random job label into the initial job data push model, and calculate the loss function. When it is determined that the loss function has not converged, calculate the first difference value between the pushed job combination data output by the job data push model and the true job combination data, transfer the first difference value back to the initial job data push model, and calculate the gradient value; use the gradient descent algorithm to update the initial model parameters according to the gradient value to minimize the value of the loss function, and re-select a second random job label from the job combination data training set, input the second random job label into the initial job data push model, and recalculate the loss function until the loss function of the initial job data push model converges or reaches a preset number of iterations, and stop the iterative training of the model.

[0146] In one embodiment, after completing the iterative training of the initial job data push model, test the initial job data push model based on the job combination data test set to obtain a test result. When the test result meets a preset test result threshold, determine the optimal job data push model, where the test result includes, but is not limited to, performance metric indicators such as accuracy and F1-score.

[0147] Step 104: Obtain the personal resume data input by the user on the first interface, call the generative pre-trained language model to perform entity extraction on the personal resume data, determine the user-preferred job label, user professional skills, user expected salary, and user educational background corresponding to the personal resume data, and generate user job combination data based on the user professional skills, the user expected salary, and the user educational background.

[0148] In one embodiment, by detecting the target input area on the first interface, when it is detected that there is an input operation in the target input area, obtain the personal resume data input by the user on the first interface, where the personal resume data includes, but is not limited to, personal resume images and personal resume texts.

[0149] In one embodiment, when the data type corresponding to the resume data is a resume image, perform image preprocessing on the resume image to obtain a preprocessed resume image; perform text recognition processing on the preprocessed resume image based on OCR technology to obtain resume text data, and perform standardization processing on the resume text data to obtain first-standard resume text data; call a generative pre-trained language model to perform feature extraction on the first-standard resume text data to obtain a resume entity node corresponding to the resume image, where the resume node includes a user-preferred position label, user professional skills, user expected salary, and user educational background.

[0150] Preferably, for the expected position in the user's resume image corresponding to the user-preferred position label, based on the expected position, determine the user-preferred position label.

[0151] In one embodiment, when the data type corresponding to the resume data is resume text, perform standardization processing on the resume text to obtain second-standard resume text data; call a generative pre-trained language model to perform feature extraction on the second-standard resume text data to obtain a resume entity node corresponding to the resume text; where the resume entity node includes professional skills, working years, position salary, and educational background.

[0152] Preferably, the process of obtaining the user-preferred position label, user professional skills, user expected salary, and user educational background corresponding to the resume data by calling a generative pre-trained language model to perform entity extraction on the resume data input by the user on the first interface is the same as the process of calling a generative pre-trained language model in step 101 above to perform entity extraction on the position recruitment data to obtain entity nodes, and will not be described in detail here.

[0153] In one embodiment, perform combination processing on the user professional skills, the user expected salary, and the user educational background to obtain user position combination data.

[0154] Specifically, perform encoding processing on the user professional skills, the user expected salary, and the user educational background respectively to obtain user professional skills encoding data, user expected salary encoding data, and user educational background encoding data; perform splicing processing on the user professional skills encoding data, the user expected salary encoding data, and the user educational background encoding data to obtain user educational background encoding data.

[0155] Preferably, the process of encoding the user's professional skills, the user's expected salary, and the user's educational background is the same as the process of encoding the user's professional skills, the user's expected salary, and the user's educational background in step 102 above, and will not be described in detail here.

[0156] Step 105: Input the user-preferred job label into the optimal job data push model, so that the optimal job data push model performs a similarity match between the user-preferred job label and the job combination data set, and outputs a target job combination data set.

[0157] In one embodiment, after determining the user-preferred job label corresponding to the user, directly use the user-preferred job label as the model input of the optimal job data push model, so that the optimal job data push model matches the user-preferred job label with each job label in the job combination data set respectively, and outputs the target job combination data set corresponding to the target job label corresponding to the user-preferred job label.

[0158] Specifically, when the optimal job data push model performs a similarity match between the user-preferred job label and the job combination data set, it obtains all job labels in the job combination data set, performs a similarity match between the user-preferred job label and all job labels respectively, and obtains the target job label corresponding to the user-preferred job label; based on the job combination data set, it obtains all job combination data sets corresponding to the target job label, and uses all job combination data sets as the target job combination data set corresponding to the user-preferred job label.

[0159] Preferably, when performing a similarity match between the user-preferred job label and all job labels, the Word Embedding model in natural language processing technology can be used to calculate the similarity between words; the Word Embedding model can represent words as vectors with semantic information, and the similarity between two words can be measured according to the distance or similarity between vectors.

[0160] Step 106: Input the target job combination data set and the user job combination data into a pre-trained data matching model, so that the data matching model outputs multiple target job combination data corresponding to the user job combination data.

[0161] In one embodiment, the job combination data set and the user job combination data are input into a pre-trained data matching model.

[0162] Specifically, the model structure of the data matching model is multi-input - multi-output.

[0163] Specifically, set the job combination dataset and the user's job combination data as multiple inputs of the data matching model.

[0164] Specifically, set the multiple target job combination data as multiple outputs of the data matching model.

[0165] In one embodiment, perform vectorization processing on the user's job combination data and each target job combination data in the target job combination dataset based on the data matching model, to obtain a user's job combination data vector and a set of target job combination data vectors.

[0166] In one embodiment, map the user's job combination data vector and each target job combination data vector in the set of target job combination data vectors to a corresponding vector space.

[0167] Specifically, by placing these vectors at appropriate positions in the feature space, a vector space is formed; in this way, different job combination data vectors are represented as different points in the vector space, and the distance and relative position between them can be used to measure the similarity between jobs.

[0168] In one embodiment, calculate the cosine similarity between the user's job combination data vector and each target job combination data vector, and based on the cosine similarity, determine a first distance between the user's job combination data vector and each target job combination data vector.

[0169] Specifically, use the cosine similarity as the first distance between the user's job combination data vector and each target job combination data vector.

[0170] In one embodiment, sort the first distances in ascending order to determine a first distance sequence, obtain the first target number of target first distances in the first distance sequence, and use the target job combination data corresponding to the first target number of target first distances as the multiple target job combination data corresponding to the user's job combination data, and output the multiple target job combination numbers.

[0171] Specifically, the first target number can be set based on user needs and can be one or more.

[0172] Step 107: Based on the multiple target job combination data, determine a talent value index corresponding to the personal resume data, and display the talent value index to the user based on the first interface.

[0173] In one embodiment, perform decoding processing on the multiple target job combination data respectively to obtain first decoding data corresponding to each target job combination data, where the first decoding data includes a decoded target job salary.

[0174] Specifically, when decoding the multiple target position combination data, each target position combination data is subjected to data splitting processing to obtain start-of-sentence splitting encoded data, in-sentence splitting encoded data, and end-of-sentence splitting encoded data; the end-of-sentence splitting encoded data is decoded to obtain the decoded target position salary.

[0175] Preferably, when decoding the end-of-sentence splitting encoded data to obtain the decoded target position salary, the end-of-sentence splitting encoded data is directly used as the decoded target position salary.

[0176] In one embodiment, based on the decoded target position salary, a target position value range is determined, the target position value range is adjusted to obtain an adjusted target position value range, and based on the adjusted target position value range, a talent value index corresponding to the user is determined.

[0177] Specifically, when determining the target position value range based on the decoded target position salary, the lowest decoded target position salary in the decoded target position salary is obtained, the highest decoded target position salary in the decoded target position salary is obtained, and based on the lowest decoded target position salary and the highest decoded target position salary, the target position value range is determined.

[0178] In one embodiment, when adjusting the target position value range to obtain an adjusted target position value range, first, based on the target position value range, a first target position value upper limit and a first target position value lower limit are determined, the decoded target position salary corresponding to each target position combination data in the multiple target position combination data is obtained, and based on the decoded target position salary, a decoded target position salary average value is calculated; when it is detected that a first difference between the decoded target position salary average value and the first target position value lower limit is greater than a second difference between the decoded target position salary average value and the first target position value upper limit, a first ratio of the first difference within the second difference is calculated, and based on the first ratio, the first target position value lower limit is adjusted to obtain an adjusted first target position value lower limit; based on the adjusted first target position value lower limit and the first target position value upper limit, the adjusted target position value range is determined; when it is detected that the first difference between the decoded target position salary average value and the first target position value lower limit is less than the second difference between the decoded target position salary average value and the first target position value upper limit, a second ratio of the first difference within the second difference is calculated, and based on the second ratio, the first target position value upper limit is adjusted to obtain an adjusted first target position value upper limit; based on the first target position value lower limit and the adjusted first target position value upper limit, the adjusted target position value range is determined.

[0179] Specifically, when determining the upper limit and the lower limit of the first target position value based on the target position value range, the highest decoded target position salary in the target position value range is used as the upper limit of the first target position value, and the lowest decoded target position salary in the target position value range is used as the lower limit of the second target position value.

[0180] Specifically, each obtained decoded target position salary is directly input into the existing mean calculation formula to calculate the mean of the decoded target position salary.

[0181] Specifically, when adjusting the lower limit of the first target position value based on the first ratio, the first ratio is multiplied by the lower limit of the first target position value to obtain the adjusted lower limit of the first target position value.

[0182] Specifically, when adjusting the upper limit of the first target position value based on the second ratio, the second ratio is multiplied by the upper limit of the first target position value to obtain the adjusted upper limit of the first target position value.

[0183] In one embodiment, when it is detected that the first difference between the mean of the decoded target position salary and the lower limit of the first target position value is equal to the second difference between the mean of the decoded target position salary and the upper limit of the first target position value, the target position value range is directly used as the adjusted target position value range.

[0184] In one embodiment, when determining the talent value index corresponding to the user based on the adjusted target position value range, the upper limit and the lower limit of the adjusted target position value corresponding to the adjusted target position value range are determined based on the adjusted target position value range; the median of the adjusted target position value of the adjusted target position value range is calculated based on the upper limit and the lower limit of the adjusted target position value; based on the preset linear weighted fusion model, the median of the adjusted target position value, the upper limit of the adjusted target position value, and the lower limit of the adjusted target position value are subjected to weighted fusion processing to obtain the talent value index corresponding to the user.

[0185] In one embodiment, the pre-constructed linear weighted fusion model is as follows:

[0186] ;

[0187] In the formula, is the talent value index, , and are weight values and satisfy , To adjust the upper limit of the target position value, To adjust the median value of the target position value, To adjust the lower limit of the target position value.

[0188] In one embodiment, after obtaining the talent value index corresponding to the user, the talent value index is used as the response result of the talent value index determination instruction, and the talent value index is displayed to the user through the first interface.

[0189] In one embodiment, after displaying the talent value index to the user in the first interface, it is considered that the response to the talent value index determination instruction has been completed. Thereafter, it is detected in real time whether the talent value index determination instruction in the target area of the first interface is triggered. After determining that the talent value index determination instruction is triggered, the talent value index determination instruction triggered by the user in the first interface is re-responded, and the above steps 101-step 107 are repeatedly executed, so that each determination of the talent value index can be based on the real-time obtained job recruitment data to update the optimal job data push model, so as to effectively respond to the situation that the needs of enterprises and positions are dynamically changing.

[0190] The applicable scenarios of a job data matching method based on large model technology provided in this embodiment include but are not limited to job seeker usage scenarios and enterprise usage scenarios. Among them, in the job seeker usage scenario, before applying for a job, a job seeker determines their personal talent value index based on personal resume data to help the individual better understand job information; in the enterprise usage scenario, after receiving the personal resume data of an applicant during recruitment, a company or enterprise determines the talent value index of the applicant based on the personal resume data to improve the recruitment decision-making efficiency of the company or enterprise.

[0191] Taking a job seeker as an example, a job data matching method based on large model technology provided in this application is specifically described: when a job seeker needs to obtain their corresponding talent value index, they can trigger the talent value index determination instruction in the first interface. At this time, the system will obtain the job recruitment data corresponding to different positions in real time from multiple data sources based on web crawler technology to ensure that the obtained job recruitment data has a certain degree of real-time nature, and call the generative pre-trained language model to perform entity extraction on the job recruitment data to obtain entity nodes. By combining the professional skills, job salary, and educational background in the entity nodes, job label combination data corresponding to job labels is generated, and then a job combination data set is generated. At this time, the job combination data set covers the job label combination data corresponding to all positions, which is used to train the optimal job data push model so that the optimal job data push model can learn the currently real-time obtained job label combination data.

[0192] After the user triggers the talent value index determination instruction in the first interface, the user can also output personal resume data to the first interface, so that the system calls the generative pre-trained language model to perform entity extraction on the personal resume data, obtaining user-preferred position tags and user position combination data; then, by calling the trained optimal position data push model, based on the user-preferred position tags in the personal resume data, it outputs a target position combination data set in the position combination data set that has a relatively high matching degree with the user-preferred position tags, and based on the data matching model, it matches multiple target position combination data corresponding to the user position combination data from the target position combination data set for calculating the talent value index; and based on the first interface, it displays the talent value index.

[0193] In this process, the user triggers the talent value index determination instruction in the first interface and inputs personal resume data to the first interface, then the acquisition of the talent value index can be realized based on the first interface, avoiding the complex data query and data analysis processes that need to be carried out manually in the prior art, and improving the real-time performance and acquisition efficiency of the talent value index.

[0194] Example 2, see Figure 2 , Figure 2 is a structural schematic diagram of an embodiment of a position data matching system provided by the present application based on large model technology. As Figure 2 shown, the system includes a job recruitment image acquisition module 201, a data combination module 202, a job data push model training module 203, a user resume data acquisition module 204, a target job combination data set push module 205, a target job combination data matching module 206, and a talent value index determination module 207, specifically as follows:

[0195] The job recruitment image acquisition module 201 is configured to respond to the talent value index determination instruction triggered by the user in the first interface, and call the generative pre-trained language model to perform entity extraction on the job recruitment data corresponding to the target job obtained in real time, obtaining entity nodes, where the entity nodes include professional skills, job salary, and educational background.

[0196] The data combination module 202 is configured to perform combination processing on the professional skills, the job salary, and the educational background, obtaining multiple job combination data, and setting job tags for the multiple job combination data according to the target job, and integrating the job tag combination data corresponding to all job tags to obtain a job combination data set.

[0197] The job data push model training module 203 is configured to perform model training on the job data push model based on the job combination data set to obtain an optimal job data push model.

[0198] The user resume data acquisition module 204 is used to acquire the personal resume data input by the user on the first interface, call the generative pre-trained language model to perform entity extraction on the personal resume data to determine the user-preferred job tags, user professional skills, user expected salary, and user educational background corresponding to the personal resume data, and generate user job combination data based on the user professional skills, the user expected salary, and the user educational background.

[0199] The target job combination dataset push module 205 is used to input the user-preferred job tags into the optimal job data push model, so that the optimal job data push model outputs a target job combination dataset.

[0200] The target job combination data matching module 206 is used to input the target job combination dataset and the user job combination data into a pre-trained data matching model, so that the data matching model performs similarity matching between the user-preferred job tags and the job combination data set, and outputs multiple target job combination data corresponding to the user job combination data.

[0201] The talent value index determination module 207 is used to determine the talent value index corresponding to the personal resume data based on the multiple target job combination data, and display the talent value index to the user based on the first interface.

[0202] In one embodiment, the job recruitment image acquisition module 201 is configured to call a generative pre-trained language model to perform entity extraction on the job recruitment data corresponding to the target job obtained in real time, so as to obtain entity nodes, where the entity nodes include professional skills, working years, job salary, and educational background. Specifically, it includes: crawling the job recruitment data corresponding to the target job from multiple data sources and determining the data type corresponding to the job recruitment data, where the data type includes job recruitment images and job recruitment texts; when the data type corresponding to the job recruitment data is the job recruitment image, performing image preprocessing on the job recruitment image to obtain a preprocessed job recruitment image; performing text recognition processing on the preprocessed job recruitment image based on OCR technology to obtain job recruitment text data, and performing standardization processing on the job recruitment text data to obtain first-standard job recruitment text data; calling a generative pre-trained language model to perform feature extraction on the first-standard job recruitment text data to obtain the entity nodes of the target job; when the data type corresponding to the job recruitment data is the job recruitment text, performing standardization processing on the job recruitment text to obtain second-standard job recruitment text data; calling a generative pre-trained language model to perform feature extraction on the second-standard job recruitment text data to obtain the entity nodes of the target job; where the entity nodes include professional skills, working years, job salary, and educational background.

[0203] In one embodiment, the data combination module 202 is configured to perform combination processing on the professional skills, the job salary, and the educational background to obtain a plurality of job combination data. Specifically, it includes: obtaining the first professional skills, the first job salary, and the first educational background corresponding to each job recruitment image in the target job; respectively performing encoding processing on the first professional skills, the first job salary, and the first educational background to obtain first professional skill encoding data, first job salary encoding data, and first educational background encoding data; performing splicing processing on the first professional skill encoding data, the first job salary encoding data, and the first educational background encoding data to obtain the job combination data corresponding to each job recruitment image in the target job; integrating the job combination data corresponding to all job recruitment images in the target job to obtain a plurality of job combination data.

[0204] In one embodiment, the job data push model training module 203 is configured to train a job data push model based on the job combination data set to obtain an optimal job data push model, which specifically includes: dividing the job combination data set according to a preset ratio to obtain a job combination data training set; constructing an initial job data push model based on a single-input multi-output model structure, where the job label corresponding to each job combination data in the job combination data training set is used as the single input of the model, and all job combination data corresponding to the job label in the job combination data training set is used as the multi-output of the model; performing iterative training on the initial job data push model based on the cross-entropy loss function and the gradient descent algorithm until the model converges or reaches a preset number of iterations to determine the optimal job data push model.

[0205] In one embodiment, the target job combination data matching module 206 is configured to input the target job combination data set and the user job combination data into a pre-trained data matching model, so that the data matching model outputs multiple target job combination data corresponding to the user job combination data, which specifically includes: inputting the job combination data set and the user job combination data into the pre-trained data matching model; performing vectorization processing on the user job combination data and each target job combination data in the target job combination data set based on the data matching model to obtain a user job combination data vector and a target job combination data vector set; mapping the user job combination data vector and each target job combination data vector in the target job combination data vector set to the corresponding word vector space; calculating the cosine similarity between the user job combination data vector and each target job combination data vector, and determining the first distance between the user job combination data vector and each target job combination data vector based on the cosine similarity; sorting the first distances in ascending order to determine a first distance sequence, obtaining the first target number of target first distances in the first distance sequence, and using the target job combination data corresponding to the first target number of target first distances as the multiple target job combination data corresponding to the user job combination data, and outputting the multiple target job combination numbers.

[0206] In one embodiment, the talent value index determination module 207 is configured to determine the talent value index corresponding to the user based on the multiple target job combination data, specifically including: respectively performing decoding processing on the multiple target job combination data to obtain first decoding data corresponding to each target job combination data, where the first decoding data includes the decoded salary of the target job; determining a target job value range based on the decoded salary of the target job, adjusting the target job value range to obtain an adjusted target job value range, and determining the talent value index corresponding to the user based on the adjusted target job value range.

[0207] In one embodiment, the talent value index determination module 207 is configured to adjust the target job value range to obtain an adjusted target job value range, specifically including: determining the first target job value upper limit and the first target job value lower limit based on the target job value range; obtaining the decoded salary of the target job corresponding to each target job combination data in the multiple target job combination data, and calculating the average decoded salary of the target job based on the decoded salary of the target job; when it is detected that a first difference between the average decoded salary of the target job and the first target job value lower limit is greater than a second difference between the average decoded salary of the target job and the first target job value upper limit, calculating a first ratio of the first difference within the second difference, and adjusting the first target job value lower limit based on the first ratio to obtain an adjusted first target job value lower limit; determining the adjusted target job value range based on the adjusted first target job value lower limit and the first target job value upper limit; when it is detected that the first difference between the average decoded salary of the target job and the first target job value lower limit is less than the second difference between the average decoded salary of the target job and the first target job value upper limit, calculating a second ratio of the first difference within the second difference, and adjusting the first target job value upper limit based on the second ratio to obtain an adjusted first target job value upper limit; determining the adjusted target job value range based on the first target job value lower limit and the adjusted first target job value upper limit.

[0208] The above job data matching device based on the large model technology can implement the job data matching method based on the large model technology in the above method embodiment. The optional items in the above method embodiment are also applicable to this embodiment and will not be elaborated here.

[0209] Figure 3 A schematic structural diagram of a terminal device. As Figure 3 shown, the terminal device 3 in this embodiment includes: at least one processor 301 ( Figure 3Only one) processor, a memory 302, and a computer program 303 stored in the memory 302 and executable on at least one processor 301 are shown. When the processor 301 executes the computer program 303, the steps in any of the above method embodiments are implemented.

[0210] The terminal device 3 may be a computing device such as a smart phone, a laptop computer, a tablet computer, and a desktop computer. The terminal device may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art can understand that Figure 3 merely examples of the terminal device 3 and do not constitute a limitation on the terminal device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0211] The so-called processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0212] The memory 302 may be an internal storage unit of the terminal device 3 in some embodiments, such as the hard disk or memory of the terminal device 3. The memory 302 may also be an external storage device of the terminal device 3 in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 3. Further, the memory 302 may also include both the internal storage unit and the external storage device of the terminal device 3. The memory 302 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 302 may also be used to temporarily store data that has been output or is to be output.

[0213] In addition, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0214] In several embodiments provided by the present application, it can be understood that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, the program segment, or the part of code includes one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved.

[0215] If the function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a terminal device to execute all or part of the steps of the methods in various embodiments of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0216] In summary, a job data matching method and system based on large model technology provided by the present application, by responding to the talent value index determination instruction triggered by the user on the first interface, calling the generative pre-trained language model for job recruitment data node extraction, and performing combined processing on the entity nodes to obtain a job combination data set, calling the generative pre-trained language model to perform entity extraction on the input personal resume data to obtain user-preferred job tags and user job combination data; inputting the user-preferred job tags into the optimal job data push model to output a target job combination data set; inputting the target job combination data set and the user job combination data into the data matching model to output multiple target job combination data; and determining the user's talent value index based on the multiple target job combination data and displaying it through the first interface; compared with the prior art, the present application can improve the real-time performance and acquisition efficiency of the talent value index.

[0217] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present application, several improvements and replacements can still be made, and these improvements and replacements should also be regarded as the protection scope of the present application.

Claims

1. A job data matching method based on large model technology, characterized in that Including: In response to a talent value index determination instruction triggered by a user on a first interface, a generative pre-trained language model is called to perform entity extraction on job recruitment data corresponding to a target position obtained in real time, obtaining entity nodes, where the entity nodes include professional skills, job salary, and educational background; Performing combination processing on the professional skills, the job salary, and the educational background to obtain multiple job combination data, and setting job tags for the multiple job combination data according to the target position, and integrating the job tag combination data corresponding to all job tags to obtain a job combination data set; Based on the job combination data set, training a job data push model to obtain an optimal job data push model; Obtaining personal resume data input by the user on the first interface, calling a generative pre-trained language model to perform entity extraction on the personal resume data, determining a user-preferred job tag, user professional skills, user expected salary, and user educational background corresponding to the personal resume data, and generating user job combination data based on the user professional skills, the user expected salary, and the user educational background; Inputting the user-preferred job tag into the optimal job data push model, so that the optimal job data push model performs similarity matching between the user-preferred job tag and the job combination data set, and outputs a target job combination data set; Inputting the target job combination data set and the user job combination data into a pre-trained data matching model, so that the data matching model outputs multiple target job combination data corresponding to the user job combination data; Based on the multiple target job combination data, determining a talent value index corresponding to the personal resume data, and displaying the talent value index to the user based on the first interface; Performing combination processing on the professional skills, the job salary, and the educational background to obtain multiple job combination data, specifically including: Obtaining the first professional skills, the first job salary, and the first educational background corresponding to each job recruitment image in the target position; Performing encoding processing on the first professional skills, the first job salary, and the first educational background respectively to obtain first professional skills encoding data, first job salary encoding data, and first educational background encoding data; Performing splicing processing on the first professional skills encoding data, the first job salary encoding data, and the first educational background encoding data to obtain job combination data corresponding to each job recruitment image in the target position; Integrating the job combination data corresponding to all job recruitment images in the target position to obtain multiple job combination data; Based on the multiple target job combination data, determining a talent value index corresponding to the user, specifically including: Performing decoding processing on the multiple target job combination data respectively to obtain first decoding data corresponding to each target job combination data, where the first decoding data includes a decoded target job salary; Based on the decoded target position salary, determine the target position value range, adjust the target position value range to obtain the adjusted target position value range, and based on the adjusted target position value range, determine the talent value index corresponding to the user.

2. The job data matching method based on large model technology according to claim 1, wherein, Call a generative pre-trained language model to perform entity extraction on the position recruitment data corresponding to the target position obtained in real time, and obtain entity nodes, where the entity nodes include professional skills, working years, position salary, and educational background, specifically including: Crawl the position recruitment image data corresponding to the target position from multiple data sources, and determine the data type corresponding to the position recruitment data, where the data type includes position recruitment images and position recruitment texts; When the data type corresponding to the position recruitment data is the position recruitment image, perform image preprocessing on the position recruitment image to obtain a preprocessed position recruitment image; Based on OCR technology, perform text recognition processing on the preprocessed position recruitment image to obtain position recruitment text data, and perform standardization processing on the position recruitment text data to obtain the first standard position recruitment text data; Call a generative pre-trained language model to perform feature extraction on the first standard position recruitment text data to obtain the entity nodes of the target position; When the data type corresponding to the position recruitment data is the position recruitment text, perform standardization processing on the position recruitment text to obtain the second standard position recruitment text data; Call a generative pre-trained language model to perform feature extraction on the second standard position recruitment text data to obtain the entity nodes of the target position; Among them, the entity nodes include professional skills, working years, position salary, and educational background.

3. The job data matching method based on large model technology according to claim 2, wherein Based on the position combination data set, perform model training on the position data push model to obtain the optimal position data push model, specifically including: Divide the position combination data set according to a preset ratio to obtain a position combination data training set; Based on the model structure of single input - multiple outputs, construct an initial position data push model, where the position label corresponding to each position combination data in the position combination data training set is used as the single input of the model, and all position combination data corresponding to the position label in the position combination data training set are used as the multiple outputs of the model; Based on the cross-entropy loss function and the gradient descent algorithm, perform iterative training on the initial position data push model until the model converges or reaches a preset number of iterations to determine the optimal position data push model.

4. The method for matching job data based on large model technology according to claim 3, wherein Input the target position combination data set and the user position combination data into a pre-trained data matching model, so that the data matching model outputs multiple target position combination data corresponding to the user position combination data, specifically including: Input the position combination data set and the user position combination data into the pre-trained data matching model; Vectorize each target job combination data in the user job combination data and the target job combination data set based on the data matching model to obtain a user job combination data vector and a set of target job combination data vectors; Map the user job combination data vector and each target job combination data vector in the set of target job combination data vectors to the corresponding word vector space; Calculate the cosine similarity between the user job combination data vector and each target job combination data vector, and determine the first distance between the user job combination data vector and each target job combination data vector based on the cosine similarity; Sort the first distances in ascending order to determine a first distance sequence, obtain the first target number of target first distances in the first distance sequence, and use the target job combination data corresponding to the first target number of target first distances as the multiple target job combination data corresponding to the user job combination data, and output the multiple target job combination numbers.

5. The method for matching job data based on large model technology according to claim 4, wherein, Adjust the target job value range to obtain an adjusted target job value range, specifically including: Based on the target job value range, determine a first target job value upper limit and a first target job value lower limit; Obtain the decoded target job salary corresponding to each target job combination data in the multiple target job combination data, and calculate the average value of the decoded target job salary based on the decoded target job salary; When it is detected that the first difference between the average value of the decoded target job salary and the first target job value lower limit is greater than the second difference between the average value of the decoded target job salary and the first target job value upper limit, calculate the first proportion of the first difference within the second difference, and based on the first proportion, adjust the first target job value lower limit to obtain an adjusted first target job value lower limit; Based on the adjusted first target job value lower limit and the first target job value upper limit, determine the adjusted target job value range; When it is detected that the first difference between the average value of the decoded target job salary and the first target job value lower limit is less than the second difference between the average value of the decoded target job salary and the first target job value upper limit, calculate the second proportion of the first difference within the second difference, and based on the second proportion, adjust the first target job value upper limit to obtain an adjusted first target job value upper limit; Based on the first target job value lower limit and the adjusted first target job value upper limit, determine the adjusted target job value range.

6. A job data matching system based on large model technology, which applies a job data matching method based on large model technology according to any one of claims 1 to 5, characterized in that, Including: A job recruitment image acquisition module, a data combination module, a job data push model training module, a user resume data acquisition module, a target job combination data set push module, a target job combination data matching module, and a talent value index determination module; The job recruitment image acquisition module is used to respond to the talent value index determination instruction triggered by the user on the first interface, call the generative pre-trained language model to perform entity extraction on the job recruitment data corresponding to the target job obtained in real time, and obtain entity nodes, where the entity nodes include professional skills, job salary, and educational background; The data combination module is used to perform combination processing on the professional skills, the job salary, and the educational background to obtain multiple job combination data, set job labels for the multiple job combination data according to the target job, and integrate the job label combination data corresponding to all job labels to obtain a job combination data set; The job data push model training module is used to perform model training on the job data push model based on the job combination data set to obtain an optimal job data push model; The user resume data acquisition module is used to acquire the personal resume data input by the user on the first interface, call the generative pre-trained language model to perform entity extraction on the personal resume data to determine the user preference job label, user professional skills, user expected salary, and user educational background corresponding to the personal resume data, and generate user job combination data based on the user professional skills, the user expected salary, and the user educational background; The target job combination data set push module is used to input the user preference job label into the optimal job data push model, so that the optimal job data push model performs similarity matching between the user preference job label and the job combination data set, and outputs a target job combination data set; The target job combination data matching module is used to input the target job combination data set and the user job combination data into a pre-trained data matching model, so that the data matching model outputs multiple target job combination data corresponding to the user job combination data; The talent value index determination module is used to determine the talent value index corresponding to the personal resume data based on the multiple target job combination data, and display the talent value index to the user on the first interface.

7. The job data matching system based on large model technology according to claim 6, wherein The job recruitment image acquisition module is used to call the generative pre-trained language model to perform entity extraction on the job recruitment data corresponding to the target job obtained in real time, and obtain entity nodes, where the entity nodes include professional skills, working years, job salary, and educational background, specifically including: Crawl the job recruitment data corresponding to the target job from multiple data sources, and determine the data type corresponding to the job recruitment data, where the data type includes job recruitment images and job recruitment texts; When the data type corresponding to the job recruitment data is the job recruitment image, perform image preprocessing on the job recruitment image to obtain a preprocessed job recruitment image; Perform text recognition processing on the preprocessed job recruitment image based on OCR technology to obtain job recruitment text data, and perform standardization processing on the job recruitment text data to obtain first-standard job recruitment text data; Call the generative pre-trained language model to extract features from the first standard job recruitment text data to obtain the entity nodes of the target job; When the data type corresponding to the job recruitment data is the job recruitment text, perform standardization processing on the job recruitment text to obtain the second standard job recruitment text data; Call the generative pre-trained language model to extract features from the second standard job recruitment text data to obtain the entity nodes of the target job; Among them, the entity nodes include professional skills, working years, job salary, and educational background.

8. The job data matching system based on large model technology according to claim 7, characterized in that The data combination module is used to perform combination processing on the professional skills, the job salary, and the educational background to obtain multiple job combination data, specifically including: Obtain the first professional skills, the first job salary, and the first educational background corresponding to each job recruitment image in the target job; Perform encoding processing on the first professional skills, the first job salary, and the first educational background respectively to obtain the first professional skills encoding data, the first job salary encoding data, and the first educational background encoding data; Perform splicing processing on the first professional skills encoding data, the first job salary encoding data, and the first educational background encoding data to obtain the job combination data corresponding to each job recruitment image in the target job; Integrate the job combination data corresponding to all job recruitment images in the target job to obtain multiple job combination data.

Images