Method and system for accurate matching and intelligent recommendation of people and sentry

Abstract

The invention discloses an intelligent recommendation method and system for accurate matching of job hunters and recruitment posts, and relates to the technical field of intelligent recommendation, and the method comprises the following steps: collecting multi-source data of the job hunters and the recruitment posts in real time, carrying out data cleaning, classification and standardization processing, and outputting unstructured text data; extracting emotional tendency of the job seeker through an emotional analysis method, and constructing an emotional portrait vector of the job seeker; by means of deep transfer learning, knowledge is transferred from a source domain to a target domain, the matching precision of posts and job seekers is optimized, and post recommendation is optimized in combination with the transferred knowledge; and generating a personalized post recommendation list in combination with the emotional demand of the job seeker and the cultural adaptation degree of the post, and preferentially recommending the post with the highest matching degree with the demand of the job seeker. According to the method, a deep transfer learning framework is utilized, knowledge of a source domain is transferred to a target domain, and the matching precision of posts and job seekers is optimized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent recommendation technology, and in particular to a method and system for intelligent recommendation of precise matching of people and jobs. Background Art

[0002] In recent years, with the development of artificial intelligence technology, especially the progress of deep learning, natural language processing (NLP) and transfer learning, intelligent recommendation systems in the recruitment field have gradually emerged. Traditional job matching methods mainly rely on manual screening, which is inefficient and greatly affected by subjective factors. Existing technologies can provide preliminary job recommendations by analyzing basic data of job seekers and job information. However, these methods often ignore the matching of job seekers' emotional needs and job culture, and cannot dynamically adapt to the complex emotional and behavioral patterns between job seekers and jobs. With the application of big data, job recommendation methods based on deep learning have been widely used, but most systems only match through simple similarity calculations, lack high-precision recommendations and personalized adjustments, and are difficult to cope with the challenges brought by data dimensions and information asymmetry.

[0003] Existing job matching technologies generally fail to fully consider the individual needs, emotional tendencies, and job culture suitability of job seekers, resulting in inaccurate and insensitive recommended matches. Therefore, how to combine the multi-dimensional characteristics of job seekers, especially emotional needs, job culture suitability, and behavioral patterns, to make accurate job recommendations has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for intelligent recommendation of accurate matching between people and jobs to solve the problem of how to accurately match job seekers and jobs, especially how to optimize job recommendations based on multi-source data and emotional needs.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for intelligent recommendation of accurate matching of people and positions, which includes collecting multi-source data of job seekers and recruitment positions in real time, and performing data cleaning, classification and standardization processing to output unstructured text data;

[0008] Extract the emotional tendencies of job seekers through sentiment analysis methods and construct the emotional portrait vector of job seekers;

[0009] Using deep transfer learning, we transfer knowledge from the source domain to the target domain, optimize the matching accuracy between positions and job seekers, and optimize job recommendations based on the transferred knowledge.

[0010] Through the multi-dimensional feature fusion algorithm, the unstructured text data of job seekers is fused with the emotional portrait vector of job seekers, the similarity between job seekers and positions is calculated, and the preliminary matching score is obtained;

[0011] Combining the emotional needs of job seekers and the cultural fit of positions, a personalized job recommendation list is generated, with priority given to positions that best match the job seekers' needs.

[0012] As a preferred solution of the method for intelligent recommendation of accurate matching of people and jobs described in the present invention, the multi-source data of job seekers and recruitment positions includes personal characteristic data, job information data, emotional demand data, job culture suitability data and job recommendation feedback data.

[0013] As a preferred solution of the method for intelligent recommendation of accurate matching of people and positions described in the present invention, the specific steps of data cleaning, classification and standardization processing and outputting unstructured text data are as follows:

[0014] Check for missing values, handle duplicate data, and handle outliers on multi-source data of job seekers and job openings;

[0015] Perform feature classification and text classification on the personal feature data and job information data in the multi-source data of job seekers and job openings, and convert them into standardized labels;

[0016] Standardize the numerical data in the processed multi-source data of job seekers and job positions, and vectorize the text data using a pre-trained language model;

[0017] Output the cleaned, classified and standardized data into unstructured text format.

[0018] As a preferred solution of the method for intelligent recommendation of accurate matching of people and jobs described in the present invention, wherein: the emotional tendency of job seekers is extracted by the sentiment analysis method to construct the emotional portrait vector of the job seeker, and the specific steps are as follows:

[0019] Segment the applicants’ unstructured text data and remove stop words, and use NLP tools to perform part-of-speech tagging;

[0020] Fine-tune the unstructured text data of job seekers based on the BERT pre-trained model, classify the text into sentiment labels through sentiment classification, and predict the sentiment intensity value of each text;

[0021] The emotion intensity value is converted into a numerical vector and concatenated with the job applicant's personal characteristic data to form an emotion portrait vector.

[0022] As a preferred solution of the method for intelligent recommendation of accurate matching of people and jobs described in the present invention, wherein: the deep transfer learning is used to transfer knowledge from the source field to the target field, the matching accuracy between positions and job seekers is optimized, and the job recommendation is optimized in combination with the transferred knowledge. The specific steps are as follows:

[0023] Extract features from source domain data through a multi-layer neural network to generate a deep feature vector representation of the source domain;

[0024] Transfer the deep feature vector representation to the target domain and build an adaptive transfer learning framework to transfer knowledge layer by layer by fine-tuning parameters;

[0025] In each layer, nonlinear mapping functions are used to fuse high-order features, thereby maintaining multi-scale information between the source and target domains;

[0026] Adopting adaptive weighting method to align the multi-layer features of source domain and target domain, during the feature alignment process, the loss function is used for optimization;

[0027] The multi-dimensional features of job seekers and positions are integrated through Gaussian kernel function to build a nonlinear fusion mechanism;

[0028] A Q-learning network is constructed based on reinforcement learning to optimize job recommendations based on personal characteristic data, job information data, emotional demand data, job culture fit data, and job recommendation feedback data.

[0029] As a preferred solution of the method for intelligent recommendation of accurate matching of people and positions described in the present invention, wherein: the unstructured text data of job seekers is fused with the emotional portrait of job seekers through a multi-dimensional feature fusion algorithm, the similarity between job seekers and positions is calculated, and a preliminary matching score is obtained. The specific steps are as follows:

[0030] The BERT model is used to extract features from the job seeker’s unstructured text data and generate a job seeker text feature vector.

[0031] Convert the predicted sentiment intensity value of each text in the sentiment portrait into a sentiment numerical vector;

[0032] Extract the job seeker's behavior data from personal feature data and job recommendation feedback data, and process it based on the LSTM time series model to capture the job seeker's behavior change trend and obtain the job seeker's behavior pattern;

[0033] Use multi-layer perceptron and deep neural network to fuse the text feature vector, sentiment value vector and behavior pattern of job applicants;

[0034] In the process of feature fusion and alignment, the weights of different features are dynamically adjusted to align and standardize the features of different data sources;

[0035] The features of different modalities are integrated layer by layer through the fully connected layer. Each layer of the neural network further optimizes the relationship between different features through weighting, activation function and nonlinear mapping to generate a comprehensive feature vector of job seekers.

[0036] Generative adversarial network matching optimization is adopted. The generator accepts the comprehensive feature vector of the job seeker, the job information data and the job culture suitability data, and generates a matching feature vector.

[0037] Based on historical job recommendation feedback data and manually labeled job matching data, deep learning methods are used for training to output true matching feature vectors;

[0038] The discriminator compares the generated matching feature vector with the real matching feature vector to evaluate the matching degree between the job seeker and the position, and outputs the matching score.

[0039] In the process of generative adversarial training, the generator and the discriminator continuously optimize the generation process of matching feature vectors through mutual competition;

[0040] After generating adversarial training to obtain the matching feature vectors of job seekers and positions, the cosine similarity is used to calculate the matching degree between job seekers and positions;

[0041] The match between the job seeker and the position is used as the preliminary match score.

[0042] As a preferred solution of the method for intelligent recommendation of accurate matching of people and jobs of the present invention, wherein: the method combines the emotional needs of job seekers and the cultural adaptability of jobs to generate a personalized job recommendation list, and gives priority to recommending jobs that have the highest matching degree with job seekers' needs. The specific steps are as follows:

[0043] Standardize and align the emotional needs data of job seekers with the job culture fit data;

[0044] Based on the matching scores, use the Top-K recommendation strategy to select the top K positions with the highest matching scores as candidate recommended positions;

[0045] Combining multi-source data of job seekers and recruitment positions, the preliminary recommendation list is re-sorted, and the positions that are most matched with the job seekers' emotional needs data and the cultural compatibility data of the positions are displayed first, and a personalized job recommendation list is output.

[0046] In a second aspect, the present invention provides a system for intelligent recommendation of precise matching of people and positions, including a data collection module, an emotion profiling module, a knowledge transfer module, a feature fusion module and a position recommendation module;

[0047] The data collection module is used to collect multi-source data of job seekers and recruitment positions in real time, and perform data cleaning, classification and standardization processing to output unstructured text data;

[0048] The emotional portrait module is used to extract the emotional tendencies of job seekers through emotional analysis methods and construct an emotional portrait vector of the job seeker;

[0049] The knowledge transfer module is used to transfer knowledge from the source domain to the target domain using deep transfer learning, optimize the matching accuracy between positions and job seekers, and optimize position recommendations based on the transferred knowledge;

[0050] The feature fusion module is used to fuse the unstructured text data of the job seeker with the emotional portrait vector of the job seeker through a multi-dimensional feature fusion algorithm, calculate the similarity between the job seeker and the position, and obtain a preliminary matching score;

[0051] The job recommendation module is used to generate a personalized job recommendation list based on the job seeker's emotional needs and the cultural adaptability of the job, and give priority to recommending the jobs that have the highest matching degree with the job seeker's needs.

[0052] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for intelligent recommendation of precise matching of people and jobs as described in the first aspect of the present invention is implemented.

[0053] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for intelligent recommendation of precise matching of people and jobs as described in the first aspect of the present invention.

[0054] The beneficial effects of the present invention are as follows: the present invention pre-processes multi-source data by adopting data cleaning, classification and standardization processing to ensure the quality and consistency of the data; extracts the emotional tendencies of job seekers through sentiment analysis, and constructs emotional portraits in combination with their personal characteristics, which can accurately capture the emotional needs of job seekers; uses the deep transfer learning framework to transfer the knowledge of the source field to the target field, optimizes the matching accuracy between positions and job seekers, and at the same time improves the effect of feature fusion through adaptive weighted feature alignment. Furthermore, through a multi-dimensional feature fusion algorithm, multiple features such as text features, emotional portraits and behavioral patterns of job seekers are integrated to calculate the similarity between job seekers and positions; the generative adversarial network continuously optimizes the matching feature vector, improves the matching accuracy and generates a personalized job recommendation list, ensuring that the recommendation results are more in line with the emotional needs of job seekers and the cultural adaptability of positions, effectively improving the matching accuracy and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0056] Figure 1 This is a flow chart of the method for intelligent recommendation of precise matching of people and positions in Example 1.

[0057] Figure 2 This is a module diagram of the system for intelligent recommendation of precise matching of people and jobs in Example 1. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0060] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0061] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a method for intelligent recommendation of precise matching of people and positions, comprising the following steps:

[0062] S1. Collect multi-source data of job seekers and job openings in real time, clean, classify and standardize the data, and output unstructured text data.

[0063] Furthermore, the multi-source data of job seekers and job openings include personal characteristic data, job information data, emotional demand data, job culture fit data, and job recommendation feedback data;

[0064] Check for missing values, handle duplicate data, and handle outliers on multi-source data of job seekers and job openings;

[0065] Among them, missing values ​​can be filled by reasonable inference (such as using the mean, mode or industry standard). If it cannot be filled, the data will be marked and ignored.

[0066] Specifically, for duplicate data processing, it is necessary to remove duplicate records based on the job seeker's unique identifier (such as mobile phone number) to retain the most complete data; for job information, the job ID and description text are used to remove duplicate jobs. For outlier processing, it is necessary to check the outliers in personal characteristic data (such as age, work experience) and job data (such as salary range). If the data exceeds the reasonable range, the mean or median is used to fill it, or the data that does not meet the specifications is deleted.

[0067] Perform feature classification and text classification on the personal feature data and job information data in the multi-source data of job seekers and job openings, and convert them into standardized labels for job recommendation and matching;

[0068] Specifically, the applicants’ education, experience and other information in the multi-source data of job seekers and recruitment positions are classified (such as education “bachelor’s degree”, “master’s degree”), and the job information is classified by industry (such as “software development”, “data analysis”) and standardized to facilitate subsequent matching. NLP technology is used to extract keywords and classify text from job descriptions and job seekers’ self-introductions, and convert them into standardized tags (such as “Python”, “product manager”) for job recommendation and matching.

[0069] Standardize the numerical data in the processed multi-source data of job seekers and recruitment positions (such as the job seekers' age, work experience, salary expectations, etc.) to ensure the unified dimension of different features and improve the accuracy of subsequent model processing.

[0070] At the same time, use pre-trained language models (such as BERT or Word2Vec) to vectorize the text data, improve the consistency and comparability of the data, and provide support for subsequent analysis and matching;

[0071] Output the data that has been cleaned, classified, and standardized into an unstructured text format, including core data such as the personal characteristics, emotional needs of job seekers, and job information, providing high-quality input for sentiment analysis and personalized job recommendations.

[0072] S2. Extract the emotional tendency of job seekers through sentiment analysis methods and construct an emotional portrait of job seekers.

[0073] Furthermore, perform word segmentation and stop word removal on the unstructured text data of job seekers, and use NLP tools such as jieba for part-of-speech tagging (such as nouns, verbs, etc.);

[0074] Among them, the purpose of Chinese word segmentation is to split continuous text into words or phrases. Stop word removal means removing common but meaningless words (such as "de", "shi", "zai", etc.) and filtering using a predefined stop word list.

[0075] Fine-tune the unstructured text data of job seekers based on the BERT pre-trained model, divide the sentiment labels of the text through sentiment classification, and predict the sentiment intensity value of each text segment;

[0076] It should be noted that BERT is a pre-trained language model based on Transformer, which can capture context information and is widely used in tasks such as text classification and sentiment analysis. BERT considers the left and right contexts of a sentence simultaneously through a bidirectional encoder, and can more accurately understand the sentiment tendency in the text.

[0077] Specifically, the fine-tuning process is as follows: Use the job seeker text dataset with sentiment labels to fine-tune the BERT model. A commonly used fine-tuning method is to optimize through the cross-entropy loss function;

[0078] The expression of the loss function L is:

[0079]

[0080] Among them, N represents the total number of samples, c represents the index variable of the sample, y c represents the sentiment label of the c-th sample (such as positive, negative, and neutral), and k(y c ) represents the prediction probability of the BERT pre-trained model for the c-th sample belonging to a certain sentiment label;

[0081] After fine-tuning, BERT will perform sentiment classification on each job applicant's text and output a sentiment label. At the same time, it will predict the sentiment intensity value (between 0 and 1) for each text, indicating the intensity of the sentiment.

[0082] The emotional intensity value is converted into a numerical vector and combined with the job seeker's personal characteristic data to form an emotional portrait vector;

[0083] The specific splicing process is as follows: vectorize the emotional labels (for example, "positive" is converted to 1) and emotional intensity values ​​(such as 0.85) obtained from sentiment analysis; standardize or encode the background characteristic data of job applicants (such as education, work experience, etc.); splice the emotional labels, emotional intensity values ​​and the background characteristics of job applicants into a large vector, namely the emotional portrait vector; the spliced ​​large vector will represent the combination of the job applicant's emotional tendency and personal background characteristics.

[0084] It should be noted that the final emotional portrait vector is composed of two parts: one is the emotional tendency and intensity, and the other is the personal background characteristics of the job seeker. At the same time, the emotional portrait vector of the job seeker not only reflects the emotional needs of the job seeker (such as the positive or negative emotional tendency), but also includes the personal background information of the job seeker, providing multi-dimensional support for subsequent job matching and recommendation.

[0085] S3. Use deep transfer learning to transfer knowledge from the source domain to the target domain, optimize the matching accuracy between positions and job seekers, and optimize job recommendations based on the transferred knowledge.

[0086] Furthermore, the data in the source domain is subjected to feature extraction through a multi-layer neural network to generate a deep feature vector representation of the source domain;

[0087] Transfer the deep feature vector representation to the target domain and build an adaptive transfer learning framework to transfer knowledge layer by layer by fine-tuning parameters;

[0088] The source domain includes job seekers and job information from historical data sets, existing recruitment platforms or other related fields. These data have been fully annotated and organized, covering the following aspects: job seekers’ personal characteristics (such as education, work experience, skills, etc.), job information (such as job requirements, job responsibilities, salary range, etc.), emotional needs (such as emotional inclinations towards the work environment, emotional needs for job matching, etc.), job culture fit data (such as company culture, team atmosphere, etc.) and historical job recommendation feedback data. The data in the source domain is used to analyze factors such as job seekers’ personal characteristics, job requirements, emotional needs, etc., and generate preliminary job matching rules based on this, which can then be applied to the target domain.

[0089] The target domain refers to the actual job seeker and job data in a specific application scenario, which usually includes new job seeker data sets and job information. It should be noted that the data in the target domain may be scarce or less dependent on annotations and historical data, so it is necessary to use the knowledge transfer of the source domain for optimization. The data in the target domain includes the job seeker's personal information, the job seeker's emotional tendencies (based on the emotional needs of the target position), and various types of job information. Through transfer learning, combined with the knowledge of the source domain, accurate job matching and recommendation optimization can be performed.

[0090] In each layer, nonlinear mapping functions are used to fuse high-order features, thereby maintaining multi-scale information between the source and target domains;

[0091] Specifically, at each layer of multi-level transfer learning, the features of the source domain and the target domain are processed through nonlinear mapping functions. The feature vectors of the source domain and the feature vectors of the target domain at each layer are processed through nonlinear mapping functions (such as the self-attention mechanism of convolutional neural networks or transformer models) to generate high-order feature vectors of the source domain. These high-order feature vectors are compared with the features of the target domain and aligned through nonlinear functions to ensure that the features of the source domain and the target domain are more consistent. The weighting function is used to adjust the mapping importance of the features of each layer to ensure that multi-scale information can be effectively transmitted and fused at each layer. Through this alternating calculation and feature fusion, the fused feature vectors of each layer are finally used for subsequent deep network training, so that the multi-scale information of the source domain and the target domain can be strengthened and the matching accuracy can be improved.

[0092] Adopting adaptive weighting method to align the multi-layer features of source domain and target domain, during the feature alignment process, the loss function is used for optimization;

[0093] The loss function is expressed as:

[0094]

[0095] in, and are the b-th layer feature representations of the source domain and the target domain, n represents the total number of feature layers, and w b Represents the weight coefficient of the b-th layer feature;

[0096] Preferably, this loss function enhances the transfer capability of the deep transfer learning framework by minimizing the distance between the features of the source domain and the target domain, and ensures the effective transfer of knowledge.

[0097] Furthermore, after the knowledge transfer and alignment of multi-level transfer learning, the features of the source domain and the target domain are then nonlinearly fused to further improve the matching accuracy between positions and job seekers. Traditional feature fusion methods mostly use simple weighted summation or linear transformation, while this method uses adaptive nonlinear mapping functions to perform higher-order feature fusion, breaking through the limitations of linear mapping;

[0098] Specifically, the multi-dimensional features of job seekers and positions are fused through the Gaussian kernel function to construct a nonlinear fusion mechanism, which is expressed as:

[0099]

[0100] Among them, F(p,j) is the fusion feature of job seeker p and job j, p i represents the i-th feature of the job seeker, j i represents the i-th feature of the position, i represents the index variable of the feature, m represents the total number of feature dimensions, γ i is the weight coefficient for each feature, is the standard deviation of the Gaussian kernel function, φ(p i ,j i ) is the adaptive mapping function between features;

[0101] Preferably, the nonlinear fusion mechanism can effectively capture the complex nonlinear relationship between job seekers and positions, avoid the linear assumption limitations of traditional methods, and significantly improve the matching accuracy.

[0102] Construct a Q-learning network based on reinforcement learning to optimize job recommendations based on personal characteristic data, job information data, emotional demand data, job culture fit data, and job recommendation feedback data;

[0103] Dynamic matching Q value update formula, the expression is:

[0104] Q(s,a)=Q(s,a)+α(r+μmaxQ(s',a')-Q(s,a));

[0105] Where Q(s,a) is the Q value of taking action a in state s, α is the learning rate, μ is the discount factor, r is the immediate reward, s' represents the next state, and a' represents the next action;

[0106] Preferably, the nonlinear fusion mechanism continuously adjusts the recommendation strategy through reinforcement learning and optimizes the matching process according to the actual feedback of job seekers, thereby improving the matching accuracy.

[0107] It should be noted that the combination of multi-level transfer learning framework, nonlinear feature fusion and reinforcement learning, as well as high-order mapping functions, greatly improves the matching accuracy between job seekers and positions. This makes it possible to handle more complex feature interactions and achieve accurate personalized job recommendations.

[0108] S4. Through the multi-dimensional feature fusion algorithm, the unstructured text data of the job seeker is fused with the emotional portrait of the job seeker, the similarity between the job seeker and the position is calculated, and the preliminary matching score is obtained.

[0109] Furthermore, the BERT model is used to extract features from the job seeker’s unstructured text data and generate a job seeker text feature vector;

[0110] Among them, the job seeker text feature vector includes the job seeker's language features, skill description, work experience and other information.

[0111] Convert the predicted sentiment intensity value of each text in the sentiment portrait into a sentiment numerical vector;

[0112] The job seeker's behavioral data (such as browsing history, application records, interview feedback, etc.) is extracted from personal characteristic data and job recommendation feedback data, and processed based on the LSTM time series model to capture the job seeker's behavioral change trend and obtain the job seeker's behavioral pattern. This process can reveal information such as job seekers' changes in interests and job search preferences.

[0113] Preferably, the LSTM time series model is used to capture the changing trend of job seekers' behavior, which not only relies on static personal information, but also introduces the dynamic factor of behavioral data. This enables the present invention to overcome the defect of lack of analysis of job seekers' dynamic behavior in traditional methods and accurately reflect the changing trend of job seekers in the time dimension.

[0114] Use multi-layer perceptron and deep neural network to fuse the text feature vector, sentiment value vector and behavior pattern of job applicants;

[0115] In the process of feature fusion and alignment, the weights of different features are dynamically adjusted, and the features of different data sources are aligned and standardized to ensure that the contribution of each feature can be reasonably measured;

[0116] The features of different modalities are integrated layer by layer through the fully connected layer. Each layer of the neural network further optimizes the relationship between different features through weighting, activation function and nonlinear mapping to generate a comprehensive feature vector of job seekers.

[0117] Specifically, the text feature vector, sentiment numerical vector, and behavior pattern data are input into the neural network. During this process, all features are first standardized to ensure that they are within the same dimensional range to prevent a certain feature from excessively affecting the training. Next, these standardized features are sent to a multi-layer fully connected neural network (MLP), where a linear transformation is performed through a weighted matrix in each layer, and nonlinear mapping is introduced through activation functions (such as ReLU) to improve the network's fitting and expression capabilities. In addition, to avoid overfitting, the network will use regularization methods (such as Dropout). Through multiple layers of weighting, activation, and nonlinear mapping, a comprehensive feature vector is finally obtained. This comprehensive feature vector integrates data from different modalities and fully captures the complex relationship between the features.

[0118] Generative adversarial network matching optimization is adopted. The generator accepts the comprehensive feature vector of the job seeker, the job information data and the job culture suitability data, and generates a matching feature vector.

[0119] Among them, the matching feature vector represents the potential matching degree between job seekers and positions.

[0120] Preferably, through generative adversarial training, the game between the generator and the discriminator can automatically optimize the matching feature vector and improve the accuracy of the matching results. This not only solves the problem of data scarcity and difficulty in modeling in traditional recommendation systems, but also effectively improves the level of intelligence of recommendations.

[0121] Based on historical job recommendation feedback data and manually labeled job matching data, deep learning methods are used for training to output true matching feature vectors;

[0122] Among them, the true matching feature vector will be regarded as the reference standard to evaluate the accuracy of the matching feature vector output by the generator.

[0123] The discriminator compares the generated matching feature vector with the real matching feature vector to evaluate the matching degree between the job seeker and the position, and outputs the matching score.

[0124] Specifically, the generator generates a matching feature vector based on the comprehensive feature vector of the job seeker and the job information data. At the same time, based on the historical job recommendation feedback data and manually annotated job matching data, a discriminator model is trained. The discriminator will accept two inputs: one is the matching feature vector output by the generator, and the other is the real matching feature vector based on historical data or manual annotation. The discriminator calculates the similarity between the two matching feature vectors, usually using metrics such as cosine similarity, Euclidean distance or Manhattan distance. In this way, the discriminator can determine whether the generated matching feature vector is close to the real matching feature vector, thereby evaluating the matching degree between the job seeker and the job. If the generated matching feature vector is closer to the real matching feature vector, the matching score output by the discriminator will be higher; otherwise, the score will be lower. In the adversarial training process, the discriminator aims to distinguish the generated matching feature vector from the real matching feature vector as accurately as possible, while the generator strives to generate matching vectors that are closer to the real features to optimize the matching accuracy. Through this game process, the generator and the discriminator evolve together, thereby improving the final job matching accuracy.

[0125] It should be noted that the relationship between the generated matching feature vector and the true matching feature vector is essentially the contrast relationship in generative adversarial training. During the training process, the generator generates a matching feature vector based on the input data, and the discriminator compares the generated matching feature vector with the true matching feature vector to evaluate the similarity between the two. The goal of the discriminator is to distinguish the generated vector from the true vector to ensure that the output of the generator is closer to the true value.

[0126] In the process of generative adversarial training, the generator and the discriminator continuously optimize the generation process of matching feature vectors through mutual competition;

[0127] After generating adversarial training to obtain the matching feature vectors of job seekers and positions, the cosine similarity is used to calculate the matching degree between job seekers and positions;

[0128] Furthermore, once the generator obtains the matching feature vectors of job seekers and positions through generative adversarial training, it is necessary to calculate the matching degree between job seekers and positions through cosine similarity.

[0129] Specifically, the matching feature vector is generated by integrating the text features, emotional profile, behavior patterns and relevant information of the job applicant, and the cosine similarity is used to measure the similarity between these feature vectors. The calculation formula of cosine similarity is:

[0130]

[0131] Where C(A,B) represents the cosine similarity between vectors A and B, A and B represent the feature vectors of job seekers and positions respectively, ||A|| and ||B|| represent the Euclidean norm of vectors A and B respectively;

[0132] It should be noted that the value of cosine similarity is between -1 and 1. When the similarity is close to 1, it means that the match between the job seeker and the position is high; when it is close to -1, it means that the match is low; when it is close to 0, it means that the match between the two is weak.

[0133] The match between the job seeker and the position is used as the preliminary match score.

[0134] S5. Generate a personalized job recommendation list based on the emotional needs of job seekers and the cultural fit of the positions, and give priority to the positions that best match the job seekers’ needs.

[0135] Going a step further, the emotional needs data of job seekers are standardized and aligned with the job culture fit data;

[0136] Based on the matching scores, use the Top-K recommendation strategy to select the top K positions with the highest matching scores as candidate recommended positions;

[0137] Among them, the Top-K recommendation strategy is derived from the common practice in information retrieval and recommendation systems, which aims to select the top K items from a large number of candidate items. For the job recommendation in the present invention, the setting of the K value can be dynamically adjusted according to the specific application scenario, such as job seeker preferences, number of positions, etc., so the rationality of the K value is closely related to the specific scenario requirements. By sorting the matching scores, the K value limits the length of the recommendation list, which can not only improve the recommendation efficiency, but also avoid excessive redundancy of the recommendation results, helping job seekers quickly find the most suitable positions.

[0138] Combining multi-source data of job seekers and job positions, the preliminary recommendation list is re-sorted, and the positions with the highest matching degree between emotional demand data and job culture adaptability data are displayed first, and a personalized job recommendation list is output;

[0139] Specifically, when performing secondary sorting, first sort the positions in the preliminary recommendation list according to the skill matching scores of job seekers, and select the top K positions as candidate positions. Next, perform secondary sorting on these candidate positions, first evaluate the cultural fit of the positions by calculating the matching degree between the job seeker's emotional needs data and the job culture fit data. In this process, the job seeker's preference for non-technical requirements such as work environment and team atmosphere is quantified, and compared with the cultural characteristic data of the position to obtain the cultural fit score of each position. Then, the cultural fit score of the position is combined with the original skill matching score, and the total matching score of each position is calculated comprehensively. Finally, the candidate positions are re-sorted according to the total matching score, and the positions that best match the job seeker's emotional needs and job culture fit are displayed first, and the final personalized job recommendation list is output to ensure that the recommendation results not only meet the job seeker's technical needs, but also meet their cultural and emotional expectations, and provide the most matching job selection.

[0140] The best one combines the emotional needs data, cultural compatibility data and other dimensional information of job seekers to generate a personalized job recommendation list that not only reflects the matching of technical capabilities, but also takes into account the emotional and cultural needs of job seekers. This cross-dimensional multi-consideration method makes the recommendation system more intelligent and personalized.

[0141] The present embodiment also provides a system for intelligent recommendation of precise matching of people and positions, including: a data collection module, an emotional portrait module, a knowledge transfer module, a feature fusion module and a position recommendation module; the data collection module is used to collect multi-source data of job seekers and recruitment positions in real time, and perform data cleaning, classification and standardization processing to output unstructured text data; the emotional portrait module is used to extract the emotional tendencies of job seekers through emotional analysis methods and construct the emotional portrait vectors of job seekers; the knowledge transfer module is used to use deep transfer learning to transfer knowledge from the source field to the target field, optimize the matching accuracy between positions and job seekers, and optimize position recommendations based on the transferred knowledge; the feature fusion module is used to fuse the unstructured text data of job seekers with the emotional portrait vectors of job seekers through a multi-dimensional feature fusion algorithm, calculate the similarity between job seekers and positions, and obtain a preliminary matching score; the position recommendation module is used to combine the emotional needs of job seekers and the cultural adaptability of positions to generate a personalized position recommendation list, and give priority to recommending positions with the highest matching degree with job seekers' needs.

[0142] This embodiment also provides a computer device, which is suitable for the method of intelligent recommendation for precise matching of people and positions, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method of intelligent recommendation for precise matching of people and positions proposed in the above embodiment.

[0143] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0144] The present embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for realizing intelligent recommendation of precise matching of people and jobs as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0145] In summary, the present invention pre-processes multi-source data by using data cleaning, classification and standardization to ensure data quality and consistency; extracts the emotional tendencies of job seekers through sentiment analysis, and constructs emotional portraits based on their personal characteristics, which can accurately capture the emotional needs of job seekers; uses a deep transfer learning framework to transfer the knowledge of the source field to the target field, optimizes the matching accuracy between positions and job seekers, and at the same time improves the effect of feature fusion through adaptive weighted feature alignment. Furthermore, through a multi-dimensional feature fusion algorithm, multiple features such as text features, emotional portraits and behavioral patterns of job seekers are integrated to calculate the similarity between job seekers and positions; the generative adversarial network continuously optimizes the matching feature vectors, improves the matching accuracy and generates a personalized job recommendation list, ensuring that the recommendation results are more in line with the emotional needs of job seekers and the cultural adaptability of the position, effectively improving the matching accuracy and user experience.

[0146] It should be noted that the above embodiments are only used to illustrate the technical solutions 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 solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligent recommendation of accurate matching of people and positions, characterized by: include, Collect multi-source data of job seekers and job openings in real time, clean, classify and standardize the data, and output unstructured text data; Extract the emotional tendencies of job seekers through sentiment analysis methods and construct the emotional portrait vector of job seekers; Using deep transfer learning, we transfer knowledge from the source domain to the target domain, optimize the matching accuracy between positions and job seekers, and optimize job recommendations based on the transferred knowledge. Through the multi-dimensional feature fusion algorithm, the unstructured text data of job seekers is fused with the emotional portrait vector of job seekers, the similarity between job seekers and positions is calculated, and the preliminary matching score is obtained; Combining the emotional needs of job seekers and the cultural fit of positions, a personalized job recommendation list is generated, with priority given to positions that best match the job seekers' needs.

2. The method for intelligent recommendation of accurate matching of people and positions according to claim 1, characterized in that: The multi-source data of job seekers and recruitment positions include personal characteristic data, position information data, emotional demand data, position culture adaptability data and position recommendation feedback data.

3. The method for intelligent recommendation of accurate matching of people and positions according to claim 2, characterized in that: The data cleaning, classification and standardization are performed to output unstructured text data. The specific steps are as follows: Check for missing values, handle duplicate data, and handle outliers on multi-source data of job seekers and job openings; Perform feature classification and text classification on the personal feature data and job information data in the multi-source data of job seekers and job openings, and convert them into standardized labels; Standardize the numerical data in the processed multi-source data of job seekers and job positions, and vectorize the text data using a pre-trained language model; Output the cleaned, classified and standardized data into unstructured text format.

4. The method for intelligent recommendation of accurate matching of people and positions according to claim 3, characterized in that: The emotional tendency of job seekers is extracted by using the emotional analysis method to construct the emotional portrait vector of job seekers. The specific steps are as follows: Segment the applicants’ unstructured text data and remove stop words, and use NLP tools to perform part-of-speech tagging; Fine-tune the unstructured text data of job seekers based on the BERT pre-trained model, classify the text into sentiment labels through sentiment classification, and predict the sentiment intensity value of each text; The emotion intensity value is converted into a numerical vector and concatenated with the job applicant's personal characteristic data to form an emotion portrait vector.

5. The method for intelligent recommendation of accurate matching of people and positions according to claim 4, characterized in that: The method uses deep transfer learning to transfer knowledge from the source domain to the target domain, optimizes the matching accuracy between positions and job seekers, and optimizes position recommendations based on the transferred knowledge. The specific steps are as follows: Extract features from source domain data through a multi-layer neural network to generate a deep feature vector representation of the source domain; Transfer the deep feature vector representation to the target domain and build an adaptive transfer learning framework to transfer knowledge layer by layer by fine-tuning parameters; In each layer, nonlinear mapping functions are used to fuse high-order features, thereby maintaining multi-scale information between the source and target domains; Adopting adaptive weighting method to align the multi-layer features of source domain and target domain, during the feature alignment process, the loss function is used for optimization; The multi-dimensional features of job seekers and positions are integrated through Gaussian kernel function to build a nonlinear fusion mechanism; A Q-learning network is constructed based on reinforcement learning to optimize job recommendations based on personal characteristic data, job information data, emotional demand data, job culture fit data, and job recommendation feedback data.

6. The method for intelligent recommendation of accurate matching of people and positions according to claim 5, characterized in that: The multi-dimensional feature fusion algorithm is used to fuse the unstructured text data of the job seeker with the emotional portrait of the job seeker, calculate the similarity between the job seeker and the position, and obtain a preliminary matching score. The specific steps are as follows: The BERT model is used to extract features from the job seeker’s unstructured text data and generate a job seeker text feature vector. Convert the predicted sentiment intensity value of each text in the sentiment portrait into a sentiment numerical vector; Extract the job seeker's behavior data from personal feature data and job recommendation feedback data, and process it based on the LSTM time series model to capture the job seeker's behavior change trend and obtain the job seeker's behavior pattern; Use multi-layer perceptron and deep neural network to fuse the text feature vector, sentiment value vector and behavior pattern of job applicants; In the process of feature fusion and alignment, the weights of different features are dynamically adjusted to align and standardize the features of different data sources; The features of different modalities are integrated layer by layer through the fully connected layer. Each layer of the neural network further optimizes the relationship between different features through weighting, activation function and nonlinear mapping to generate a comprehensive feature vector of job seekers. Generative adversarial network matching optimization is adopted. The generator accepts the comprehensive feature vector of the job seeker, the job information data and the job culture suitability data, and generates a matching feature vector. Based on historical job recommendation feedback data and manually labeled job matching data, deep learning methods are used for training to output true matching feature vectors; The discriminator compares the generated matching feature vector with the real matching feature vector to evaluate the matching degree between the job seeker and the position, and outputs the matching score. In the process of generative adversarial training, the generator and the discriminator continuously optimize the generation process of matching feature vectors through mutual competition; After generating adversarial training to obtain the matching feature vectors of job seekers and positions, the cosine similarity is used to calculate the matching degree between job seekers and positions; The match between the job seeker and the position is used as the preliminary match score.

7. The method for intelligent recommendation of accurate matching of people and positions according to claim 6, characterized in that: The above method combines the emotional needs of job seekers and the cultural adaptability of positions to generate a personalized job recommendation list, and prioritizes the positions that have the highest degree of match with job seekers' needs. The specific steps are as follows: Standardize and align the emotional needs data of job seekers with the job culture fit data; Based on the matching scores, use the Top-K recommendation strategy to select the top K positions with the highest matching scores as candidate recommended positions; Combining multi-source data of job seekers and recruitment positions, the preliminary recommendation list is re-sorted, and the positions that are most matched with the job seekers' emotional needs data and the cultural compatibility data of the positions are displayed first, and a personalized job recommendation list is output.

8. A system for intelligent recommendation of precise matching of people and jobs, based on the method for intelligent recommendation of precise matching of people and jobs as claimed in any one of claims 1 to 7, characterized in that: Including data collection module, emotion profiling module, knowledge transfer module, feature fusion module and job recommendation module; The data collection module is used to collect multi-source data of job seekers and recruitment positions in real time, and perform data cleaning, classification and standardization processing to output unstructured text data; The emotional portrait module is used to extract the emotional tendencies of job seekers through emotional analysis methods and construct an emotional portrait vector of the job seeker; The knowledge transfer module is used to transfer knowledge from the source domain to the target domain using deep transfer learning, optimize the matching accuracy between positions and job seekers, and optimize position recommendations based on the transferred knowledge; The feature fusion module is used to fuse the unstructured text data of the job seeker with the emotional portrait vector of the job seeker through a multi-dimensional feature fusion algorithm, calculate the similarity between the job seeker and the position, and obtain a preliminary matching score; The job recommendation module is used to generate a personalized job recommendation list based on the job seeker's emotional needs and the cultural adaptability of the job, and give priority to recommending the jobs that have the highest matching degree with the job seeker's needs.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for intelligent recommendation of precise matching of people and jobs described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligent recommendation of precise matching of people and jobs described in any one of claims 1 to 7 are implemented.

Images