A method for pushing employment information based on user behavior analysis

Abstract

The present invention discloses a method for pushing employment information based on user behavior analysis, which belongs to the field of semantic processing technology and is implemented by the following steps: firstly, extracting the historical submissions and job positions of users and constructing corresponding job positions sets; then comparing the user to be pushed with other users and identifying similar users in work and similar submissions; then, constructing a set of unemployed and unsubmitted positions and their preference coefficients according to similar users; then, calculating the employment intention weight by analyzing the correlation between the browsed positions of the user to be pushed and the unemployed and unsubmitted positions; finally, processing these job positions sets using an employment push model, adjusting based on the employment intention weight, and finally generating personalized employment information push. This method effectively improves the accuracy and diversity of recommendations and provides users with job information that better meets their career development needs.

Description

Technical Field

[0001] The present invention relates to the field of semantic processing technology, and in particular to a method for pushing employment information based on user behavior analysis. Background Art

[0002] In the current rapidly changing job market, traditional job information recommendation systems face severe challenges. Existing recommendation technologies mainly rely on simple keyword matching, job classification tags or user static portraits, which often fail to accurately capture users' complex career development intentions and career transformation needs.

[0003] Existing technologies mainly predict the employment preferences of target users through their historical employment behaviors and historical employment search behaviors. However, this method has serious limitations: First, recommendations based solely on historical behaviors are prone to form information cocoons, which confine users to a narrow career development track and make it impossible to effectively discover cross-field and cross-industry career opportunities. Second, existing methods rely too much on historical data, ignoring the dynamic nature of users' career development and potential transformation needs. Employment recommendations are severely limited by users' historical behaviors, and there is a serious problem of homogeneity in recommended content. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for pushing employment information based on user behavior analysis, which solves the problem that the employment recommendations in the prior art are severely limited by the user's historical behavior and there is serious homogeneity in the recommended content.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for pushing employment information based on user behavior analysis, comprising the following steps:

[0006] S1. Extract the user's historical job positions and build a job position set; extract the user's historical job positions and build a job position set;

[0007] S2. Compare the user to be pushed with other users in the job position set to find users with similar jobs, where the other users are users on the same platform except the user to be pushed;

[0008] S3. Calculate the behavior similarity between the user to be pushed and other users based on the set of job positions, and find similar users;

[0009] S4. Constructing a set of unemployed positions and a set of preference coefficients for unemployed positions, and a set of undelivered positions and a set of preference coefficients for undelivered positions based on users with similar jobs and users with similar submissions;

[0010] S5, obtaining employment intention weights according to the relevance of the browsed positions of the user to be pushed and the set of unemployed positions and the set of undelivered positions;

[0011] S6. Use the employment push model to process the set of unemployed positions and the set of preference coefficients for unemployed positions, as well as the set of undelivered positions and the set of preference coefficients for undelivered positions, and obtain pushed employment information based on the adjustment of employment intention weights.

[0012] Further, S2 includes the following sub-steps:

[0013] S21, using other users as elements to construct a comparison user set, and setting an iteration parameter i, where the initial value of i is 1;

[0014] S22, when the i-th job position in the job position set of the user to be pushed is different from the i-th job position in the job position set of the user in the comparison user set, the different users in the comparison user set are eliminated;

[0015] S23. Determine whether i is equal to N. If so, take the remaining users in the comparison user set as users with similar jobs. If not, i is incremented by 1 and jump to step S22. N is the number of job positions in the job position set of the user to be pushed.

[0016] Furthermore, S3 includes the following sub-steps:

[0017] S31, taking the intersection of the posting positions set of the push user and other users;

[0018] S32. Calculate the behavior similarity based on the number of positions posted in the intersection: , where θ is the behavior similarity, z j is the number of job postings in the jth intersection, z total is the number of posted positions in the posted position set of the user to be pushed, j is a positive integer;

[0019] S33. Filter other users whose behavior similarity is greater than 0.5 as delivery-similar users.

[0020] Further, S4 includes the following sub-steps:

[0021] S41, removing the job positions of the user to be pushed from the job position sets corresponding to the users with similar jobs, to obtain unemployed positions;

[0022] S42, calculating the unemployed position preference coefficient for each unemployed position;

[0023] S43, according to the size of the unemployed position preference coefficient, the unemployed positions are arranged in descending order, the first M unemployed positions are extracted, and a set of unemployed positions is constructed, where M is a positive integer;

[0024] S44, taking the unemployed position preference coefficients corresponding to the unemployed position set as elements, and constructing an unemployed position preference coefficient set;

[0025] S45, removing the posted positions of the user to be pushed from the posted position sets corresponding to the users with similar posts, to obtain the unposted positions;

[0026] S46, calculating the undelivered position preference coefficient for each undelivered position;

[0027] S47, according to the size of the preference coefficient of the undelivered positions, the undelivered positions are sorted in descending order, the first M undelivered positions are extracted, and a set of undelivered positions is constructed;

[0028] S48. Use the undelivered position preference coefficients corresponding to the undelivered position set as elements to construct an undelivered position preference coefficient set.

[0029] Furthermore, the formula for calculating the unemployed position preference coefficient in S42 is: , where μ notw,n is the preference coefficient of the nth unemployed position, f notw,n is the number of times the nth unemployed position appears in the set of jobs corresponding to all users with similar jobs, where n is a positive integer;

[0030] The formula for calculating the preference coefficient of undelivered positions in S46 is: , where μ notd,n is the preference coefficient of the nth undelivered position, f notd,n It is the number of times the nth undelivered job appears in the set of delivered jobs corresponding to all users who have delivered similar jobs.

[0031] Further, S5 includes the following sub-steps:

[0032] S51, performing word segmentation processing on the browsed positions of the to-be-pushed user, the unemployed positions in the unemployed positions set, and the undelivered positions in the undelivered positions set;

[0033] S52, counting the frequency of each word in the job browsed by the user to be pushed in the set of unemployed jobs, and obtaining a first relevance;

[0034] S53, counting the frequency of each word in the job browsed by the user to be pushed in the set of undelivered jobs, and obtaining a second relevance;

[0035] S54. Calculate the employment intention weight of the set of unemployed positions and the employment intention weight of the set of undelivered positions respectively according to the first correlation degree and the second correlation degree.

[0036] Furthermore, the formula for obtaining the first degree of association in S52 is: , where C1 is the first correlation, P notw,m is the frequency of the mth word in the job browsed by the user to be pushed appearing in the set of unemployed jobs, L is the number of words in the job browsed by the user to be pushed, and m is a positive integer;

[0037] The formula for obtaining the second correlation degree in S53 is: , where C2 is the second correlation degree, P notd,m The frequency of the mth word in the job browsed by the user to be pushed appearing in the set of undelivered jobs.

[0038] Furthermore, the formula for calculating the employment intention weight of the set of unemployed positions in S54 is: , the formula for calculating the employment intention weight of the undelivered job set is: , where γ notw is the employment intention weight of the unemployed positions, γ notd is the employment intention weight of the set of undelivered positions.

[0039] Further, the employment push model in S6 includes: an unemployed occupation feature extraction unit, an undelivered occupation feature extraction unit, a multiplier M1, a multiplier M2, a multiplier M3, a multiplier M4, a first BiLSTM unit, a second BiLSTM unit, a concatenation unit, and a CRF unit;

[0040] The input end of the unemployed occupation feature extraction unit is used to input the unemployed position set;

[0041] The input end of the undelivered job feature extraction unit is used to input the undelivered job set;

[0042] The first input terminal of the multiplier M1 is connected to the output terminal of the unemployed occupation feature extraction unit, the second input terminal thereof is used to input the unemployed position preference coefficient set, and the output terminal thereof is connected to the input terminal of the first BiLSTM unit;

[0043] The first input terminal of the multiplier M2 is connected to the output terminal of the undelivered occupation feature extraction unit, the second input terminal thereof is used to input the undelivered occupation preference coefficient set, and the output terminal thereof is connected to the input terminal of the second BiLSTM unit;

[0044] A first input terminal of the multiplier M3 is connected to the output terminal of the first BiLSTM unit, and a second input terminal thereof is used to input the employment intention weight of the unemployed position set;

[0045] The first input terminal of the multiplier M4 is connected to the output terminal of the second BiLSTM unit, and the second input terminal thereof is used to input the employment intention weight of the undelivered position set;

[0046] The input end of the splicing unit is connected to the output end of the multiplier M3 and the output end of the multiplier M4 respectively, and the output end thereof is connected to the input end of the CRF unit;

[0047] The output of the CRF unit serves as the output of the job push model.

[0048] Further, the unemployed occupation feature extraction unit and the undelivered occupation feature extraction unit both include: a plurality of LSTM networks and a plurality of fully connected layers;

[0049] In the unemployed occupation feature extraction unit, each LSTM network is used to input an unemployed position in the unemployed position set to obtain a feature value, and each fully connected layer is used to weight the feature value corresponding to an unemployed position to obtain the unemployed position feature;

[0050] In the undelivered job feature extraction unit, each LSTM network is used to input an undelivered job in the undelivered job set to obtain a feature value, and each fully connected layer is used to weight the feature value corresponding to an undelivered job to obtain the undelivered job feature;

[0051] The multiplier M1 is used for element-wise multiplication of the unemployed position preference coefficient set and the output of the unemployed occupation feature extraction unit;

[0052] The multiplier M2 is used to multiply the undelivered job preference coefficient set and the output of the undelivered job feature extraction unit element by element.

[0053] The beneficial effects of the present invention are:

[0054] 1. The present invention compares the user to be pushed with other users on the job position set to find out users with similar jobs, and then finds out users with similar delivery according to the delivery status of the user to be pushed and other users. Based on these two users, a set of unemployed positions and a set of preference coefficients for unemployed positions are constructed, reflecting the occupations not involved in the user to be pushed. The present invention breaks through the limitations of historical behavior, reduces the homogeneity of recommended content, and can more effectively capture the user's employment development direction by finding two users with similar behaviors and then finding out the occupations not involved in the user to be pushed from these two users.

[0055] 2. The unemployed position preference coefficient set and the undelivered position preference coefficient set constructed by the present invention reflect each user's preference for positions, making it easier to select the best employment information.

[0056] 3. The present invention adopts an employment push model to process a set of unemployed positions and a set of preference coefficients for unemployed positions, as well as a set of undelivered positions and a set of preference coefficients for undelivered positions, and then uses employment intention weights for adjustment to improve the push accuracy of employment information. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a flow chart of a method for pushing employment information based on user behavior analysis;

[0058] Figure 2 This is a schematic diagram of the structure of the employment push model;

[0059] Figure 3 It is a schematic diagram of the structure of the unemployed occupation feature extraction unit and the undelivered occupation feature extraction unit. DETAILED DESCRIPTION

[0060] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0061] like Figure 1 As shown, a method for pushing employment information based on user behavior analysis includes the following steps:

[0062] S1. Extract the user's historical job positions and build a job position set; extract the user's historical job positions and build a job position set;

[0063] S2. Compare the user to be pushed with other users in the job position set to find users with similar jobs, where the other users are users on the same platform except the user to be pushed;

[0064] S3. Calculate the behavior similarity between the user to be pushed and other users based on the set of job positions, and find similar users;

[0065] S4. Constructing a set of unemployed positions and a set of preference coefficients for unemployed positions, and a set of undelivered positions and a set of preference coefficients for undelivered positions based on users with similar jobs and users with similar submissions;

[0066] S5, obtaining employment intention weights according to the relevance of the browsed positions of the user to be pushed and the set of unemployed positions and the set of undelivered positions;

[0067] S6. Use the employment push model to process the set of unemployed positions and the set of preference coefficients for unemployed positions, as well as the set of undelivered positions and the set of preference coefficients for undelivered positions, and obtain pushed employment information based on the adjustment of employment intention weights.

[0068] In this embodiment, S2 includes the following sub-steps:

[0069] S21, using other users as elements to construct a comparison user set, and setting an iteration parameter i, where the initial value of i is 1;

[0070] S22, when the i-th job position in the job position set of the user to be pushed is different from the i-th job position in the job position set of the user in the comparison user set, the different users in the comparison user set are eliminated;

[0071] S23. Determine whether i is equal to N. If so, take the remaining users in the comparison user set as users with similar jobs. If not, i is incremented by 1 and jump to step S22. N is the number of job positions in the job position set of the user to be pushed.

[0072] The present invention compares each job title in the job title set of the user to be pushed with that of other users, screens out user groups with highly consistent career trajectories, extracts the positions these users have worked in, and is more instructive for the career development of the user to be pushed.

[0073] In this embodiment, S3 includes the following sub-steps:

[0074] S31, taking the intersection of the posting positions set of the push user and other users;

[0075] S32. Calculate the behavior similarity based on the number of positions posted in the intersection: , where θ is the behavior similarity, z j is the number of job postings in the jth intersection, z total is the number of posted positions in the posted position set of the user to be pushed, j is a positive integer;

[0076] S33. Filter other users whose behavior similarity is greater than 0.5 as delivery-similar users.

[0077] The present invention obtains the intersection of the user to be pushed and other users in the set of posted positions, finds out users with similar posting behaviors, extracts the positions posted by these users, and provides more guidance for the posting direction of the user to be pushed.

[0078] In this embodiment, S4 includes the following sub-steps:

[0079] S41, removing the job positions of the user to be pushed from the job position sets corresponding to the users with similar jobs, to obtain unemployed positions;

[0080] S42, calculating the unemployed position preference coefficient for each unemployed position;

[0081] S43, according to the size of the unemployed position preference coefficient, the unemployed positions are arranged in descending order, the first M unemployed positions are extracted, and a set of unemployed positions is constructed, where M is a positive integer;

[0082] S44, taking the unemployed position preference coefficients corresponding to the unemployed position set as elements, and constructing an unemployed position preference coefficient set;

[0083] S45, removing the posted positions of the user to be pushed from the posted position sets corresponding to the users with similar posts, to obtain the unposted positions;

[0084] S46, calculating the undelivered position preference coefficient for each undelivered position;

[0085] S47, according to the size of the preference coefficient of the undelivered positions, the undelivered positions are sorted in descending order, the first M undelivered positions are extracted, and a set of undelivered positions is constructed;

[0086] S48. Use the undelivered position preference coefficients corresponding to the undelivered position set as elements to construct an undelivered position preference coefficient set.

[0087] The present invention selects positions that the user to be pushed has not worked for from the job position set of users with similar work, selects positions that the user to be pushed has not submitted for from the posting position set of users with similar submission, and then calculates the preference coefficient of the position to reflect the popularity of the position.

[0088] In this embodiment, the formula for calculating the unemployed position preference coefficient in S42 is: , where μ notw,n is the preference coefficient of the nth unemployed position, f notw,n is the number of times the nth unemployed position appears in the set of jobs corresponding to all users with similar jobs, where n is a positive integer.

[0089] In this embodiment, the formula for calculating the undelivered position preference coefficient in S46 is: , where μ notd,n is the preference coefficient of the nth undelivered position, f notd,n It is the number of times the nth undelivered job appears in the set of delivered jobs corresponding to all users who have delivered similar jobs.

[0090] In the present invention, the more users have selected the unemployed positions and the undelivered positions, the greater the preference coefficients of the unemployed positions and the undelivered positions.

[0091] In this embodiment, S5 includes the following sub-steps:

[0092] S51, performing word segmentation processing on the browsed positions of the to-be-pushed user, the unemployed positions in the unemployed positions set, and the undelivered positions in the undelivered positions set;

[0093] S52, counting the frequency of each word in the job browsed by the user to be pushed in the set of unemployed jobs, and obtaining a first relevance;

[0094] S53, counting the frequency of each word in the job browsed by the user to be pushed in the set of undelivered jobs, and obtaining a second relevance;

[0095] S54. Calculate the employment intention weight of the set of unemployed positions and the employment intention weight of the set of undelivered positions respectively according to the first correlation degree and the second correlation degree.

[0096] The word segmentation process of step S51 is to divide the job positions into words or phrases, for example: the browsing job positions to be pushed to users: "Data Analyst" → word segmentation result: ["data", "analyst"]; a set of unemployed jobs: "Data Scientist" → ["data", "scientist"], "Data Engineer" → ["data", "engineer"], "Market Analyst" → ["market", "analyst"]; a set of undelivered jobs: "Business Analyst" → ["business", "analyst"], "Product Manager" → ["product", "manager"], "Data Mining Engineer" → ["data", "mining", "engineer"].

[0097] In this embodiment, the formula for obtaining the first degree of association in S52 is: , where C1 is the first correlation, P notw,m is the frequency of the mth word in the job browsed by the user to be pushed appearing in the set of unemployed jobs, L is the number of words in the job browsed by the user to be pushed, and m is a positive integer;

[0098] The formula for obtaining the second correlation degree in S53 is: , where C2 is the second correlation degree, P notd,m The frequency of the mth word in the job browsed by the user to be pushed appearing in the set of undelivered jobs.

[0099] In this embodiment, there can be multiple positions to be browsed. When C1 and C2 are 0, the employment intention weight of the unemployed position set and the employment intention weight of the undelivered position set are assigned the same value, both set to 0.5.

[0100] In this embodiment, the formula for calculating the employment intention weight of the unemployed position set in S54 is: , the formula for calculating the employment intention weight of the undelivered job set is: , where γ notw is the employment intention weight of the unemployed positions, γ notd is the employment intention weight of the set of undelivered positions.

[0101] The present invention counts the frequency of occurrence of each word in the browsed positions in the unemployed positions set and the undelivered positions set, calculates the employment intention weight, and reflects the preference degree of the user to be pushed to the unemployed positions set and the undelivered positions set.

[0102] like Figure 2 As shown, the employment push model in S6 includes: an unemployed occupation feature extraction unit, an undelivered occupation feature extraction unit, a multiplier M1, a multiplier M2, a multiplier M3, a multiplier M4, a first BiLSTM unit, a second BiLSTM unit, a concatenation unit, and a CRF unit;

[0103] The input end of the unemployed occupation feature extraction unit is used to input the unemployed position set;

[0104] The input end of the undelivered job feature extraction unit is used to input the undelivered job set;

[0105] The first input terminal of the multiplier M1 is connected to the output terminal of the unemployed occupation feature extraction unit, the second input terminal thereof is used to input the unemployed position preference coefficient set, and the output terminal thereof is connected to the input terminal of the first BiLSTM unit;

[0106] The first input terminal of the multiplier M2 is connected to the output terminal of the undelivered occupation feature extraction unit, the second input terminal thereof is used to input the undelivered occupation preference coefficient set, and the output terminal thereof is connected to the input terminal of the second BiLSTM unit;

[0107] The first input terminal of the multiplier M3 is connected to the output terminal of the first BiLSTM unit, and the second input terminal thereof is used to input the employment intention weight of the unemployed position set;

[0108] The first input terminal of the multiplier M4 is connected to the output terminal of the second BiLSTM unit, and the second input terminal thereof is used to input the employment intention weight of the undelivered position set;

[0109] The input end of the splicing unit is connected to the output end of the multiplier M3 and the output end of the multiplier M4 respectively, and the output end thereof is connected to the input end of the CRF unit;

[0110] The output of the CRF unit serves as the output of the job push model.

[0111] like Figure 3 As shown, the unemployed occupation feature extraction unit and the undelivered occupation feature extraction unit both include: multiple LSTM networks and multiple fully connected layers;

[0112] In the unemployed occupation feature extraction unit, each LSTM network is used to input an unemployed position in the unemployed position set to obtain a feature value, and each fully connected layer is used to weight the feature value corresponding to an unemployed position to obtain the unemployed position feature;

[0113] In the undelivered job feature extraction unit, each LSTM network is used to input an undelivered job in the undelivered job set to obtain a feature value, and each fully connected layer is used to weight the feature value corresponding to an undelivered job to obtain the undelivered job feature;

[0114] The multiplier M1 is used for element-wise multiplication of the unemployed position preference coefficient set and the output of the unemployed occupation feature extraction unit, wherein the output of the unemployed occupation feature extraction unit is a vector concatenated from multiple unemployed position features;

[0115] The multiplier M2 is used to multiply the undelivered position preference coefficient set and the output of the undelivered occupation feature extraction unit element by element, wherein the output of the undelivered occupation feature extraction unit is a vector concatenated from multiple undelivered position features.

[0116] Each cell unit in the LSTM network of the present invention is used to input a code value corresponding to an unemployed position or an undelivered position. The expression of the fully connected layer is: , where y is the feature output by the fully connected layer, x k is the kth eigenvalue corresponding to a position, ω k For x k The weight of b k For x k , k is a positive integer, and K is the number of eigenvalues ​​corresponding to a position.

[0117] For the unemployed occupation feature extraction unit, the number of LSTM networks and fully connected layers is equal to the number of unemployed positions in the unemployed position set; for the unapplied occupation feature extraction unit, the number of LSTM networks and fully connected layers is equal to the number of unapplied positions in the unapplied position set.

[0118] The input of the first BiLSTM unit is the output of the multiplier M1, and the input of the second BiLSTM unit is the output of the multiplier M2.

[0119] In this embodiment, the job push model can be trained by the existing gradient descent method.

[0120] The employment intention weights of the set of unemployed positions are applied to the output of the first BiLSTM unit, and the employment intention weights of the set of undelivered positions are applied to the output of the second BiLSTM unit. The sizes of the two eigenvalues ​​are adjusted, and then the features are concatenated through the concatenation unit, and the CRF unit outputs the employment information.

[0121] Employment information may include: employment occupation, employment direction, job recruitment, etc.

[0122] The present invention compares the user to be pushed with other users on the job position set to find out users with similar jobs, and then finds out users with similar delivery according to the delivery status of the user to be pushed and other users, and constructs a set of unemployed positions and a set of preference coefficients for unemployed positions based on these two users, reflecting the occupations not involved in the user to be pushed. The present invention breaks through the limitations of historical behavior, reduces the homogeneity of recommended content, and can more effectively capture the user's employment development direction by finding two users with similar behaviors and then finding out the occupations not involved in the user to be pushed from these two users.

[0123] The unemployed position preference coefficient set and the undelivered position preference coefficient set constructed by the present invention reflect the preferences of each user for positions, making it easier to select the best employment information.

[0124] The present invention adopts an employment push model to process a set of unemployed positions and a set of unemployed position preference coefficients, as well as a set of undelivered positions and a set of undelivered position preference coefficients, and then adopts employment intention weights for adjustment to improve the push accuracy of employment information.

[0125] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for pushing employment information based on user behavior analysis, characterized in that: The following steps are involved: S1. Extract the user's historical job positions and build a job position set; extract the user's historical job positions and build a job position set; S2. Compare the user to be pushed with other users in the job position set to find users with similar jobs, where the other users are users on the same platform except the user to be pushed; S3. Calculate the behavior similarity between the user to be pushed and other users based on the set of job positions, and find similar users; S4. Constructing a set of unemployed positions and a set of preference coefficients for unemployed positions, and a set of undelivered positions and a set of preference coefficients for undelivered positions based on users with similar jobs and users with similar submissions; S5, obtaining employment intention weights according to the relevance of the browsed positions of the user to be pushed and the set of unemployed positions and the set of undelivered positions; S6. Using the employment push model to process the set of unemployed positions and the set of preference coefficients for unemployed positions, as well as the set of undelivered positions and the set of preference coefficients for undelivered positions, obtain pushed employment information based on the adjustment of employment intention weights; The S4 comprises the following sub-steps: S41, removing the job positions of the user to be pushed from the job position sets corresponding to the users with similar jobs, to obtain unemployed positions; S42. Calculate the unemployed position preference coefficient for each unemployed position: , where μ notw,n is the preference coefficient of the nth unemployed position, f notw,n is the number of times the nth unemployed position appears in the set of jobs corresponding to all users with similar jobs, where n is a positive integer; S43, according to the size of the preference coefficient of the unemployed position, the unemployed positions are arranged in descending order, the first M unemployed positions are extracted, and a set of unemployed positions is constructed, where M is a positive integer; S44, taking the unemployed position preference coefficients corresponding to the unemployed position set as elements, and constructing an unemployed position preference coefficient set; S45, removing the posted positions of the user to be pushed from the posted position sets corresponding to the users with similar posts, to obtain the unposted positions; S46. Calculate the undelivered position preference coefficient for each undelivered position: , where μ notd,n is the preference coefficient of the nth undelivered position, f notd,n is the number of times the nth undelivered job appears in the set of delivered jobs corresponding to all similar users; S47, according to the size of the preference coefficient of the undelivered positions, the undelivered positions are sorted in descending order, the first M undelivered positions are extracted, and a set of undelivered positions is constructed; S48. Use the undelivered position preference coefficients corresponding to the undelivered position set as elements to construct an undelivered position preference coefficient set.

2. The method for pushing employment information based on user behavior analysis according to claim 1, characterized in that: The S2 comprises the following sub-steps: S21, using other users as elements to construct a comparison user set, and setting an iteration parameter i, where the initial value of i is 1; S22, when the i-th job position in the job position set of the user to be pushed is different from the i-th job position in the job position set of the user in the comparison user set, the different users in the comparison user set are eliminated; S23. Determine whether i is equal to N. If so, take the remaining users in the comparison user set as users with similar jobs. If not, i is incremented by 1 and jump to step S22. N is the number of job positions in the job position set of the user to be pushed.

3. The method for pushing employment information based on user behavior analysis according to claim 1, characterized in that: The S3 comprises the following sub-steps: S31, taking the intersection of the posting positions set of the push user and other users; S32. Calculate the behavior similarity based on the number of positions posted in the intersection: , where θ is the behavior similarity, z j is the number of job postings in the jth intersection, z total is the number of posted positions in the posted position set of the user to be pushed, j is a positive integer; S33. Filter other users whose behavior similarity is greater than 0.5 as delivery-similar users.

4. The method for pushing employment information based on user behavior analysis according to claim 1, characterized in that: The S5 comprises the following sub-steps: S51, performing word segmentation processing on the browsed positions of the to-be-pushed user, the unemployed positions in the unemployed positions set, and the undelivered positions in the undelivered positions set; S52, counting the frequency of each word in the job browsed by the user to be pushed in the set of unemployed jobs, and obtaining a first relevance; S53, counting the frequency of each word in the job browsed by the user to be pushed in the set of undelivered jobs, and obtaining a second relevance; S54. Calculate the employment intention weight of the set of unemployed positions and the employment intention weight of the set of undelivered positions respectively according to the first correlation degree and the second correlation degree.

5. The method for pushing employment information based on user behavior analysis according to claim 4 is characterized in that: The formula for obtaining the first degree of association in S52 is: , where C1 is the first correlation, P notw,m is the frequency of the mth word in the job browsed by the user to be pushed appearing in the set of unemployed jobs, L is the number of words in the job browsed by the user to be pushed, and m is a positive integer; The formula for obtaining the second degree of association in S53 is: , where C2 is the second correlation degree, P notd,m The frequency of the mth word in the job browsed by the user to be pushed appearing in the set of undelivered jobs.

6. The method for pushing employment information based on user behavior analysis according to claim 5, characterized in that: The formula for calculating the employment intention weight of the unemployed position set in S54 is: , the formula for calculating the employment intention weight of the undelivered job set is: , where γ notw is the employment intention weight of the unemployed positions, γ notd is the employment intention weight of the set of undelivered positions.

7. The method for pushing employment information based on user behavior analysis according to claim 1, characterized in that: The employment push model in S6 includes: an unemployed occupation feature extraction unit, an undelivered occupation feature extraction unit, a multiplier M1, a multiplier M2, a multiplier M3, a multiplier M4, a first BiLSTM unit, a second BiLSTM unit, a concatenation unit and a CRF unit; The input end of the unemployed occupation feature extraction unit is used to input the unemployed position set; The input end of the undelivered job feature extraction unit is used to input a set of undelivered positions; The first input end of the multiplier M1 is connected to the output end of the unemployed occupation feature extraction unit, the second input end is used to input the unemployed position preference coefficient set, and the output end is connected to the input end of the first BiLSTM unit; The first input end of the multiplier M2 is connected to the output end of the undelivered occupation feature extraction unit, the second input end is used to input the undelivered occupation preference coefficient set, and the output end is connected to the input end of the second BiLSTM unit; The first input terminal of the multiplier M3 is connected to the output terminal of the first BiLSTM unit, and the second input terminal thereof is used to input the employment intention weight of the unemployed position set; The first input terminal of the multiplier M4 is connected to the output terminal of the second BiLSTM unit, and the second input terminal thereof is used to input the employment intention weight of the undelivered position set; The input end of the splicing unit is connected to the output end of the multiplier M3 and the output end of the multiplier M4 respectively, and the output end thereof is connected to the input end of the CRF unit; The output end of the CRF unit serves as the output end of the employment push model.

8. The method for pushing employment information based on user behavior analysis according to claim 7, characterized in that: The unemployed occupation feature extraction unit and the undelivered occupation feature extraction unit both include: a plurality of LSTM networks and a plurality of fully connected layers; In the unemployed occupation feature extraction unit, each LSTM network is used to input an unemployed position in the unemployed position set to obtain a feature value, and each fully connected layer is used to weight the feature value corresponding to an unemployed position to obtain the unemployed position feature; In the undelivered job feature extraction unit, each LSTM network is used to input an undelivered job in the undelivered job set to obtain a feature value, and each fully connected layer is used to weight the feature value corresponding to an undelivered job to obtain the undelivered job feature; The multiplier M1 is used to multiply the unemployed position preference coefficient set and the output of the unemployed occupation feature extraction unit element by element; The multiplier M2 is used for element-wise multiplication of the undelivered position preference coefficient set and the output of the undelivered occupation feature extraction unit.

Images