Deep learning intelligent man-post matching method and system based on large language model and multiple Prompts

Abstract

The invention discloses a deep learning intelligent person and post matching method and system based on a large language model and multiple Prompts, and relates to the technical field of big data, a resume data set and a post data set are collected, and a resume embedding vector and a post embedding vector are constructed by taking the resume data set and the post data set as input and taking a resume embedding vector and a post embedding vector as output; taking the matching similarity of the embedded vector as a double-tower model of a prediction target, training the double-tower model, constructing a matching model taking the resume data set and the post data set as input and taking the matching degree as output, training the matching model, collecting man-post matching requests, and providing a primary screening set for the man-post matching requests through the double-tower model; and screening out a final matching result from the preliminary screening set by using a matching model. Through cooperative work of the two models, efficient and accurate man-post matching is realized.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a deep learning intelligent person-job matching method and system based on a large language model and multiple prompts. Background Art

[0002] With the rapid growth of the global economy and the continuous advancement of technology, the human resources market is also developing rapidly, and companies are increasingly demanding efficient and accurate job matching. Traditional recruitment processes often rely on manual screening and interviews, which is not only time-consuming and labor-intensive, but also difficult to process a large number of resumes and job requirements in a short period of time. Therefore, how to use advanced technical means to improve the efficiency and accuracy of job matching has become an important research topic.

[0003] Traditional matching methods have the following limitations: Traditional matching methods mainly rely on keyword matching or simple text similarity calculation. Although these methods are simple and easy to use, they have the following significant limitations: Lack of semantic understanding: Traditional keyword matching methods cannot capture the deep semantic information in the text, resulting in inaccurate matching results.

[0004] Limited information processing capabilities: Resumes and job requirements usually contain a large amount of information. Traditional matching methods have difficulty in effectively processing this complex information and performing efficient matching.

[0005] A Chinese patent with the authorization announcement number CN111144723B discloses a person-job matching recommendation method and system, and storage medium, wherein the method includes the following steps: obtaining the position text information entered by the employer and the resume text information entered by the job seeker, wherein the position text information is the relevant information for the position to be recruited, and the resume text information is the resume of the job seeker; performing text parsing on the position text information and resume text information based on a text parsing algorithm to obtain label information; processing the label information in combination with a position similarity algorithm and a position resume matching algorithm to recommend the most matching resume for the current position. However, this method fails to utilize the technology of multiple prompts, making it difficult to grasp the correlation between the resume and the position; To this end, the present invention proposes a deep learning intelligent person-job matching method and system based on a large language model and multiple prompts. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a deep learning intelligent person-job matching method and system based on a large language model and multiple prompts, and realizes efficient and accurate person-job matching through the collaborative work of the two models.

[0007] To achieve the above objectives, a deep learning intelligent job matching method based on a large language model and multiple prompts is proposed, which includes the following steps: Step 1: Collect resume data set and job data set; Step 2: Build a dual-tower model that takes the resume dataset and the job dataset as input, the resume embedding vector and the job embedding vector as output, and the matching similarity of the embedding vector as the prediction target, and train the dual-tower model; Step 3: Build a matching model with resume dataset and job dataset as input and matching degree as output, and train the matching model; Step 4: Collect job matching requests, and use the dual-tower model to provide a preliminary screening set for the job matching requests. Then use the matching model to filter out the final matching results from the preliminary screening set. The method of collecting resume data set and job data set is: Read the resume data uploaded by all job seekers from the background database of the recruitment site to form a resume data set, and read the job data uploaded by all companies to form a job data set; The method of constructing the dual-tower model that takes the resume dataset and the job dataset as input and takes the resume embedding vector and the job embedding vector as output is as follows: The dual-tower model includes a job tower and a resume tower; The post tower includes an input layer, a slicing layer, an embedding model, a fully connected layer, an attention pooling layer, and an output layer; The resume tower includes an input layer, a slicing layer, an embedding model, a Transformer encoding layer, an attention pooling layer, and an output layer; The input layer of the job tower takes the job description of each job in the job data set as input; The slicing layer and embedding model of the job tower are used to segment and vectorize the job description to obtain the initial embedding vector of the job; The fully connected layer and the attention pooling layer of the post tower are conventional technical means in the art, and the initial post embedding vector is converted into a post embedding vector of length d through the fully connected layer and the attention pooling layer; d is the preset embedding vector length; The input layer of the resume tower takes the resume text of each job seeker in the resume dataset as input; The slicing layer and embedding model of the resume tower are used to segment and vectorize the resume text to obtain the initial embedding vector of the resume; The method for training the dual-tower model is: Manually match positions and resumes and annotate each matching pair; Use the labeled matching pairs to train the twin tower model; The method of constructing a matching model that takes the resume dataset and the job dataset as input and takes the matching degree as output is as follows: The matching model includes a job description processing part, a resume text processing part and a merging and matching part; The processing of the job description processing part includes: Input the job description of each job in the job data set into the segmentation layer to obtain M core contents; Input each core content into the embedding model to obtain the embedding vector corresponding to the core content; Input the embedding vector of each core content into the fully connected layer to obtain the prompt embedding vector of the corresponding core content; Stack the prompt embeddings of each core content to obtain the job prompt embedding; The processing process of the resume text processing part includes: Input the resume text of each job seeker in the resume dataset into the segmentation layer and the embedding model in turn to obtain the resume initial embedding vector output by the embedding model; Then input the resume initial embedding vector into the Transformer encoding layer and the attention pooling layer in sequence to obtain the resume embedding vector; The processing process of the merge matching part includes: Input the job description embedding and resume embedding vectors into the multi-head attention layer to obtain the merged embedding vector; Input the merged embedding vector into the fully connected layer to obtain the intermediate result vector; The intermediate result vector is input into the sigmoid function layer, and the output range is a real value [0, 1], that is, the classification probability value, which serves as the confidence of the classification result and also as the matching degree; Set the classification threshold. Samples with output greater than or equal to the classification threshold are classified as positive, which means the resume and the position match; samples with output less than the classification threshold are classified as negative, which means the resume and the position do not match; The method of training the matching model is: Use the labeled matching pairs to train the matching model; Due to the imbalance of sample categories, Focal Loss is used to calculate the training loss and update the parameter weights of the matching model; The collecting of job matching requests is to collect the resumes of job seekers through the background of the recruitment site and generate job matching requests for the job seekers, and to collect the job descriptions of enterprises and generate resume matching requests for the enterprises; The method of providing a preliminary screening set for a person-job matching request through the dual-tower model is: For job matching requests, the resume embedding vector corresponding to the job seeker's resume and the job embedding vectors of all jobs are obtained through the dual-tower model. The cosine similarity between the resume embedding vector and the embedding vector of each job is calculated, and the jobs with cosine similarity greater than the preset similarity threshold are grouped into a preliminary screening set. For resume matching requests, the dual-tower model is used to obtain the job embedding vector corresponding to the company's job description and the resume embedding vectors of all resumes. The cosine similarity between the job embedding vector and the embedding vector of each resume is then calculated, and the resumes with cosine similarity greater than the preset similarity threshold are grouped into a preliminary screening set. The method of using the matching model to screen out the final matching result from the preliminary screening set is: For job matching requests, the job seeker's resume and each job description in the initial screening set are input into the matching model to obtain the matching degree between the resume and each job in the initial screening set. The job with the highest matching degree is the final matching result. For resume matching requests, the company's job description and each resume in the initial screening set are input into the matching model to obtain the matching degree between the job description and each resume in the initial screening set. The resume with the highest matching degree is the final matching result.

[0008] A deep learning intelligent job matching system based on a large language model and multiple prompts is proposed, including a data set collection module, a dual-tower model training module, a matching model training module, and a matching module; wherein each module is electrically connected; A data set collection module collects resume data sets and job data sets, and sends the resume data sets and job data sets to a dual-tower model training module and a matching model training module; The dual-tower model training module constructs a dual-tower model with resume dataset and job dataset as input, resume embedding vector and job embedding vector as output, and matching similarity of embedding vector as prediction target, trains the dual-tower model, and sends the trained dual-tower model to the matching module; The matching model training module builds a matching model with the resume data set and the job data set as input and the matching degree as output, trains the matching model, and sends the trained matching model to the matching module; The matching module collects job matching requests, provides a preliminary screening set for the job matching requests through the dual-tower model, and then uses the matching model to filter out the final matching results from the preliminary screening set.

[0009] An electronic device is proposed, comprising: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the above-mentioned deep learning intelligent person-job matching method based on a large language model and multiple prompts by calling the computer program stored in the memory.

[0010] A computer-readable storage medium is provided, on which a rewritable computer program is stored; When the computer program runs on a computer device, the computer device executes the above-mentioned deep learning intelligent person-job matching method based on a large language model and multiple prompts.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The present invention collects resume data sets and job data sets, constructs a double-tower model with the resume data sets and job data sets as inputs, resume embedding vectors and job embedding vectors as outputs, and uses the matching similarity of the embedding vectors as a prediction target, and trains the double-tower model; constructs a matching model with the resume data sets and job data sets as inputs and uses the matching degree as output, and trains the matching model; collects person-job matching requests, and provides a preliminary screening set for the person-job matching requests through the double-tower model, and then uses the matching model to screen out the final matching results from the preliminary screening set; by training the double-tower model and the matching model, wherein the double-tower model is used to generate context vectors of job positions and resumes, and stores them in a database to support rapid retrieval of resumes and job positions, various different ways of recall can be flexibly performed, such as "resume vs job", "resume vs resume", "job vs job", The matching model is used to predict the matching of positions and resumes. The fragmentation of resume and position text ensures that the detailed information of positions and resumes can be maintained even when the resume text is long. The position is integrated with the resume in the form of prompt for matching prediction, which helps to enhance the semantic understanding of the model and improve the prediction performance of the model, making it more suitable for complex and diverse job and resume matching scenarios. Through the collaborative work of the two models, efficient and accurate person-job matching is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flowchart of the deep learning intelligent person-job matching method based on a large language model and multiple prompts in Example 1 of the present invention; Figure 2 This is a schematic diagram of a double tower model in Example 1 of the present invention; Figure 3 Schematic diagram of the matching model in Embodiment 1 of the present invention; Figure 4 This is a module connection relationship diagram of the deep learning intelligent job matching system based on a large language model and multiple prompts in Example 2 of the present invention. DETAILED DESCRIPTION

[0013] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] Example 1 like Figure 1 As shown in the figure, the deep learning intelligent job matching method based on a large language model and multiple prompts includes the following steps: Step 1: Collect resume data set and job data set; Step 2: Build a dual-tower model that takes the resume dataset and the job dataset as input, the resume embedding vector and the job embedding vector as output, and the matching similarity of the embedding vector as the prediction target, and train the dual-tower model; Step 3: Build a matching model with resume dataset and job dataset as input and matching degree as output, and train the matching model; Step 4: Collect job matching requests, and use the dual-tower model to provide a preliminary screening set for the job matching requests, and then use the matching model to filter out the final matching results from the preliminary screening set.

[0015] The method of collecting the resume dataset and the job dataset is as follows: Read the resume data uploaded by all job seekers from the background database of the recruitment site to form a resume data set, and read the job data uploaded by all companies to form a job data set; Specifically, the recruitment sites include but are not limited to human resources markets, recruitment websites and other offline or online sites that provide job matching services; The resume data includes but is not limited to: Job applicants’ personal information: name, gender, age, education background, etc.; Work experience: previous positions, years of work experience, job responsibilities, project experience, etc.; Skills: Proficient in programming languages, frameworks, tools, etc. Award certificates: including education and career certificates, awards, etc.; Personal description: statement about your abilities, career plans, etc.; The job data set includes but is not limited to company description, job description, and job requirements.

[0016] It should be noted that the job data and resume data involved in this application have been obtained with full consent and authorization, and the collection, use and processing of relevant information must comply with the relevant laws, regulations and standards of the relevant countries and regions; Furthermore, the method of constructing a dual-tower model with a resume dataset and a job dataset as input and a resume embedding vector and a job embedding vector as output is: like Figure 2 The schematic diagram of the dual-tower model shown in the figure is a model training method that maps data of different modalities into the same vector space. The resume and job description are considered as two different modalities. By ensuring that the embedding vectors of the resume and job description have the same dimension, the model is able to map them into the same vector space. The dual-tower model includes a job tower and a resume tower; The post tower includes an input layer, a slicing layer, an embedding model, a fully connected layer, an attention pooling layer, and an output layer; The resume tower includes an input layer, a slicing layer, an embedding model, a Transformer encoding layer, an attention pooling layer, and an output layer; The input layer of the position tower takes the position description of each position in the position data set as input; The slicing layer and embedding model of the job tower are used to segment and vectorize the job description to obtain the initial embedding vector of the job; The fully connected layer and attention pooling layer of the position tower are conventional technical means in this field. The fully connected layer and attention pooling layer are used to convert the initial position embedding vector into a position embedding vector of length d; d is the preset embedding vector length.

[0017] Specifically, the method of segmenting and vectorizing the job description to obtain the initial embedding vector of the job is as follows: The slice layer of the job tower reads M core contents of each job from the job data set; M is the number of preset core contents; Generally, M is set to 3, and the core contents described in item M correspond to the company description, job description, and job requirements of each position; Input the M core contents into the embedding model of the job tower, and obtain the description embedding sequence of length M output by the embedding model as the initial embedding vector of the job; The input layer of the resume tower takes the resume text of each job seeker in the resume dataset as input; The slicing layer and embedding model of the resume tower are used to segment and vectorize the resume text respectively to obtain the initial embedding vector of the resume.

[0018] Specifically, the resume text is segmented and vectorized to obtain the resume initial embedding vector as follows: The slicing layer of the resume tower divides the resumes of all job seekers in the resume dataset into resume slice sequences according to the preset resume text segmentation rules; Input the resume slice sequence into the resume tower's embedding model to obtain the resume initial embedding vector output by the embedding model; It should be noted that due to the large differences in the length and structure of different resumes, it is not easy to segment according to a fixed structure. Therefore, a set of segmentation rules needs to be preset to perform segmentation, such as relatively simple fixed text length segmentation, or more complex segmentation using natural language processing technology and machine learning models, etc., which needs to be evaluated based on factors such as project resource investment and experimental results. Since the length of each resume varies, the length of the resulting resume slice sequence is also different; In machine learning and natural language processing, embedding models refer to the process of mapping high-dimensional data (such as text, images, and videos) to low-dimensional space. Simply put, an embedding vector is a multidimensional real-valued vector that represents the input data as a point in a continuous numerical space. The mainstream embedding vector construction methods include but are not limited to the Bert model, CimCSE model, etc. The Transformer encoding layer and attention pooling layer of the resume tower are conventional technical means in this field. Transformer Encoder is used to process resume vector sequences of non-fixed length, and attention pooling is used to compress the variable-length sequence into an embedding vector of fixed length d.

[0019] Furthermore, the method of training the dual-tower model is: Manually match positions and resumes and annotate each matching pair; It is understandable that the method of marking each matching pair may be to manually select the matching pairs of positions and resumes that are more consistent and mark them as 1, and mark the remaining matching pairs of positions and resumes as 0; Use the labeled matching pairs to train the twin tower model; Specifically, a distance metric (such as dot product, cosine similarity, Euclidean distance) is used to calculate the matching similarity between the resume embedding vector and the job embedding vector; Considering that the resumes that are screened and finally receive offers only account for a small part of all the resumes received, it is reflected in the problem of sample imbalance in the data set. The following mixed loss function is used to calculate the model loss and update the model weight accordingly: ; in, is the model weight, is the contrast loss, is Focal Loss, and All are preset weight coefficients; Among them, the contrast loss ensures that the distance between matching sample pairs (positive class) is smaller, and the distance between mismatching sample pairs (negative class) is larger. The formula is as follows: ; Where N is the number of all matching pairs, are the resume embedding vector and job embedding vector in the matching pair respectively, To match the labeled label value, margin is a set boundary value that controls the minimum distance of negative samples; max is the maximum value function, which is used to take the maximum value among several values; d is the distance metric function, which can be the Euler distance; Among them, Focal Loss solves the problem of imbalanced resume matching categories by reducing the loss contribution of simple samples and increasing the weight of difficult-to-classify samples. The formula is as follows: ; in, is the weight of the category, usually the inverse of the number of positive or negative samples (labeled 1 or 0); It is an adjustment parameter that controls the focus of Focal Loss on difficult samples, usually set to 2; is the matching probability output by the twin tower model, is the label value actually annotated for each matching pair; The above-mentioned Two-Tower Model is a common architecture, which is particularly suitable for processing two different types of inputs (such as resumes and job requirements, which are both texts but have very different content structures). Each tower independently encodes one type of input and generates its own embedding vector.

[0020] This structure has the following advantages: - Independent Optimization: Each tower can be optimized for a specific type of input, thereby better capturing the unique properties of the respective input.

[0021] - Efficient computing: Once the embedding vectors are generated, they can be matched through simple vector operations (such as cosine similarity), greatly improving the response speed of online services.

[0022] - Scalability: Embedding vectors can be pre-computed and stored offline, allowing the system to handle large-scale datasets.

[0023] Furthermore, multi-prompt technology refers to the use of multiple prompts to enhance the model's semantic understanding and context awareness when processing text. By integrating job requirements with resumes in the form of prompts for matching predictions, the correlation between the two can be better captured, thereby improving the accuracy and robustness of matching.

[0024] Therefore, the method of constructing a matching model that takes the resume dataset and the job dataset as input and takes the matching degree as output is as follows: like Figure 3 The matching model schematic diagram shown in the figure includes a job description processing part, a resume text processing part and a merging matching part; The processing of the job description processing part includes: Input the job description of each job in the job data set into the segmentation layer to obtain M core contents; In this embodiment, M is 3, and the corresponding core contents are company description, job description, and job requirements; Input each core content into the embedding model to obtain the embedding vector corresponding to the core content; In this embodiment, Figure 3 Embedding company description, job description and job requirements in the content; It can be understood that the functions of the slice layer and embedding model in the job description processing part are consistent with those of the slice layer and embedding model in the job tower in the dual tower model; Input the embedding vector of each core content into the fully connected layer to obtain the prompt embedding vector of the corresponding core content; In this embodiment, Figure 3 Embed company description prompts, job description prompts, and job requirements prompts in the Stack the prompt embeddings of each core content to obtain the job prompt embedding.

[0025] Furthermore, the processing of the resume text processing part includes: Input the resume text of each job seeker in the resume dataset into the segmentation layer and the embedding model in turn to obtain the resume initial embedding vector output by the embedding model; Then input the resume initial embedding vector into the Transformer encoding layer and the attention pooling layer in sequence to obtain the resume embedding vector; It can be understood that the segmentation layer, embedding model, Transformer encoding layer and attention pooling layer of the resume text processing part have the same functions as the segmentation layer, embedding model, Transformer encoding layer and attention pooling layer of the resume tower in the dual-tower model.

[0026] Furthermore, the processing of merging the matching parts includes: Input the job description embedding and resume embedding vectors into the multi-head attention layer to obtain the merged embedding vector; Input the merged embedding vector into the fully connected layer to obtain the intermediate result vector; The intermediate result vector is input into the sigmoid function layer, and the output range is a real value [0, 1], that is, the classification probability value, which serves as the confidence of the classification result and also as the matching degree; Set the classification threshold. Samples with output greater than or equal to the classification threshold are classified as positive, that is, the resume and the position match; samples with output less than the classification threshold are classified as negative, that is, the resume and the position do not match.

[0027] Furthermore, the matching model is trained in the following manner: Use the labeled matching pairs to train the matching model; Due to the imbalance of sample categories, Focal Loss is used to calculate the training loss and update the parameter weights of the matching model; The collecting of job matching requests is to collect the resumes of job seekers through the background of the recruitment site and generate job matching requests for the job seekers, as well as to collect the job descriptions of enterprises and generate resume matching requests for the enterprises.

[0028] The method of providing a preliminary screening set for job matching requests through the dual-tower model is: For job matching requests, the resume embedding vector corresponding to the job seeker's resume and the job embedding vectors of all jobs are obtained through the dual-tower model. The cosine similarity between the resume embedding vector and the embedding vector of each job is calculated, and the jobs with cosine similarity greater than the preset similarity threshold are grouped into a preliminary screening set. For resume matching requests, the dual-tower model is used to obtain the job embedding vector corresponding to the company's job description and the resume embedding vectors of all resumes. The cosine similarity between the job embedding vector and the embedding vector of each resume is then calculated, and the resumes with cosine similarity greater than the preset similarity threshold are formed into an initial screening set.

[0029] In a further preferred embodiment, for the twin tower model, various retrieval methods can be supported by calculating cosine similarity, including: Use job search resumes to list all potential job candidates; Use resume to search for positions and find all available positions; Use the job search function to view other similar jobs; Use Resume Search to view similar candidates.

[0030] Furthermore, the method of using the matching model to screen out the final matching result from the preliminary screening set is: For job matching requests, the job seeker's resume and each job description in the initial screening set are input into the matching model to obtain the matching degree between the resume and each job in the initial screening set. The job with the highest matching degree is the final matching result. For resume matching requests, the company's job description and each resume in the initial screening set are input into the matching model to obtain the matching degree between the job description and each resume in the initial screening set. The resume with the highest matching degree is the final matching result.

[0031] The following are some practical application scenarios of the present invention, showing how to provide convenient and fast assistance to job seekers and enterprises through the dual tower model and matching model: Scenario 1: Online recruitment platform - recommended positions Application scenario: Match job seekers’ resumes with job descriptions posted by companies and recommend jobs to job seekers; Specific implementation: The platform collects a large amount of resume and job description data, uses the dual-tower model to generate context vectors for resumes and job descriptions, and stores them in the database.

[0032] When a job seeker uploads a resume, the resume embedding vector is obtained through the resume tower of the dual-tower model. The platform can use the resume embedding vector to quickly search and find preliminary matching job descriptions.

[0033] The matching model is further used to make matching predictions for the retrieved job descriptions and resumes, and the matching prediction results and matching confidence levels are given.

[0034] Filter out job descriptions that do not match or do not meet the confidence threshold, and recommend suitable jobs to job seekers.

[0035] Scenario 2: Online recruitment platform - Recommended resume Application scenario: Matching job descriptions posted by companies with job seekers’ resumes and recommending resumes to companies; Specific implementation: The platform collects a large amount of resume and job description data, uses the dual-tower model to generate context vectors for resumes and job descriptions, and stores them in the database.

[0036] When an enterprise posts a job, it obtains the job embedding vector through the job tower of the dual-tower model. The platform uses the job embedding vector to quickly search and find preliminary matching resumes.

[0037] The matching model is further used to make matching predictions for the retrieved resumes and job descriptions, and the matching prediction results and matching confidence are given.

[0038] Filter out resumes that do not match or do not meet the confidence threshold, and recommend suitable resumes to companies.

[0039] Scenario 3: Human Resources Management System Application scenario: The human resource management system within an enterprise can use the method of the present invention to optimize internal talent allocation; Specific implementation: The company collects employees' resumes and internal job descriptions, uses the twin-tower model to generate context vectors for the employee resumes, and stores them in the database.

[0040] When there is a new job requirement, the job embedding vector is obtained through the job tower of the dual-tower model. The platform uses the job embedding vector to quickly search and find preliminary matching employee resumes.

[0041] The matching model is further used to make matching predictions and output the prediction results and prediction confidence.

[0042] The system displays matching results and confidence levels to help the HR department make decisions.

[0043] Similarly, employee resumes can be used to match vacancies in internal positions, thereby facilitating the transfer of internal talent and quickly finding matching internal positions for employees who become idle after the project is completed.

[0044] Scenario 4: Career consultation and planning Application scenario: Career consulting agencies can use the method of the present invention to help job seekers with career planning and job recommendations; Specific implementation: The consulting agency collects job descriptions on the job market, uses the dual-tower model to generate job description context vectors, and stores them in the database.

[0045] When job seekers seek career advice, the system can quickly search through the job seekers' resumes to find preliminary matching positions.

[0046] The matching model is further used to make matching predictions and output the prediction results and prediction confidence.

[0047] The system displays matching results and confidence levels to help job seekers better understand their career development direction.

[0048] Scenario 5: Education and training industry Application scenarios: The education and training industry can use the method of the present invention to recommend career paths and courses suitable for students; Specific implementation: Educational institutions collect students’ resumes and job descriptions of market demand, use the dual-tower model to generate context vectors, and store them in the database.

[0049] When trainees complete their training, the system can quickly search for preliminary matching positions for them.

[0050] The matching model is further used to make matching predictions and output the prediction results and prediction confidence.

[0051] The system displays the matching results and confidence levels, and recommends the career path and related courses that best suit the students.

[0052] Example 2 like Figure 4 As shown, the deep learning intelligent job matching system based on a large language model and multiple prompts includes a data set collection module, a dual-tower model training module, a matching model training module, and a matching module; wherein each module is electrically connected; A data set collection module collects resume data sets and job data sets, and sends the resume data sets and job data sets to a dual-tower model training module and a matching model training module; The dual-tower model training module constructs a dual-tower model with resume dataset and job dataset as input, resume embedding vector and job embedding vector as output, and matching similarity of embedding vector as prediction target, trains the dual-tower model, and sends the trained dual-tower model to the matching module; The matching model training module builds a matching model with the resume data set and the job data set as input and the matching degree as output, trains the matching model, and sends the trained matching model to the matching module; The matching module collects job matching requests, provides a preliminary screening set for the job matching requests through the dual-tower model, and then uses the matching model to filter out the final matching results from the preliminary screening set.

[0053] Example 3 According to another aspect of the present application, an electronic device is provided. The electronic device may include one or more processors and one or more memories. The memories store computer-readable codes, and when the computer-readable codes are executed by one or more processors, the deep learning intelligent person-job matching method based on a large language model and multiple prompts as described above may be executed.

[0054] The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, an input / output component, a hard disk, etc. A storage device in the electronic device, such as a ROM or a hard disk, may store the deep learning intelligent person-job matching method based on a large language model and multiple prompts provided in the present application.

[0055] Furthermore, the electronic device may also include a user interface. Of course, this architecture is only exemplary, and when implementing different devices, one or more components in the electronic device may be omitted according to actual needs.

[0056] Example 4 A computer-readable storage medium according to an embodiment of the present application. Computer-readable instructions are stored on the computer-readable storage medium. When the computer-readable instructions are executed by the processor, the deep learning intelligent person-job matching method based on a large language model and multiple prompts according to the embodiment of the present application described with reference to the above figures can be executed. The computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0057] In addition, according to the implementation of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, the non-transitory machine-readable storage medium stores machine-readable instructions, the machine-readable instructions can be run by a processor to execute instructions corresponding to the method steps provided by the present application, and when the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0058] The methods, apparatuses, and devices of the present application may be implemented in many ways. For example, the methods, apparatuses, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0059] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0060] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0061] The above preset parameters or preset thresholds are all set by technicians in this field according to actual conditions or obtained through large-scale data simulation.

[0062] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A deep learning intelligent job matching method based on a large language model and multiple prompts, characterized by: The following steps are involved: Step 1: Collect resume data set and job data set; Step 2: Build a dual-tower model that takes the resume dataset and the job dataset as input, the resume embedding vector and the job embedding vector as output, and the matching similarity of the embedding vector as the prediction target, and train the dual-tower model; Step 3: Build a matching model with resume dataset and job dataset as input and matching degree as output, and train the matching model; Step 4: Collect job matching requests, and use the dual-tower model to provide a preliminary screening set for the job matching requests, and then use the matching model to filter out the final matching results from the preliminary screening set.

2. According to claim 1, the deep learning intelligent person-job matching method based on a large language model and multiple prompts is characterized in that: The method of collecting resume data set and job data set is: The resume data uploaded by all job seekers are read from the background database of the recruitment site to form a resume data set, and the job data uploaded by all companies are read to form a job data set.

3. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 2 is characterized in that: The method of constructing the dual-tower model that takes the resume dataset and the job dataset as input and takes the resume embedding vector and the job embedding vector as output is as follows: The dual-tower model includes a job tower and a resume tower; The post tower includes an input layer, a slicing layer, an embedding model, a fully connected layer, an attention pooling layer, and an output layer; The resume tower includes an input layer, a slicing layer, an embedding model, a Transformer encoding layer, an attention pooling layer, and an output layer; The input layer of the job tower takes the job description of each job in the job data set as input; The slicing layer and embedding model of the job tower are used to segment and vectorize the job description to obtain the initial embedding vector of the job; The input layer of the resume tower takes the resume text of each job seeker in the resume dataset as input; The slicing layer and embedding model of the resume tower are used to segment and vectorize the resume text respectively to obtain the initial embedding vector of the resume.

4. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 3 is characterized in that: The method for training the dual-tower model is: Manually match positions and resumes and annotate each matching pair; Use the labeled matching pairs to train the twin tower model.

5. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 4 is characterized in that: The method of constructing a matching model that takes the resume dataset and the job dataset as input and takes the matching degree as output is as follows: The matching model includes a job description processing part, a resume text processing part and a merging and matching part.

6. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 5 is characterized in that: The processing of the job description processing part includes: Input the job description of each job in the job data set into the segmentation layer to obtain M core contents; Input each core content into the embedding model to obtain the embedding vector corresponding to the core content; Input the embedding vector of each core content into the fully connected layer to obtain the prompt embedding vector of the corresponding core content; Stack the prompt embeddings of each core content to obtain the job prompt embedding.

7. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 6 is characterized in that: The processing process of the resume text processing part includes: Input the resume text of each job seeker in the resume dataset into the segmentation layer and the embedding model in turn to obtain the resume initial embedding vector output by the embedding model; Then input the resume initial embedding vector into the Transformer encoding layer and the attention pooling layer in turn to obtain the resume embedding vector.

8. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 7 is characterized in that: The processing process of the merge matching part includes: Input the job description embedding and resume embedding vectors into the multi-head attention layer to obtain the merged embedding vector; Input the merged embedding vector into the fully connected layer to obtain the intermediate result vector; The intermediate result vector is input into the sigmoid function layer, and the output range is a real value [0, 1], that is, the classification probability value, which serves as the confidence of the classification result and also as the matching degree; Set the classification threshold. Samples with output greater than or equal to the classification threshold are classified as positive, that is, the resume and the position match; samples with output less than the classification threshold are classified as negative, that is, the resume and the position do not match.

9. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 8 is characterized in that: The collecting of job matching requests is to collect the resumes of job seekers through the background of the recruitment site and generate job matching requests for the job seekers, as well as to collect the job descriptions of enterprises and generate resume matching requests for the enterprises.

10. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 9 is characterized in that: The method of providing a preliminary screening set for a person-job matching request through the dual-tower model is: For job matching requests, the resume embedding vector corresponding to the job seeker's resume and the job embedding vectors of all jobs are obtained through the dual-tower model. The cosine similarity between the resume embedding vector and the embedding vector of each job is calculated, and the jobs with cosine similarity greater than the preset similarity threshold are grouped into a preliminary screening set. For resume matching requests, the dual-tower model is used to obtain the job embedding vector corresponding to the company's job description and the resume embedding vectors of all resumes. The cosine similarity between the job embedding vector and the embedding vector of each resume is then calculated, and the resumes with cosine similarity greater than the preset similarity threshold are formed into an initial screening set.

11. The deep learning intelligent person-job matching method based on a large language model and multiple prompts according to claim 10 is characterized in that: The method of using the matching model to screen out the final matching result from the preliminary screening set is: For job matching requests, the job seeker's resume and each job description in the initial screening set are input into the matching model to obtain the matching degree between the resume and each job in the initial screening set. The job with the highest matching degree is the final matching result. For resume matching requests, the company's job description and each resume in the initial screening set are input into the matching model to obtain the matching degree between the job description and each resume in the initial screening set. The resume with the highest matching degree is the final matching result.

12. A deep learning intelligent job matching system based on a large language model and multiple prompts, which is used to implement the deep learning intelligent job matching method based on a large language model and multiple prompts as described in any one of claims 1-11, characterized in that: It includes a data set collection module, a dual-tower model training module, a matching model training module and a matching module; wherein each module is electrically connected; A data set collection module collects resume data sets and job data sets, and sends the resume data sets and job data sets to a dual-tower model training module and a matching model training module; The dual-tower model training module constructs a dual-tower model with resume dataset and job dataset as input, resume embedding vector and job embedding vector as output, and matching similarity of embedding vector as prediction target, trains the dual-tower model, and sends the trained dual-tower model to the matching module; The matching model training module builds a matching model with the resume data set and the job data set as input and the matching degree as output, trains the matching model, and sends the trained matching model to the matching module; The matching module collects job matching requests, provides a preliminary screening set for the job matching requests through the dual-tower model, and then uses the matching model to filter out the final matching results from the preliminary screening set.

13. An electronic device, characterized in that: include: A processor and a memory, wherein: The memory stores a computer program that can be called by the processor; The processor executes the deep learning intelligent person-job matching method based on a large language model and multiple prompts as described in any one of claims 1-11 in the background by calling the computer program stored in the memory.

14. A computer-readable storage medium, characterized in that: A rewritable computer program is stored thereon; When the computer program runs on a computer device, the computer device executes the deep learning intelligent person-job matching method based on a large language model and multiple prompts as described in any one of claims 1 to 11 in the background.

Images