Resume classification method, device, electronic device and storage medium

Abstract

This application relates to natural language processing in the field of artificial intelligence, and specifically discloses a resume classification method, device, electronic device, and storage medium. The method includes: obtaining a resume to be classified; extracting resume feature words from the resume to be classified; generating a feature coding vector for the resume to be classified based on the resume feature words; calculating the association probability between each sample resume and the resume to be classified based on the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in a resume library; determining a target sample resume that matches the resume to be classified based on the association probability between each sample resume and the resume to be classified; and determining the job resume set to which the resume to be classified belongs based on the target job category corresponding to the target sample resume. This solution realizes automatic resume classification and improves resume classification efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing of artificial intelligence, and more specifically, to a resume classification method, device, electronic device and storage medium. Background Art

[0002] During a company's recruitment process, after receiving a job applicant's resume, HR staff must manually sort the resume into a collection of resumes for the appropriate job category, or filter out unqualified resumes. This facilitates subsequent resume screening and the selection of candidates for interviews. Manual resume sorting is time-consuming and labor-intensive, resulting in low efficiency. Summary of the Invention

[0003] In view of the above problems, the embodiments of the present application propose a resume classification method, device, electronic device and storage medium to solve the problem of low efficiency of resume classification in the prior art.

[0004] According to one aspect of an embodiment of the present application, a resume classification method is provided, the method comprising: obtaining a resume to be classified; extracting resume feature words from the resume to be classified; generating a feature coding vector for the resume to be classified based on the resume feature words; calculating an association probability between each of the sample resumes and the resume to be classified based on the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in a resume library; determining a target sample resume that matches the resume to be classified based on the association probability between each of the sample resumes and the resume to be classified; and determining a job resume set to which the resume to be classified belongs based on a target job category corresponding to the target sample resume.

[0005] According to one aspect of an embodiment of the present application, a resume classification device is provided, which includes: an acquisition module for acquiring resumes to be classified; a resume feature word extraction module for extracting resume feature words from the resumes to be classified; a feature coding vector generation module for generating feature coding vectors of the resumes to be classified based on the resume feature words; an association probability calculation module for calculating the association probability between each sample resume and the resume to be classified based on the feature coding vector of the resume to be classified and the feature coding vectors of each sample resume in a resume library; a target sample resume determination module for determining a target sample resume that matches the resume to be classified based on the association probability between each sample resume and the resume to be classified; and a job resume set determination module for determining the job resume set to which the resume to be classified belongs based on the target job category corresponding to the target sample resume.

[0006] In some embodiments of the present application, based on the aforementioned scheme, the feature coding vector generation module includes: a feature word vector calculation unit, used to calculate the feature word vector corresponding to the resume feature word; and a character vector calculation unit, used to calculate the character vector of each character in the resume feature word; a splicing unit, used to splice the resume feature word with the character vector of each character in the resume feature word respectively, to obtain at least two splicing vectors corresponding to the resume feature word; a combination unit, used to combine at least two splicing vectors corresponding to all the resume feature words in the resume to be classified, to obtain the feature coding vector of the resume to be classified.

[0007] In some embodiments of the present application, based on the aforementioned scheme, the association probability calculation module includes: an input unit, used to input the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume library into the association prediction model; an association probability prediction unit, used to perform association probability prediction based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume by the association prediction model, to obtain the association probability between the sample resume and the resume to be classified.

[0008] In some embodiments of the present application, based on the aforementioned scheme, the association prediction model includes a converter network layer, a fusion processing layer and a classification layer; in this embodiment, the association probability prediction unit includes: a correlation feature extraction unit, which is used by the converter network layer to extract correlation features based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume to obtain the correlation features between the resume to be classified and the sample resume; a fusion processing unit, which is used by the fusion processing layer to fuse the correlation features to obtain fusion features; an association probability classification unit, which is used by the classification layer to perform association probability classification based on the fusion features to obtain the association probability between the sample resume and the resume to be classified.

[0009] In some embodiments of the present application, based on the aforementioned scheme, the resume feature word extraction module includes: a word type determination unit, used to determine the word type of each candidate word in the resume to be classified according to a preset word classification strategy; an importance weight calculation unit, used to calculate the importance weight corresponding to each candidate word according to the word category to which each candidate word belongs; candidate word screening, used to screen candidate words according to the importance weight corresponding to each candidate word, and determine the first target candidate word; a first resume feature word determination unit, used to determine the first target candidate word as the resume feature word.

[0010] In some embodiments of the present application, based on the aforementioned scheme, the word category includes subtypes under at least two classification dimensions; the importance weight calculation unit includes: a subweight calculation unit, used to calculate the subweight of each candidate word under each classification dimension according to the subtype to which each candidate word belongs under each classification dimension; a calculation unit, used to calculate the importance weight corresponding to each candidate word according to the subweight of each candidate word under each classification dimension.

[0011] In other embodiments of the present application, based on the aforementioned scheme, the resume feature word extraction module includes: a search unit, used to search for candidate words in the resume to be classified in the job category keyword library, and determine a second target candidate word located in the job category keyword library; the job category keyword library is constructed based on the words in the sample resumes in the resume library; a second resume feature word determination unit, used to determine the second target candidate word as the resume feature word.

[0012] In some embodiments of the present application, based on the aforementioned scheme, the resume database includes a sub-resume database corresponding to each job category; in this embodiment, the resume classification device also includes: a word frequency calculation module, used to calculate the word frequency of each first candidate word in all sample resumes in the sub-resume database; the first candidate word is a candidate word in the sample resume; an inverse document frequency index calculation module, used to calculate the inverse document frequency index of each first candidate word in the resume database; a first weight calculation module, used to calculate the first weight corresponding to each first candidate word according to the word frequency of each first candidate word and the inverse document frequency index of each first candidate word; a first candidate word screening module, used to screen the first candidate word according to the first weight corresponding to each first candidate word; a first adding module, used to add the screened first candidate word to the job category keyword library.

[0013] In some embodiments of the present application, based on the aforementioned scheme, the resume feature word extraction module also includes: a named entity recognition module, which is used to perform named entity recognition on each candidate word in the resume to be classified, and determine the candidate word identified as a named entity; a third resume feature word determination module, which is used to determine the candidate word identified as a named entity as the resume feature word.

[0014] In some embodiments of the present application, based on the aforementioned scheme, the resume classification device further includes: a block module, which is used to block the resume to be classified according to the writing framework of the resume to be classified, and determine the resume blocks included in the resume to be classified and the block resume texts corresponding to each resume block; a first filtering module, which is used to filter the resume blocks included in the resume to be classified, and determine the candidate resume blocks; a second filtering module, which is used to filter information on the block resume texts corresponding to the candidate resume blocks, and determine the candidate texts; a candidate word determination module, which is used to determine the words in the candidate text as candidate words of the resume to be classified.

[0015] In some embodiments of the present application, based on the aforementioned solution, the target sample resume determination module includes: a first determination unit for determining, based on the association probabilities between each sample resume and the resume to be classified, the sample resume corresponding to the maximum association probability; and a second determination unit for determining the sample resume corresponding to the maximum association probability as the target sample resume.

[0016] In some embodiments of the present application, based on the aforementioned scheme, the resume classification device also includes: a second adding module, which is used to determine the resume to be classified as an unqualified resume if the association probability between each of the sample resumes and the resume to be classified is less than the association probability threshold, and add the resume to be classified to the specified resume collection.

[0017] According to one aspect of an embodiment of the present application, an electronic device is provided, including: a processor; a memory, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processor, the resume classification method described above is implemented.

[0018] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the resume classification method described above is implemented.

[0019] In the scheme of the present application, resume feature words are extracted from the resume to be classified, and a feature coding vector of the resume to be classified is generated based on the resume feature words. Then, the correlation between the resume to be classified and the sample resume is mined based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume in the resume library. Then, the job resume set to which the resume to be classified belongs is determined based on the job category corresponding to the target sample resume that has a high correlation with the resume to be classified. This realizes automatic classification of resumes according to job categories based on the job categories corresponding to the sample resumes in the known resume library and the correlation between the resume to be classified and the sample resume, thereby improving the efficiency of resume classification and realizing automatic job category matching for resumes. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification, are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0021] Figure 1 A schematic diagram shows an exemplary system architecture to which the technical solutions of the embodiments of the present application can be applied.

[0022] Figure 2 A flowchart of a resume classification method according to an embodiment of the present application is shown.

[0023] Figure 3 FIG. 2 is a flowchart of step 230 according to an embodiment of the present application.

[0024] Figure 4 It is a schematic diagram showing the concatenation of word feature vectors and character vectors according to a specific embodiment of the present application.

[0025] Figure 5 This is a flowchart of performing association probability prediction according to an embodiment of the present application.

[0026] Figure 6 is a schematic structural diagram of a converter neural network according to a specific embodiment.

[0027] Figure 7 It is a structural diagram of an association prediction model according to a specific embodiment of the present application.

[0028] Figure 8 FIG. 2 is a flowchart of step 220 according to an embodiment of the present application.

[0029] Figure 9 This is a flowchart of constructing a job category keyword library according to an embodiment of the present application.

[0030] Figure 10 This is a flowchart of the steps before step 220 according to an embodiment of the present application.

[0031] Figure 11 It is a flowchart of a resume classification method according to a specific embodiment of the present application.

[0032] Figure 12 It is a block diagram of a resume classification device according to an embodiment of the present application.

[0033] Figure 13A schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0034] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0035] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner.In the following description, many specific details are provided so as to provide a full understanding of the embodiments of the present application. However, it will be appreciated by those skilled in the art that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps etc. can be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0036] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be executed in the order described. For example, some operations / steps may be further decomposed, while others may be combined or partially combined, so the actual execution order may vary depending on the actual situation.

[0037] It should be noted that the term "plurality" used in this document refers to two or more. "And / or" describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. The character " / " generally indicates an "or" relationship between the associated objects.

[0038] During a company's recruitment process, after receiving a job applicant's resume, HR staff must manually sort it into a collection of resumes for the appropriate job category, or screen out unqualified resumes. This facilitates subsequent resume screening and the selection of candidates for interviews. Manual resume sorting is time-consuming and labor-intensive, resulting in low efficiency. In this solution, natural language processing technology is applied to resume sorting to improve both efficiency and accuracy.

[0039] Natural language processing (NLP) is a key area of ​​research in computer science and artificial intelligence. It studies the theories and methods that enable effective communication between humans and computers using natural language. Natural language processing (NLP) integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language we use in everyday life—and is closely linked to the study of linguistics. Natural language processing technologies typically include text processing, semantic understanding, machine translation, robotic question answering, and knowledge graphs.

[0040] Figure 1 Schematic diagram showing an exemplary system architecture to which the technical solution of the embodiment of the present application can be applied. Figure 1 As shown, the system architecture may include a terminal 110 and a server 120 in communication with the terminal 110. The communication connection established between the terminal 110 and the server 120 may be a wired or wireless network connection.

[0041] The terminal 110 may be a smart phone, tablet computer, laptop computer, desktop computer, or other electronic device that can interact with a user. The terminal 110 may display a first user interface, in which a resume classification control is provided. The user may trigger the resume classification control to send a resume classification request to the server 120. The resume classification request indicates the resume identification information of the resume to be classified. Thus, after receiving the resume classification request, the server 120 obtains the corresponding resume to be classified according to the resume identification information, classifies the resume to be classified, determines the job resume set to which the resume to be classified belongs, and accordingly adds the resume to be classified to the determined job resume set.

[0042] In some embodiments of the present application, the user can also set a job resume set according to the job category set by the company in the user interface provided by terminal 110. One job category uniquely corresponds to one job resume set. Therefore, after receiving a resume, the job resume set to which the resume belongs is determined from among the pre-set multiple job resume sets. It is understandable that because different companies set different positions and require different positions, the job resume sets set by users of different companies may differ.

[0043] The server 120 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. It is worth mentioning that the resume classification method provided in this application is not limited to being executed by the server 120. In other embodiments, the method of this application may also be executed by the terminal 110, or the terminal 110 and the server 120 may interact to implement the method of this application to achieve automatic resume classification.

[0044] The following is a detailed description of the implementation details of the technical solution of the embodiment of the present application:

[0045] Figure 2 The flowchart of the resume classification method according to one embodiment of the present application is shown. The method can be executed by a computer device with processing capabilities, such as a server, a terminal with processing capabilities, etc., which is not specifically limited here. Figure 2 As shown, the method includes at least steps 210 to 260, which are described in detail as follows:

[0046] Step 210: Obtain the resume to be classified.

[0047] Step 220: extract resume feature words from the resume to be classified.

[0048] Resume feature words are words extracted from the resume to be classified that have the function of distinguishing job categories. In some embodiments of the present application, a feature word library can be pre-built, and the words in this feature word library are words that have the function of distinguishing job categories. Based on this, the words in the resume to be classified can be searched in the feature word library. If the word is found, it can be determined as a resume feature word. For example, artificial intelligence, machine learning, data mining, AI, big data, data analysis, deep learning, neural networks, language models, Python, SCI papers, etc. can be added to the feature word library.

[0049] In some embodiments of the present application, the resume feature words may be keywords in the education section, the work experience section, the project experience section, the professional skills section, etc. The resume feature words extracted in step 220 may be one or more. In a specific embodiment, to improve the accuracy of resume classification, the resume feature words are multiple.

[0050] It is understandable that in step 220, the text content in the resume to be classified needs to be segmented, and resume feature words are extracted based on the segmentation. Since the segmentation result directly affects the accuracy of the extracted resume feature words, the resume as a sample can be used to train the segmentation tool so that the segmentation tool is suitable for segmenting the text content in the resume. For example, for the word "data analysis", the commonly used segmentation tool may divide the word into two words, namely "data" and "analysis", and in order to classify resumes according to job categories, using "data analysis" as a word has a more obvious effect on distinguishing job categories. Therefore, before step 220, the segmentation tool is trained in a targeted manner using the resume as a sample so that the segmentation tool is suitable for segmenting the text content in the resume.

[0051] Step 230: Generate a feature coding vector of the resume to be classified based on the resume feature words.

[0052] The feature encoding vector is a vectorized representation of a resume's characteristic words. If the resume to be classified includes multiple resume characteristic words, you can first encode each resume characteristic word separately to obtain a word feature vector for each resume characteristic word. Then, you can concatenate the word feature vectors corresponding to the multiple resume characteristic words in the resume to be classified and use the concatenated vector as the feature encoding vector for the resume to be classified.

[0053] In some embodiments of the present application, each resume feature word can be encoded by one-hot encoding to obtain a word feature vector for each resume feature word. In other embodiments, the embedding vector of the resume feature word can be output by a trained first word vector model, and the embedding vector of the resume feature word can be used as the word feature vector of the resume feature word. The first word vector model can be trained by a word2vec (word vector) model.

[0054] It can be understood that a resume feature word includes at least two characters, and the semantics of the characters in the resume feature word may be quite different from the semantics of the resume feature word. In some embodiments of the present application, in order to make full use of the information expressed by the resume feature word, the resume feature word can be further divided into characters, and a feature coding vector of the resume to be classified is generated based on the characters included in the resume feature word and the resume feature word.

[0055] In some embodiments, as Figure 3As shown, step 230 further includes: step 310, calculating the feature word vector corresponding to the resume feature word; and step 320, calculating the word vector of each character in the resume feature word; step 330, splicing the resume feature word with the word vectors of each character in the resume feature word respectively to obtain at least two spliced vectors corresponding to the resume feature word; step 340, combining at least two spliced vectors corresponding to all the resume feature words in the to-be-classified resume to obtain the feature encoding vector of the to-be-classified resume.

[0056] The word vector is the vectorized representation of a word. In some embodiments of the present application, the word2vec (word vector) model can be trained respectively to obtain a first word vector model with words as single items and a first word vector model with characters as single items. Then, the resume feature word is input into the first word vector model, and the word feature vector of the resume feature word is output by the first word vector model; each character in the resume feature word is input into the first word vector model, and the word vector of each character in the resume feature word is output by the first word vector model.

[0057] For each resume feature word, it includes at least two characters. Therefore, in step 330, the word feature vector of the resume feature word is spliced with the word vectors of each character in the resume feature word respectively to obtain at least two spliced vectors corresponding to the resume feature word. It can be understood that the number of spliced vectors obtained for each resume feature word is equal to the number of characters included in the resume feature word.

[0058] Figure 4 is a schematic diagram showing the splicing of the word feature vector and the word vector according to a specific embodiment of the present application. For example, if the resume feature word includes "data mining", the characters in the resume feature word "data mining" include numbers, data, digging, and excavation. After obtaining the word feature vector A of the resume feature word "data mining" and the word vectors of each character (the word vector of "number" is B1, the word vector of "data" is B2, the word vector of "digging" is B3, and the word vector of "excavation" is B4), according to Figure 4 As shown, the word feature vector A corresponding to "data mining" is spliced with the word vectors B1, B2, B3, and B4 respectively to obtain four spliced vectors.

[0059] In some embodiments of the present application, in step 340, it can be to perform secondary splicing on the spliced vectors obtained for all the resume feature words, and use the vector obtained by the secondary splicing as the feature encoding vector of the to-be-classified resume.

[0060] In this solution, since the word feature vector of the resume feature word is mixed with the word vectors of each character in the resume feature word, the information expressed by the resume feature word is fully utilized, providing more information for subsequent associated probability prediction.

[0061] Please continue reading Figure 2 , step 240, calculate the association probability between each sample resume and the resume to be classified based on the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume library.

[0062] The sample resume refers to a resume in a resume library. In some embodiments, the resume library may include qualified resumes under multiple job categories, thereby using the sample resumes in the resume library as the data basis for classifying resumes according to job categories.

[0063] In a specific embodiment, the resume database can be constructed based on the resumes of current and former employees of the company, thereby tailoring the resume database to the company. Since the resume database is constructed based on the resumes of the company's employees, it can be ensured that the resumes include resumes for each position category set by the company, ensuring that the resume database matches the company's set position categories. It is understandable that the position categories covered by the sample resumes in the resume database may vary from company to company.

[0064] In one specific embodiment, each position in a company is defined as a position category. Resumes of employees who have previously held or are currently employed in that position are then divided into positive and negative sample sets for each position category according to preset rules. Resumes in the positive sample set are qualified resumes for the corresponding position category, also known as positive samples; resumes in the negative sample set are unqualified resumes for the corresponding position category, also known as negative samples. Resumes can then be selected from the positive sample set and added to the resume database. For example, all resumes in the positive sample set can be added to the resume database, or a subset of resumes can be selected from the positive sample set and added to the resume database.

[0065] In some embodiments of the present application, the preset rule may be a rule set based on the assessment results of current employees and past employees in the company, and / or their years of service in the company. For example, the resumes of employees whose number of qualified assessment results in the company accounts for no less than a first threshold of all assessment results may be set as positive samples; or the resumes of employees whose time in the company exceeds a first preset number of years may be set as positive samples.

[0066] In a specific embodiment, the preset rule may be: the resumes of employees who have been in the position for more than a preset number of years (for example, the preset number of years is set to 3 years) and the number of times they were qualified during their tenure as a percentage of all assessment times is not less than a first assessment threshold (for example, the first assessment threshold is set to 1, 0.8, 0.9, etc.) are used as positive samples; the resumes of employees who left the position within a set number of years (which can be set to 2 years) or who were in the position but the number of times they were unqualified during their tenure as a percentage of all assessment times is less than a second assessment threshold (wherein the second assessment threshold is not greater than the first assessment threshold, for example, the second assessment threshold is set to 0.5) are used as negative samples.

[0067] Then, based on the positive samples identified and the job categories corresponding to them, a resume database is constructed. In this database, the sample resumes serving as positive samples are associated with the job categories corresponding to them. For example, they are stored in the form of a binary array of <job category, qualified resume>. Job categories can be labeled with category IDs, such as 0, 1, 2, etc. The constructed resume database can be shown in Table 1 below:

[0068] Table 1

[0069]

[0070]

[0071] In a specific embodiment, in order to ensure the accuracy of resume classification, a resume quantity threshold can be set, and resumes for each job category can be screened from the positive sample set according to the set quantity threshold and added to the resume library, so that the number of sample resumes included in each job category in the resume library is not less than the resume quantity threshold.

[0072] It is understandable that the feature coding vector of each sample resume in the resume database is also determined in the same way as the feature coding vector of the resume to be classified, and the process of determining the feature coding vector of the sample will not be described in detail here.

[0073] In some embodiments of the present application, a similarity calculation may be performed between the feature coding vector of the resume to be classified and the feature coding vector of the sample resume, and the calculated similarity is determined as the feature coding vector of the resume to be classified and the feature coding vector of the sample resume. The similarity calculation may be, for example, cosine similarity, Euclidean distance, etc., which may be specifically set as needed.

[0074] In some embodiments of the present application, since the feature coding vector of the resume to be classified is the initial vectorized representation of the resume feature words in the resume to be classified, and there may be correlation between different resume feature words, the association probability calculation can be performed after deep feature extraction of the feature coding vector of the resume to be classified.

[0075] In some embodiments of the present application, the trained association probability prediction model can also be used to calculate the association probability between the feature coding vector of the resume to be classified and the sample resume based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume. In this case, step 240 further includes: inputting the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume library into the association prediction model; and the association prediction model performs association probability prediction based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume to obtain the association probability between the sample resume and the resume to be classified.

[0076] Wherein, the association prediction model is obtained by training the initial neural network with training data, and the initial neural network can be based on convolutional neural network (CNN), long short-term memory network (LSTM, Long Short-Term Memory), neural network based on transformer network (Transformer model), etc., which are not specifically limited here. In a specific embodiment, the training data of the association prediction model can be the resumes in the positive sample set and negative sample set constructed according to the resumes of salt workers in the company as above.

[0077] Step 250: Determine a target sample resume that matches the resume to be classified based on the association probability between each sample resume and the resume to be classified.

[0078] In some embodiments of the present application, step 250 may include: determining the sample resume corresponding to the maximum association probability based on the association probability between each of the sample resumes and the resumes to be classified; and determining the sample resume corresponding to the maximum association probability as the target sample resume.

[0079] In some embodiments of the present application, a sample resume whose corresponding association probability exceeds a preset association probability threshold may be determined as a target sample resume. In this case, there may be multiple target sample resumes.

[0080] Step 260: Determine the job resume set to which the resume to be classified belongs based on the target job category corresponding to the target sample resume.

[0081] The target job category refers to the job category corresponding to the target sample resume. As described above, the sample resumes in the resume library are qualified resumes in the corresponding job category. Therefore, the sample resumes are associated with the job category.

[0082] This scheme classifies resumes according to job categories. Since the probability of association between the target sample resume and the resume to be classified is high, the abilities, experiences and other relevant requirements of the corresponding job seekers reflected in the resume to be classified are similar to those reflected in the target sample resume. Therefore, the job resume set to which the resume to be classified belongs can be determined based on the job category corresponding to the target sample resume.

[0083] In some embodiments of the present application, if the determined target sample resume is one, the job resume set corresponding to the target job category corresponding to the target sample resume can be determined as the job resume set to which the resume to be classified belongs.

[0084] In some embodiments of the present application, if there are multiple target sample resumes determined, the target job categories corresponding to the multiple target sample resumes are determined based on the target job categories corresponding to each target sample resume, and the number of target sample resumes corresponding to each target job category is determined, and the job resume set corresponding to the target job category with the largest number of corresponding target sample resumes is determined as the job resume set to which the resume to be classified belongs.

[0085] In some embodiments, due to the comprehensiveness of the applicant's abilities as reflected in a resume, the resume may be suitable for multiple job categories. In this case, multiple job resume sets are determined for the resume to be classified. For example, if multiple target sample resumes are determined for the resume to be classified, the job resume sets corresponding to the target job category corresponding to each target sample resume can be determined as the job resume sets to which the resume to be classified belongs.

[0086] In some embodiments of the present application, after step 240, the method further includes: if the association probability between each of the sample resumes and the resume to be classified is less than the association probability threshold, the resume to be classified is determined to be an unqualified resume, and the resume to be classified is added to the designated resume set. For example, if the association probability between a sample resume and the resume to be classified is 0.7, assuming that the association probability threshold is 0.8, the resume to be classified is determined to be an unqualified resume. The designated resume set is used to store unqualified resumes, and the designated resume set is different from the position resume set. In this embodiment, unqualified resumes are automatically screened out by setting the association probability threshold, thereby preventing unqualified resumes from flowing into subsequent links.

[0087] In the scheme of the present application, resume feature words are extracted from the resume to be classified, and a feature coding vector of the resume to be classified is generated based on the resume feature words. Then, the correlation between the resume to be classified and the sample resume is mined based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume in the resume library. Then, the job resume set to which the resume to be classified belongs is determined based on the job category corresponding to the target sample resume that has a high correlation with the resume to be classified. This realizes automatic classification of resumes according to job categories based on the job categories corresponding to the sample resumes in the known resume library and the correlation between the resume to be classified and the sample resume, thereby improving the efficiency of resume classification and realizing automatic job category matching for resumes.

[0088] The solution of this application can be widely applied in scenarios such as human resources system construction, targeted talent mining, and target group recruitment. By setting qualified sample resumes in the resume library, resumes can be classified according to job categories. For example, in targeted talent mining, qualified resumes can be screened from the targeted job categories to build a resume library. In the subsequent process, resumes that match the targeted job categories can be automatically classified based on the constructed resume library.

[0089] In some embodiments of the present application, the association prediction model includes a converter network layer, a fusion processing layer, and a classification layer; Figure 5 As shown, the association prediction model performs association probability prediction based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume to obtain the association probability between the sample resume and the resume to be classified, including:

[0090] In step 510, the converter network layer extracts correlation features based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume to obtain the correlation features between the resume to be classified and the sample resume.

[0091] The transformer network layer consists of one or more cascaded transformer neural networks, also known as Transformer models. Compared to convolutional neural networks, Transformer models can capture global information. Furthermore, because they utilize a self-attention mechanism for rapid parallelization, they can be trained faster than recurrent neural networks.

[0092] Figure 6 The schematic diagram of the structure of the converter neural network is shown in Figure 6 As shown, the converter neural network includes a multi-head self-attention layer 610, a first summation and normalization layer 620, a feedforward neural network layer 630, and a second summation and normalization layer 640.

[0093] The multi-head self-attention layer 610 is composed of multiple self-attention heads with the same structure but different weight matrices. Multiple self-attention heads calculate the attention scores in parallel. The output of the multi-head self-attention layer is then input into the first summation and normalization layer 620, which performs summation and normalization processing. Figure 6 As shown, there is a residual connection between the input of the multi-head self-attention layer 610 and the first summation and normalization layer 620. That is, the input of the multi-head self-attention layer 610 is input into the first summation and normalization layer 620, so that the input of the first summation and normalization layer 620 includes the output of the multi-head self-attention layer 610 and the input of the multi-head self-attention layer 610. By performing a residual connection, the problems of gradient dissipation and gradient explosion caused by increasing network depth can be avoided.

[0094] Please continue reading Figure 6 As shown, the output of the first summation and normalization layer 620 is input to the feedforward neural network layer 630, processed by the feedforward neural network layer 630, and then passed to the second summation and normalization layer 640. Like the first summation and normalization layer 620, there is also a residual connection between the input of the feedforward neural network layer 630 and the second summation and normalization layer 640.

[0095] The input to the transformer network layer is the feature encoding vector of the resume to be classified and the feature encoding vector of the sample resume. For example, the feature encoding vector of "artificial intelligence" is: [0.255, -1.41, ...]; the feature encoding vector of "machine learning" is: [5.472, 6.2109, ...], etc. The output of the transformer network layer is a vector representation of the correlation features between the feature encoding vectors of the two resumes, for example: [1.231, 0.552, ... 1.762].

[0096] Due to the multi-head attention mechanism in the converter neural network, which includes multiple layers of self-attention heads with the same structure but different weight matrices, each attention head can focus on different features, which can prevent the association prediction model from focusing on only part of the input features. As a result, the converter network layer will pay attention to more features and learn the correlation between different feature encoding vectors, that is, from the feature encoding vector of the resume to be classified and the feature encoding vector of the sample resume, the correlation features between the resume to be classified and the sample resume are extracted.

[0097] In step 520, the fusion processing layer fuses the correlation features to obtain fusion features. In step 530, the classification layer performs association probability classification based on the fusion features to obtain the association probability between the sample resume and the resume to be classified.

[0098] In some embodiments of the present application, the fusion processing layer may include an RGU (Gate Recurrent Unit) layer and a feedforward neural network layer. In other embodiments, the fusion processing layer may also be a multi-layer feedforward neural network layer. GRU is a type of recurrent neural network that is proposed to solve problems such as gradients in long-term and back-propagation. The GRU layer is for deep feature extraction. Of course, when the fusion processing layer is a multi-layer feedforward neural network layer, effective feature fusion processing can also be achieved.

[0099] In a specific embodiment, the classification layer can calculate the association probability using a sigmoid function. The value range of the sigmoid function is (0, 1). Thus, the classification layer maps the fusion features obtained by the fusion processing layer to between (0, 1) to obtain the association probability between the sample resume and the resume to be classified.

[0100] Figure 7 is a structural diagram of an association prediction model according to a specific embodiment of the present application, such as Figure 7 As shown, the association prediction model includes a cascaded converter network layer 710, a fusion processing layer 720 and a classification layer 730, wherein the fusion processing layer 720 includes a gated recurrent unit layer and a feedforward neural network layer.

[0101] like Figure 7 As shown, the correlation features output by the converter network layer 710 are input to the gated recurrent unit layer 721, which extracts deep features. The extracted deep features are then input to the feedforward neural network layer for fusion processing. The classification layer 730 uses the sigmoid function to perform association probability classification.

[0102] In some embodiments of the present application, the loss function of the association prediction model may be a standard cross entropy loss function, whose formula is as follows:

[0103]

[0104] Where y is the true value of the association probability, and p(x) is the association probability actually output by the model during training. In a specific embodiment, the association prediction model can be obtained by using samples with association relationships (i.e., qualified resumes of the same job category) in the positive sample set constructed above as training samples for model training.

[0105] In some embodiments of the present application, Figure 8 As shown, step 220 includes:

[0106] Step 810: Determine the word type of each candidate word in the resume to be classified according to a preset word classification strategy.

[0107] The word classification strategy indicates the multiple word types included and the principles for determining each word type. Thus, based on the word classification strategy, the word type to which each candidate word in the resume to be classified belongs can be determined. The preset word classification strategy can be set according to actual needs and is not specifically limited here.

[0108] Among them, the preset word classification strategy can be to classify candidate words under one classification dimension, or to classify candidate words under at least two classification dimensions. The classification dimensions, for example, classify according to the role of words in distinguishing job categories, classify according to parts of speech, classify according to whether they are proper nouns, etc., and can be specifically set according to the actual application scenario. It can be understood that if the preset word classification strategy is set to classify under at least two classification dimensions, the word types to which each candidate word belongs will also correspond to multiple. For the sake of convenience, if the preset word classification strategy is to classify under at least two classification dimensions, the word type under each classification dimension will be called a subtype, that is, the word type includes the subtype under each classification dimension in at least two classification dimensions.

[0109] In some embodiments of the present application, in order to determine the subtype to which a candidate word in a resume to be classified belongs under a classification dimension, a feature word library can be pre-built for each subtype under the classification dimension. Therefore, in step 810, if it is to determine the subtype to which a candidate word belongs under the target classification dimension, the candidate word is searched in the feature word library built for each subtype under the target classification dimension. If the candidate word is found in a feature word library, the subtype corresponding to the feature word library containing the candidate word is determined as the subtype to which the candidate word belongs under the target classification dimension.

[0110] For example, if the subtypes under the target classification dimension include subtype I, subtype II, and subtype III, and corresponding feature vocabulary T1 corresponding to subtype I, feature vocabulary T2 corresponding to subtype II, and feature vocabulary T3 corresponding to subtype III are constructed, if the candidate word "artificial intelligence" is found in the feature vocabulary T2, then it is determined that the subtype to which the candidate word "artificial intelligence" belongs under the target classification dimension is subtype II.

[0111] In some embodiments of the present application, a feature word library corresponding to each subtype under a classification dimension can be constructed based on words in sample resumes in a resume library.

[0112] In some embodiments of the present application, the candidate words in the resume to be classified may refer to all words in the resume to be classified. In other embodiments, since the resume to be classified may contain information irrelevant to the job matching, such as the name, contact information, address, etc. in the job seeker's basic personal information, the classified resumes may be filtered first, and then the words in the filtered resume to be classified may be used as candidate words.

[0113] Step 820: Calculate the importance weight corresponding to each candidate word according to the word category to which each candidate word belongs.

[0114] In some embodiments of the present application, the calculation method for the importance weight corresponding to each word type can be pre-set, so that after determining the word category to which the candidate word belongs, the importance weight corresponding to the candidate word is calculated according to the calculation method corresponding to the word category to which the candidate word belongs.

[0115] In some embodiments of the present application, if the word classification strategy indicates that classification is performed under at least two classification dimensions, a sub-weight corresponding to each sub-type under each classification dimension may be set, and then the importance weight corresponding to each candidate word is calculated according to the sub-weight corresponding to each sub-type under each classification dimension. For ease of distinction, if the word classification strategy indicates that classification is performed under one classification dimension, the sub-weight corresponding to the sub-type to which the candidate word belongs may be regarded as the importance weight corresponding to the candidate word.

[0116] In some embodiments of the present application, the sub-weights corresponding to each sub-type under a classification dimension can be pre-set. If a candidate word belongs to a sub-type under a classification dimension, the sub-weight corresponding to the sub-type can be determined as the sub-weight of the candidate word under the classification dimension.

[0117] For example, if the word classification strategy includes a classification dimension based on whether it is a noun, and the subtypes under the classification dimension based on whether it is a noun include a subtype indicating that it is a noun and a subtype indicating that it is not a noun, the subweight corresponding to the subtype indicating that it is a noun can be set (for example, δ noun (w)), and set a subweight (e.g., 1) corresponding to a subtype indicating that it is not a noun.

[0118] For another example, if the word classification strategy includes a classification dimension based on whether it is a proper noun, and the subtypes under the classification dimension based on whether it is a proper noun include a subtype indicating that it is a proper noun and a subtype indicating that it is not a proper noun, the subweight corresponding to the subtype corresponding to the proper noun can be set (for example, δ prop (w)), and set a sub-weight (e.g., 1) corresponding to a sub-type indicating that the sub-type is not a proper noun.

[0119] In some embodiments of the present application, the word classification strategy may be to divide word types according to the distinguishing effect of words on job categories in the resume. For example, according to the distinguishing effect of words on job categories in the resume from large to small, the word categories set in the word classification strategy include domain words, common words and irrelevant words. Among them, candidate words belonging to domain words may be words that can distinguish different job categories in the resume, irrelevant words may be words that have a smaller distinguishing effect on job categories in the resume, and other words in the resume except domain words and irrelevant words may be regarded as common words. For ease of description, if the word category to which a candidate word belongs is common words, the candidate word is called a common word; similarly, if the word category to which a candidate word belongs is domain words, the candidate word is called a domain word; if the word category to which a candidate word belongs is irrelevant words, the candidate word is called an irrelevant word.

[0120] In some embodiments of the present application, the importance weight of irrelevant words in a resume can be set to 0. The sub-weights of common words can be calculated according to the following formula:

[0121]

[0122] Among them, P(W c ) represents a common word W c The sub-weight of tf c Represents a common word W c Frequency of appearance in all sample resumes in the resume database; max df k The document frequency with the largest value among the document frequencies of all common words in the resume database; df c To include common words W c The document frequency of the sample resume in the resume database.

[0123] The sub-weights belonging to domain words can be calculated according to the following formula:

[0124]

[0125] Among them, P(W f ) represents the domain word W f Sub-weight of maxPW c is the maximum value of the importance weights of all common words in the common vocabulary; tf f domain word W f The frequency of appearance in all sample resumes in the resume database, df l The document frequency of the sample resume that includes the lth domain word in the domain vocabulary in the resume database; maxdf l The maximum value of the document frequency corresponding to all domain words in the domain vocabulary; df f domain word W fThe frequency of a document appearing in all sample resumes in the resume database.

[0126] In some embodiments of the present application, the word category includes subtypes under at least two classification dimensions; step 820 includes: calculating the sub-weight of each candidate word under each classification dimension according to the sub-type to which each candidate word belongs under each classification dimension; and calculating the importance weight corresponding to each candidate word according to the sub-weight of each candidate word under each classification dimension.

[0127] In some embodiments, the sub-weights of the candidate word under all classification dimensions may be added together, and the result of the addition may be determined as the importance weight corresponding to the candidate word. In other embodiments, the sub-weights of the candidate word under all classification dimensions may be multiplied together, and the result of the multiplication may be determined as the importance weight corresponding to the candidate word.

[0128] In a specific embodiment, the word category includes subtypes under the three classification dimensions listed above (i.e., word types are divided into domain words, common words, and irrelevant words according to the role of words in the resume in distinguishing job categories; classification according to whether they are proper nouns; and classification according to whether they are nouns). The subweights of the candidate words under the three classification dimensions can be multiplied, and the multiplication result is used as the importance weight corresponding to the candidate word. For example, according to the above example, the importance weight corresponding to a candidate word w is:

[0129] W topic (w)=PW×δ1(w)×δ2(w); (Formula 4)

[0130] Among them, PW is the sub-weight corresponding to the sub-type of the candidate word w under the classification dimension of classification based on the role of words in the resume in distinguishing job categories. That is, if it is an irrelevant word, PW is 0; if it is an ordinary word, PW is the result calculated according to the above formula 3; if it is a domain word, PW is the result calculated according to the above formula 2; δ1(w) is the sub-weight corresponding to the classification dimension of whether it is a proper noun. If it is a proper noun, δ1(w) is δ prop (w), if it is not a proper noun, δ1(w) is 1. δ2(w) is the sub-weight corresponding to the classification dimension of whether it is a noun. If it is a noun, δ2(w) is δ noun (w), if not a noun, δ noun (w) is 1.

[0131] According to Formula 4, if a candidate word w is a noun and a proper noun, the importance weight corresponding to the candidate word w is:

[0132] W topic (w)=PW×δ prop (w)×δnoun (w); (Formula 5)

[0133] In this embodiment, when the candidate word is a noun and / or proper noun, the weight of the PW is increased. Because proper nouns are more likely to represent a subject than common nouns, and nouns are more likely to represent a subject than other words such as verbs and adjectives, increasing the weight of the PW when the candidate word is a proper noun and / or a noun can effectively increase the importance of the candidate word in representing the subject, thereby reflecting the candidate word's contribution to job classification.

[0134] Step 830 , screening candidate words according to the importance weights corresponding to the candidate words, and determining the first target candidate word.

[0135] Step 840: Determine the first target candidate word as the resume feature word.

[0136] In some embodiments of the present application, the candidate words in the resume to be classified can be sorted according to the calculated importance weights, in descending order of importance weights, and the candidate words that are at the top of the sorting by a set number are determined as the first target candidate words. In other embodiments, the candidate words whose importance weights are greater than the first weight threshold can also be determined as the first target candidate words according to a set first weight threshold. It is understandable that the determined first target candidate words can be one or more.

[0137] Through the above embodiment, resume feature words are selected according to the importance weights corresponding to word categories. The importance weights of candidate words can reflect the contribution of the candidate words to the topic to a certain extent. Therefore, the above embodiment is equivalent to implementing the selection of resume feature words based on the importance weights of the topics.

[0138] In some other embodiments of the present application, step 220 includes: searching for candidate words in the resume to be classified in the job category keyword library, and determining a second target candidate word located in the job category keyword library; the job category keyword library is constructed based on words in the sample resumes in the resume library; and determining the second target candidate word as the resume feature word.

[0139] The words in the job category keyword library are job category keywords that distinguish job categories. The second target candidate word refers to a candidate word in the job category keyword library. That is, if a candidate word in the resume to be classified is found in the job category keyword library, the candidate word is determined to be the second target candidate word. In this embodiment, the candidate word in the resume to be classified that is in the job category keyword library is determined as the resume feature word of the resume to be classified.

[0140] In some embodiments of the present application, a position category keyword library can be constructed by mining keywords from sample resumes of different position categories in a resume library. It can be understood that if a word appears frequently in sample resumes of a certain position category but less frequently in sample resumes of other position categories, it indicates that the word has a distinguishing effect on the position category. Therefore, based on the frequency of occurrence of the word in the position category, words that have a distinguishing effect on the position category can be screened out to construct a position category keyword library.

[0141] In some embodiments of the present application, the TF-IDF (term frequency–inverse document frequency) algorithm can be used to mine sample resumes in the resume database for keywords that distinguish different job categories. Specifically, the resume database includes a sub-resume database corresponding to each job category; in this embodiment, the resume database can be mined according to the following algorithm: Figure 9 The process shown is used to build a job category keyword library, such as Figure 9 Shown, including:

[0142] Step 910, calculate the word frequency of each first candidate word in all sample resumes in the sub-resume database; the first candidate word is the candidate word in the sample resume.

[0143] Each sub-resume database uniquely corresponds to a job category, and the sample resumes in that sub-resume database are qualified resumes for the job category corresponding to that sub-resume database. The sub-resume database where the first candidate word is located refers to the sub-resume database to which the sample resume from which the first candidate word originates belongs.

[0144] For each candidate word (i.e., the first candidate word) in each sample resume, the word frequency of the first candidate word in the sub-resume database can be calculated according to the following formula:

[0145]

[0146] Among them, TF(W i,j ) represents the first candidate word W of the i-th resume in the sample resume under the sub-resume database corresponding to the j-th job category i,j The word frequency in the sub-resume database corresponding to the j-th job category;

[0147] P(W i,j ) represents the first candidate word W i,j The number of times the j-th job category appears in the sample resumes under the sub-resume database; N j It represents the total number of first candidate words in the sample resumes under the sub-resume database corresponding to the j-th job category.

[0148] Step 920: Calculate the inverse document frequency index of each of the first candidate words in the resume database.

[0149] The inverse document frequency index of the first candidate word in the resume database can be calculated according to the following formula:

[0150]

[0151] Among them, IDF(W i,j ) represents the first candidate word W i,j The inverse document frequency index in the resume database; N is the total number of sample resumes in the resume database; Q(W i,j ) is the first candidate word W i,j In all other sub-resume databases except the sub-resume database, including the first candidate word W i,j The number of sample resumes.

[0152] Step 930 : Calculate the first weight corresponding to each of the first candidate words according to the word frequency of each of the first candidate words and the inverse document frequency index of each of the first candidate words.

[0153] Specifically, the word frequency of the first candidate word is multiplied by the inverse document frequency index of the first candidate word to obtain the word frequency-inverse document frequency index of the first candidate word, that is:

[0154] TF-IDF(W i,j )=TF(W i,j )×IDF(W i,j ); (Formula 8)

[0155] The calculated word frequency-inverse document frequency index of the first candidate word is used as the first weight of the first candidate word.

[0156] Step 940: Screen the first candidate words according to the first weights corresponding to the first candidate words.

[0157] Step 950: Add the filtered first candidate word to the job category keyword library.

[0158] In some embodiments of the present application, a second weight threshold can be set to screen the first candidate words, and the first candidate words whose first weight exceeds the second weight are added to the job category keyword library. In other embodiments of the present application, the first candidate words are sorted from large to small according to the second weight, and the first candidate words ranked in the top second set number are screened and added to the job category keyword library.

[0159] In the above embodiment, full use is made of the frequency of occurrence of words in resumes of different job categories to distinguish job categories, and resume feature words are screened through the word frequency of candidate words in the resume database and the inverse document frequency index, thereby ensuring that the extracted job feature words have a distinguishing effect on job categories.

[0160] In some embodiments of the present application, step 220 further includes: performing named entity recognition on each candidate word in the resume to be classified to determine the candidate word identified as a named entity; and determining the candidate word identified as a named entity as the resume feature word.

[0161] Named Entity Recognition (NER), also known as "proper name recognition", refers to the identification of entities with specific meanings in text, mainly including names of people, places, institutions, conferences, magazines, competitions, proper nouns, etc.

[0162] In some embodiments of the present application, since the competitions a job applicant has participated in, the papers published in magazines, the institutions where he / she has interned, the companies he / she has worked for, and the schools he / she has attended can reflect the applicant's abilities to a certain extent, and different job categories have different emphases on the applicant's abilities, the named entities in the resume to be classified can, to a certain extent, reflect the job category to which the job applicant is suitable. That is, the named entities in the resume have a distinguishing effect on the job category. Therefore, in this embodiment, the named entities identified from the resume to be classified are also used as resume feature words.

[0163] In some embodiments of the present application, a named entity recognition tool (NER tool) can be used to perform naming recognition and determine candidate words identified as named entities in the resume to be classified as resume feature words.

[0164] The named entities in the resumes to be classified can reflect the ability of job seekers to a certain extent and have a certain distinguishing effect on different job categories. Therefore, by extracting the candidate words identified as named entities in the resumes to be classified as resume feature words, the number and types of resume feature words are enriched, which provides more information for subsequent association probability calculation, improves the accuracy of association probability calculation, and thus improves the accuracy of resume classification.

[0165] In some embodiments of the present application, the resume feature words extracted from the resume to be classified are not limited to including the first target candidate word and the candidate word identified as a named entity, or including the second target candidate word and the candidate word identified as a named entity, but may also include the first target candidate word, the second target candidate word, and the candidate word identified as a named entity. By extracting resume feature words in three dimensions, the comprehensiveness of the extracted resume feature words can be effectively guaranteed, providing sufficient information basis for performing association probability matching, thereby ensuring the accuracy of association probability calculation.

[0166] In some embodiments of the present application, before step 220, as Figure 10 As shown, the method further includes:

[0167] Step 1010: Divide the resume to be classified into blocks based on the writing structure of the resume to be classified, and determine the resume blocks included in the resume to be classified and the block resume text corresponding to each resume block. Step 1020: Filter the resume blocks included in the resume to be classified to determine candidate resume blocks. Step 1030: Filter the block resume text corresponding to the candidate resume blocks to determine candidate text. Step 1040: Determine the words in the candidate text as candidate words for the resume to be classified.

[0168] The significant difference between resume text and other texts is that resume text has a hierarchical and modular structure. For example, a resume text generally includes important parts such as personal basic information, job search intentions, educational background, work experience (project experience), self-evaluation, professional skills, awards, etc. Moreover, each part will be marked with a block title in the resume text.

[0169] In a specific embodiment, based on the hierarchical and modular nature of resume text, the resume can be segmented by identifying the segment titles within the resume text. Each segment title corresponds to a resume segment, and the text content under a segment title is the segmented resume text corresponding to that segment. In other words, segmented resume text refers to the text content included in a resume segment, for example, the text content in the basic information segment, the text content in the self-evaluation segment, etc.

[0170] In some embodiments of the present application, title keywords corresponding to the resume blocks to be filtered can be pre-set. If the block title of a resume block includes the title keyword, the resume block can be filtered out.

[0171] In some embodiments of the present application, the resume blocks filtered out may be resume blocks that are not significantly relevant to the job category match, such as personal basic information (name, gender, birthday, address, mobile phone number, email address, etc.), job-seeking intentions (intended position, expected salary, etc.), etc.

[0172] In related technologies, in order to improve the efficiency of resume classification, resumes are generally classified according to the job-seeking intentions in the resumes. However, in practice, due to the inaccurate self-positioning of job seekers, or the job seekers are suitable for multiple job categories but are limited to writing only one intended position in their resumes, the resume classification based on job-seeking intentions will result in low resume classification accuracy.

[0173] In this embodiment, since information such as job-seeking intention is filtered out, the extracted resume feature words do not include words related to the intended position. This can avoid the situation where resume classification accuracy is low due to classification according to the intended position.

[0174] In a resume, except that there is the resume segmentation that is not significantly related to the job category matching, in some resume segments that are relevant to the job category matching, there may also be some information that is not very effective for the job category matching. Therefore, the information that is not very effective for the job category matching in this resume segmentation text can be filtered by step 1030. In a specific embodiment, the time information in the resume segmentation text is not very effective for the job category matching, and time information is such as the time information in the work experience, the time information in the project experience, the time information in the study experience, etc. But there may be more time information in each resume segmentation. Therefore, the time information in the segmentation resume text in the resume segmentation remaining after filtering can be filtered by step 1030. In some embodiments of the application, the time information in each resume segmentation text can be filtered based on regular matching by regular expression. Regular expression describes a pattern of string matching, which can be used to check whether a character string includes a certain character or a certain character segment.

[0175] In the above embodiment, since the information that has little effect on matching the job category in the resume to be classified is filtered out before extracting the resume feature words, the words in the candidate text obtained after filtering out the information that has little effect on matching the job category are determined as candidate words, thereby improving the efficiency of extracting resume feature words, reducing the interference of the information that has little effect on matching the job category in the resume to be classified on the extraction of resume feature words, and improving the accuracy of the extracted resume feature words.

[0176] Figure 11 is a flowchart of a resume classification method according to a specific embodiment of the present application, such as Figure 11 As shown, specifically including:

[0177] Step 1110: divide the positive and negative samples according to the preset rules. Specifically, according to the method in the above embodiment, qualified resumes in each job category can be added to the positive sample set, and unqualified resumes in each job category can be added to the negative sample set.

[0178] Step 1120: Build a resume database. Specifically, all or part of the resumes can be selected from the positive sample set and added to the resume database.

[0179] Step 1130: Extract resume feature words from the resume to be classified and perform vectorization. The process of extracting resume feature words is described above. Vectorization refers to generating a feature encoding vector for the resume to be classified based on the resume feature words extracted from the resume to be classified.

[0180] Step 1140: mining the association probability between the resume to be classified and the sample resumes in the resume database based on the association probability prediction model.

[0181] Step 1150: Determine the job category that matches the resume to be classified. In a specific embodiment, the job category corresponding to the sample resume with the highest probability of association with the resume to be classified can be determined as the job category that matches the resume to be classified, and the job resume set to which the resume to be classified belongs can be determined to be the job resume set corresponding to the job category that matches the resume to be classified.

[0182] In this embodiment, the association probability prediction model is used to mine the association probability between the resume to be classified and the qualified resumes (i.e., sample resumes) under each job category in the resume database. The association probability reflects the correlation between the resume to be classified and the sample resume. The job category that matches the resume to be classified is determined based on the size of the correlation, and the job resume set to which the resume to be classified belongs is correspondingly determined, thereby realizing automatic resume classification according to job category and effectively improving the efficiency of resume classification.

[0183] The following describes an embodiment of the device of the present application, which can be used to perform the method described in the above embodiment of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the above method embodiment of the present application.

[0184] Figure 12 is a block diagram of a resume classification device according to an embodiment. Figure 12As shown, the resume classification device includes: an acquisition module 1210, used to obtain the resume to be classified; a resume feature word extraction module 1220, used to extract resume feature words from the resume to be classified; a feature coding vector generation module 1230, used to generate the feature coding vector of the resume to be classified based on the resume feature words; an association probability calculation module 1240, used to calculate the association probability between each sample resume and the resume to be classified based on the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume library; a target sample resume determination module 1250, used to determine the target sample resume that matches the resume to be classified based on the association probability between each sample resume and the resume to be classified; a job resume set determination module 1260, used to determine the job resume set to which the resume to be classified belongs based on the target job category corresponding to the target sample resume.

[0185] In some embodiments of the present application, the feature coding vector generation module 1230 includes: a feature word vector calculation unit, used to calculate the feature word vector corresponding to the resume feature word; and a character vector calculation unit, used to calculate the character vector of each character in the resume feature word; a splicing unit, used to splice the resume feature word with the character vector of each character in the resume feature word respectively, to obtain at least two splicing vectors corresponding to the resume feature word; a combination unit, used to combine at least two splicing vectors corresponding to all the resume feature words in the resume to be classified, to obtain the feature coding vector of the resume to be classified.

[0186] In some embodiments of the present application, the association probability calculation module 1240 includes: an input unit for inputting the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume library into an association prediction model; an association probability prediction unit for performing association probability prediction based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume by the association prediction model to obtain the association probability between the sample resume and the resume to be classified.

[0187] In some embodiments of the present application, the association prediction model includes a converter network layer, a fusion processing layer and a classification layer; in this embodiment, the association probability prediction unit includes: a correlation feature extraction unit, which is used by the converter network layer to extract correlation features based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume to obtain the correlation features between the resume to be classified and the sample resume; a fusion processing unit, which is used by the fusion processing layer to fuse the correlation features to obtain fusion features; an association probability classification unit, which is used by the classification layer to perform association probability classification based on the fusion features to obtain the association probability between the sample resume and the resume to be classified.

[0188] In some embodiments of the present application, the resume feature word extraction module 1220 includes: a word type determination unit, which is used to determine the word type of each candidate word in the resume to be classified according to a preset word classification strategy; an importance weight calculation unit, which is used to calculate the importance weight corresponding to each candidate word according to the word category to which each candidate word belongs; candidate word screening, which is used to screen candidate words according to the importance weight corresponding to each candidate word and determine the first target candidate word; and a first resume feature word determination unit, which is used to determine the first target candidate word as the resume feature word.

[0189] In some embodiments of the present application, the word category includes subtypes under at least two classification dimensions; the importance weight calculation unit includes: a subweight calculation unit, used to calculate the subweight of each candidate word under each classification dimension according to the subtype to which each candidate word belongs under each classification dimension; a calculation unit, used to calculate the importance weight corresponding to each candidate word according to the subweight of each candidate word under each classification dimension.

[0190] In some embodiments of the present application, the resume feature word extraction module 1220 includes: a search unit, used to search for candidate words in the resume to be classified in the job category keyword library, and determine a second target candidate word located in the job category keyword library; the job category keyword library is constructed based on the words in the sample resumes in the resume library; a second resume feature word determination unit, used to determine the second target candidate word as the resume feature word.

[0191] In some embodiments of the present application, the resume database includes a sub-resume database corresponding to each job category; in this embodiment, the resume classification device also includes: a word frequency calculation module, used to calculate the word frequency of each first candidate word in all sample resumes in the sub-resume database; the first candidate word is a candidate word in the sample resume; an inverse document frequency index calculation module, used to calculate the inverse document frequency index of each first candidate word in the resume database; a first weight calculation module, used to calculate the first weight corresponding to each first candidate word according to the word frequency of each first candidate word and the inverse document frequency index of each first candidate word; a first candidate word screening module, used to screen the first candidate words according to the first weight corresponding to each first candidate word; a first adding module, used to add the screened first candidate words to the job category keyword library.

[0192] In some embodiments of the present application, the resume feature word extraction module 1220 also includes: a named entity recognition module, which is used to perform named entity recognition on each candidate word in the resume to be classified, and determine the candidate word identified as a named entity; a third resume feature word determination module, which is used to determine the candidate word identified as a named entity as the resume feature word.

[0193] In some embodiments of the present application, the resume classification device also includes: a block module, which is used to block the resume to be classified according to the writing framework of the resume to be classified, and determine the resume blocks included in the resume to be classified and the block resume texts corresponding to each resume block; a first filtering module, which is used to filter the resume blocks included in the resume to be classified, and determine the candidate resume blocks; a second filtering module, which is used to filter the information of the block resume texts corresponding to the candidate resume blocks, and determine the candidate texts; a candidate word determination module, which is used to determine the words in the candidate text as candidate words of the resume to be classified.

[0194] In some embodiments of the present application, the target sample resume determination module 1250 includes: a first determination unit configured to determine, based on the association probabilities between the sample resumes and the resumes to be classified, the sample resume corresponding to the maximum association probability; and a second determination unit configured to determine the sample resume corresponding to the maximum association probability as the target sample resume.

[0195] In some embodiments of the present application, the resume classification device also includes: a second adding module, which is used to determine the resume to be classified as an unqualified resume if the association probability between each of the sample resumes and the resume to be classified is less than the association probability threshold, and add the resume to be classified to the specified resume collection.

[0196] Figure 13 The following is a schematic diagram showing the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application. Figure 13 The computer system 1300 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0197] like Figure 13As shown, computer system 1300 includes a central processing unit (CPU) 1301, which can perform various appropriate actions and processes according to programs stored in read-only memory (ROM) 1302 or programs loaded from storage unit 1308 into random access memory (RAM) 1303, such as executing the methods in the above embodiments. Various programs and data required for system operation are also stored in RAM 1303. CPU 1301, ROM 1302, and RAM 1303 are connected to each other via bus 1304. Input / output (I / O) interface 1305 is also connected to bus 1304.

[0198] The following components are connected to the I / O interface 1305: an input section 1306 including a keyboard, a mouse, and the like; an output section 1307 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 1308 including a hard disk; and a communication section 1309 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 1309 performs communication processing via a network such as the Internet. A drive 1310 is also connected to the I / O interface 1305 as needed. Removable media 1311, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 1310 as needed, so that computer programs read from the removable media can be installed in the storage section 1308 as needed.

[0199] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1309, and / or installed from a removable medium 1311. When the computer program is executed by the central processing unit (CPU) 1301, the various functions defined in the system of the present application are executed.

[0200] It should be noted that the computer-readable medium shown in the embodiments of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0201] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. Among them, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0202] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. In some cases, the names of these units do not constitute limitations on the units themselves.

[0203] As another aspect, the present application further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments, or may exist independently without being incorporated into the electronic device. The computer-readable storage medium carries computer-readable instructions, and when the computer-readable instructions are executed by a processor, the method of any of the above embodiments is implemented.

[0204] According to one aspect of the present application, an electronic device is also provided, which includes: a processor; a memory, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processor, the method in any of the above embodiments is implemented.

[0205] According to one aspect of an embodiment of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any of the above embodiments.

[0206] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the application, the features and functions of two or more modules or units described above can be concretized in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0207] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

[0208] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art that are not disclosed herein.

[0209] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A resume classification method, characterized in that: include: Get resumes to be classified; Extracting resume feature words from the resume to be classified; The resume feature words are words that have the function of distinguishing job categories; Generating a feature coding vector of the resume to be classified according to the resume feature words; Calculating the association probability between each sample resume and the resume to be classified based on the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in a resume database; the resume database includes qualified resumes under multiple job categories, and the qualified resumes serve as the sample resumes; Determining a target sample resume that matches the resume to be classified based on the association probability between each sample resume and the resume to be classified; According to the target job category corresponding to the target sample resume, the job resume set to which the resume to be classified belongs is determined; the job resume set to which the resume to be classified belongs is the job resume set corresponding to the target job category; the job resume set is used for subsequent resume screening.

2. The method according to claim 1, characterized in that Generating the feature coding vector of the resume to be classified according to the resume feature words includes: Calculating the feature word vector corresponding to the resume feature word; and Calculate the word vector of each word in the resume feature word; Concatenate the resume feature word with the word vector of each word in the resume feature word to obtain at least two concatenated vectors corresponding to the resume feature word; At least two concatenated vectors corresponding to all resume feature words in the resume to be classified are combined to obtain a feature coding vector of the resume to be classified.

3. The method according to claim 1, characterized in that The step of calculating the association probability between each sample resume and the resume to be classified based on the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume database includes: Inputting the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume database into an association prediction model; The association prediction model performs association probability prediction based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume to obtain the association probability between the sample resume and the resume to be classified.

4. The method according to claim 3, characterized in that The association prediction model includes a converter network layer, a fusion processing layer and a classification layer; The association prediction model performs association probability prediction based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume to obtain the association probability between the sample resume and the resume to be classified, including: The converter network layer extracts correlation features based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume to obtain correlation features between the resume to be classified and the sample resume; The fusion processing layer performs fusion processing on the correlation features to obtain fusion features; The classification layer performs association probability classification based on the fusion features to obtain the association probability between the sample resume and the resume to be classified.

5. The method according to claim 1, wherein The step of extracting resume feature words from the resume to be classified includes: Determine the word type of each candidate word in the resume to be classified according to a preset word classification strategy; Calculating the importance weight of each candidate word according to the word category to which each candidate word belongs; Screening the candidate words according to the importance weights corresponding to the candidate words to determine the first target candidate word; The first target candidate word is determined as the resume feature word.

6. The method according to claim 5, characterized in that The word category includes subtypes under at least two classification dimensions; Calculating the importance weight of each candidate word according to the word category to which each candidate word belongs includes: Calculating the sub-weight of each candidate word under each classification dimension according to the sub-type to which each candidate word belongs under each classification dimension; The importance weight corresponding to each candidate word is calculated based on the sub-weight of each candidate word under each of the classification dimensions.

7. The method according to claim 1, characterized in that The step of extracting resume feature words from the resume to be classified includes: Searching for candidate words in the resume to be classified in a job category keyword library, and determining a second target candidate word in the job category keyword library; the job category keyword library is constructed based on words in sample resumes in the resume library; The second target candidate word is determined as the resume feature word.

8. The method according to claim 7, characterized in that The resume database includes a sub-resume database corresponding to each job category; The method further comprises searching for candidate words in the resume to be classified in the job category keyword library and determining that the candidate words are located before the second target candidate words in the job category keyword library: Calculating the word frequency of each first candidate word in all sample resumes in the sub-resume database; the first candidate word is a candidate word in the sample resume; Calculating the inverse document frequency index of each of the first candidate words in the resume database; Calculating first weights corresponding to the first candidate words according to the word frequencies of the first candidate words and the inverse document frequency indexes of the first candidate words; Screening the first candidate words according to the first weights corresponding to the first candidate words; The first candidate word selected is added to the job category keyword library.

9. The method according to any one of claims 5 to 8, characterized in that The step of extracting resume feature words from the resume to be classified further includes: Performing named entity recognition on each candidate word in the resume to be classified to determine the candidate word recognized as a named entity; The candidate words identified as named entities are determined as the resume feature words.

10. The method according to any one of claims 5 to 8, characterized in that Before extracting resume feature words from the resume to be classified, the method further includes: Divide the resume to be classified into blocks according to the writing framework of the resume to be classified, and determine the resume blocks included in the resume to be classified and the block resume texts corresponding to each resume block; Filtering the resume blocks included in the resume to be classified to determine candidate resume blocks; Filtering information on the resume text blocks corresponding to the candidate resume blocks to determine candidate texts; The words in the candidate text are determined as candidate words for the resume to be classified.

11. The method according to claim 1, wherein Determining a target sample resume that matches the resume to be classified based on the association probability between each sample resume and the resume to be classified includes: Determine the sample resume corresponding to the maximum association probability based on the association probability between each sample resume and the resume to be classified; The sample resume corresponding to the maximum association probability is determined as the target sample resume.

12. The method according to claim 1, characterized in that After calculating the association probability between each sample resume and the resume to be classified based on the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume database, the method further includes: If the association probability between each of the sample resumes and the resume to be classified is less than the association probability threshold, the resume to be classified is determined to be an unqualified resume, and the resume to be classified is added to the designated resume set.

13. A resume classification device, characterized in that: include: Acquisition module, used to obtain resumes to be classified; A resume feature word extraction module, used to extract resume feature words from the resume to be classified; The resume feature words are words that have the function of distinguishing job categories; A feature coding vector generating module, configured to generate a feature coding vector of the resume to be classified according to the resume feature words; an association probability calculation module, configured to calculate an association probability between each sample resume and the resume to be classified based on a feature coding vector of the resume to be classified and a feature coding vector of each sample resume in a resume library; the resume library includes qualified resumes under multiple job categories, and the qualified resumes serve as the sample resumes; A target sample resume determination module is used to determine a target sample resume that matches the resume to be classified based on the association probability between each sample resume and the resume to be classified; A job resume set determination module is used to determine the job resume set to which the resume to be classified belongs based on the target job category corresponding to the target sample resume; The job resume set to which the resume to be classified belongs is the job resume set corresponding to the target job category; the job resume set is used for subsequent resume screening.

14. The device according to claim 13, characterized in that The feature coding vector generation module includes: A feature word vector calculation unit, configured to calculate a feature word vector corresponding to the resume feature word; and A word vector calculation unit, used to calculate the word vector of each word in the resume feature word; a concatenation unit, configured to concatenate the resume feature word with the word vector of each word in the resume feature word to obtain at least two concatenation vectors corresponding to the resume feature word; The combining unit is used to combine at least two concatenated vectors corresponding to all the resume feature words in the resume to be classified to obtain a feature coding vector of the resume to be classified.

15. The device according to claim 13, characterized in that The association probability calculation module includes: An input unit, configured to input the feature coding vector of the resume to be classified and the feature coding vector of each sample resume in the resume database into an association prediction model; The association probability prediction unit is used to perform association probability prediction based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume by the association prediction model to obtain the association probability between the sample resume and the resume to be classified.

16. The device according to claim 15, characterized in that The association prediction model includes a converter network layer, a fusion processing layer and a classification layer; the association probability prediction unit includes: A correlation feature extraction unit, configured to extract correlation features based on the feature coding vector of the resume to be classified and the feature coding vector of the sample resume by the converter network layer, to obtain correlation features between the resume to be classified and the sample resume; a fusion processing unit, configured to perform fusion processing on the correlation features by the fusion processing layer to obtain fusion features; The association probability classification unit is used to perform association probability classification according to the fusion feature by the classification layer to obtain the association probability between the sample resume and the resume to be classified.

17. The device according to claim 13, characterized in that The resume feature word extraction module includes: A word type determination unit, configured to determine the word type to which each candidate word in the resume to be classified belongs according to a preset word classification strategy; An importance weight calculation unit, configured to calculate the importance weight corresponding to each candidate word according to the word category to which each candidate word belongs; Candidate word screening, for screening candidate words according to the importance weights corresponding to the candidate words, and determining the first target candidate word; The first resume feature word determination unit is configured to determine the first target candidate word as the resume feature word.

18. The device according to claim 17, characterized in that The word category includes subtypes under at least two classification dimensions; the importance weight calculation unit includes: a sub-weight calculation unit, configured to calculate the sub-weight of each candidate word in each classification dimension according to the sub-type to which each candidate word belongs in each classification dimension; The calculation unit is used to calculate the importance weight corresponding to each candidate word according to the sub-weight of each candidate word under each of the classification dimensions.

19. The device according to claim 13, characterized in that The resume feature word extraction module includes: a search unit configured to search a job category keyword library for candidate words in the resume to be classified, and determine a second target candidate word in the job category keyword library; the job category keyword library is constructed based on words in sample resumes in the resume library; The second resume feature word determination unit is configured to determine the second target candidate word as the resume feature word.

20. The device according to claim 19, characterized in that The resume database includes a sub-resume database corresponding to each job category; the resume classification device also includes: A word frequency calculation module, configured to calculate the word frequency of each first candidate word in all sample resumes in the sub-resume database; the first candidate word is a candidate word in the sample resume; an inverse document frequency index calculation module, configured to calculate an inverse document frequency index of each of the first candidate words in the resume database; A first weight calculation module, configured to calculate a first weight corresponding to each of the first candidate words according to the word frequency of each of the first candidate words and the inverse document frequency index of each of the first candidate words; A first candidate word screening module, configured to screen first candidate words according to first weights corresponding to the first candidate words; The first adding module is used to add the screened first candidate word to the job category keyword library.

21. The device according to any one of claims 17 to 20, characterized in that The resume feature word extraction module also includes: A named entity recognition module is used to perform named entity recognition on each candidate word in the resume to be classified, and determine the candidate word recognized as a named entity; The third resume feature word determination module is used to determine the candidate word identified as the named entity as the resume feature word.

22. The device according to any one of claims 17 to 20, characterized in that The resume classification device further includes: A block division module is used to divide the resume to be classified into blocks according to the writing framework of the resume to be classified, and determine the resume blocks included in the resume to be classified and the block resume text corresponding to each resume block; A first filtering module is used to filter the resume blocks included in the resume to be classified to determine candidate resume blocks; The second filtering module is used to filter the resume texts corresponding to the candidate resume blocks to determine the candidate texts; The candidate word determination module is used to determine the words in the candidate text as candidate words for the resume to be classified.

23. The device according to claim 13, characterized in that The target sample resume determination module includes: A first determining unit is configured to determine a sample resume corresponding to a maximum association probability based on association probabilities between the sample resumes and the resumes to be classified; The second determining unit is used to determine the sample resume corresponding to the maximum association probability as the target sample resume.

24. The device according to claim 13, characterized in that The resume classification device also includes: The second adding module is used to determine the resume to be classified as an unqualified resume if the association probability between each of the sample resumes and the resume to be classified is less than the association probability threshold, and add the resume to be classified to the designated resume set.

25. An electronic device, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 12 is implemented.

26. A computer-readable storage medium having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by a processor, the method according to any one of claims 1 to 12 is implemented.

27. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the method according to any one of claims 1 to 12.

Images