A method of job matching based on student preferences

Abstract

The present invention discloses a method for employment matching based on student preferences, which belongs to the field of semantic processing technology. The present invention constructs a first interest value set by analyzing the page reading situation of students under each search content, and forms a second interest value set based on each search content. Subsequently, a dual interest enhanced employment prediction model is used to perform semantic analysis on the interest content set, and at the same time, the attention applied by the first interest value set and the second interest value set is combined to effectively enhance the characteristics of students' interests, thereby significantly improving the accuracy of employment and career matching.

Description

Technical Field

[0001] The present invention relates to the field of semantic processing technology, and in particular to an employment matching method based on student preferences. Background Art

[0002] In today's rapidly developing job market, students are faced with increasingly fierce competition and diverse career choices. Traditional job matching methods often rely on simple keyword searches and static resume analysis, which cannot fully consider students' personalized needs and dynamically changing career preferences. This method often leads to inaccurate matching between students and positions, affecting job search efficiency and career satisfaction.

[0003] Most of the existing job matching technologies are based on static data and lack real-time analysis of students’ active behaviors during the job search process. Many platforms rely only on the resumes and basic information provided by students, ignoring the real interests shown by students during the actual search and browsing process. This single-dimensional matching method often fails to reflect students’ career preferences, resulting in a large gap between recommended positions and students’ expectations.

[0004] In addition, when dealing with students' search behavior, existing technologies often only focus on search keywords, while ignoring students' reading behavior and interest depth under different search contents. Due to the lack of multi-level analysis of search contents, many potential career opportunities cannot be accurately identified and recommended, further exacerbating students' employment difficulties. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a job matching method based on student preferences to solve the problem of low job matching accuracy in the prior art.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for employment matching based on student preferences, comprising the following steps:

[0007] S1. Extract the search content entered by students on the employment platform and construct a search content set in chronological order;

[0008] S2, calculating the first interest value of each search content according to the reading pages of the students under each search content;

[0009] S3, removing search contents whose first interest values ​​are less than a threshold from the search content set, and constructing an interest content set;

[0010] S4. Constructing a first interest value set from the first interest values ​​corresponding to the interest content set;

[0011] S5, calculating a second interest value for each search content in the interest content set, and constructing a second interest value set;

[0012] S6. Use the dual interest enhanced employment prediction model to perform semantic analysis on the interest content set, and obtain matching occupations based on the attention applied by the first interest value set and the second interest value set.

[0013] Further, S2 includes the following sub-steps:

[0014] S21. Under each search content, mark the pages that students have read as reading pages;

[0015] S22, calculating the interest value of the reading page according to the page to which the student jumps when clicking a link on the reading page;

[0016] S23. Under each search content, calculate the first interest value of each search content according to the number of reading pages and the interest value of each reading page.

[0017] Furthermore, the formula for calculating the interest value of the reading page in S22 is: , where I i is the interest value of the i-th reading page, t i,read is the reading time of the i-th reading page, t i,j is the reading time of jumping to the jth page through a link in the i-th reading page, i is the number of the reading page, j is the number of the jump page, and N is the number of jump pages;

[0018] The formula for calculating the first interest value of each search content in S23 is: , where θ 1 is the first interest value of the search content, and M is the number of pages read under each search content.

[0019] Further, S5 includes the following sub-steps:

[0020] S51, counting the frequency of each word in each search content in the interest content set appearing in the interest content set;

[0021] S52, calculating the Jaccard similarity between each search content in the interest content set and the adjacent search content;

[0022] S53, calculating a second interest value according to the frequency of occurrence of each word in the search content and the Jaccard similarity;

[0023] S54: Use the second interest value of each search content in the interest content set as an element to construct a second interest value set.

[0024] Furthermore, the formula for calculating the second interest value in S53 is: , where θ 2 is the second interest value of the search content, S n,m is the Jaccard similarity between the nth search content and the adjacent mth search content, f n,k is the frequency of the kth word in the nth search content appearing in the interest content set, H is the length of the selected adjacent search content, K is the number of words in the search content, n is the number of the search content, m is the number of the adjacent search content, and k is the number of the word.

[0025] Further, the dual interest enhanced employment prediction model in S6 includes: a plurality of search content semantic extraction units, a first splicing layer, a second splicing layer, a first interest enhancement layer E1, a second interest enhancement layer E2, a first BiLSTM layer, a second BiLSTM layer, an adder A1, and a CRF layer;

[0026] The input end of each search content semantic extraction unit is used to input a search content in the interest content set, and the number of search content semantic extraction units is equal to the number of search contents in the interest content set;

[0027] The input end of the first splicing layer is respectively connected to the first output ends of all search content semantic extraction units;

[0028] The input end of the second splicing layer is respectively connected to the second output ends of all search content semantic extraction units;

[0029] A first input end of the first interest enhancement layer E1 is connected to an output end of the first splicing layer, a second input end thereof is used to input a first interest value set, and an output end thereof is connected to an input end of the first BiLSTM layer;

[0030] A first input end of the second interest enhancement layer E2 is connected to the output end of the second splicing layer, a second input end thereof is used to input a second interest value set, and an output end thereof is connected to the input end of the second BiLSTM layer;

[0031] A first input terminal of adder A1 is connected to the output terminal of the first BiLSTM layer, a second input terminal thereof is connected to the output terminal of the second BiLSTM layer, and an output terminal thereof is connected to the input terminal of the CRF layer;

[0032] The output of the CRF layer is used as the output of the dual interest enhanced employment prediction model.

[0033] Further, the search content semantic extraction unit includes: an embedding layer, a first word semantic extraction layer, a second word semantic extraction layer, a first associated semantic extraction layer, and a second associated semantic extraction layer;

[0034] The input end of the embedding layer is used to input the search content, and the output end thereof is connected to the input end of the first word semantic extraction layer and the input end of the second word semantic extraction layer respectively;

[0035] The input end of the first association semantic extraction layer is connected to the output end of the first word semantic extraction layer, and its output end serves as the first output end of the search content semantic extraction unit;

[0036] The input end of the second association semantic extraction layer is connected to the output end of the second word semantic extraction layer, and its output end serves as the second output end of the search content semantic extraction unit.

[0037] Furthermore, the expression of the first word semantic extraction layer is: , where R 1,s is the sth word feature value output by the first word semantic extraction layer, σ is the Sigmoid function, x s,r is the rth element in the sth word vector, ω 1,x,s,r is the first word semantic extraction layer for x s,r The weight assigned, b 1,x,s,r is the first word semantic extraction layer for x s,r The assigned bias, R is the number of elements in the word vector, r is the element number, and s is the number of the word feature value;

[0038] The expression of the second word semantic extraction layer is: , where R 2,s is the sth word feature value output by the second word semantic extraction layer, Tanh is the hyperbolic tangent activation function, ω 2,x,s,r is the semantic extraction layer for x in the second word s,r The weight assigned, b 2,x,s,r is the semantic extraction layer for x in the second word s,r Assigned bias.

[0039] Furthermore, the expression of the first associated semantic extraction layer is: , where Y 1 is the output feature value of the first associated semantic extraction layer, ω R,1,s For R 1,s The weight of b R,1,s For R 1,s , L is the number of word feature values;

[0040] The expression of the second associative semantic extraction layer is: , where Y 2 is the output feature value of the second associated semantic extraction layer, ω R,2,s For R 2,s The weight of b R,2,s For R 2,s The bias.

[0041] Furthermore, the expression of the first interest enhancement layer E1 is: , where X 1 is the first enhanced eigenvector, C 1 is the output vector of the first concatenation layer, D 1 is the first interest value set, is element-wise multiplication;

[0042] The expression of the second interest enhancement layer E2 is: , where X 2 is the second enhanced eigenvector, C 2 is the output vector of the second concatenation layer, D 2 is the second interest value set.

[0043] The beneficial effects of the present invention are:

[0044] 1. By extracting students' search content and reading behavior on the employment platform, the present invention can dynamically reflect students' changes in interest, thereby achieving more accurate career matching and reducing the gap between recommended positions and students' expectations.

[0045] 2. The present invention fully considers the individual needs and career preferences of students, and can provide job recommendations that are more in line with their career development potential based on their actual search and browsing behaviors, thereby improving the pertinence and effectiveness of job hunting.

[0046] 3. The present invention realizes a multi-dimensional evaluation of students' interests by calculating the first interest value and the second interest value, so that the matching process not only depends on keywords, but also comprehensively considers the students' attention to different contents, thereby enhancing the reliability of the matching results.

[0047] 4. The present invention can provide corresponding employment opportunities according to students' preferences more effectively through semantic analysis of the dual interest enhancement model, help students find more positions that match their abilities and interests, and alleviate employment difficulties.

[0048] 5. The present invention constructs a first interest value set by analyzing the page reading of students under each search content, and forms a second interest value set based on each search content. Subsequently, a dual interest enhanced employment prediction model is used to perform semantic analysis on the interest content set, and the attention applied by the first interest value set and the second interest value set is combined to effectively enhance the characteristics of students' interests, thereby significantly improving the accuracy of employment and career matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flowchart of a method for job matching based on student preferences;

[0050] Figure 2Schematic diagram of the structure of the employment prediction model for dual interest enhancement;

[0051] Figure 3 Schematic diagram of the structure of the search content semantic extraction unit. DETAILED DESCRIPTION

[0052] 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.

[0053] like Figure 1 As shown, a method for employment matching based on student preferences includes the following steps:

[0054] S1. Extract the search content entered by students on the employment platform and construct a search content set in chronological order;

[0055] S2, calculating the first interest value of each search content according to the reading pages of the students under each search content;

[0056] S3, removing search contents whose first interest values ​​are less than a threshold from the search content set, and constructing an interest content set;

[0057] S4. Constructing a first interest value set from the first interest values ​​corresponding to the interest content set;

[0058] S5, calculating a second interest value for each search content in the interest content set, and constructing a second interest value set;

[0059] S6. Use the dual interest enhanced employment prediction model to perform semantic analysis on the interest content set, and obtain matching occupations based on the attention applied by the first interest value set and the second interest value set.

[0060] In this embodiment, S2 includes the following sub-steps:

[0061] S21. Under each search content, mark the pages that students have read as reading pages;

[0062] S22, calculating the interest value of the reading page according to the page to which the student jumps when clicking a link on the reading page;

[0063] S23. Under each search content, calculate the first interest value of each search content according to the number of reading pages and the interest value of each reading page.

[0064] The present invention can accurately identify students' actual reading behavior by marking the pages that students have read under each search content. First, the interest value is calculated according to the page that the student jumps to when clicking a link in the reading page to obtain the interest value of the reading page, which reflects the students' attention to the reading page. Then, by combining the number of reading pages and the interest value of each page, the attractiveness of each search content and the students' satisfaction with the search content can be comprehensively evaluated.

[0065] In this embodiment, the types of search content include: position type search, industry search, skill search, geographic location search, etc. For example, for position type search, the search content may be software engineer; for industry search, the search content may be financial industry internship; for skill search, the search content may be digital marketing positions; for geographic location search, the search content may be product manager positions in Beijing.

[0066] In this embodiment, the formula for calculating the interest value of the reading page in S22 is: , where I i is the interest value of the i-th reading page, t i,read is the reading time of the i-th reading page, t i,j is the reading time of jumping to the jth page through a link in the i-th reading page, i is the number of the reading page, j is the number of the jump page, and N is the number of jump pages;

[0067] In this embodiment, the formula for calculating the first interest value of each search content in S23 is: , where θ 1 is the first interest value of the search content, and M is the number of pages read under each search content.

[0068] The present invention counts the time spent on reading each reading page and the time spent on jumping to the page. The longer the time spent, the greater the interest value of the reading page. The interest values ​​of all reading pages under a search content are then counted to obtain the first interest value of the search content.

[0069] In this embodiment, the range of the first interest value is 0-1, so the threshold in step S3 can be set to 0.5, and the specific threshold value can be adjusted according to needs.

[0070] In this embodiment, S5 includes the following sub-steps:

[0071] S51, counting the frequency of each word in each search content in the interest content set appearing in the interest content set;

[0072] S52, calculating the Jaccard similarity between each search content in the interest content set and the adjacent search content;

[0073] S53, calculating a second interest value according to the frequency of occurrence of each word in the search content and the Jaccard similarity;

[0074] S54: Use the second interest value of each search content in the interest content set as an element to construct a second interest value set.

[0075] In this embodiment, step S52 specifically includes: performing word segmentation on the search content, the search content corresponds to a word set, and the Jaccard similarity between a word set and an adjacent word set is calculated to reflect the existence of the same words in the two word sets.

[0076] In this embodiment, cosine similarity may also be used in step S52.

[0077] In this embodiment, the formula for calculating the second interest value in S53 is: , where θ 2 is the second interest value of the search content, S n,m is the Jaccard similarity between the nth search content and the adjacent mth search content, f n,k is the frequency of the kth word in the nth search content appearing in the interest content set, H is the length of the selected adjacent search content, K is the number of words in the search content, n is the number of the search content, m is the number of the adjacent search content, and k is the number of the word.

[0078] In this embodiment, H is equal to 3, 4 or 5, etc. When H is equal to 3, the Jaccard similarity between the current search content and the three adjacent search contents is calculated.

[0079] The present invention counts the frequency of each word in each search content appearing in the interest content set. The higher the frequency, the higher the importance of the corresponding word. The present invention calculates the Jaccard similarity between each search content and the adjacent search content. Since step S1 arranges the search content in chronological order, the Jaccard similarity can reflect the relevance of the students in the input search content. The greater the Jaccard similarity, the higher the students' attention to the search content. Combining the frequency of the word and the Jaccard similarity, the second interest value is calculated to reflect the interest value of the students in the search content itself.

[0080] like Figure 2 As shown, the dual interest enhanced employment prediction model in S6 includes: multiple search content semantic extraction units, a first splicing layer, a second splicing layer, a first interest enhancement layer E1, a second interest enhancement layer E2, a first BiLSTM layer, a second BiLSTM layer, an adder A1 and a CRF layer;

[0081] The input end of each search content semantic extraction unit is used to input a search content in the interest content set, and the number of search content semantic extraction units is equal to the number of search contents in the interest content set;

[0082] The input end of the first splicing layer is respectively connected to the first output ends of all search content semantic extraction units;

[0083] The input end of the second splicing layer is respectively connected to the second output ends of all search content semantic extraction units;

[0084] A first input end of the first interest enhancement layer E1 is connected to an output end of the first splicing layer, a second input end thereof is used to input a first interest value set, and an output end thereof is connected to an input end of the first BiLSTM layer;

[0085] A first input end of the second interest enhancement layer E2 is connected to the output end of the second splicing layer, a second input end thereof is used to input a second interest value set, and an output end thereof is connected to the input end of the second BiLSTM layer;

[0086] A first input terminal of adder A1 is connected to the output terminal of the first BiLSTM layer, a second input terminal thereof is connected to the output terminal of the second BiLSTM layer, and an output terminal thereof is connected to the input terminal of the CRF layer;

[0087] The output of the CRF layer is used as the output of the dual interest enhanced employment prediction model.

[0088] The present invention processes a search content through a search content semantic extraction unit to generate two feature outputs. First, all the first feature outputs are spliced ​​into a vector using the first splicing layer, and then all the second feature outputs are spliced ​​into another vector through the second splicing layer. Then, the first interest enhancement layer E1 and the second interest enhancement layer E2 enhance the two vectors respectively to strengthen the students' interest direction. The dual interest enhancement mechanism ensures that the model can make full use of the students' interest characteristics. Next, the first BiLSTM layer processes the features strengthened by the first interest value set, while the second BiLSTM layer processes the features strengthened by the second interest value set. This process further enhances the model's attention to students' interests and ensures that the recommendation results are more in line with students' actual needs. Finally, the adder A1 effectively fuses the outputs from the two BiLSTM layers to ensure the full use of features from different sources, thereby improving the accuracy of employment and occupation matching.

[0089] like Figure 3 As shown, the search content semantic extraction unit includes: an embedding layer, a first word semantic extraction layer, a second word semantic extraction layer, a first associated semantic extraction layer, and a second associated semantic extraction layer;

[0090] The input end of the embedding layer is used to input the search content, and the output end thereof is connected to the input end of the first word semantic extraction layer and the input end of the second word semantic extraction layer respectively;

[0091] The input end of the first association semantic extraction layer is connected to the output end of the first word semantic extraction layer, and its output end serves as the first output end of the search content semantic extraction unit;

[0092] The input end of the second association semantic extraction layer is connected to the output end of the second word semantic extraction layer, and its output end serves as the second output end of the search content semantic extraction unit.

[0093] In this embodiment, the expression of the first word semantic extraction layer is: , where R 1,s is the sth word feature value output by the first word semantic extraction layer, σ is the Sigmoid function, x s,r is the rth element in the sth word vector, ω 1,x,s,r is the first word semantic extraction layer for x s,r The weight assigned, b 1,x,s,r is the first word semantic extraction layer for x s,r The assigned bias, R is the number of elements in the word vector, r is the element number, and s is the number of the word feature value;

[0094] The expression of the second word semantic extraction layer is: , where R 2,s is the sth word feature value output by the second word semantic extraction layer, Tanh is the hyperbolic tangent activation function, ω 2,x,s,r is the semantic extraction layer for x in the second word s,r The weight assigned, b 2,x,s,r is the semantic extraction layer for x in the second word s,r Assigned bias.

[0095] The first word semantic extraction layer and the second word semantic extraction layer are used to extract word feature values ​​of the input word vector. Different activation functions are used in the two word semantic extraction layers. The model can extract features from different angles. The first layer pays more attention to the activation degree of the features, while the second layer can capture the relative strength and directionality of the features, making the feature expression diverse.

[0096] In this embodiment, the expression of the first associated semantic extraction layer is: , where Y 1 is the output feature value of the first associated semantic extraction layer, ω R,1,s For R 1,s The weight of b R,1,s For R 1,s , L is the number of word feature values;

[0097] The expression of the second associative semantic extraction layer is: , where Y 2 is the output feature value of the second associated semantic extraction layer, ω R,2,s For R 2,s The weight of b R,2,s For R 2,s The bias.

[0098] In the associated semantic extraction layer, the word feature values ​​output by the previous word semantic extraction layer are combined to obtain the corresponding output feature values, which reflect the semantic features of a search content. The first splicing layer and the second splicing layer splice the semantic features corresponding to all search contents in the interest content set.

[0099] In this embodiment, the expression of the first interest enhancement layer E1 is: , where X 1 is the first enhanced eigenvector, C 1 is the output vector of the first concatenation layer, D 1 is the first interest value set, is element-wise multiplication;

[0100] The expression of the second interest enhancement layer E2 is: , where X 2 is the second enhanced eigenvector, C 2 is the output vector of the second concatenation layer, D 2 is the second interest value set.

[0101] Each cell unit in the first BiLSTM layer inputs an element of the first enhanced feature vector, and each cell unit in the second BiLSTM layer inputs an element of the second enhanced feature vector.

[0102] In this embodiment, the weights and biases in the dual interest enhanced employment prediction model are trained using the existing gradient descent method.

[0103] By extracting students' search content and reading behavior on the employment platform, the present invention can dynamically reflect students' changes in interest, thereby achieving more accurate career matching and reducing the gap between recommended positions and students' expectations.

[0104] The present invention fully considers the individual needs and career preferences of students, and can provide job recommendations that are more in line with their career development potential based on their actual search and browsing behaviors, thereby improving the pertinence and effectiveness of job hunting.

[0105] The present invention realizes a multi-dimensional evaluation of students' interests by calculating the first interest value and the second interest value, so that the matching process not only depends on keywords, but also comprehensively considers the students' attention to different contents, thereby enhancing the reliability of the matching results.

[0106] The present invention can provide corresponding employment opportunities according to students' preferences more effectively through semantic analysis of the dual interest enhancement model, help students find more positions that match their abilities and interests, and alleviate employment difficulties.

[0107] The present invention constructs a first interest value set by analyzing the page reading of students under each search content, and forms a second interest value set based on each search content. Subsequently, a dual interest enhanced employment prediction model is used to perform semantic analysis on the interest content set, and the attention applied by the first interest value set and the second interest value set is combined to effectively enhance the characteristics of students' interests, thereby significantly improving the accuracy of employment and career matching.

[0108] 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 job matching based on student preferences, characterized in that: The following steps are involved: S1. Extract the search content entered by students on the employment platform and construct a search content set in chronological order; S2, calculating the first interest value of each search content according to the reading pages of the students under each search content; S3, removing search contents whose first interest values ​​are less than a threshold from the search content set, and constructing an interest content set; S4. Constructing a first interest value set from the first interest values ​​corresponding to the interest content set; S5, calculating a second interest value for each search content in the interest content set, and constructing a second interest value set; S6. Use a dual interest enhanced employment prediction model to perform semantic analysis on the interest content set, and obtain matching occupations based on the attention applied by the first interest value set and the second interest value set, wherein the dual interest enhanced employment prediction model includes: multiple search content semantic extraction units, a first splicing layer, a second splicing layer, a first interest enhancement layer E1, a second interest enhancement layer E2, a first BiLSTM layer, a second BiLSTM layer, an adder A1 and a CRF layer; The input end of each search content semantic extraction unit is used to input a search content in the interest content set, and the number of search content semantic extraction units is equal to the number of search contents in the interest content set; the input end of the first splicing layer is respectively connected to the first output ends of all search content semantic extraction units; the input end of the second splicing layer is respectively connected to the second output ends of all search content semantic extraction units; the first input end of the first interest enhancement layer E1 is connected to the output end of the first splicing layer, its second input end is used to input the first interest value set, and its output end is connected to the input end of the first BiLSTM layer; the first input end of the second interest enhancement layer E2 is connected to the output end of the second splicing layer, its second input end is used to input the second interest value set, and its output end is connected to the input end of the second BiLSTM layer; the first input end of the adder A1 is connected to the output end of the first BiLSTM layer, its second input end is connected to the output end of the second BiLSTM layer, and its output end is connected to the input end of the CRF layer; the output end of the CRF layer serves as the output end of the dual interest enhancement employment prediction model.

2. The employment matching method based on student preferences according to claim 1, characterized in that: The S2 comprises the following sub-steps: S21. Under each search content, mark the pages that students have read as reading pages; S22, calculating the interest value of the reading page according to the page to which the student jumps when clicking a link on the reading page; S23. Under each search content, calculate the first interest value of each search content according to the number of reading pages and the interest value of each reading page.

3. The employment matching method based on student preferences according to claim 2, characterized in that: The formula for calculating the interest value of the reading page in S22 is: , where I i is the interest value of the i-th reading page, t i,read is the reading time of the i-th reading page, t i,j is the reading time of jumping to the jth page through a link in the i-th reading page, i is the number of the reading page, j is the number of the jump page, and N is the number of jump pages; The formula for calculating the first interest value of each search content in S23 is: , where θ1 is the first interest value of the search content, and M is the number of pages read under each search content.

4. The employment matching method based on student preferences according to claim 1, characterized in that: The S5 comprises the following sub-steps: S51, counting the frequency of each word in each search content in the interest content set appearing in the interest content set; S52, calculating the Jaccard similarity between each search content in the interest content set and the adjacent search content; S53, calculating a second interest value according to the frequency of occurrence of each word in the search content and the Jaccard similarity; S54: Use the second interest value of each search content in the interest content set as an element to construct a second interest value set.

5. The employment matching method based on student preferences according to claim 4, characterized in that: The formula for calculating the second interest value in S53 is: , where θ2 is the second interest value of the search content, S n,m is the Jaccard similarity between the nth search content and the adjacent mth search content, f n,k is the frequency of the kth word in the nth search content appearing in the interest content set, H is the length of the selected adjacent search content, K is the number of words in the search content, n is the number of the search content, m is the number of the adjacent search content, and k is the number of the word.

6. The employment matching method based on student preferences according to claim 1, characterized in that: The search content semantic extraction unit includes: an embedding layer, a first word semantic extraction layer, a second word semantic extraction layer, a first associated semantic extraction layer, and a second associated semantic extraction layer; The input end of the embedding layer is used to input the search content, and the output end thereof is connected to the input end of the first word semantic extraction layer and the input end of the second word semantic extraction layer respectively; The input end of the first association semantic extraction layer is connected to the output end of the first word semantic extraction layer, and its output end serves as the first output end of the search content semantic extraction unit; The input end of the second association semantic extraction layer is connected to the output end of the second word semantic extraction layer, and its output end serves as the second output end of the search content semantic extraction unit.

7. The method for job matching based on student preferences according to claim 6, characterized in that: The expression of the first word semantic extraction layer is: , where R 1,s is the sth word feature value output by the first word semantic extraction layer, σ is the Sigmoid function, x s,r is the rth element in the sth word vector, ω 1,x,s,r is the first word semantic extraction layer for x s,r The weight assigned, b 1,x,s,r is the first word semantic extraction layer for x s,r The assigned bias, R is the number of elements in the word vector, r is the element number, and s is the number of the word feature value; The expression of the second word semantic extraction layer is: , where R 2,s is the sth word feature value output by the second word semantic extraction layer, Tanh is the hyperbolic tangent activation function, ω 2,x,s,r is the semantic extraction layer for x in the second word s,r The weight assigned, b 2,x,s,r is the semantic extraction layer for x in the second word s,r Assigned bias.

8. The method for job matching based on student preferences according to claim 7, characterized in that: The expression of the first association semantic extraction layer is: , where Y1 is the output feature value of the first associated semantic extraction layer, ω R,1,s For R 1,s The weight of b R,1,s For R 1,s The bias of L is the number of word feature values; the expression of the second associative semantic extraction layer is: , where Y2 is the output feature value of the second associated semantic extraction layer, ω R,2,s For R 2,s The weight of b R,2,s For R 2,s The bias.

9. The method for job matching based on student preferences according to claim 1, characterized in that: The expression of the first interest enhancement layer E1 is: , where X1 is the first enhanced feature vector, C1 is the output vector of the first concatenated layer, and D1 is the first interest value set. is element-wise multiplication; The expression of the second interest enhancement layer E2 is: , where X2 is the second enhanced feature vector, C2 is the output vector of the second concatenated layer, and D2 is the second interest value set.

Images