Intelligent information screening method and system for human resource system

Abstract

The invention discloses an intelligent information screening method and system for a human resource system, and relates to the technical field of human resource systems. The problem that a traditional matching method is difficult to deeply mine semantic association and cannot accurately measure the matching degree of candidates and posts is solved. According to the method, through multi-step intelligent data processing and optimization, the candidate screening efficiency and accuracy are remarkably improved, a semantic analysis technology is adopted to construct a multi-layer topic tree and an emotion analysis model, and deep semantic matching of candidate resumes and position requirements is achieved; meanwhile, candidates are intelligently screened by using a bidirectional semantic matching network, the screening effect is continuously optimized by mining potential high-matching candidates and historical data feedback, bidirectional recommendation is realized through a block chain communication link, the security and efficiency of data transmission are guaranteed, complex enterprise recruitment requirements can be effectively adapted, the recruitment period is shortened, and the recruitment efficiency is improved. The man-post matching accuracy is improved, and the human resource management cost is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of human resource systems, and in particular to an intelligent information screening method and system for human resource systems. Background Art

[0002] In the recruitment process, the massive number of candidate resumes and diverse job requirements lead to inefficient information processing. Recruiters need to spend a lot of time screening resumes, which can easily lead to missing out on potential excellent candidates. At the same time, resumes and job requirements are mostly unstructured data, and traditional matching methods are difficult to deeply explore semantic associations and cannot accurately measure the match between candidates and positions. In addition, the implicit needs of recruiters are difficult to capture with existing technologies, such as corporate culture fit, emotional tendencies, etc., resulting in less than ideal recommendation results. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent information screening method and system for a human resources system, which adopts semantic analysis technology to construct a multi-layer topic tree and sentiment analysis model to achieve deep semantic matching between candidate resumes and job requirements; at the same time, it uses a two-way semantic matching network to intelligently screen candidates, and realizes two-way recommendation through a blockchain communication link, which can automatically process massive recruitment data and achieve efficient and accurate person-job matching. At the same time, it has a continuous optimization and feedback mechanism to meet the dynamic recruitment needs of enterprises, so as to solve the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions: The intelligent information screening method for a human resources system comprises the following steps: Step 1: Data collection: Obtain multi-source candidate resumes and job requirements data based on preset interface rules, and perform data cleaning and data normalization on the acquired data; Step 2: Semantic analysis: Analyze the processed candidate resumes and job requirement data based on natural language processing technology to extract keywords from the candidate resumes and job requirement data; Step 3: Candidate information screening: Keyword feature recognition is performed based on the extracted keywords, and intelligent matching and screening is performed through a two-way semantic matching network in combination with job requirements, and candidates are sorted by matching degree; Step 4: Optimize screening results: Optimize the screening results, explore potential high-matching candidates, evaluate candidates outside the screening boundary, and optimize the two-way semantic matching network based on the company's historical recruitment data; Step 5: Generate a recommendation list: Based on the preliminary screening and optimization results, generate a list of priority recommended candidates according to the matching degree, obtain the recruitment party's selection behavior, analyze the acceptance rate of the recommendation results, and provide data feedback based on the analysis results; The step 2: semantic analysis also includes: sentiment and intention analysis of the candidate's resume and job requirement data, and analysis of the candidate's interest and attitude towards the position. The specific steps are as follows: Extract emotional content from candidate resumes and emotional and motivational language from job requirements; Construct a sentiment dictionary related to the recruitment field, including positive sentiment words, negative sentiment words and neutral words, and identify the sentiment tendency in the text based on the extraction results combined with the sentiment dictionary; Classify the candidate resumes and job requirement data based on the recognition results, and predict the sentiment tendency of the candidate resumes and job requirement data; Extract keywords related to intention, combine the predicted sentiment tendency to determine the candidate's specific focus and attitude towards the keywords, and evaluate the intensity of interest by combining word frequency and context association strength.

[0005] Furthermore, the data cleaning and data normalization processing specifically includes: Establish standardized data interfaces with various data sources, identify the resume and job requirements data formats in different data sources based on interface protocols and calling rules, and mark the structured and unstructured data in the data; Based on the labeling results, the corresponding unified data format conversion is performed. At the same time, missing values ​​in the data are detected and filled according to the data format; Based on the field information in the candidate's resume, establish an industry vocabulary and terminology library, and normalize the field information content to convert the natural language description into standardized information fields.

[0006] Furthermore, the step 2: semantic analysis specifically includes: Perform semantic segmentation on the processed candidate resumes and job requirements data, perform semantic stratification on the candidate resumes and job requirements data according to the semantic structures in the resumes and job requirements, and construct a multi-layer topic tree; Each layer of the topic tree contains several clusters, each cluster represents a semantic field, and the sub-data information corresponding to each layer is clustered to obtain several subsets, and each subset is assigned to the corresponding level node of the topic tree; Segment the text content in resumes and job requirements, extract keywords in type sentences, and extract the features of words and the features of the sentences they are in; Calculate the semantic similarity between words based on the synonym dictionary and word vector model, calculate the comprehensive similarity based on sentence features, remove duplicates from words with comprehensive similarity higher than the preset threshold, retain words with strong semantic uniqueness, and output the target extracted word set; The words with consecutive positions in the target extracted word set are concatenated into complete keywords, and the concatenated keyword set is cleaned according to the statistical characteristics of part of speech; According to the semantic scope of the cleaned target keywords, the corresponding layer nodes of the topic tree are allocated to build a multi-layer semantic keyword system.

[0007] Furthermore, the process of obtaining the comprehensive similarity specifically includes: Build a refined synonym dictionary and use the synonym relationship settings in the dictionary to calculate the synonym similarity between words; Use the pre-trained word vector model to calculate the vector cosine similarity between words; Obtaining word-level similarity using the synonym similarity and vector cosine similarity; The word-level similarity is obtained by the following formula: ; Among them, S word (w1, w2) represents the word-level similarity between word vectors w1 and w2; S syn (w1, w2) represents the synonym similarity between word vectors w1 and w2; S cos (w1, w2) represents the vector cosine similarity between the word vectors w1 and w2; a, b and c represent the first adjustable parameters, which are used to adjust the weight and decay speed of the synonym similarity and the vector cosine similarity in the calculation; Using the sentence features to obtain sentence-level similarity; The word-level similarity and sentence-level similarity are used to obtain the comprehensive similarity.

[0008] Furthermore, using the sentence features to obtain sentence-level similarity includes: Retrieve sentence features; Constructing a sentence vector using the sentence features; Calculate the cosine similarity between each sentence vector; Retrieve the weight value corresponding to the sentence structure feature; Comparing the cosine similarity between each of the sentence vectors with a preset sentence similarity threshold; Assigning the same weight value to the sentence vectors whose cosine similarity between the sentence vectors exceeds the preset sentence similarity threshold; Assigning a weight value corresponding to the sentence structure feature to which the sentence vector belongs to the sentence vector whose cosine similarity between the sentence vectors does not exceed the preset sentence similarity threshold; The sentence-level similarity is obtained by using the cosine similarity between all sentence vectors with weight values; The sentence-level similarity is obtained by the following formula: ; Among them, S sent (v1, v2) represents the sentence-level similarity between sentence vectors v1 and v2; S cos (v1, v2) represents the cosine similarity between sentence vectors v1 and v2; Y represents the preset sentence similarity threshold; k1 and k2 represent the weight values ​​corresponding to sentence vectors v1 and v2; λ, σ, μ and ξ represent the adjustable second parameters.

[0009] Further, the word-level similarity and sentence-level similarity are used to obtain a comprehensive similarity, including: Encoding sentence vectors and word vectors using a neural network, and then calculating attention weights; wherein the attention weights reflect the contribution of each word and sentence in the calculation of the comprehensive similarity; Retrieve the attention weights corresponding to every two word vectors; Using the attention weights in combination with the word-level similarity and the sentence-level similarity to obtain a word adjustment factor and a sentence adjustment factor; The word adjustment factor is obtained by the following formula: ; Among them, F w A represents the word adjustment factor; w1 and A w2 represents the attention weights corresponding to the word vectors w1 and w2; S word (w1, w2) represents the word-level similarity between word vectors w1 and w2; A wp Represents the average of the attention weights corresponding to the word vectors w1 and w2; Furthermore, the statement adjustment factor is obtained by the following formula: ; Among them, F s A represents the sentence adjustment factor; v1 and A v2 A represents the attention weights corresponding to the sentence vectors v1 and v2; vp represents the average of the attention weights corresponding to the sentence vectors v1 and v2; S sent (v1, v2) represents the sentence-level similarity between sentence vectors v1 and v2; The word adjustment factor and the sentence adjustment factor are used to combine the word-level similarity and the sentence-level similarity to obtain a comprehensive similarity corresponding to the semantic similarity; The comprehensive similarity corresponding to the semantic similarity is obtained by the following formula: ; Among them, S z represents the comprehensive similarity corresponding to the semantic similarity; n and m represent the logarithm of the word vector and the logarithm of the sentence vector contained in the semantic analysis of a text; S word (w1, w2) i represents the word-level similarity between the i-th pair of word vectors w1 and w2; F wi represents the word adjustment factor corresponding to the i-th pair of word vectors w1 and w2; S sent (v1, v2) i represents the sentence-level similarity between the i-th pair of sentence vectors v1 and v2; F si represents the sentence adjustment factor corresponding to the i-th pair of sentence vectors v1 and v2; x, y and g represent the preset growth rate influencing factors.

[0010] Furthermore, in step 3, intelligent matching and screening are performed through a bidirectional semantic matching network, specifically including: Read the recruitment task attributes in the job requirement information, generate a recruitment task plan based on the reading results, and extract the recruitment task categories, key positions, and corresponding target recruitment numbers; Analyze the urgency and importance of recruitment tasks and the priority strategy of enterprise resource allocation, use weighted scoring method to sort the recruitment tasks to be executed, and give priority to positions with higher urgency and matching requirements; Analyze the corresponding job requirements based on the attribute information of the recruitment task, and build a keyword set corresponding to each job requirement; Based on the recruitment task, a blockchain node for the recruitment position is created, including the position description and requirement information, the network addresses of the recruiter and the candidate are obtained, and a communication link between the recruiter and the candidate is formed in the blockchain; Building a bidirectional semantic matching network based on the communication link to perform bidirectional semantic matching between the extracted keywords and sentiment tendency in the candidate's resume and the keyword set in the job requirements; Based on the matching results, the matching scores between the candidates and the positions are output, and the candidates meeting the requirements are screened out based on the matching degree, and based on the matching results, two-way recommendations are made to the recruiter and the candidates through the communication link.

[0011] Furthermore, a bidirectional semantic matching network is constructed based on the communication link, specifically including: Analyze the needs of the employer: obtain the employer's specific needs for the position, and calculate the employer's preference weight for the specific content in the resume based on the specific needs of the position, and clarify the employer's emphasis on each requirement; Evaluate candidate preferences: Based on the results of keyword feature recognition and sentiment and intention analysis, compare them with the needs of the employer one by one to determine the degree of match between the candidate and the employer's needs. Based on the degree of match for each need, comprehensively evaluate the candidate's overall performance in meeting the employer's needs. Calculate the employer's satisfaction: Evaluate the employer's overall satisfaction with the candidate based on the matching performance between the employer's demand priorities and the candidate's preferences. If the key items in the candidate's resume that match the employer's needs are low, the candidate is considered to be unqualified and removed from the recommendation list. Otherwise, the candidate's performance is considered to meet or exceed the employer's core requirements for the resume and the candidate is retained in the recommendation list.

[0012] Furthermore, in step 5, the acceptance rate of the recommendation results is analyzed, including: According to the recommendation priority of candidates and the behavioral record of recruiters, evaluation indicators are set, including the matching score of candidates and the actual acceptance behavior of recruiters; By comparing candidate recommendations with the recruitment party's selection behavior, the effectiveness of candidate recommendations is gradually analyzed; Determine the matching parameters between candidates and job requirements based on the analysis results, and determine the acceptance rate of the recommendation results based on the employer's feedback on the recommendation results; Determine whether the employer has non-explicit requirements that are not captured by the two-way semantic matching network, and optimize the weight distribution in the two-way semantic matching network based on the employer's feedback and judgment results.

[0013] The present invention provides another technical solution, which is an intelligent information screening system for a human resources system, comprising: The data collection module is used to establish standardized interfaces with various data sources to obtain multi-source candidate resumes and job requirements data, clean and normalize the collected data, and mark structured and unstructured data; The semantic analysis module is used to perform semantic segmentation on the processed candidate resumes and job requirements data, extract keywords, build a multi-layer topic tree, and combine sentiment analysis and intention analysis to evaluate the semantic match between candidates and job requirements; The keyword matching and screening module is used to use the bidirectional semantic matching network to calculate the matching scores between candidates and job requirements based on the extracted keywords and sentiment analysis results, and to screen candidates by matching ranking; Screening result optimization module, used to optimize screening results, explore potential high-matching candidates, evaluate resumes outside the matching boundary, adjust algorithm weights based on historical recruitment data, and improve screening accuracy and matching effect; The recommendation generation module is used to comprehensively screen and optimize the results, generate a list of priority recommended candidates, transmit the recommendation results bidirectionally through the blockchain communication link, and record the recruitment party's selection behavior data; The recommendation result analysis and feedback module is used to analyze the recommendation acceptance rate and evaluate the effectiveness of the recommendation logic and matching parameters based on the candidate matching score and the recruiter's behavior records.

[0014] Compared with the prior art, the present invention has the following beneficial effects: Through multi-step intelligent data processing and optimization, the efficiency and accuracy of candidate screening have been significantly improved. Multi-source data is collected based on standardized interfaces, and data consistency is ensured through data cleaning and normalization. Semantic analysis technology is used to build a multi-layer topic tree and sentiment analysis model to achieve deep semantic matching between candidate resumes and job requirements. At the same time, a two-way semantic matching network is used to intelligently screen candidates, and the screening effect is continuously optimized by mining potential high-matching candidates and historical data feedback. Two-way recommendations are achieved through blockchain communication links to ensure the security and efficiency of data transmission. It can effectively adapt to complex corporate recruitment needs, shorten recruitment cycles, improve the accuracy of job matching, greatly reduce human resource management costs, and provide companies with efficient and intelligent recruitment solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the intelligent information screening method for a human resources system according to the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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.

[0017] In order to solve the technical problems of low recruitment efficiency, inaccurate matching, difficulty in capturing hidden needs, and unsatisfactory recommendation results of existing technologies, please refer to Figure 1 , this embodiment provides the following technical solutions: The intelligent information screening method for a human resources system comprises the following steps: Step 1: Data collection: Obtain multi-source candidate resumes and job requirements data based on preset interface rules, and perform data cleaning and data normalization on the acquired data; In this embodiment, the data cleaning and data normalization processing specifically includes: Establish standardized data interfaces with data sources such as internal enterprise databases, recruitment platforms, and social networks, identify the data formats of resumes and job requirements in different data sources based on interface protocols and calling rules, and mark out structured data (such as tables and databases) and unstructured data (such as text descriptions) in the data; Perform the corresponding unified data format conversion based on the labeling results. At the same time, detect missing values ​​in the data, including null values, abnormal values ​​or error values, and fill in the missing values ​​according to the data format (such as using the mean, median, mode or based on model prediction); Based on the skills, job titles and other field information in the candidate's resume, establish an industry vocabulary and terminology library, and perform normalization conversion on the field information content, such as synonym merging and skill label standardization, to convert natural language descriptions into standardized information fields; Step 2: Semantic analysis: Analyze the processed candidate resumes and job requirement data based on natural language processing technology to extract keywords from the candidate resumes and job requirement data; Step 3: Candidate information screening: Keyword feature recognition is performed based on the extracted keywords, and intelligent matching and screening is performed through a two-way semantic matching network in combination with job requirements, and candidates are sorted by matching degree; Step 4: Optimize screening results: Optimize the screening results, explore potential high-matching candidates, evaluate candidates outside the screening boundary, and optimize the two-way semantic matching network based on the company's historical recruitment data; Step 5: Generate a recommendation list: Based on the preliminary screening and optimization results, generate a priority list of recommended candidates according to the degree of match, obtain the recruitment party's selection behavior, analyze the acceptance rate of the recommendation results, and provide data feedback based on the analysis results.

[0018] In this embodiment, by establishing a standardized data interface, resumes and job requirement data can be efficiently integrated from multiple data sources, reducing the workload of manual data entry and improving the speed and accuracy of data processing. By extracting keywords and performing semantic analysis through natural language processing technology, the skills and experience of candidates, as well as job requirements, can be more accurately understood, thereby improving the accuracy of candidate screening. The use of a two-way semantic matching network makes candidate screening more intelligent, and can sort candidates according to the degree of match between job requirements and candidate capabilities, greatly improving recruitment efficiency, and can continuously improve recruitment strategies to make them more in line with the actual needs of the enterprise. Data feedback based on the acceptance rate of recommendation results provides the recruitment system with opportunities for self-learning and optimization, making the system more intelligent and efficient in continuous iterations.

[0019] In this embodiment, the step 2: semantic analysis specifically includes: Perform semantic segmentation on the processed candidate resumes and job requirements data, perform semantic stratification on the candidate resumes and job requirements data according to the semantic structures in the resumes and job requirements, and construct a multi-layer topic tree; Each layer of the topic tree contains several clusters, each cluster represents a semantic field, and the sub-data information corresponding to each layer (skills, job descriptions, etc.) is clustered to obtain several subsets, and each subset is assigned to the corresponding level node of the topic tree to ensure the semantic relevance and hierarchical structure of the data; Segment the text content in resumes and job requirements, extract keywords from sentences such as skill names and job responsibilities, and extract word features (such as part of speech, word frequency, context position) and sentence features (such as sentence weight and sentence type); Calculate the semantic similarity between words based on the synonym dictionary and word vector model, calculate the comprehensive similarity based on sentence features, remove duplicates from words with comprehensive similarity higher than the preset threshold, retain words with strong semantic uniqueness, and output the target extracted word set; Concatenate consecutive words (such as "data" and "analysis") in the target word set into complete keywords (such as "data analysis"), and clean the concatenated keyword set according to the statistical characteristics of parts of speech (such as nouns and verbs) to remove low-frequency words and irrelevant words; According to the semantic scope of the cleaned target keywords, the corresponding layer nodes of the topic tree are assigned (such as "Python" is assigned to the "Programming Skills" layer) to build a multi-layer semantic keyword system.

[0020] In this embodiment, the constructed multi-layer semantic keyword system provides accurate input data for the bidirectional semantic matching network. Keywords extracted from job requirements and candidate resumes are assigned to different semantic levels, so as to gradually calculate the matching degree between candidates and positions in different semantic fields, thereby enhancing the accuracy of screening. In this embodiment, the process of obtaining the comprehensive similarity specifically includes: Build a refined synonym dictionary and use the synonym relationship settings in the dictionary to calculate the synonym similarity between words; Use pre-trained word vector models (such as BERT, GPT, etc.) to calculate the vector cosine similarity between words; Obtaining word-level similarity using the synonym similarity and vector cosine similarity; The word-level similarity is obtained by the following formula: ; Among them, S word (w1, w2) represents the word-level similarity between word vectors w1 and w2; S syn(w1, w2) represents the synonym similarity between word vectors w1 and w2; S cos (w1, w2) represents the vector cosine similarity between the word vectors w1 and w2; a, b and c represent adjustable first parameters, respectively, for adjusting the weight and decay rate of the synonym similarity and the vector cosine similarity in the calculation, and a, b and c are determined by experiments combined with application requirements; Using the sentence features to obtain sentence-level similarity; The word-level similarity and sentence-level similarity are used to obtain the comprehensive similarity.

[0021] The technical effect of the above technical solution is: by constructing a fine synonym dictionary and setting the synonym similarity between words using the synonym relationship in the dictionary, the semantic connection between words can be captured, especially those words with the same or similar meanings but different forms. Using a pre-trained word vector model (such as BERT, GPT, etc.) to calculate the vector cosine similarity between words can capture the positional relationship between words in the vector space, thereby reflecting the semantic similarity between words. Combining synonym similarity and vector cosine similarity to calculate word-level similarity can comprehensively consider the semantic connection and vector space position of words, thereby improving the accuracy of similarity calculation. The introduction of adjustable first parameters a, b and c allows the weight and decay rate of synonym similarity and vector cosine similarity in the calculation to be adjusted according to experimental and application requirements. This design makes the similarity calculation process more flexible and can adapt to different application scenarios and text features. The technical solution not only calculates word-level similarity, but also further uses sentence features to obtain sentence-level similarity, and finally combines word-level similarity and sentence-level similarity to obtain comprehensive similarity. This multi-level similarity calculation method can more comprehensively reflect the similarity between texts and is suitable for more complex text matching and comparison tasks. By accurately calculating the similarity between texts, this technical solution can be applied to a variety of text processing and analysis tasks, such as text classification, information retrieval, text summarization, machine translation, etc. Improving the accuracy of similarity calculation will help improve the performance and effectiveness of these tasks. This technical solution provides an effective similarity calculation method for natural language processing and artificial intelligence applications, which will help promote the development and innovation of these fields. By continuously optimizing and improving similarity calculation technology, the intelligence level and user experience of natural language processing and artificial intelligence applications can be further improved.

[0022] In summary, this technical solution achieves accurate, flexible and comprehensive calculation of text similarity by combining a sophisticated synonym dictionary and a pre-trained word vector model, as well as introducing adjustable parameters and multi-level similarity calculation methods, providing strong support for text processing and analysis.

[0023] In this embodiment, using the sentence features to obtain sentence-level similarity includes: Retrieve sentence features; Constructing a sentence vector using the sentence features; Calculate the cosine similarity between each sentence vector; Retrieve the weight value corresponding to the sentence structure feature; Comparing the cosine similarity between each of the sentence vectors with a preset sentence similarity threshold; Assigning the same weight value to the sentence vectors whose cosine similarity between the sentence vectors exceeds the preset sentence similarity threshold; Assigning a weight value corresponding to the sentence structure feature to which the sentence vector belongs to the sentence vector whose cosine similarity between the sentence vectors does not exceed the preset sentence similarity threshold; The sentence-level similarity is obtained by using the cosine similarity between all sentence vectors with weight values; The sentence-level similarity is obtained by the following formula: ; Among them, S sent (v1, v2) represents the sentence-level similarity between sentence vectors v1 and v2; S cos (v1, v2) represents the cosine similarity between sentence vectors v1 and v2; Y represents a preset sentence similarity threshold; k1 and k2 represent weight values ​​corresponding to sentence vectors v1 and v2; λ, σ, μ and ξ represent adjustable second parameters, which are used to adjust the influence and decay rate of each part such as cosine similarity, weight value and exponential function part, and the determination of λ, σ, μ and ξ is determined through experiments combined with application requirements.

[0024] The technical effect of the above technical solution is: by calling the sentence features and constructing the sentence vector, the technical solution can capture the representation of the sentence in the vector space, and provide a basis for calculating the similarity between sentences. Using cosine similarity to measure the similarity between sentence vectors is a common and effective method that can reflect the direction and angle relationship between vectors. By introducing the weight value corresponding to the sentence structure feature, the technical solution can consider the influence of different sentence structures on the similarity calculation and further improve the accuracy of the calculation. In particular, for sentence vectors whose cosine similarity exceeds the preset threshold, the same weight value is assigned, and it can be considered that these sentences have a high similarity in content; while for sentence vectors that do not exceed the threshold, different weight values ​​are assigned according to their sentence structure features to reflect their different importance in the similarity calculation. The technical solution introduces adjustable second parameters λ, σ, μ and ξ, which allow adjustment according to experimental and application requirements to optimize the effect of similarity calculation. This design makes the technical solution more flexible and can adapt to different application scenarios and text features. By adjusting these parameters, the influence of various parts such as cosine similarity, weight value and exponential function in the calculation can be balanced, so as to achieve fine control of the similarity calculation results. This technical solution not only calculates word-level similarity, but also further calculates sentence-level similarity, forming a multi-level similarity calculation system. This multi-level calculation method can more comprehensively reflect the similarity between texts, which helps to improve the accuracy and efficiency of text processing and analysis. By accurately calculating sentence-level similarity, this technical solution can be applied to a variety of text processing and analysis tasks, such as text classification, information retrieval, text summarization, sentiment analysis, etc. Improving the accuracy of similarity calculation helps to improve the performance and effect of these tasks, thereby promoting the intelligent level of text processing and analysis. This technical solution provides an effective sentence-level similarity calculation method for natural language processing and artificial intelligence applications, which helps to promote the development and innovation of these fields. By continuously optimizing and improving similarity calculation technology, the intelligence level and user experience of natural language processing and artificial intelligence applications can be further improved.

[0025] In summary, this technical solution achieves accurate, flexible and comprehensive calculation of sentence-level similarity by combining multiple factors such as sentence features, sentence vectors, cosine similarity and adjustable parameters, providing strong support for text processing and analysis.

[0026] In this embodiment, the word-level similarity and sentence-level similarity are used to obtain the comprehensive similarity, including: Encode sentence vectors and word vectors using a neural network (such as LSTM), and then calculate attention weights; wherein the attention weights reflect the contribution of each word and sentence in the calculation of the comprehensive similarity; Retrieve the attention weights corresponding to every two word vectors; Using the attention weights in combination with the word-level similarity and the sentence-level similarity to obtain a word adjustment factor and a sentence adjustment factor; The word adjustment factor is obtained by the following formula: ; Among them, F w A represents the word adjustment factor; w1 and A w2 represents the attention weights corresponding to the word vectors w1 and w2; S word (w1, w2) represents the word-level similarity between word vectors w1 and w2; A wp Represents the average of the attention weights corresponding to the word vectors w1 and w2; Furthermore, the statement adjustment factor is obtained by the following formula: ; Among them, F s A represents the sentence adjustment factor; v1 and A v2 A represents the attention weights corresponding to the sentence vectors v1 and v2; vp represents the average of the attention weights corresponding to the sentence vectors v1 and v2; S sent (v1, v2) represents the sentence-level similarity between sentence vectors v1 and v2; The word adjustment factor and the sentence adjustment factor are used to combine the word-level similarity and the sentence-level similarity to obtain a comprehensive similarity corresponding to the semantic similarity; The comprehensive similarity corresponding to the semantic similarity is obtained by the following formula: ; Among them, S z represents the comprehensive similarity corresponding to the semantic similarity; n and m represent the logarithm of the word vector and the logarithm of the sentence vector contained in the semantic analysis of a text; S word (w1, w2) i represents the word-level similarity between the i-th pair of word vectors w1 and w2; F wi represents the word adjustment factor corresponding to the i-th pair of word vectors w1 and w2; S sent (v1, v2) i represents the sentence-level similarity between the i-th pair of sentence vectors v1 and v2; F si It represents the sentence adjustment factor corresponding to the i-th pair of sentence vectors v1 and v2; x, y and g represent the preset growth rate influencing factors, and the specific values ​​are obtained through the combination of experiments and application requirements.

[0027] The technical effect of the above technical solution is: by using a neural network (such as LSTM) to encode the sentence vector and the word vector and calculate the attention weight, the technical solution can capture the contribution of each word and sentence in the comprehensive similarity calculation. This attention mechanism helps to highlight important words and sentences, thereby improving the accuracy of the calculation. Combining word-level similarity and sentence-level similarity, the word adjustment factor and the sentence adjustment factor are calculated using the attention weight, and the similarity calculation process is further refined. This multi-level and multi-factor consideration method makes the calculation of comprehensive similarity more comprehensive and accurate. The technical solution introduces adjustable growth rate influencing factors x, y and g, which can be adjusted according to experimental and application requirements to optimize the effect of similarity calculation. This design makes the technical solution more flexible and can adapt to different application scenarios and text features. By adjusting parameters such as attention weight, word adjustment factor, sentence adjustment factor and growth rate influencing factor, the similarity calculation results can be finely regulated to meet different application requirements. The technical solution not only calculates word-level similarity and sentence-level similarity, but also further combines attention weight and adjustment factor to calculate comprehensive similarity, forming a multi-level similarity calculation system. This multi-level calculation method can more comprehensively reflect the semantic similarity between texts, which helps to improve the accuracy and efficiency of text processing and analysis. By accurately calculating the comprehensive similarity, this technical solution can be applied to a variety of text processing and analysis tasks, such as text classification, information retrieval, text summarization, machine translation, sentiment analysis, etc. Improving the accuracy of similarity calculation will help improve the performance and effect of these tasks, thereby promoting the intelligent level of text processing and analysis. This technical solution provides an effective comprehensive similarity calculation method for natural language processing and artificial intelligence applications, which helps to promote the development and innovation of these fields. By continuously optimizing and improving similarity calculation technology, the intelligence level and user experience of natural language processing and artificial intelligence applications can be further improved.

[0028] In summary, this technical solution achieves accurate, flexible and comprehensive calculation of comprehensive similarity by combining multiple factors such as attention weight, word-level similarity, sentence-level similarity and adjustable parameters, providing strong support for text processing and analysis. This technical solution not only improves the accuracy of similarity calculation, but also enhances the flexibility and adaptability of calculation, providing strong guarantee for the development of natural language processing and artificial intelligence applications.

[0029] In this embodiment, the step 2: semantic analysis, further includes: performing sentiment and intention analysis on the candidate resume and job requirement data, analyzing the candidate's interest and attitude towards the job, and the specific steps are as follows: Extract emotion-related content from candidate resumes, such as self-evaluation, job-seeking intentions, career planning, etc., and extract emotional and motivational language from job requirements, such as corporate culture description, benefits, job advantages, etc.; Construct a sentiment dictionary related to the recruitment field, including positive sentiment words (such as "enthusiasm", "expectation", "like"), negative sentiment words (such as "rejection", "unsuitability", "disappointment") and neutral words, and identify the sentiment tendency in the text based on the extraction results combined with the sentiment dictionary; Classify the candidate resumes and job requirement data based on the recognition results, and predict the sentiment tendency of the candidate resumes and job requirement data; Extract keywords related to the intention, such as job titles, skill requirements, career development paths, etc., combine the predicted sentiment tendency to determine the candidate's specific focus and attitude towards the keywords, and evaluate the intensity of interest by combining word frequency and context association strength.

[0030] Furthermore, in step 3, intelligent matching and screening are performed through a bidirectional semantic matching network, specifically including: Read the recruitment task attributes in the job requirement information, including job requirements, target recruitment numbers, skill requirements, etc., generate a recruitment task plan based on the reading results, and extract the recruitment task categories, key positions, and corresponding target recruitment numbers; Analyze the urgency and importance of recruitment tasks and the priority strategy of enterprise resource allocation, use weighted scoring method to sort the recruitment tasks to be executed, and give priority to positions with higher urgency and matching requirements; Analyze the corresponding job requirements based on the attribute information of the recruitment task, including key features such as skill requirements, job responsibilities, and years of experience, and build a keyword set corresponding to each job requirement; Based on the recruitment task, a blockchain node for the recruitment position is created, including the position description and requirement information, and the network addresses (such as IP addresses) of the recruiter and the candidate are obtained to form a communication link between the recruiter and the candidate in the blockchain; A two-way semantic matching network is constructed based on the communication link to perform two-way semantic matching between the extracted keywords and sentiment tendencies (interests, attitudes) in the candidate's resume and the keyword set (skills, responsibilities, benefits) in the job requirements; Output the matching scores between the candidates and the positions based on the matching results, select the candidates that meet the requirements based on the matching degree, and make two-way recommendations to the recruiter and the candidates through the communication link based on the matching results; In this embodiment, building a bidirectional semantic matching network based on the communication link specifically includes: Analyze the employer's needs: Obtain the employer's specific needs for the position, such as key skills, years of experience, job responsibilities, etc., and calculate the employer's preference weight for specific content in the resume based on the specific needs of the position, such as specific educational background, certificates, project experience, etc., to clarify the employer's emphasis on each requirement; Evaluate candidate preferences: Based on the results of keyword feature recognition and sentiment and intention analysis, compare them with the needs of the employer one by one to determine the degree of match between the candidate and the employer's needs. Based on the degree of match for each need, comprehensively evaluate the candidate's overall performance in meeting the employer's needs, such as whether the candidate is good at a certain skill, whether he or she has relevant industry experience, etc. Calculate the employer's satisfaction: Evaluate the employer's overall satisfaction with the candidate based on the matching performance between the employer's demand priorities and the candidate's preferences. If the key items in the candidate's resume that match the employer's needs are low, the candidate is considered to be unqualified and removed from the recommendation list. Otherwise, the candidate's performance is considered to meet or exceed the employer's core requirements for the resume and the candidate is retained in the recommendation list.

[0031] In this embodiment, through sentiment and intention analysis, a deeper understanding of the interests and attitudes of candidates is achieved, the assessment of the emotional fit between candidates and positions is improved, and it is helpful to screen out candidates who are truly interested in the positions. The two-way semantic matching network combines the emotional tendencies and skill matching of candidates to make the matching results more accurate and reduce the possibility of mismatching. By building a communication link through blockchain technology, the transparency and trust in the recruitment process are enhanced, and the interactive experience between recruiters and candidates is improved. Recruitment tasks are sorted based on their urgency and importance, ensuring priority recruitment for key positions and improving recruitment efficiency. The needs of recruiters and candidate preferences are comprehensively considered to achieve personalized candidate recommendations and improve the success rate of recruitment.

[0032] In this embodiment, the acceptance rate of the recommendation results is analyzed in step 5, specifically including: According to the recommendation priority of candidates and the behavior record of the recruiter, set evaluation indicators, including the candidate's matching score and the recruiter's actual acceptance behavior (such as reviewing resumes, inviting interviews or hiring); By comparing candidate recommendations with the employer's selection behavior, the effectiveness of candidate recommendations is gradually analyzed. If high-priority candidates are fully accepted by the employer (such as hired), it means that the recommendation results meet the needs. If the priority recommended candidates are ignored, but the subsequent low-priority candidates are accepted, there may be a recommendation priority deviation. If all recommended candidates are not accepted, it is necessary to evaluate the accuracy of the recommendation logic and whether it is fully matched with the employer's needs; Determine the matching parameters between candidates and job requirements based on the analysis results, such as the job requirements for skills, experience, and cultural fit, and determine the acceptance rate of the recommendation results based on the employer's feedback on the recommendation results; Determine whether the employer has non-explicit requirements (such as personality, interests, etc.) that are not captured by the two-way semantic matching network, and optimize the weight distribution in the two-way semantic matching network based on the employer's feedback and judgment results. If a certain type of requirement (such as skill matching) is ignored multiple times, the weight of this type of requirement will be enhanced. If the feedback involves emotional or intention mismatches, the analysis algorithm for emotional tendencies will be optimized.

[0033] In this embodiment, by analyzing the comparison between the recommendation results and the actual acceptance behavior of the employer, it is possible to promptly discover and correct deviations in the recommendation priority, improve the accuracy of the recommendation system, and adjust the matching parameters and weight distribution according to the feedback from the employer, so that the recommendation system is more in line with actual recruitment needs, improve the adaptability of the matching algorithm, accurately analyze the recommendation acceptance rate, and improve the matching degree between candidates and positions, thereby improving the success rate of recruitment and providing more accurate and personalized recruitment services for companies and candidates.

[0034] In order to better implement the intelligent information screening method of the human resources system, the present invention provides an intelligent information screening system for the human resources system, comprising: The data collection module is used to establish standardized interfaces with data sources such as the company's internal database, recruitment platform, and social network to obtain multi-source candidate resumes and job requirements data, clean and normalize the collected data, and mark structured and unstructured data; The semantic analysis module is used to perform semantic segmentation on the processed candidate resumes and job requirements data, extract keywords, build a multi-layer topic tree, and combine sentiment analysis and intention analysis to evaluate the semantic match between candidates and job requirements; Keyword matching and screening module, which uses a two-way semantic matching network to calculate the matching scores between candidates and job requirements based on the extracted keywords and sentiment analysis results, and screens candidates by matching ranking; Screening result optimization module, used to optimize screening results, explore potential high-matching candidates, evaluate resumes outside the matching boundary, adjust algorithm weights based on historical recruitment data, and improve screening accuracy and matching effect; The recommendation generation module is used to comprehensively screen and optimize the results, generate a list of priority recommended candidates, transmit the recommendation results bidirectionally through the blockchain communication link, and record the recruitment party's selection behavior data; The recommendation result analysis and feedback module is used to analyze the recommendation acceptance rate and evaluate the effectiveness of the recommendation logic and matching parameters based on the candidate matching score and the recruiter's behavior records.

[0035] In this embodiment, the data acquisition module obtains candidate resumes and job requirement data, and after cleaning and normalization, passes the data to the semantic analysis module. The semantic analysis module extracts keywords and sentiment information, and passes the extraction results to the keyword matching and screening module. The keyword matching and screening module intelligently screens the candidate information and generates preliminary matching results. The screening result optimization module analyzes and optimizes the preliminary screening results, supplements potential high-matching candidates, and adjusts the algorithm weights. The recommendation generation module comprehensively screens and optimizes the results, generates a recommendation list and passes the results to the recruiter, while recording the recruiter's behavior. The recommendation result analysis and feedback module analyzes the recruiter's behavioral data, evaluates the recommendation acceptance rate, and uses the feedback results to improve the screening algorithm. This system can flexibly adapt to enterprise needs, improve the efficiency and accuracy of human resource screening, and optimize the recommendation effect through continuous feedback.

[0036] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent information screening method for a human resources system, characterized in that: The following steps are involved: Step 1: Data collection: Obtain multi-source candidate resumes and job requirements data based on preset interface rules, and perform data cleaning and data normalization on the acquired data; Step 2: Semantic analysis: Analyze the processed candidate resumes and job requirement data based on natural language processing technology to extract keywords from the candidate resumes and job requirement data; Step 3: Candidate information screening: Keyword feature recognition is performed based on the extracted keywords, and intelligent matching and screening is performed through a two-way semantic matching network in combination with job requirements, and candidates are sorted by matching degree; Step 4: Optimize screening results: Optimize the screening results, explore potential high-matching candidates, evaluate candidates outside the screening boundary, and optimize the two-way semantic matching network based on the company's historical recruitment data; Step 5: Generate a recommendation list: Based on the preliminary screening and optimization results, generate a list of priority recommended candidates according to the matching degree, obtain the recruitment party's selection behavior, analyze the acceptance rate of the recommendation results, and provide data feedback based on the analysis results; The step 2: semantic analysis also includes: sentiment and intention analysis of the candidate's resume and job requirement data, and analysis of the candidate's interest and attitude towards the position. The specific steps are as follows: Extract emotional content from candidate resumes and emotional and motivational language from job requirements; Construct a sentiment dictionary related to the recruitment field, including positive sentiment words, negative sentiment words and neutral words, and identify the sentiment tendency in the text based on the extraction results combined with the sentiment dictionary; Classify the candidate resumes and job requirement data based on the recognition results, and predict the sentiment tendency of the candidate resumes and job requirement data; Extract keywords related to intention, combine the predicted sentiment tendency to determine the candidate's specific focus and attitude towards the keywords, and evaluate the intensity of interest by combining word frequency and context association strength.

2. The intelligent information screening method for a human resources system according to claim 1, characterized in that: The data cleaning and data normalization processing specifically include: Establish standardized data interfaces with various data sources, identify the resume and job requirements data formats in different data sources based on interface protocols and calling rules, and mark the structured and unstructured data in the data; Based on the labeling results, the corresponding unified data format conversion is performed. At the same time, missing values ​​in the data are detected and filled according to the data format; Based on the field information in the candidate's resume, establish an industry vocabulary and terminology library, and normalize the field information content to convert the natural language description into standardized information fields.

3. The intelligent information screening method for a human resources system according to claim 2, characterized in that: The step 2: semantic analysis, specifically includes: Perform semantic segmentation on the processed candidate resumes and job requirements data, perform semantic stratification on the candidate resumes and job requirements data according to the semantic structures in the resumes and job requirements, and construct a multi-layer topic tree; Each layer of the topic tree contains several clusters, each cluster represents a semantic field, and the sub-data information corresponding to each layer is clustered to obtain several subsets, and each subset is assigned to the corresponding level node of the topic tree; Segment the text content in resumes and job requirements, extract keywords in type sentences, and extract the features of words and the features of the sentences they are in; Calculate the semantic similarity between words based on the synonym dictionary and word vector model, calculate the comprehensive similarity based on sentence features, remove duplicates from words with comprehensive similarity higher than the preset threshold, retain words with strong semantic uniqueness, and output the target extracted word set; The words with consecutive positions in the target extracted word set are concatenated into complete keywords, and the concatenated keyword set is cleaned according to the statistical characteristics of part of speech; According to the semantic scope of the cleaned target keywords, the corresponding layer nodes of the topic tree are allocated to build a multi-layer semantic keyword system.

4. The intelligent information screening method for a human resources system according to claim 3, characterized in that: The process of obtaining the comprehensive similarity specifically includes: Build a refined synonym dictionary and use the synonym relationship settings in the dictionary to calculate the synonym similarity between words; Use the pre-trained word vector model to calculate the vector cosine similarity between words; Obtaining word-level similarity using the synonym similarity and vector cosine similarity; The word-level similarity is obtained by the following formula: ； Among them, S word (w1, w2) represents the word-level similarity between word vectors w1 and w2; S syn (w1, w2) represents the synonym similarity between word vectors w1 and w2; S cos (w1, w2) represents the vector cosine similarity between the word vectors w1 and w2; a, b and c represent the first adjustable parameters, which are used to adjust the weight and decay speed of the synonym similarity and the vector cosine similarity in the calculation; The sentence features are used to obtain sentence-level similarity; and the word-level similarity and sentence-level similarity are used to obtain comprehensive similarity.

5. The intelligent information screening method for human resource system according to claim 4, characterized in that: Using the sentence features to obtain sentence-level similarity includes: retrieving sentence features; constructing sentence vectors using the sentence features; calculating the cosine similarity between each sentence vector; retrieving the weight value corresponding to the sentence structure feature; Comparing the cosine similarity between each of the sentence vectors with a preset sentence similarity threshold; Assigning the same weight value to the sentence vectors whose cosine similarity between the sentence vectors exceeds the preset sentence similarity threshold; Assigning a weight value corresponding to the sentence structure feature to which the sentence vector belongs to the sentence vector whose cosine similarity between the sentence vectors does not exceed the preset sentence similarity threshold; The sentence-level similarity is obtained by using the cosine similarity between all sentence vectors with weight values; wherein the sentence-level similarity is obtained by the following formula: ； Among them, S sent (v1, v2) represents the sentence-level similarity between sentence vectors v1 and v2; S cos (v1, v2) represents the cosine similarity between sentence vectors v1 and v2; Y represents the preset sentence similarity threshold; k1 and k2 represent the weight values ​​corresponding to sentence vectors v1 and v2; λ, σ, μ and ξ represent the adjustable second parameters.

6. The intelligent information screening method for human resource system according to claim 5, characterized in that: The comprehensive similarity is obtained by using the word-level similarity and sentence-level similarity, including: Encoding sentence vectors and word vectors using a neural network, and then calculating attention weights; wherein the attention weights reflect the contribution of each word and sentence in the calculation of the comprehensive similarity; Retrieve the attention weights corresponding to every two word vectors; Using the attention weights in combination with the word-level similarity and the sentence-level similarity to obtain a word adjustment factor and a sentence adjustment factor; The word adjustment factor is obtained by the following formula: ； Among them, F w A represents the word adjustment factor; w1 and A w2 represents the attention weights corresponding to the word vectors w1 and w2; S word (w1, w2) represents the word-level similarity between word vectors w1 and w2; A wp represents the average of the attention weights corresponding to the word vectors w1 and w2; and the sentence adjustment factor is obtained by the following formula: ； Among them, F s A represents the sentence adjustment factor; v1 and A v2 A represents the attention weights corresponding to the sentence vectors v1 and v2; vp represents the average of the attention weights corresponding to the sentence vectors v1 and v2; S sent (v1, v2) represents the sentence-level similarity between sentence vectors v1 and v2; The word adjustment factor and the sentence adjustment factor are used to combine the word-level similarity and the sentence-level similarity to obtain a comprehensive similarity corresponding to the semantic similarity; The comprehensive similarity corresponding to the semantic similarity is obtained by the following formula: ； Among them, S z represents the comprehensive similarity corresponding to the semantic similarity; n and m represent the logarithm of the word vector and the logarithm of the sentence vector contained in the semantic analysis of a text; S word (w1, w2) i represents the word-level similarity between the i-th pair of word vectors w1 and w2; F wi represents the word adjustment factor corresponding to the i-th pair of word vectors w1 and w2; S sent (v1, v2) i represents the sentence-level similarity between the i-th pair of sentence vectors v1 and v2; F si represents the sentence adjustment factor corresponding to the i-th pair of sentence vectors v1 and v2; x, y and g represent the preset growth rate influencing factors.

7. The intelligent information screening method for human resource system according to claim 6, characterized in that: In step 3, intelligent matching and screening are performed through a bidirectional semantic matching network, specifically including: Read the recruitment task attributes in the job requirement information, generate a recruitment task plan based on the reading results, and extract the recruitment task categories, key positions, and corresponding target recruitment numbers; Analyze the urgency and importance of recruitment tasks and the priority strategy of enterprise resource allocation, use weighted scoring method to sort the recruitment tasks to be executed, and give priority to positions with higher urgency and matching requirements; Analyze the corresponding job requirements based on the attribute information of the recruitment task, and build a keyword set corresponding to each job requirement; Based on the recruitment task, a blockchain node for the recruitment position is created, including the position description and requirement information, the network addresses of the recruiter and the candidate are obtained, and a communication link between the recruiter and the candidate is formed in the blockchain; Building a bidirectional semantic matching network based on the communication link to perform bidirectional semantic matching between the extracted keywords and sentiment tendency in the candidate's resume and the keyword set in the job requirements; Based on the matching results, the matching scores between the candidates and the positions are output, and the candidates meeting the requirements are screened out based on the matching degree, and based on the matching results, two-way recommendations are made to the recruiter and the candidates through the communication link.

8. The intelligent information screening method for a human resources system according to claim 7, characterized in that: Building a bidirectional semantic matching network based on the communication link specifically includes: Analyze the needs of the employer: obtain the employer's specific needs for the position, and calculate the employer's preference weight for the specific content in the resume based on the specific needs of the position, and clarify the employer's emphasis on each requirement; Evaluate candidate preferences: Based on the results of keyword feature recognition and sentiment and intention analysis, compare them with the needs of the employer one by one to determine the degree of match between the candidate and the employer's needs. Based on the degree of match for each need, comprehensively evaluate the candidate's overall performance in meeting the employer's needs. Calculate the employer's satisfaction: Evaluate the employer's overall satisfaction with the candidate based on the matching performance between the employer's demand priorities and the candidate's preferences. If the key items in the candidate's resume that match the employer's needs are low, the candidate is considered to be unqualified and removed from the recommendation list. Otherwise, the candidate's performance is considered to meet or exceed the employer's core requirements for the resume and the candidate is retained in the recommendation list.

9. The intelligent information screening method for a human resources system according to claim 8, characterized in that: In step 5, the acceptance rate of the recommended results is analyzed, including: According to the candidate's recommendation priority and the employer's behavioral record, set evaluation indicators, including the candidate's matching score and the employer's actual acceptance behavior; By comparing candidate recommendations with the recruitment party's selection behavior, the effectiveness of candidate recommendations is gradually analyzed; Determine the matching parameters between candidates and job requirements based on the analysis results, and determine the acceptance rate of the recommendation results based on the employer's feedback on the recommendation results; Determine whether the employer has non-explicit requirements that are not captured by the two-way semantic matching network, and optimize the weight distribution in the two-way semantic matching network based on the employer's feedback and judgment results.

10. An intelligent information screening system for a human resources system, applied in the intelligent information screening method for a human resources system according to claim 9, characterized in that: include: The data collection module is used to establish standardized interfaces with various data sources to obtain multi-source candidate resumes and job requirements data, clean and normalize the collected data, and mark structured and unstructured data; The semantic analysis module is used to perform semantic segmentation on the processed candidate resumes and job requirements data, extract keywords, build a multi-layer topic tree, and combine sentiment analysis and intention analysis to evaluate the semantic match between candidates and job requirements; Keyword matching and screening module, which uses a two-way semantic matching network to calculate the matching scores between candidates and job requirements based on the extracted keywords and sentiment analysis results, and screens candidates by matching ranking; Screening result optimization module, used to optimize screening results, explore potential high-matching candidates, evaluate resumes outside the matching boundary, adjust algorithm weights based on historical recruitment data, and improve screening accuracy and matching effect; The recommendation generation module is used to comprehensively screen and optimize the results, generate a list of priority recommended candidates, transmit the recommendation results bidirectionally through the blockchain communication link, and record the recruitment party's selection behavior data; The recommendation result analysis and feedback module is used to analyze the recommendation acceptance rate and evaluate the effectiveness of the recommendation logic and matching parameters based on the candidate matching score and the recruiter's behavior records.

Images