People and post matching method and system based on bidirectional collaborative filtering algorithm

Abstract

The invention discloses a person and post matching method and system based on a bidirectional collaborative filtering algorithm, and the method comprises the steps: assisting a job seeker to clearly input job-seeking basic information and transmit job-seeking appeals in a flexible manner through an intelligent interaction interface integrating voice input and large language model analysis; the method comprises the following steps: collecting data through interaction information collection, resume information analysis and recruitment demand decomposition, carrying out data cleaning and label mining, generating a job seeker dominant label, a job seeker implicit label, a recruiter dominant label and a recruiter implicit label based on an ontology, preliminarily constructing an interestingness preference model of the job seeker and the recruiter, and establishing an interestingness preference model of the job seeker and the recruiter; and then calculating a job-seeking label data set and a recruitment position label data set based on an improved collaborative filtering algorithm, calculating a person-position bidirectional matching coefficient by integrating dominant demands and implicit demands of both parties, recommending a working position with a higher bidirectional matching degree for a job-seeker position, and improving the accuracy of position recommendation.

Description

Technical Field

[0001] This invention relates to the field of intelligent computing technology, specifically to a method and system for matching people to positions based on a bidirectional collaborative filtering algorithm. Background Technology

[0002] Currently, human resources and social security departments across the country are vigorously promoting public employment service networks and developing online employment service platforms to facilitate job matching and precise employment through the "Internet + employment" model. However, most of these platforms rely on companies or employers posting job requirements, and job seekers search for jobs through portals or mobile applications and communicate to reach an agreement. This traditional online job search and recruitment system suffers from the following problems: (1) Keyword search dependence leads to "double high" barriers (high operation threshold and high accuracy of demand description): The interaction between job postings and job search information depends on keyword matching, which requires users to have extremely high information extraction capabilities. Furthermore, due to the ambiguity of keywords and semantic deviations, the matching accuracy is low (low result precision).

[0003] (2) "Blind spots in demand" of explicit label profiles: Label profiles based on resumes or job descriptions only capture explicit information (such as job position, gender, age, skills, experience), making it difficult to identify the implicit needs of job seekers (such as career expectations, adaptability potential, distance factors) and the implicit thresholds of the job (such as job stability, whether business trips are possible, whether night shifts are possible), resulting in job matching remaining at the level of superficial label matching and missing the possibility of in-depth adaptation.

[0004] (3) "Semantic gap" of mechanical tag matching: The matching of job positions and resumes relies solely on the overlap of tags, without incorporating in-depth calculations such as semantic analysis and contextual association. This fails to solve implicit matching problems such as "differences in professional terminology" and "lack of soft conditions," exacerbating the risk of supply and demand mismatch.

[0005] (4) “Ability exclusion” in manual resume generation: Traditional template resumes require users to extract information themselves, which creates operational obstacles for job seekers (such as those with low education levels and weak digital skills). At the same time, standardized templates are difficult to meet personalized job seeker needs (such as flexible employment and part-time job suitability), further weakening job competitiveness. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a job matching method and system based on a bidirectional collaborative filtering algorithm. Through an intelligent interactive interface integrating voice input and large language model analysis, it assists job seekers in flexibly and clearly inputting basic job information and conveying their job requirements. Data is collected through interactive information collection, resume information analysis, and recruitment requirement decomposition. Data cleaning and tag mining are then performed to generate ontology-based explicit tags, implicit tags, explicit tags, and implicit tags for both job seekers and recruiters. A preliminary interest preference model between job seekers and recruiters is constructed. Then, based on an improved collaborative filtering algorithm, the job tag dataset and the job position tag dataset are calculated. The bidirectional matching coefficient is calculated by combining the explicit and implicit requirements of both parties, recommending jobs with higher bidirectional matching degrees to job seekers and improving the accuracy of job recommendations.

[0007] To achieve the above objectives, the present invention provides a person-job matching method based on a bidirectional collaborative filtering algorithm, the person-job matching method comprising: S1. Obtain job information through an intelligent interactive interface, extract keywords and save them; the intelligent interactive interface includes at least one of the following input methods: text, voice, image, gesture or preset options; the job information includes several items such as job seeker's basic information, professional skills, job requirements, employment history, work resume, etc. S2. Construct a set of job search requirement tags for users, which includes explicit job search tags (also known as explicit job search tags) and implicit job search tags (also known as implicit preference tags). S3. Based on the published job requirements and historical job postings, extract keywords from the job postings and save them; The recruitment information includes several items such as the recruiter's job position, basic information and professional skills, and salary. S4. Construct a set of recruitment demand tags for employers, which includes explicit tags for recruiters (also known as explicit recruitment tags) and implicit tags for recruiters (also known as implicit recruitment preference tags). S5. Based on a bidirectional collaborative filtering algorithm, calculate the bidirectional matching value between job seekers and recruiters, and push the results to job seekers in order of matching value. This includes: calculating the similarity between the job requirement tag set and the recruitment requirement tag set based on explicit tags, and forming a preliminary set of job positions that meet the criteria based on the similarity. ; Calculate the matching score of explicit and implicit tags for both job seekers and recruiters; calculate the job-person matching coefficient by combining the explicit and implicit needs of both parties, and push the results to job seekers in order of matching coefficient. Among them, S1~S2 and S3~S4 have no order, meaning that the entry or collection of job seeker and recruiter information can be carried out at different times or simultaneously.

[0008] Furthermore, the explicit tags for job seekers include at least three of the following: job title W1, age W2, professional skills W3, gender W4, education W5, salary W6, city W7, and years of work experience W8. The implicit tags for job seekers include at least two of the following: personnel attributes R1, salary model preference R2, work address preference R3, job stability preference R4, company size preference R5, work intensity R6, and marital / childbearing status R7. Users can add or delete specific explicit and implicit tags according to their own actual situation. In general, explicit tags can more clearly reflect the basic and necessary needs of both job seekers and recruiters, while implicit tags reflect some needs that are not easy to express or some deeper desires.

[0009] Furthermore, the explicit labels of the recruiter include at least three items from the following: job position J1, age J2, professional skills J3, gender J4, education J5, salary J6, city J7, and years of work experience J8. The implicit labels of the recruiter include at least two items from the following: personnel attribute Y1, salary model Y2, work address Y3, job stability Y4, company size Y5, work intensity Y6, and marital / childbearing status Y7. And so on. Moreover, the explicit labels of the recruiter correspond to the explicit labels of the job seeker, and the implicit labels of the recruiter correspond to the implicit labels of the job seeker.

[0010] Furthermore, the label classification includes text and numerical categories, and the collaborative filtering algorithm includes an improved bidirectional collaborative filtering algorithm. Step S5, after decomposition, includes: S5.1 For text-based tags, the similarity between the corresponding job requirement tag set and the explicit tags in the recruitment requirement tag set is calculated based on the cosine similarity algorithm. The cosine similarity calculation method for job tag is as follows: ; in, It is the i-th vector element after the tag text is vectorized, and there are a total of n vector elements; the similarity calculation method for other text tags such as professional skills (W3, J3), gender (W4, J4), etc. is the same as above; For numerical tags, the similarity calculation method includes a deterministic calculation method or a fuzzy calculation method; the deterministic calculation method includes a similarity of 1 when the job seeker's tag value meets the range requirement of the recruiter's tag value, and 0 otherwise; the fuzzy calculation method includes calculating the similarity of the actual value based on the boundary similarity value and the degree of closeness between the actual value and the boundary. Calculate the similarity value between the explicit labels of job seekers and the explicit labels of each recruiter; The closer the similarity is to 1, the higher the similarity.

[0011] S5.2 Select job positions whose label similarity is greater than the given first threshold a1 to form a preliminary set of job positions that meet the criteria. ; ; in, Let k be the kth explicit label, and the set refers to all explicit labels or at least all specified explicit labels in all job postings; S5.3. Using job seekers as the benchmark, calculate the job set by combining explicit and implicit labels. Matching scores of job seekers for various job postings in China Where N is the set of job positions. The number of job openings and the matching score calculation formula for each job opening to job seekers are as follows: ; and ; Where N≥j≥1, each position has A explicit labels and B implicit labels; , These are the weighting coefficients. , These are the matching scores for recruiters relative to job seekers' explicit and implicit labels, respectively. S5.4. Using recruiters as the benchmark, calculate the job set by combining explicit and implicit tags. Matching scores of job seekers for various positions The calculation formula is as follows: ; in, , These are the matching scores of job seekers relative to the explicit and implicit labels of recruiters; S5.5 Calculate the person-job matching coefficient U j (Matching coefficient between the j-th recruiter and job seeker), comprehensive job set Matching scores of each job posting for the job seeker And the matching scores of job seekers for each position. The calculation methods include: ; S5.6, to For each recruitment position, according to Sort the values ​​from smallest to largest (person-job fit from largest to smallest). The smaller the value, the higher the person-job fit. S5.3 and S5.4 have no specific order.

[0012] Furthermore, in the matching scores of recruiters relative to job seekers' explicit and implicit labels, the education matching score... Salary matching score Work experience matching score Work address preference matching score At least one of the following criteria shall be calculated as follows: if the other criteria are met, the result shall be 1; otherwise, the result shall be 0. S5.3.1, Educational Background Matching Score The calculation method includes: starting from the job seeker's perspective (i.e., considering from the job seeker's point of view), a perfect match scores 1, and scores decrease progressively for those whose educational qualifications are higher or lower than the recruiter's educational requirements; Another method for setting experience points is to set the score to 1 when the job seeker's education level is higher than the job requirements, that is, the bottom left element of the main diagonal of the table is 1.

[0013] S5.3.2 Salary Matching Score The calculation methods include: Extract the upper limit of job seekers' expected salary W6 max With W6 min The upper and lower limits of the salary for the positions offered by the recruiting company J6 max and J6 min Using the median salary of position J6 L =(J6 max -J6 min Rate the position of the salary relative to the expected salary range. 1) When the midline J6 L If the salary is at or below the lower limit of the expected salary, the score is 0; 2) When the median line J6 L A score of 1 is given when the salary is at or above the upper limit of the expected salary. 3) When the median line J6 L When the salary falls within the expected range, calculate using the following formula: =(J6 L -W6 min ) / (W6 max -W6 min ); S5.3.3, Work Experience Matching Score The calculation method includes: starting with the job seeker, a perfect match scores 1, and scores decrease progressively for those whose educational qualifications are higher or lower than the recruiter's educational requirements; Another method for setting experience points is to set the score to 1 when the job seeker's years of work experience exceed the job requirements, that is, the bottom left element of the main diagonal of the table above is all 1.

[0014] S5.3.4, Work Address Preference Matching Score The calculation method includes: starting from the residence of the job seeker, calculating the transportation distance D to the location of the job position of the recruitment unit, and assigning scores in segments according to the distance or interpolating scores continuously within a certain range; The continuous interpolation score assignment method includes: denoting the lower bound and upper bound of the distance as D1 and D2 respectively, then when D2≥D≥D1 =(D - D1) / (D2 - D1); When D>D2, =1; when D<D1, =0; There is no order of precedence for S5.3.1 to S5.3.4.

[0015] Furthermore, in the matching scores of the explicit and implicit labels of the job seeker relative to the recruiter, the matching score for education level , the matching score for salary , and the matching score for work experience are calculated by at least one of the following methods, and the rest of the labels are 1 if they match and 0 if they do not match; S5.4.1. The matching score for education level The calculation method includes: starting from the recruiter (i.e., considering from the perspective of the recruiter), the perfect matching score increases from 0.8 to 1 from the lowest education level to the highest education level, and the increase includes continuous increase and / or segmented increase. When the education level of the job seeker is lower than the education level requirement of the recruiter, the score decreases successively. When the education level of the job seeker is higher than the education level requirement of the recruiter, the score first increases and then decreases; S5.4.2. The matching score for salary The value calculation method includes: Extracting the upper limit of the expected salary W6 of the job seeker max and W6 min , the upper and lower limits of the salary of the job position of the recruitment unit J6 max and J6 min , and using the median line of the expected salary W6 L =(W6 max -W6 min ) / 2 to score according to the position in the salary range of the job position; 1) When the median line W6 L is at or above the upper limit of the salary of the job position, the score is 0; 2) When the median line W6 L is at or below the lower limit of the salary of the job position, the score is 1; 3) When the median line W6 L is within the salary range of the job position, calculate according to the following formula.

[0016] Calculation formula: =( J6 max -W6 L ) / (J6 max -J6 min ); Whether it's the job seeker's expected salary or the employer's salary for the position, if only the base salary information is available, then use that as the lower limit, and take twice that as the upper limit.

[0017] S5.4.3, Work Experience Matching Score The calculation method includes: starting from the recruiter, the score for a perfect match increases from 0.8 to 1 as the job seeker's years of work experience are lower than the recruiter's requirements, and the score increases first and then decreases as the job seeker's years of work experience are higher than the recruiter's requirements. S5.4.1, S5.4.2, and S5.4.3 have no specific order.

[0018] On the other hand, a job matching system based on a bidirectional collaborative filtering algorithm is provided, wherein the job matching system includes a computer system implemented based on any of the above-mentioned job matching methods based on a bidirectional collaborative filtering algorithm.

[0019] The advantages and beneficial effects of this invention are as follows: Specifically, it collects data through interactive information collection, resume information analysis, and recruitment demand decomposition, performs data cleaning and tag mining, generates explicit tags, implicit tags, and implicit tags of job seekers based on ontology, initially constructs an interest preference model between job seekers and recruiters, and then calculates the job tag dataset and job posting tag dataset based on an improved collaborative filtering algorithm to obtain the matching results between job seekers and job postings and recommends them to job seekers and / or recruiters. This is a reciprocal recommendation technology, and based on this technology, it integrates a large language model to develop interactive interfaces and application functions on an online employment service platform. This technology calculates a two-way matching coefficient between job seekers and employers by integrating explicit and implicit needs from both parties. It recommends jobs with a higher degree of two-way matching to job seekers, improving the accuracy of job recommendations. At the same time, it can also recommend more suitable candidates to employers. By improving the collaborative filtering algorithm, it reduces the performance consumption of calculating the matching degree between job seekers and employers, improves the response speed of the interactive interface, and enhances the accuracy of job matching. The system improves the ease of use and user-friendliness of flexible employment job seekers through an intelligent interactive interface developed using a large language model. It can be widely integrated into the online employment service platforms built by human resources and social security departments in various provinces and cities, and has innovativeness and high practical value.

[0020] First, it defines and constructs explicit and implicit job-seeking and recruitment labels corresponding to job seekers and recruiters respectively. By judging the conformity of basic conventional information of explicit labels, the amount of subsequent fine matching scoring calculation is greatly reduced. The definition of implicit labels pays more attention to some implicit needs of job seekers and recruiters, and greatly facilitates the information collection of groups with difficulties in job seeking, especially the extraction of implicit information, through intelligent interaction.

[0021] Secondly, considering the performance requirements arising from the cold start problem, the traditional one-stage algorithm was changed to a three-stage algorithm. The first stage calculates similarity based on explicit limited tags, including job title, age, gender, education requirements, salary, city, etc., to initially screen and form a job set, thus narrowing down the scope of recommended jobs and greatly reducing the amount of subsequent matching calculations. The second stage calculates a two-way matching score for both job seekers and recruiters within the job set. The third stage calculates the person-job matching coefficient by combining the two-way scores of explicit and implicit needs from both parties, thereby taking into account the interests of both job seekers and recruiters and recommending a more suitable win-win result.

[0022] Third, the matching algorithms for salary requirements, education level, years of work experience, and address requirements have been improved. Traditional algorithms only calculate whether the job seeker's salary requirements and education level fall within the recruiter's salary and education ranges, and whether the job seeker's residential address and the job address are in the same city. This invention calculates and scores the upper and lower limits of the job seeker's salary requirements and the recruiter's job salary range, the job seeker's education requirements and the recruiter's education range, and the job seeker's years of work experience and the recruiter's years of work experience range, making the matching more reasonable. At the same time, based on GIS maps, the distance between the job seeker's residential address and the recruiter's job address is calculated, and scores are assigned according to the distance, which better reflects the implicit preferences of people facing employment difficulties for job addresses.

[0023] Fourth, it breaks through the limitations of one-way modeling by comprehensively calculating the job seeker's intentions and preferences, as well as the job matching coefficient of the recruiter's needs and preferences, thus making the recommendation more accurate. Attached Figure Description

[0024] Figure 1 This is a flowchart of the person-job matching method based on the bidirectional collaborative filtering algorithm of the present invention. Detailed Implementation

[0025] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and examples. The following examples are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0026] The development of artificial intelligence, especially the emergence of large language models, has provided a new solution to the problem that key groups, especially impoverished laborers, are not good at using computers or mobile phones for searching and communication. Through intelligent interactive interfaces that integrate voice input and large language model analysis, they have the basic functions of conveniently assisting these groups to clearly input basic job information and convey their job requirements.

[0027] Collaborative filtering is a class of recommendation algorithms, including item-rating-based and user-rating-based algorithms. Its core principle is to analyze user groups' shared interests or historical behaviors, utilizing neighborhood methods to provide responses and record them. It then uses similarity to predict positions a job seeker might be interested in, thus enabling intelligent recommendations. However, in the job search process, there is often a massive number of job openings and job requests. Current recommendations typically only match the direct needs of job seekers, rarely considering the uncertainty and implicit preferences of those facing employment difficulties, nor adequately considering the implicit preferences of recruiters. This results in low suitability of job recommendations, and a low probability of leading to hiring intentions and successful employment.

[0028] Example 1: like Figure 1 As shown, this invention is a person-job matching method based on a bidirectional collaborative filtering algorithm, the person-job matching method comprising: S1. Obtain job information through an intelligent interactive interface, extract keywords and save them; the intelligent interactive interface includes at least one of the input methods such as text, voice, image, gesture or predetermined options. This embodiment focuses on special job-seeking groups with difficulties, and adds voice, image and gesture input methods on the basis of the usual text and predetermined option input methods, and uses voice technology, image analysis technology and large language model to build an effective information acquisition module; the job information includes several items such as job seeker basic information, professional skills, job requirements, employment history, work resume, etc., as detailed in the table below; Table 1: Job Information Classification and Explanation

[0029] The basic information provided by job seekers includes gender, age, education, home address, family situation, and health status.

[0030] S2. Construct a user's job search requirement tag set, which includes explicit job search tags and implicit job search tags (also known as implicit preference tags), as shown in Table 2 below. The tag set is generally constructed through data cleaning and tag mining. One of the important innovations of this invention is that, compared with general recommendation algorithms that only extract tags based on job title, basic information, expected salary, etc., it adds tag extraction and weighted calculation of implicit preferences of job seekers. Table 2: Classification and Explanation of Explicit and Implicit Labels of Job Seekers

[0031] S3. Based on the published job requirements and historical job postings, extract keywords from the job postings and save them; The recruitment information includes several items such as the recruiter's job position, basic information and professional skills, and salary, as detailed in the table below; the recruiter's basic information includes gender, age, education (consistent with the categories of basic information provided by the job seeker), work address, and personnel attributes; Table 3: Classification and Explanation of Recruitment Information

[0032] S4. Construct a tag set for employer recruitment needs. This tag set includes explicit tags for recruiters (also known as explicit recruitment tags) and implicit tags for recruiters (also known as implicit recruitment preference tags), as shown in Table 2 below. Generally, the tag set is constructed through data cleaning and tag mining to generate explicit job tags and implicit preference tags for employers based on job requirements. A second key innovation of this invention is that, compared to general recommendation algorithms that only extract explicit requirements such as job title, salary range, age, and gender based on recruitment needs, it adds the extraction of implicit preference tags such as personnel attributes and job stability, which generally correspond to the explicit and implicit tags of job seekers. Table 4: Classification and Explanation of Explicit and Implicit Labels of Recruiters

[0033] S5. Based on a bidirectional collaborative filtering algorithm, calculate the bidirectional matching value between job seekers and recruiters, and push the results to job seekers in order of matching value. This includes: calculating the similarity between the job requirement tag set and the recruitment requirement tag set based on explicit tags, and forming a preliminary set of job positions that meet the criteria based on the similarity. ; Calculate the matching score of explicit and implicit tags for both job seekers and recruiters; calculate the job-person matching coefficient by combining the explicit and implicit needs of both parties, and push the results to job seekers in order of matching coefficient. Among them, S1~S2 and S3~S4 have no order, meaning that the entry or collection of job seeker and recruiter information can be carried out at different times or simultaneously.

[0034] Preferably, the explicit tags for job seekers include at least three of the following: job title W1, age W2, professional skills W3, gender W4, education level W5, salary W6, city W7, and years of work experience W8. The implicit tags for job seekers include at least two of the following: personnel attribute R1, salary model preference R2, work address preference R3, job stability preference R4, company size preference R5, work intensity R6, and marital / childbearing status R7. This embodiment includes all of the above. Users can add or delete specific explicit and implicit tags according to their own actual needs. Generally speaking, explicit tags can more clearly reflect the basic and necessary needs of both job seekers and recruiters, while implicit tags reflect some needs that are not easy to express or some deeper desires.

[0035] Preferably, the explicit labels of the recruiter include at least three items from the following: job position J1, age J2, professional skills J3, gender J4, education J5, salary J6, city J7, and years of work experience J8. The implicit labels of the recruiter include at least two items from the following: personnel attribute Y1, salary model Y2, work address Y3, job stability Y4, company size Y5, work intensity Y6, and marital and childbearing status Y7. Others are similar. This embodiment includes all of the above items, and the explicit labels of the recruiter correspond to the explicit labels of the job seeker, and the implicit labels of the recruiter correspond to the implicit labels of the job seeker.

[0036] Preferably, the tag classification includes text-based and numerical-based tags. For example, job tags and professional skill tags can be classified as text-based, while age tags, salary tags, and years of service can be classified as numerical-based. Some text-based tags, if their characteristics can be represented by definite, finite numbers, can also be processed in the same way as numerical tags, such as gender tags and education tags. The collaborative filtering algorithm includes an improved bidirectional collaborative filtering algorithm, and step S5, after decomposition, includes: S5.1 For text-based tags, the similarity between the corresponding job requirement tag set and the explicit tags in the recruitment requirement tag set is calculated based on the cosine similarity algorithm. The cosine similarity calculation method for job tag is as follows:

[0037] in, It is the i-th vector element after the tag text is vectorized, and there are a total of n vector elements. The text vectorization method is an existing technology, such as Word2Vec text vectorization technology, Sentence-BERT sentence vectorization technology, etc.; the similarity calculation method for other professional skills (W3, J3), gender (W4, J4), etc., text tags is the same as above; For numerical tags, the similarity calculation method includes deterministic calculation or fuzzy calculation. This embodiment adopts deterministic calculation, which means that the similarity is 1 when the job seeker's tag value meets the range requirement of the recruiter's tag value, and 0 otherwise. For example, if the age tag (W2, J2) requires the recruiter to be 18 years old or older and no more than 35 years old, the similarity is 1 if the job seeker is 25 years old, and 0 if the job seeker is 36 years old. Calculate the similarity value between the explicit labels of job seekers and the explicit labels of each recruiter; The closer the similarity is to 1, the higher the similarity.

[0038] S5.2 Select job positions whose label similarity is greater than the given first threshold a1 to form a preliminary set of job positions that meet the criteria. In this embodiment, the first threshold a1 is set to 0.5.

[0039] in, Let k be the kth explicit label, and the set refers to all explicit labels or at least all specified explicit labels in all job postings; S5.3. Using job seekers as the benchmark, calculate the job set by combining explicit and implicit labels. Matching scores of job seekers for various job postings in China Where N is the set of job positions. The number of job openings and the matching score calculation formula for each job opening to job seekers are as follows:

[0040] and

[0041] Where N≥j≥1, each position has A explicit labels and B implicit labels; , These are the weighting coefficients. , These are the matching scores of the recruiter relative to the job seeker's explicit and implicit tags, respectively. The matching score includes preset values ​​or calculated values. In a simplified case, the score is preset to 1 when the two match and 0 when they do not match. In this embodiment, A=8 and B=7, and the weight values ​​are shown in Tables 2 and 4. S5.4. Using recruiters as the benchmark, calculate the job set by combining explicit and implicit tags. Matching scores of job seekers for various positions The calculation formula is as follows:

[0042] in, , These are the matching scores of the job seeker relative to the recruiter's explicit and implicit tags, respectively. As above, in the simplified case, the matching score is 1, and the non-matching score is 0. S5.5 Calculate the person-job matching coefficient U j (Matching coefficient between the j-th recruiter and job seeker), comprehensive job set Matching scores of each job posting for the job seeker And the matching scores of job seekers for each position. The calculation methods include:

[0043] S5.6, to For each recruitment position, according to Sort the values ​​from smallest to largest (person-job fit from largest to smallest). The smaller the value, the higher the person-job fit. S5.3 and S5.4 have no specific order.

[0044] Preferably, in the matching scores of recruiters relative to job seekers' explicit and implicit labels, the education matching score... Salary matching score Work experience matching score Work address preference matching score At least one of the following criteria shall be calculated as follows: if the other criteria are met, the result shall be 1; otherwise, the result shall be 0. S5.3.1, Educational Background Matching Score The calculation method includes: starting from the job seeker's perspective (i.e., considering from the job seeker's point of view), a perfect match score is 1, and the score decreases successively for scores that are higher or lower than the recruiter's educational requirements. The specific score is given based on experience. This embodiment gives the preset values ​​of various matching scores under the current educational standards, as shown in the table below. Table 5: Educational Background Matching Score from the Job Seeker's Perspective

[0045] S5.3.2 Salary Matching Score The calculation methods include: Extract the upper limit of job seekers' expected salary W6 max With W6 min The upper and lower limits of the salary for the positions offered by the recruiting company J6 max and J6 min Using the median salary of position J6 L =(J6 max -J6 min Rate the position of the salary relative to the expected salary range. 1) When the midline J6 LIf the salary is at or below the lower limit of the expected salary, the score is 0; 2) When the median line J6 L A score of 1 is given when the salary is at or above the upper limit of the expected salary. 3) When the median line J6 L When the salary falls within the expected range, calculate using the following formula: =(J6 L -W6 min ) / (W6 max -W6 min ); Whether it's the job seeker's expected salary or the employer's salary for the position, if only the base salary information is available, then this should be used as the lower limit, and the upper limit should be twice that amount, or other reasonable settings. This also aligns with the usual scoring and evaluation logic.

[0046] S5.3.3, Work Experience Matching Score The calculation method includes: starting from the job seeker, the score for a perfect match is 1, and the score decreases successively for those whose educational qualifications are higher or lower than the recruiter's educational requirements. The specific score is given based on experience. This embodiment gives the preset values ​​of various matching scores under the general work experience division standard, as shown in the table below. Table 6: Work Experience Matching Score from the Job Seeker's Perspective

[0047] S5.3.4, Work Address Preference Matching Score The calculation method includes: taking the job seeker's place of residence as the starting point, this embodiment calculates the transportation distance D between the job location and the employer's workplace based on GIS, and assigns scores in segments according to the distance or continuously interpolates scores within a certain interval; this embodiment uses segmented scoring, based on empirical values, as shown in the table below: Table 7: Job seeker's work address preference matching score

[0048] S5.3.1 to S5.3.4 have no specific order.

[0049] Preferably, in the matching scores of job seekers relative to the explicit and implicit labels of recruiters, the education matching score... Salary matching score Work experience matching score At least one of the following criteria must be calculated as follows; if the remaining criteria are met, the result is 1; otherwise, the result is 0. S5.4.1 Educational Background Matching Score The calculation method includes: starting from the recruiter's perspective (i.e., considering the recruiter's point of view), the perfect match score increases from 0.8 to 1 from the lowest to the highest educational level. The increase includes continuous increase and / or segmented increase. When the job seeker's educational level is lower than the recruiter's educational level requirement, the score decreases successively. When the job seeker's educational level is higher than the recruiter's educational level requirement, the score increases first and then decreases. The specific score is given based on experience. This embodiment gives the preset values ​​of various matching scores under the current educational level standard, as shown in the table below. Table 8: Educational Background Matching Score from the Recruiter's Perspective

[0050] S5.4.2 Salary Matching Score Value calculation methods include: Extract the upper limit of job seekers' expected salary W6 max With W6 min The upper and lower limits of the salary for the positions offered by the recruiting company J6 max and J6 min Using the median expected salary W6 L =(W6 max -W6 min The score is based on the position's relative position within the salary range. 1) When the midline W6 L If the salary is at or above the maximum salary for the position, the score is 0. 2) Among the midline W6 L If the salary is at or below the minimum level for the position, the score is 1. 3) When the midline W6 L If the salary falls within the range of the job position, calculate according to the formula below.

[0051] Calculation formula: =( J6 max -W6 L ) / (J6 max -J6 min ); Whether it's the job seeker's expected salary or the employer's salary for the position, if only the base salary information is available, use that as the lower limit, and take twice that amount as the upper limit, or set it as another reasonable amount.

[0052] S5.4.3, Work Experience Matching Score The calculation method includes: starting from the recruiter, the score for a perfect match increases from 0.8 to 1 as the work experience of the job seeker is lower than the work experience requirement of the recruiter. When the work experience of the job seeker is higher than the work experience requirement of the recruiter, the score increases first and then decreases. The specific score is given based on experience. This embodiment gives the preset values ​​of various matching scores under the general work experience classification standard, as shown in the table below. Table 9: Work Experience Matching Score from the Recruiter's Perspective

[0053] Steps S5.4.1, S5.4.2, and S5.4.3 have no specific order.

[0054] Example 2: The difference from Example 1 is that the implicit tags for job seekers in this example also include benefits R8 and corporate culture R9, and the implicit tags for recruiters also include benefits Y8 and corporate culture Y9. These are text-based tags, with A=8 and B=9. The tag matching score is calculated based on the text similarity result, or by pre-defining fixed option expressions. The system determines whether the job seeker's and recruiter's needs match based on the fixed options selected by the user. A score of 1 is given when they match, and a score of 0 is given when they do not match. Furthermore, the weights of R6 to R9 and Y6 to Y9 are each 0.025.

[0055] Example 3: The difference from Example 1 is that in this example, the first threshold a1 is set to 0.8.

[0056] Another method for setting experience values ​​for Table 5 is to set the score to 1 when the job seeker's education level is higher than the recruitment requirements, that is, the bottom left element of the main diagonal of the table is 1.

[0057] Another method for setting experience values ​​for Table 6 is to set the score to 1 when the job seeker's years of work experience exceed the recruitment requirements, that is, the bottom left element of the main diagonal of the table is 1.

[0058] Example 4: The difference from Example 1 is that, for numerical tags, the similarity calculation method in this example adopts fuzzy calculation, which includes calculating the similarity of the actual value based on the boundary similarity value and the closeness of the actual value to the boundary. For example, for the salary tag (W6, J6), the job requirements are a monthly income of no less than 5,000 yuan, and the monthly salary of the job posting is 6,000 yuan. With the similarity of 0.5 for a monthly salary of 5,000 yuan and 1 for a monthly salary of 10,000 yuan as the boundary, linear interpolation can be used to calculate that the similarity of a monthly salary of 4,000 yuan is 0.4, the similarity of a monthly salary of 8,000 yuan is 0.8, and the similarity of a monthly salary of 12,000 yuan is 1.

[0059] Example 5: The difference from Example 1 is that the working address preference matching score in this example... The calculation method uses continuous interpolation, which involves setting the lower and upper bounds of the distance as D1 and D2, respectively. Then, when D2 ≥ D ≥ D1, ... =(D-D1) / (D2-D1); When D>D2, = 1; when D < D1, = 0.

[0060] Embodiment 6: A human-post matching system based on a bidirectional collaborative filtering algorithm, the human-post matching system comprising a computer system implemented by the human-post matching method based on the bidirectional collaborative filtering algorithm according to any one of the above embodiments and combinations of their preferred solutions.

[0061] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A person-job matching method based on a bidirectional collaborative filtering algorithm, characterized in that, The person-job matching method includes: S1. Obtain job information through an intelligent interactive interface, extract keywords and save them; the intelligent interactive interface includes at least one of the following input methods: text, voice, image, gesture or predetermined options; the job information includes several items from the following: job seeker's basic information, professional skills, job requirements, employment history, and work resume. S2. Construct a set of job search requirement tags for users, which includes explicit tags and implicit tags for job seekers; S3. Based on the published job requirements and historical job postings, extract keywords from the job postings and save them; The recruitment information includes several items such as the recruiter's job position, basic information and professional skills, and salary. S4. Construct a set of recruitment demand tags for employers, which includes explicit tags and implicit tags for recruiters; S5. Based on a bidirectional collaborative filtering algorithm, calculate the bidirectional matching value between job seekers and recruiters, and push the results to job seekers in order of matching value. This includes: calculating the similarity value of the job seeker's explicit tag relative to each recruiter's explicit tag, and forming a preliminary set of job positions that meet the threshold conditions based on the similarity. ; Calculate the matching score of recruiters relative to job seekers' explicit and implicit labels based on job seekers. The matching score of job seekers relative to the explicit and implicit labels of recruiters is calculated based on the recruiter's criteria. Where j represents the job set The j-th job in the list; combining the above two matching scores, calculate the job-person matching coefficient, and push the results to job seekers in sorted order of the matching coefficient. The job-person matching coefficient includes... and The distance difference; The execution order of S1 to S4 includes either the order of S1, S2, S3, S4 or the order of S3, S4, S1, S2.

2. The person-job matching method based on a bidirectional collaborative filtering algorithm according to claim 1, characterized in that, The explicit labels for job seekers include at least three of the following: job title (W1), age (W2), professional skills (W3), gender (W4), education level (W5), salary (W6), city (W7), and years of work experience (W8). The implicit labels for job seekers include at least two of the following: personnel attributes (R1), salary model preference (R2), work location preference (R3), job stability preference (R4), company size preference (R5), work intensity (R6), and marital / childbearing status (R7).

3. The person-job matching method based on a bidirectional collaborative filtering algorithm according to claim 1, characterized in that, The explicit labels for recruiters include at least three of the following: job title J1, age J2, professional skills J3, gender J4, education J5, salary J6, city J7, and years of work experience J8. The implicit labels for recruiters include at least two of the following: personnel attribute Y1, salary model Y2, work address Y3, job stability Y4, company size Y5, work intensity Y6, and marital / childbearing status Y7.

4. The person-job matching method based on a bidirectional collaborative filtering algorithm according to claim 1, characterized in that, The label classification includes text and numerical categories, and the collaborative filtering algorithm includes an improved bidirectional collaborative filtering algorithm. Step S5, after decomposition, includes: S5.1 For text-based tags, the similarity between the corresponding job requirement tag set and the explicit tags in the recruitment requirement tag set is calculated based on the cosine similarity algorithm. The cosine similarity calculation method for job tag is as follows: ； in, It is the i-th vector element after the tag text is vectorized, and there are a total of n vector elements; the similarity calculation method for other text-type tags is the same as above; For numerical tags, the similarity calculation method includes a deterministic calculation method or a fuzzy calculation method; the deterministic calculation method includes a similarity of 1 when the job seeker's tag value meets the range requirement of the recruiter's tag value, and 0 otherwise; the fuzzy calculation method includes calculating the similarity of the actual value based on the boundary similarity value and the degree of closeness between the actual value and the boundary. Calculate the similarity value between the explicit labels of job seekers and the explicit labels of each recruiter; S5.2 Select job positions whose label similarity is greater than the given first threshold a1 to form a preliminary set of job positions that meet the criteria. ; ； in, Let k be the kth explicit label, and the set refers to all explicit labels or at least all specified explicit labels in all job postings; S5.

3. Using job seekers as the benchmark, calculate the job set by combining explicit and implicit labels. Matching scores of job seekers for various job postings in China Where N is the set of job positions. The number of job openings and the matching score calculation formula for each job opening to job seekers are as follows: ； and ; Where N≥j≥1, each position has A explicit labels and B implicit labels; , These are the weighting coefficients. , These are the matching scores for recruiters relative to job seekers' explicit and implicit labels, respectively. S5.

4. Using recruiters as the benchmark, calculate the job set by combining explicit and implicit tags. Matching scores of job seekers for various positions The calculation formula is as follows: ； in, , These are the matching scores of job seekers relative to the explicit and implicit labels of recruiters; S5.5 Calculate the person-job matching coefficient U j Comprehensive job collection Matching scores of each job posting for the job seeker And the matching scores of job seekers for each position. The calculation methods include: ； S5.6, to For each recruitment position, according to Sort the values ​​from smallest to largest.

5. The person-job matching method based on a bidirectional collaborative filtering algorithm according to claim 4, characterized in that, In the matching scores of recruiters relative to job seekers' explicit and implicit labels, the education matching score is... Salary matching score Work experience matching score Work address preference matching score At least one of them shall be calculated according to the following methods; S5.3.1 Educational Background Matching Score The calculation method includes: starting with the job seeker, a perfect match scores 1, and scores decrease progressively for those whose educational qualifications are higher or lower than the recruiter's educational requirements; S5.3.2 Salary Matching Score The calculation methods include: Extract the upper limit of job seekers' expected salary W6 max With W6 min The upper and lower limits of the salary for the positions offered by the recruiting company J6 max and J6 min Using the median salary of position J6 L =(J6 max -J6 min Rate the position of the salary relative to the expected salary range. 1) When the midline J6 L If the salary is at or below the lower limit of the expected salary, the score is 0; 2) When the median line J6 L A score of 1 is given when the salary is at or above the upper limit of the expected salary. 3) When the median line J6 L When the salary falls within the expected range, calculate using the following formula: =(J6 L -W6 min ) / (W6 max -W6 min ); S5.3.3, Work Experience Matching Score The calculation method includes starting with the job seeker, with a perfect match score of 1, and the score decreasing for each subsequent match if the applicant's educational level is higher or lower than the recruiter's educational requirements. S5.3.4, Work Address Preference Matching Score The calculation method includes: taking the job seeker's residence as the starting point, calculating the transportation distance D between the job seeker's residence and the location of the employer's workplace, and assigning scores based on the distance in segments or through continuous interpolation within a certain interval; the continuous interpolation scoring method includes setting the lower and upper bounds of the distance as D1 and D2, respectively, and then calculating the distance when D2 ≥ D ≥ D1. =(D- D1) / ( D2- D1); When D > D2, = 1; when D < D1, = 0.

6. The person-job matching method based on a bidirectional collaborative filtering algorithm according to claim 4, characterized in that, In the matching scores of job seekers relative to explicit and implicit labels of recruiters, the education matching score is... Salary matching score Work experience matching score At least one of them shall be calculated as follows; S5.4.1 Educational Background Matching Score The calculation method includes: starting from the recruiter, the score for a perfect match from the lowest to the highest educational level increases from 0.8 to 1. When the job seeker's educational level is lower than the recruiter's educational level requirement, the score decreases step by step. When the job seeker's educational level is higher than the recruiter's educational level requirement, the score first increases and then decreases. S5.4.2 Salary Matching Score Value calculation methods include: Extract the upper limit of job seekers' expected salary W6 max With W6 min The upper and lower limits of the salary for the positions offered by the recruiting company J6 max and J6 min Using the median expected salary W6 L =(W6 max -W6 min The score is based on the position's relative position within the salary range. 1) When the midline W6 L If the salary is at or above the maximum salary for the position, the score is 0. 2) Among the midline W6 L If the salary is at or below the minimum level for the position, the score is 1. 3) When the midline W6 L If the salary falls within the range of the job position, calculate according to the formula below; Calculation formula: =( J6 max -W6 L ) / (J6 max -J6 min ); S5.4.3, Work Experience Matching Score The calculation method includes starting from the recruiter, with the score for a perfect match increasing from 0.8 to 1 as the job seeker's years of work experience increases from low to high. When the job seeker's years of work experience are lower than the recruiter's requirements, the score decreases successively. When the job seeker's years of work experience are higher than the recruiter's requirements, the score first increases and then decreases.

7. A person-job matching system based on a bidirectional collaborative filtering algorithm, characterized in that, The job matching system includes a computer system implemented based on the job matching method based on the bidirectional collaborative filtering algorithm according to any one of claims 1 to 6.

Images