A resume retrieval method and device

Abstract

This application provides a resume retrieval method and apparatus. The resume retrieval method includes: acquiring resume retrieval text; performing entity recognition on the resume retrieval text to identify each entity content; determining target search keywords associated with each entity content using preset association rules; and retrieving target resumes from a resume database based on each target search keyword. Thus, by determining the target search keywords corresponding to the entity content in the resume retrieval text, more search keywords can be obtained, making the retrieval more comprehensive, avoiding missing resumes that meet the criteria, and eliminating the need for multiple searches, thereby improving the accuracy and efficiency of resume retrieval.

Description

Technical Field

[0001] This application relates to the field of text processing technology, and in particular to a resume retrieval method and apparatus, computing device and computer-readable storage medium. Background Technology

[0002] With the development of online recruitment, more and more people are looking for jobs through online platforms. The amount of talent information in online recruitment products is becoming increasingly massive, and the talent levels are becoming increasingly complex. Recruiting companies often get lost in the vast space of talent information and are unable to find the talent they need.

[0003] Typically, resumes can be retrieved from a resume database based on the resume search text. For example, if the resume search text is "people who graduated from Peking University with a degree in Computer Science," this text contains two keywords: Peking University and Computer Science. Existing methods use character matching to search, identifying candidates whose resumes include "Peking University" and "Computer Science." However, this approach may miss some qualified candidates, resulting in an insufficient search. Therefore, it may be necessary to modify the resume search text for further retrieval, reducing the accuracy and efficiency of the resume search. Summary of the Invention

[0004] In view of this, embodiments of this application provide a resume retrieval method and apparatus, a computing device and a computer-readable storage medium to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this application, a resume retrieval method is provided, including:

[0006] Obtain the resume search text;

[0007] Entity recognition is performed on the resume search text to identify the content of each entity in the resume search text;

[0008] For each entity content, target search keywords associated with that entity content are determined using preset association rules;

[0009] Based on the target search keywords, the system retrieves target resumes from the resume database.

[0010] According to a second aspect of the embodiments of this application, a resume retrieval device is provided, comprising:

[0011] The acquisition module is configured to retrieve resume search text;

[0012] The recognition module is configured to perform entity recognition on the resume search text and recognize the content of each entity in the resume search text.

[0013] The determination module is configured to determine the target search keywords associated with each entity content using preset association rules;

[0014] The search module is configured to retrieve target resumes from the resume database based on various target search keywords.

[0015] According to a third aspect of the embodiments of this application, a computing device is provided, including a memory, a processor, and computer instructions stored in the memory and executable on the processor, wherein the processor executes the instructions to implement the steps of the resume retrieval method.

[0016] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided that stores computer instructions, which, when executed by a processor, implement the steps of the resume retrieval method.

[0017] According to a fifth aspect of the present application, a chip is provided that stores computer instructions, which, when executed by the chip, implement the steps of the resume retrieval method.

[0018] In this embodiment, a resume retrieval text is obtained; entity recognition is performed on the resume retrieval text to identify the content of each entity; for each entity, target search keywords associated with that entity are determined using preset association rules; and target resumes are retrieved from the resume database based on each target search keyword. Thus, by determining the target search keywords corresponding to the entity content in the resume retrieval text, more search keywords can be obtained, making the retrieval more comprehensive, avoiding missing resumes that meet the criteria, and eliminating the need for multiple searches, thereby improving the accuracy and efficiency of resume retrieval. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of a resume retrieval system provided according to an embodiment of this application;

[0020] Figure 2 This is a flowchart of a resume retrieval method provided according to an embodiment of this application;

[0021] Figure 3 This is a flowchart of another resume retrieval method provided according to an embodiment of this application;

[0022] Figure 4 This is a flowchart illustrating a method for determining target retrieval keywords according to an embodiment of this application;

[0023] Figure 5 This is a flowchart of another method for determining target retrieval keywords according to an embodiment of this application;

[0024] Figure 6 This is a flowchart illustrating another method for determining target retrieval keywords according to an embodiment of this application;

[0025] Figure 7 This is a flowchart of another method for determining target retrieval keywords according to an embodiment of this application;

[0026] Figure 8 This is a flowchart of yet another resume retrieval method provided according to an embodiment of this application;

[0027] Figure 9 This is a flowchart of another resume retrieval method provided according to an embodiment of this application;

[0028] Figure 10 This is a flowchart of another resume retrieval method provided according to an embodiment of this application;

[0029] Figure 11 This is a schematic diagram of a resume retrieval method provided according to an embodiment of this application;

[0030] Figure 12 This is a schematic diagram of a resume retrieval device according to an embodiment of this application;

[0031] Figure 13 This is a structural block diagram of a computing device provided according to an embodiment of this application. Detailed Implementation

[0032] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0033] The terminology used in one or more embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this application. The singular forms “a,” “the,” and “the” used in one or more embodiments of this application and in the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” used in one or more embodiments of this application refers to and includes any or all possible combinations of one or more associated listed items.

[0034] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this application, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "in response to a determination".

[0035] First, the terminology used in one or more embodiments of the present invention will be explained.

[0036] Resume retrieval: refers to searching a resume database for target resumes that meet the conditions of a search query.

[0037] Resume search text: The text used for resume searching, including the conditions that the target resume must meet. The text content of this resume search text can be in the form of search statements.

[0038] The first category of entity content: Entity content related to additional experience information, which can be additional experiences or additional content about other experiences. For example, additional experiences can be skills acquired, and additional content about other experiences can be graduation time in education or working time in work experience.

[0039] The second category of entity content: Entities related to educational experience, which can be content typically written in the educational experience. For example, the second category of entity content can be the school, major, and subject in the educational experience.

[0040] Abbreviation-type entity content: includes entity content described in both abbreviation and full name.

[0041] Multi-level entity content: includes entity content with multiple levels such as first-level and second-level, and there are relationships between multiple levels.

[0042] BERT: Bidirectional Encoder Representation from Transformers, is an open-source pre-trained language model that trains a language representation model using a large amount of data.

[0043] NER (Named Entity Recognition) refers to the identification of entities with specific meanings in text, mainly including names of people, places, organizations, proper nouns, and times.

[0044] CRF (Conditional Random Field algorithm) is an undirected graphical model that combines the features of maximum entropy model and hidden Markov model. It can be used in sequence labeling tasks such as word segmentation, part-of-speech tagging, and named entity recognition.

[0045] CNN (Convolutional Neural Network) is used to extract features from inputs and then classify, recognize, predict, etc., based on the extracted features.

[0046] Word embedding is a collective term for a set of language modeling and feature learning techniques in Word Embedded Natural Language Processing (NLP), in which words or phrases from the vocabulary are mapped to vectors of real numbers.

[0047] Word embedding refers to the process of embedding a high-dimensional space containing the number of all words into a continuous vector space with a much lower dimension, where each word or phrase is mapped to a vector in the real number field.

[0048] Word units: Before any actual processing is performed on the input text, it needs to be divided into language units such as characters, punctuation marks, numbers, or letters. These language units are called word units. For English text, a word unit can be a word, a punctuation mark, a number, etc.; for Chinese text, the smallest word unit can be a character, a punctuation mark, a number, etc.

[0049] word2vec is a method for word embedding, developed by Mikolov based on the Bengio Neural Network Language Model (NNLM). It's an efficient word vector training method that allows for word embedding of text to obtain word vectors.

[0050] BiLSTM (Bi-directional Long Short-Term Memory) is a combination of forward LSTM and backward LSTM, and is often used in NLP to model contextual information.

[0051] Next, the application scenarios of the resume retrieval method provided in the embodiments of this application will be described.

[0052] Current resume retrieval methods typically rely on character matching based on the resume search text. However, the search text may not contain the full name of the university (e.g., "Jiaotong University" or "Peking University"), while the resume usually contains the full name. Current character matching methods can only search for abbreviations, potentially failing to retrieve resumes including the full name. Similarly, the input of abbreviations or alternative names for the major (e.g., "Computer Science and Technology" could be "Computer," "Computer Science," "Computer Science Category," etc.) may not yield suitable resumes using current character matching algorithms. Likewise, the input of the undergraduate major may differ significantly between undergraduate and graduate programs (e.g., "Automation" for undergraduate and "Control Science and Technology" for graduate). Searching solely by specialization may not identify all eligible resumes. Therefore, current character matching methods may miss some suitable resumes, resulting in insufficient retrieval and reduced accuracy and efficiency.

[0053] Therefore, this application provides a resume retrieval method to solve the above problems. For specific implementation details, please refer to the relevant descriptions of the following embodiments.

[0054] This application provides a resume retrieval method and apparatus, a computing device and a computer-readable storage medium, which will be described in detail in the following embodiments.

[0055] Figure 1 This is a schematic diagram of the structure of a resume retrieval system provided according to an embodiment of this application.

[0056] The execution entity of the resume retrieval method provided in this application embodiment can be a server or a terminal, and this application embodiment does not limit the type of execution. Furthermore, the terminal can be any electronic product capable of human-computer interaction with the user, such as a PC (Personal Computer), mobile phone, PPC (Pocket PC), tablet computer, etc. The server can be a single server, a server cluster composed of multiple servers, or a cloud computing service center, and this application embodiment does not limit the type of server either.

[0057] Taking a terminal as the executing entity as an example, after obtaining the resume search text, the terminal performs entity recognition on the text, identifying each entity content within it. For each entity content, it uses preset association rules to determine the target search keywords associated with that entity content. Based on these target search keywords, it retrieves the target resume from the resume database. Furthermore, in determining the target search keywords associated with the entity content, a first similarity model and a second similarity model may be used. These first and second similarity models can be trained on a server and sent to the terminal.

[0058] Taking a server as the executing entity as an example, after obtaining the resume search text, the server performs entity recognition on the text, identifying each entity content within it. For each entity content, it uses preset association rules to determine the target search keywords associated with that entity content. Based on these target search keywords, it retrieves the target resume from the resume database. Furthermore, the server can independently train a first similarity model and a second similarity model based on training samples, and uses the trained first and second similarity models when determining the target search keywords.

[0059] In this embodiment of the application, by determining the target search keywords corresponding to the entity content in the resume search text, more search keywords can be obtained, making the search more comprehensive. This avoids missing resumes that meet the conditions and eliminates the need for multiple searches, thereby improving the accuracy and efficiency of resume retrieval.

[0060] Those skilled in the art should understand that the above-described terminals and servers are merely examples. Other existing or future terminals or servers that are applicable to the embodiments of this application should also be included within the scope of protection of the embodiments of this application, and are hereby incorporated by reference.

[0061] Figure 2 This is a flowchart of a resume retrieval method provided according to an embodiment of this application, including steps 202 to 208.

[0062] Step 202: Obtain the resume search text.

[0063] As an example, resume search text is text used to filter target resumes. This text includes the criteria the target resume must meet, or the hiring requirements of the recruiting company, and the content of the resume search text can be in the form of a search query. For example, the content of the resume search text could be "People who graduated from Tsinghua University in 2020 with a degree in Computer Science and know Java."

[0064] In some embodiments, the resume search text can be set by the recruiting company according to its recruitment needs or conditions, and can be stored on the terminal, a server, or a search text database. When a resume search is needed, the resume search text can be retrieved from the local terminal, the server, or the search text database.

[0065] In this embodiment, recruitment requirements or conditions are set in the form of resume search text. When searching for resumes, it is not necessary to filter resumes one by one according to the recruitment conditions. Resumes can be searched directly based on the resume search text, which improves the efficiency of resume retrieval.

[0066] Step 204: Perform entity recognition on the resume search text to identify the content of each entity in the resume search text.

[0067] In implementation, after obtaining the resume search text, it is necessary to determine the search terms that can be used for resume retrieval from the resume search text. In this embodiment, entity recognition can be used to determine the entity content in the resume search text, that is, to first determine what entity content exists in the resume search text from a literal perspective.

[0068] In some embodiments, entity recognition models can be used to perform entity recognition on resume search text to determine the entity content within the resume search text. The entity recognition model can be any model capable of recognizing entities in text; by taking the resume search text as input, the output of the entity recognition model is the content of each entity within the resume search text.

[0069] As an example, the entity recognition model can be a NER model. Exemplarily, the NER model can consist of a BERT model and a CRF layer, or it can consist of an RNN model and a CRF layer.

[0070] For example, the NER model can be a BERT model with a CRF layer added after it. The CRF is a conditional random field algorithm. In this NER model, the CRF is a downstream task layer used to constrain the relationship between label transition probabilities.

[0071] As an example, regardless of the structure of the entity recognition model, it can be trained as follows: Obtain sample text and assign labels to it. These labels can indicate whether each character in the sample text is part of an entity content and the corresponding entity category. For instance, characters in the sample text can be labeled using the B, I, O method to indicate whether they are part of an entity content. Here, B can be called Begin, representing the first word of the entity content; I can be called Inside, representing the second to last words of the entity content; and O can be called Outside, representing non-entity content. Furthermore, a category label is added to each character. For example, if a character belongs to the school category, it means it is part of the entity content of the school category; if it belongs to the major category, it means it is part of the entity content of the major category; if it belongs to the time category, it means it is part of the entity content of the time category; if it belongs to the skill category, it means it is part of the entity content of the skill category; and if the category is empty, it means it is not part of the entity content. Labeled sample text can be input into an entity recognition model. The model can then output a predicted label for each character in the sample text. A loss value is determined based on the predicted and ground truth labels for each character. If the loss value is less than a preset loss threshold, the difference between the predicted and actual results is considered small, indicating that the model has achieved a good performance. Therefore, model training can be stopped, resulting in a trained entity recognition model. Furthermore, different labels can be assigned to the sample text for different entity recognition tasks, meaning different entity recognition models can be trained for different tasks.

[0072] For example, suppose the resume search text is "people who graduated from Peking University's Computer Science program in 2021". Inputting this resume search text into an entity recognition model, it can be determined that the entity content in the resume search text includes Peking University, Computer Science, and 2021. Specifically, Peking University belongs to the school category, Computer Science belongs to the major category, and 2021 belongs to the time category. For instance, if the input to the entity recognition model is "people who graduated from Peking University's Computer Science program in 2021", its output could be [(Peking University, school), (Computer Science, major), (2021, time)].

[0073] Furthermore, for time-related entities, the context can be used to determine whether the entity represents graduation date in education or employment date in work experience. This allows for more precise time-based filtering during subsequent resume searches, resulting in more accurate target resumes.

[0074] As an example, if the context of a time-related entity includes characters related to "graduation," then the entity content of that time-related entity can be identified as the graduation date in the educational experience; if the context of a time-related entity includes characters related to "work," then the entity content of that time-related entity can be identified as the work date in the work experience.

[0075] For example, suppose the resume search text is "someone who graduated from Peking University with a computer science degree in 2021". The entity content of this resume search text includes Peking University, computer science, and 2021. Peking University belongs to the school category, computer science belongs to the major category, and 2021 belongs to the time category. Since the word after 2021 is "graduation", it can be determined that 2021 is the graduation time in the education experience. Alternatively, suppose the resume search text is "someone with 3 years of front-end development experience". The entity content of this resume search text includes front-end development and 3 years. Front-end development belongs to the job category, and 3 years belongs to the time category. Since the word after 3 years is "work experience", it can be determined that 3 years is the work time in the work experience.

[0076] In this embodiment, entity recognition is performed on the resume retrieval text using a NER model, identifying all important content (i.e., entity content) in the resume retrieval text. This filters out some unimportant words in the resume retrieval text, allowing for focused retrieval based on entity content, thus improving the efficiency of resume retrieval to some extent.

[0077] Step 206: For each entity content, use preset association rules to determine the target search keywords associated with that entity content.

[0078] Because the input text in the resume search settings may not be strictly standardized—for schools, it might be an abbreviation or alternative name; for majors, it might be an alternative name; or, for some majors, the undergraduate and graduate major names might differ—it's impossible to filter out some qualified resumes solely based on the entity content identified in the resume search text. To address this, this embodiment utilizes preset association rules to determine target search keywords associated with the entity content. For example, it might associate the abbreviation of a school or major with its full name, or associate an undergraduate major with a graduate major, and so on. By identifying words related to the entity content, more qualified resumes can be filtered out during the search, improving the accuracy of resume screening.

[0079] In some embodiments, a preset association rule is a rule that associates entity content with target search keywords. This preset association rule determines the target search keywords associated with each entity content. Furthermore, the preset association rule may differ for different entity content. For example, if the entity content is a school name, the preset association rule could be the association between the school's abbreviation or alternative name and its full name. Similarly, if the entity content is a major name, the preset association rule could be the association between the major's abbreviation or alternative name and its full name. Alternatively, if the entity content is a major name, the preset association rule could be the association between a first-level discipline and N-level disciplines. Or, if the entity content is a major name, the preset association rule could be the association between the major's abbreviation or alternative name and its full name, as well as the association between a first-level discipline and N-level disciplines. Here, N is a positive integer greater than 1.

[0080] For example, suppose the entity content includes "Peking University" and "Computer Science," where "Peking University" is the school category and "Computer Science" is the major category. For the school name, assuming that in the preset association rules, the entity associated with "Peking University" is Peking University, then "Peking University" can be identified as the target search keyword for the entity content of the school category. For the major name, assuming that in the preset association rules, the majors associated with "Computer Science" include Computer Science and Technology, Software Development, Software Engineering, and Computer Network Technology, then "Computer Science and Technology, Software Development, Software Engineering, and Computer Network Technology" can be identified as the target search keywords for the entity content of the major category.

[0081] In this embodiment of the application, by pre-setting association rules, target search keywords associated with the content of each entity are determined, which can obtain more search keywords related to the search conditions, thereby identifying more resumes that meet the conditions and making the search more comprehensive.

[0082] Step 208: Retrieve target resumes from the resume database based on the target search keywords.

[0083] In implementation, after determining the target search keywords associated with each entity's content, resumes containing each target search keyword can be retrieved from the resume database as target resumes. Because the target search keywords are relatively abundant, a large number of resumes meeting the criteria can be retrieved, reducing the chance of missing eligible resumes.

[0084] In some embodiments, since there are multiple target search keywords associated with certain entity content, but these multiple keywords may be in a parallel relationship, such as multiple majors, a resume can be identified as a target resume as long as it includes at least one target search keyword for each category.

[0085] As an example, suppose the target search keywords include "instrument science and technology," "precision instruments and machinery," and "testing and measurement technology and instruments." Since these three are all professional search keywords, resumes containing "instrument science and technology," "precision instruments and machinery," or "testing and measurement technology and instruments" can be retrieved from the resume database as target resumes. For example, this can be done using Elastic Search.

[0086] As an example, suppose the target search keywords include "Peking University," "Computer Science and Technology," and "Computer Networking and Technology." Since both "Computer Science and Technology" and "Computer Networking and Technology" are professional fields, resumes containing "Peking University" and "Computer Science and Technology" can be retrieved from the resume database as target resumes, or vice versa. For example, this can be done using Elastic Search.

[0087] As an example, resume retrieval can be performed using both Elastic Search and MySQL. For target keywords categorized by school, major, and skills, Elastic Search can be used; for target keywords categorized by time, MySQL can be used. Taking the intersection of the resumes retrieved using both methods will yield the target resume. Alternatively, one can first use Elastic Search to obtain candidate resumes, and then use MySQL to retrieve the target resume from those candidate resumes.

[0088] Elasticsearch is suitable for storing unstructured document data and supports intelligent word segmentation and fuzzy matching for queries. For example, in the product search bar of an e-commerce website, when a user enters a string separated by spaces (e.g., "home appliances" or "television"), the Elasticsearch database search engine will perform word segmentation and fuzzy matching queries on the unstructured data stored in the database based on the user's input, returning the top N records that meet the matching criteria to the user. MySQL is an open-source relational database suitable for storing and querying structured data. In a resume retrieval scenario, it can return all resumes that meet the matching criteria.

[0089] As an example, assuming the target search keywords include "Tsinghua University," "Computer Science and Technology," and "≤2020," we can use MySQL to retrieve candidate resumes that meet the condition "≤2020," and then use Elastic Search to retrieve resumes containing "Tsinghua University" and "Computer Science and Technology" from these candidate resumes. The resumes retrieved in this final search will be identified as the target resumes. For instance, if a resume includes "Tsinghua University," "Computer Science and Technology," and "studied from 2016-2020," then that resume can be identified as the target resume.

[0090] In this embodiment of the application, by determining the target search keywords corresponding to the entity content in the resume search text, more search keywords can be obtained, which can identify more resumes that meet the conditions, making the search more comprehensive, avoiding missing resumes that meet the conditions, and eliminating the need for multiple searches, thereby improving the accuracy and efficiency of resume retrieval.

[0091] Figure 3 This is a flowchart of another resume retrieval method provided according to an embodiment of this application, including steps 302 to 308.

[0092] Step 302: Obtain the resume search text.

[0093] Step 304: Perform entity recognition on the resume search text to identify the content of each entity in the resume search text.

[0094] It should be noted that the specific implementation of steps 302-304 can be found in the relevant descriptions of steps 202-204 above, and will not be repeated here in this embodiment.

[0095] Step 306: Identify the first type of entity content as the target search keywords.

[0096] The first category of entity content consists of entities related to additional experience information within the resume search text. This can be additional experiences or additional content about other experiences. For example, additional experiences could be skills acquired, while additional content about other experiences could include graduation dates in education or work dates in work experience. For instance, additional experiences could include proficiency in CAD (Computer-Aided Design) or proficiency in various office software.

[0097] In some embodiments, the first type of entity content typically does not have abbreviations or alternative names. Therefore, for the first type of entity content, the preset association rule can be that the first type of entity content is the target search keyword, that is, the first type of entity content can be directly determined as the target search keyword. For example, assuming that the first type of entity content includes java, then java can be used as the target search keyword associated with the first type of entity content.

[0098] In this embodiment, for entity content related to additional experience information, since there is no inconsistency between the description in the resume and the description in the resume search text, the first type of entity content can be directly identified as the target search keyword. This reduces the operation of processing the first type of entity content to determine the target search keyword and improves the efficiency of determining the target search keyword.

[0099] It should be noted that step 306 is a specific implementation of step 206 above.

[0100] Step 308: Based on the target search keywords, retrieve the target resumes from the resume database.

[0101] As an example, assuming the target search keywords include "Tsinghua University", "Computer Science and Technology", and "java", then the Elastic Search method can be used to retrieve resumes containing "Tsinghua University", "Computer Science and Technology", and "java" from the resume database as target resumes.

[0102] It should be noted that the specific implementation of step 308 can be found in the relevant description of step 208 above, and will not be repeated here in this embodiment.

[0103] In this embodiment, after recognizing the obtained resume search text, the first type of entity content is directly determined as the target search keyword without any other operations, which improves the efficiency of determining the target search keyword. Moreover, the target search keyword is the original word in the resume search text and has a higher degree of correlation with the search conditions. Therefore, the target resume retrieved from the resume database based on the target search keyword is more in line with the user's needs.

[0104] Figure 4 This is a flowchart of a method for determining target retrieval keywords according to an embodiment of this application, including steps 402 to 404.

[0105] Step 402: If the time-related entity content is determined to be a time point based on the preset template, then the time-related entity content is determined as the target search keyword.

[0106] The entity content in the resume search text includes a first category of entity content related to additional experience information, which includes time-related entity content. For example, this time-related entity content could be the graduation date in educational experience or the work dates in work experience.

[0107] The preset template can be a pre-set template used to determine whether the content of a time-related entity is a point in time or a time period.

[0108] In some embodiments, after determining the entity content of the resume search text, for time-related entity content, it can be determined whether the time-related entity content is a point in time or a time period according to a preset template, and different methods are used to determine the target search keywords for point in time and time period.

[0109] As an example, the preset template can be a specific expression related to time, such as "before xx", "after xx", "xx-xx", "xx-present", "xx year", "xx month", etc. "Before xx", "after xx", and "xx-xx year" all indicate that the time-related entity content is a time period, while "xx year" and "xx month" indicate that the time-related entity content is a point in time. For example, assuming the time-related entity content is "before 2020", it can be determined that it conforms to the specific expression "xx years ago", thus determining that "before 2020" is a time period. Or, assuming the time-related entity content is "2021", it can be determined that it conforms to the specific expression "xx year", thus determining that "2021" is a point in time.

[0110] As another example, the preset template may include keywords or key symbols, such as "before," "after," "-," "to," etc. It is determined that time-related entity content including these keywords or key symbols is a time period, while time-related entity content excluding these keywords or key symbols is a point in time. For example, assuming the time-related entity content is "before 2020," it can be determined that it includes the keyword "before," thus determining that "before 2020" is a time period. Alternatively, assuming the time-related entity content is "2021," it can be determined that it does not include the keywords and key symbols included in the preset template, thus determining that "2021" is a point in time.

[0111] In some embodiments, if the content of a time-related entity is determined to be a point in time based on a preset template, then during retrieval, as long as the point in time exists in the resume, it can be determined that the target search keyword exists. Therefore, the content of the time-related entity can be directly determined as the target search keyword.

[0112] Continuing with the example above, assuming the time-related entity content is "2021", the preset template can determine that the time-related entity content is a point in time, so "2021" can be identified as the target search keyword associated with the time-related entity content.

[0113] In this embodiment of the application, when it is determined that the content of a time-type entity is a time point, the time point is directly determined as the target search keyword, which can improve the efficiency of determining search keywords.

[0114] Step 404: If the time-related entity content is determined to be a time period based on the preset template, convert the time-related entity content into a standard time search expression, and determine the standard time search expression as the target search keyword for the time-related entity content.

[0115] Among them, the standard time search expression is a standard search format that search tools need to use when performing time-related search processing.

[0116] In some embodiments, if the time-related entity content is determined to be a time period based on a preset template, a standard search expression may be required to perform the search operation. Therefore, it is necessary to convert the time-related entity content into a standard time search expression and determine the standard time search expression as the target search keyword for the time-related entity content.

[0117] As an example, the following conversion rules can be used to convert time-related entity content into standard time search expressions: If the expression of time-related entity content is "xx years ago", then its corresponding standard time search expression is "≤xx years"; if the expression of time-related entity content is "xx years later", then its corresponding standard time search expression is "≥xx years"; if the expression of time-related entity content is "xx years - xx years", then its corresponding standard time search expression is "xx years - xx years"; if the expression of time-related entity content is "xx years to present", then its corresponding standard time search expression is "xx years to present".

[0118] Continuing with the example above, assuming the time-related entity content is "before 2020", based on the preset template, it can be determined that "before 2020" is a time period, so it can be converted into the standard time search expression "≤2020", and "≤2020" can be determined as the target search keyword for the time-related entity content.

[0119] In this embodiment of the application, when it is determined that the content of a time-type entity is a time period, it is converted into a standard search time format for use with the search tool, which facilitates subsequent search processing.

[0120] It should be noted that steps 402-404 are a specific implementation method when the first type of entity content in step 306 is a time-related entity content. Furthermore, steps 402 and 404 are parallel, and in the specific implementation process, one of steps 402 and 404 is executed.

[0121] Step 406: Retrieve target resumes from the resume database based on target search keywords.

[0122] In this embodiment of the application, since the retrieval methods used for time points and time periods may be different, it is possible to determine whether the time-related entity content is a time point or a time period based on a preset template. The different methods for determining target retrieval keywords for time points and time periods can determine more accurate target retrieval keywords that meet the conditions for using the retrieval tool, thereby improving the efficiency of determining retrieval keywords and thus improving the accuracy of resume retrieval.

[0123] Figure 5 This is a flowchart of another method for determining target retrieval keywords according to an embodiment of this application, including steps 502 to 506.

[0124] Step 502: Query the content of the second type of entity from the keyword database.

[0125] The second type of entity content consists of entity content related to educational experience, which is included in the resume retrieval text.

[0126] As an example, entity content related to educational experience can include abbreviation-type entity content and multi-sub-level entity content. Abbreviation-type entity content is entity content that has a full name, abbreviation, and alternative name, while multi-sub-level entity content is entity content that has multiple sub-levels such as first-level disciplines and second-level disciplines.

[0127] For example, if a school has an abbreviation, an alternative name, and a full name, then the content of the abbreviation-type entity can include the content of the school-type entity. For the same major, the undergraduate major name and the graduate major name may be different. The undergraduate major name is a first-level discipline, and the graduate major name is a second-level discipline. Then the content of the multi-level sub-type entity can be the content of the major-type entity.

[0128] In some embodiments, the keyword library is a pre-defined thesaurus containing multiple keywords, and these keywords are associated with the second type of entity content and conform to the specified description method. Since the second type of entity content has a specified description method and is recorded in the keyword library, after identifying the second type of entity content, it is possible to first query whether the second type of entity content exists in the keyword library, and then determine the target search keywords using different methods based on the query results.

[0129] For example, assuming the second type of entity content includes "Peking University" and "Software Engineering", then we can query the keyword library corresponding to the second type of entity content to see if "Peking University" and "Software Engineering" exist.

[0130] In other embodiments, the keyword database may include a keyword database corresponding to school-related entities and a keyword database corresponding to major-related entities. Thus, after identifying the second type of entity content, one can query for school-related entities within the second type of entity content from the keyword database corresponding to school-related entities, and query for major-related entities within the second type of entity content from the keyword database corresponding to major-related entities. That is, by recording keywords representing the standardized descriptions of schools and majors in different keyword databases, and by querying within different keyword databases, query accuracy can be improved.

[0131] Step 504: If the second type of entity content is found, the second type of entity content is identified as the target search keyword.

[0132] In some embodiments, if the second type of entity content is found, it means that the second type of entity content in the resume search text is described in a standardized way. Therefore, the second type of entity content can be directly identified as the target search keyword.

[0133] Continuing with the example above, suppose the second type of entity content includes "Peking University" (school category) and "Software Engineering" (major category), and "Peking University" and "Software Engineering" are found in the keyword database corresponding to the second type of entity content. Then we can determine that "Peking University" is the full name of the school and "Software Engineering" is the full name of the major. Therefore, "Peking University" and "Software Engineering" can be identified as target search keywords associated with the second type of entity content.

[0134] In this embodiment of the application, when determining that the keyword database is in the second type of entity content, the second type of entity content can be considered as a relatively standardized description. Therefore, the second type of entity content can be directly determined as the target search keyword, reducing the operation of processing the second type of entity content to determine the target search keyword and improving the efficiency of determining the target search keyword.

[0135] Step 506: If no second type of entity content is found, use the preset matching rules to determine the target search keywords associated with the second type of entity content.

[0136] If no second type of entity content is found during implementation, it means that the second type of entity content in the resume search text is not described in a standardized way. Therefore, in order to retrieve more relevant resumes, a preset matching rule can be used to determine the target search keywords associated with the second type of entity content.

[0137] The preset matching rule can be a rule for matching the content of the second type of entity with the target search keywords. This preset matching rule can be used to determine the target search keywords associated with the content of the second type of entity.

[0138] In some embodiments, the preset matching rule can be a rule that associates second-type entity content with target search keywords. Each second-type entity content in the preset matching rule can correspond to one or more target search keywords. For example, assuming the second-type entity content includes "Peking University" and "Computer Science," and "Peking University" and "Computer Science" are not found in the keyword database corresponding to the second-type entity content, it can be determined that neither "Peking University" nor "Computer Science" is a standard descriptive term. Assuming that the preset matching rule includes "Beijing University" corresponding to "Peking University" and "Computer Science" and "Software Engineering" corresponding to "Computer Science," it can be determined that the target search keywords associated with this second-type entity content include "Peking University," "Computer Science and Technology," and "Software Engineering."

[0139] As an example, when the second type of entity content is an abbreviation-based entity content, the preset matching rule can be an abbreviation mapping rule. This abbreviation mapping rule is a rule that can associate abbreviations or aliases with full names, and each abbreviation or alias can correspond to one or more full names. For example, assuming the second type of entity content is "Tsinghua", then in the abbreviation mapping rule, only "Tsinghua University" corresponds to "Tsinghua", so it can be determined that the target search keyword associated with this second type of entity content is "Tsinghua University".

[0140] In this embodiment of the application, when it is determined that there is no second type of entity content in the keyword library, the second type of entity content can be considered as an abbreviation or alias. Therefore, the target search keyword that the second type of entity content matches in the preset matching rules can be determined and used as the target search keyword associated with the second type of entity content. This can avoid the impact of non-standard names (abbreviations or aliases) on resume retrieval.

[0141] It should be noted that steps 502-506 are a specific implementation of step 206 above.

[0142] In this embodiment of the application, for the second type of entity content, a query is first performed in the keyword database. If the query finds the content, it indicates that the second type of entity content is standardized and can be directly used as a target keyword for resume retrieval. If the query does not find the content, it indicates that the second type of entity content is non-standard. Therefore, it is necessary to determine the corresponding standardized target keyword and perform a search based on that target keyword. This can avoid the impact of non-standard names on resume retrieval and improve the accuracy of resume retrieval.

[0143] Figure 6This is a flowchart of another method for determining target retrieval keywords according to an embodiment of this application, including steps 602 to 606.

[0144] Step 602: If no second type of entity content is found, determine the candidate search keywords corresponding to the abbreviation-type entity content based on the abbreviation mapping rule.

[0145] The second type of entity content consists of entity content related to educational experience, which is included in the resume retrieval text.

[0146] As an example, entity content related to educational experience can include abbreviation-type entity content and multi-sub-level entity content. Abbreviation-type entity content is entity content that has a full name, abbreviation, and alternative name, while multi-sub-level entity content is entity content that has multiple sub-levels such as first-level disciplines and second-level disciplines.

[0147] For example, if a school has an abbreviation, an alternative name, and a full name, then the content of the abbreviation-type entity can include the content of the school-type entity. For the same major, the undergraduate major name and the graduate major name may be different. The undergraduate major name is a first-level discipline, and the graduate major name is a second-level discipline. Then the content of the multi-level sub-type entity can be the content of the major-type entity.

[0148] In some embodiments, if no second type of entity content is found, it means that the second type of entity content in the resume search text is not described in a standard way. Therefore, it is necessary to determine the target search keywords associated with the second type of entity content.

[0149] As an example, an abbreviation mapping rule can be a rule that associates abbreviations or aliases with full names, where each abbreviation or alias can correspond to one or more full names. For instance, assuming the abbreviation entity content is a school name, then the abbreviation entity content is the school's abbreviation or alias. Based on this abbreviation mapping rule, the full name of the school can be determined, and the full name of the school can be identified as a candidate search keyword.

[0150] For example, assuming the abbreviation-type entity content is "Jiaoda", and the full names corresponding to "Jiaoda" in the abbreviation mapping rules include "Shanghai Jiaotong University", "Xi'an Jiaotong University" and "Southwest Jiaotong University", then the candidate search keywords corresponding to the abbreviation-type entity content can be determined to include "Shanghai Jiaotong University", "Xi'an Jiaotong University" and "Southwest Jiaotong University".

[0151] Step 604: Determine the similarity between each candidate search keyword and the content of the abbreviation-type entity.

[0152] In practice, word embedding can be performed on each candidate search keyword and abbreviation-type entity content to obtain word vectors for each candidate search keyword and word vectors for each abbreviation-type entity content. Then, based on the word vectors of each candidate keyword and the word vectors of the abbreviation-type entity content, the similarity between each candidate search keyword and the abbreviation-type entity content can be determined.

[0153] In some embodiments, the word2vec algorithm can be used to perform word embedding processing on each candidate search keyword and abbreviation-type entity content to obtain word vectors. Alternatively, each candidate search keyword and abbreviation-type entity content can be input into a word vector model, and the word vector model can be used to determine the word vectors of each candidate search keyword and abbreviation-type entity content. This word vector model can be a trained BERT model.

[0154] In some embodiments, the similarity between two word vectors can be determined by calculating the cosine distance, Euclidean distance, Manhattan distance, or edit distance, thereby obtaining the similarity between each candidate search keyword and the content of the abbreviation-type entity.

[0155] Furthermore, before word embedding, each candidate search keyword can be segmented to obtain multiple first word units, and the abbreviation-type entity content can be segmented to obtain multiple second word units. Then, word embedding is performed on the multiple first word units and multiple second word units to obtain word vectors.

[0156] For example, assuming the candidate search keywords for the abbreviation-type entity "Jiaoda" include "Shanghai Jiaotong University", "Xi'an Jiaotong University", and "Southwest Jiaotong University", we can perform word segmentation on "Jiaoda", "Shanghai Jiaotong University", "Xi'an Jiaotong University", and "Southwest Jiaotong University" respectively, resulting in the following word segmentation results: [Jiaoda] for "Jiaoda", [Shanghai], [Jiaotong], [University] for "Shanghai Jiaotong University", [Xi'an], [Jiaotong], [University] for "Xi'an Jiaotong University", and [Xi'an], [Jiaotong], [University] for "Southwest Jiaotong University". Then, the word2vec algorithm is used to perform word embedding processing on the word segmentation results to obtain the word vector of each word segmentation result. The word vectors of the word segmentation results of the same candidate search keyword are concatenated to obtain the word vector of the candidate search keyword. Then, based on the word vectors of "Jiaoda", "Shanghai Jiaotong University", "Xi'an Jiaotong University", and "Southwest Jiaotong University", the similarity between "Jiaoda" and "Shanghai Jiaotong University", "Jiaoda" and "Xi'an Jiaotong University", and "Jiaoda" and "Southwest Jiaotong University" are determined respectively.

[0157] It should be noted that the word segmentation algorithm used in the embodiments of this application can be any one of the forward maximum matching word segmentation algorithm, the reverse maximum matching word segmentation algorithm, the dictionary-based word segmentation algorithm, etc., and the embodiments of this application do not limit it.

[0158] Step 606: Identify the candidate search keywords with the highest similarity as the target search keywords associated with the abbreviation-type entity content.

[0159] In some embodiments, after determining the similarity between each candidate search keyword and the abbreviation-type entity content, since the candidate search keyword with the highest similarity is closest to the abbreviation-type entity content and also closest to the search conditions, the candidate search keyword with the highest similarity can be determined as the target search keyword associated with the abbreviation-type entity content. For example, after determining the similarity between "Jiaoda" and "Shanghai Jiaotong University", "Xi'an Jiaotong University" and "Southwest Jiaotong University", assuming that "Jiaoda" and "Shanghai Jiaotong University" have the highest similarity, the target search keyword can be determined to be "Shanghai Jiaotong University".

[0160] In other embodiments, after determining multiple target search keywords, the similarity may not be determined. Instead, a web crawler may be used to randomly select one of the multiple candidate search keywords as the target search keyword. Alternatively, the candidate search keyword corresponding to the university with the highest ranking may be determined as the target search keyword based on the ranking of the universities corresponding to the multiple candidate search keywords.

[0161] It should be noted that the above-mentioned similarity-based methods, crawler-based random selection methods, and university ranking methods can be used in any combination to determine target search keywords associated with abbreviation-type entity content. This application embodiment does not limit this.

[0162] It should be noted that steps 602-606 are a specific implementation of step 506 above.

[0163] In this embodiment of the application, for abbreviation-type entity content, any or any combination of abbreviation mapping rules, word vector similarity, university ranking, and web crawling methods are used to determine the target search keywords. This can determine the full name with the highest similarity to the abbreviation of the school name, thus solving the problem that school abbreviations affect resume retrieval.

[0164] Figure 7 This is a flowchart of another method for determining target retrieval keywords according to an embodiment of this application, including steps 702 to 708.

[0165] Step 702: If no second type of entity content is found, input the multi-sub-level entity content into the first similarity model to determine the first-level candidate search keywords related to the multi-sub-level entity content.

[0166] In practice, if no second-category entity content is found in the keyword database, it indicates that the second-category entity content in the resume search text is not described using a standardized descriptive method. Therefore, it is necessary to determine the target search keywords associated with the second-category entity content. As an example, when multiple sub-category entity contents are professional category entity contents, the primary candidate search keyword can be a first-level discipline or undergraduate major name.

[0167] In some embodiments, the first similarity model can output the full name based on the input abbreviation or alias. For example, the first similarity model can be a trained BERT model.

[0168] As an example, the training process of this first similarity model is as follows: Multiple sets of training samples are obtained, each set including two sample texts, and each set carries a true label. The true label of each training sample indicates whether the two sample texts in that set are similar; if similar, the true label is 1, and if dissimilar, the true label is 0. The training samples are input into the BERT model, which outputs the predicted label for each set of training samples. Based on the predicted label and the true label, a loss value is determined. If the loss value is greater than or equal to a preset loss threshold, it indicates a large gap between the predicted and actual results, and the model's performance is not good enough. Therefore, the model continues to be trained. If the loss value is less than the preset loss threshold, it can be considered that the gap between the predicted and actual results is small, meaning the model has achieved a relatively good effect, and model training can be stopped, resulting in the completed first similarity model.

[0169] For example, taking the content of multiple child-level entities as professional entity content, assuming that the professional entity content is "computer", then inputting "computer" into the first similarity model, the model can output multiple professional fields related to computer as first-level candidate search keywords.

[0170] Step 704: Input the first-level candidate search keywords into the second similarity model to determine the second-level candidate search keywords corresponding to the first-level candidate search keywords.

[0171] In some embodiments, the second similarity model can determine the corresponding second-level candidate search keywords based on the input first-level candidate search keywords. For example, the second similarity model can be a BiLSTM model.

[0172] In some embodiments, the training process of the second similarity model is as follows: First-level text and its corresponding second-level text are obtained, and each set of first-level text and each second-level text is grouped into a training sample, resulting in multiple sets of training samples. Each set of training samples carries a true label, which indicates whether the first-level text and the second-level text in that set are in a hierarchical relationship. If they are, the true label is 1; otherwise, it is 0. The training samples are input into a BiLSTM model, which outputs the predicted label for each set of training samples. Based on the predicted label and the true label, a loss value is determined. If the loss value is greater than or equal to a preset loss threshold, it indicates a large gap between the predicted result and the actual result, and the model's performance is not good enough. Therefore, the model continues to be trained. If the loss value is less than the preset loss threshold, it can be considered that the gap between the predicted result and the actual result is small, meaning the model has achieved a relatively good effect. Model training can then be stopped, resulting in the trained second similarity model. As an example, the first-level text can be a first-level subject, and the second-level text can be a second-level subject; or the first-level text can be an undergraduate major, and the second-level text can be a graduate major.

[0173] For example, assuming that the primary candidate search keywords include "instrument science and technology", inputting "instrument science and technology" into the second similarity model can output the corresponding secondary disciplines "precision instruments and machinery" and "testing and measurement technology and instruments".

[0174] In other embodiments, taking the abbreviation of a professional name as an example, the second-level discipline corresponding to the first-level discipline can be determined based on a pre-created correspondence table of first-level and second-level disciplines, or the graduate major corresponding to the undergraduate major can be determined based on a pre-created correspondence table of undergraduate and graduate majors. That is, the second-level candidate search keywords corresponding to the first-level candidate search keywords are determined based on the correspondence table of disciplines or the correspondence table of majors.

[0175] Step 706: Determine the union of the first-level candidate search keywords and the second-level candidate search keywords as the target search keywords associated with the content of multi-child class entities.

[0176] In some embodiments, primary candidate search keywords include undergraduate majors, and secondary candidate search keywords include graduate majors. Both undergraduate and graduate majors meet the professional requirements in the resume search text. Therefore, the union of primary and secondary candidate search keywords can be determined as the target search keyword, which can avoid missing some search keywords.

[0177] It should be noted that steps 702-706 are a specific implementation of step 506 above.

[0178] In this embodiment, for multi-sub-level entity content, the first similarity model and the second similarity model are combined to determine the target search keywords. This can determine the full name of the major corresponding to the abbreviated major name, and can also obtain the graduate major corresponding to the undergraduate major. This solves the problem that the abbreviation of the major, the first and second-level disciplines, and the difference between undergraduate and graduate major names affect resume retrieval.

[0179] Figure 8 This is a flowchart of another resume retrieval method provided according to an embodiment of this application, including steps 802 to 806.

[0180] Step 802: Determine the similarity between the target search keywords and the content of multi-level class entities.

[0181] In practice, after identifying the target search keywords associated with the content of multi-level subclass entities, since these target search keywords belong to the same category, the resume is considered to meet the search criteria of that category as long as any one of these target search keywords (i.e., the target search keywords associated with the content of multi-level subclass entities) exists in the resume. However, the relevance between these target search keywords and the search criteria is different. When searching resumes, it is usually desirable to quickly filter out the most suitable target resumes. Therefore, the similarity between the target search keywords and the content of multi-level subclass entities can be determined, and resumes can be searched according to the degree of similarity. In this way, the target resumes that best meet the recruitment requirements can be filtered out as quickly as possible, improving the efficiency of resume retrieval.

[0182] In practice, word embedding processing can be performed on the target search keywords and the content of multiple child-level entities to obtain the word vectors of the target search keywords and the word vectors of the content of multiple child-level entities. Then, based on the word vectors of the target keywords and the word vectors of the content of multiple child-level entities, the similarity between the target search keywords and the content of multiple child-level entities can be determined.

[0183] In some embodiments, the word2vec algorithm can be used to embed the target search keywords and multi-level class entity content into words to obtain word vectors. Alternatively, the target search keywords and multi-level class entity content can be input into a word vector model, which determines the word vectors of the target search keywords and multi-level class entity content. This word vector model can be a trained BERT model.

[0184] In some embodiments, the similarity between two word vectors can be determined by calculating the cosine distance, Euclidean distance, Manhattan distance, or edit distance, thereby obtaining the similarity between the target retrieval keyword and the content of multiple child-level entity classes.

[0185] Furthermore, before word embedding, the target search keywords can be segmented to obtain multiple third word units, and the content of multiple child-level entities can be segmented to obtain multiple fourth word units. Then, word embedding is performed on the multiple third word units and multiple fourth word units to obtain word vectors.

[0186] For example, suppose the multi-level entity content is "instrument category", and the associated target search keywords include "instrument science and technology", "precision instruments and machinery", and "testing and measurement technology and instruments". We can perform word segmentation on "instrument category", "instrument science and technology", "precision instruments and machinery", and "testing and measurement technology and instruments" respectively, resulting in the following word segmentation results: [instrument category] for "instrument category", [instrument], [and], [technology] for "instrument science and technology", [precision], [instrument], [and], [machinery] for "precision instruments and machinery", and [testing and measurement technology and instruments], [testing], [measurement]. The words are categorized as "[technology]", "[and]", and "[instrument]". The word2vec algorithm is then used to embed the word segments, resulting in word vectors for each segment. The word vectors of the same candidate search keyword are concatenated to form the word vector for that keyword. Based on the word vectors for "instrument category", "instrument science and technology", "precision instruments and machinery", and "testing and measurement technology and instruments", the similarity between "instrument category" and "instrument science and technology", "instrument category" and "precision instruments and machinery", and "instrument category" and "testing and measurement technology and instruments" are determined.

[0187] In other embodiments, the similarity between the first-level target search keywords and the content of multiple sub-level class entities can be determined by the word vector similarity method, and the similarity between each second-level target search keyword and the content of multiple sub-level class entities can be determined, as well as the similarity between the first-level target search keywords corresponding to each second-level target search keyword and the content of multiple sub-level class entities can be determined.

[0188] For example, taking a professional entity content X as an example, the first-level target search keyword can be a first-level discipline, and the second-level target search keyword can be a second-level discipline. Assuming that first-level discipline A corresponds to three second-level disciplines A1, A2, and A3, and first-level discipline B corresponds to one second-level discipline B1, and the similarity between first-level discipline A and professional entity content X is 'a', and the similarity between first-level discipline B and professional entity content X is 'b', then it is determined that the similarity between second-level disciplines A1, A2, and A3 and professional entity content X is 'a', and the similarity between second-level discipline B1 and professional entity content X is 'b'.

[0189] It should be noted that step 802 can be a step performed after step 506 or step 708.

[0190] Step 804: Based on the similarity ranking, search the resume database sequentially using the target search keywords to obtain target resumes containing the target search keywords.

[0191] In this embodiment of the application, the resume database can be searched sequentially using the target search keywords in descending order of similarity, or the resume database can be searched sequentially using the target search keywords in ascending order of similarity.

[0192] In the first possible implementation, target resumes containing the target search keywords with the highest similarity can be retrieved from the resume database first. Then, target resumes containing the target search keywords with the second highest similarity can be retrieved from the resume database, and so on. In this way, target resumes containing the target search keywords can be retrieved, and the target resumes are obtained in descending order of their degree of compliance with the search criteria.

[0193] As an example, the target search keywords can be sorted before the search to obtain a list of target search keywords. Then, target resumes can be retrieved from the resume database according to the order of the target search keywords in the list. Specifically, the target search keywords can first be sorted in descending order of similarity to obtain a list of target search keywords for multi-level entity content. Then, the resume database can be searched sequentially using the target search keywords according to the order of the target search keywords for multi-level entity content in the list to obtain target resumes that include the target search keywords.

[0194] The target search keyword list is a table obtained by sorting the target search keywords in descending order of similarity. For example, see Table 1, which is an exemplary target search keyword list provided in an embodiment of this application. As can be seen, the target search keyword list includes multiple target search keywords, each with a corresponding serial number, and the serial number of each target search keyword is a sorting of the similarity between that target search keyword and the content of multiple child-level entity classes.

[0195] Table 1 List of target search keywords

[0196]

[0197]

[0198] In some embodiments, candidate resumes can be first determined, including target search keywords associated with abbreviated entity content and target search keywords associated with first type of entity content. Then, target resumes including the first ranked target search keyword in the target search keyword list can be determined from the candidate resumes. Next, target resumes including the second ranked target search keyword in the target search keyword list can be determined from the candidate resumes, and so on, until all target search keywords in the target search keyword list have been searched and the target resumes are obtained, that is, multiple target resumes are obtained.

[0199] For example, suppose the target search keyword associated with the abbreviation-type entity content is "Tsinghua University," the target search keywords associated with the first type of entity content include "java" and "2021," and the target search keyword list includes the first-ranked target search keyword "Instrument Science and Technology," the second-ranked target search keyword "Precision Instruments and Machinery," and the third-ranked target search keyword "Testing and Measurement Technology and Instruments." When conducting a resume search, resumes containing "Tsinghua University," "java," and "2021" can be identified as candidate resumes. Then, resume A containing "Instrument Science and Technology" can be retrieved from the candidate resumes. Next, resume B containing "Precision Instruments and Machinery" can be retrieved from the candidate resumes. Finally, resume C containing "Testing and Measurement Technology and Instruments" can be retrieved from the candidate resumes. Resumes A, B, and C can then be identified as the target resumes.

[0200] In other embodiments, resumes that include the first ranked target search keyword in the target search keyword list can be determined first, then resumes that include the second ranked target search keyword in the target search keyword list can be determined, and so on, until all target search keywords in the target search keyword list have been searched and candidate resumes are obtained. Then, target resumes that include target search keywords associated with abbreviation-type entity content and target search keywords associated with first-type entity content can be determined from the candidate resumes.

[0201] As an example, resume retrieval can be performed using both Elastic Search and MySQL. For target keywords categorized by school, major, and skills, Elastic Search can be used; for target keywords categorized by time, MySQL can be used. Taking the intersection of the resumes retrieved using both methods will yield the target resume. Alternatively, one can first use Elastic Search to obtain candidate resumes, and then use MySQL to retrieve the target resume from those candidate resumes.

[0202] In the second possible implementation, target resumes containing the least similar keywords can be retrieved from the resume database first. Then, target resumes containing the second most similar keywords can be retrieved, and so on. This way, target resumes containing the target keywords can be retrieved, and the target resumes are obtained in ascending order of their similarity to the requirements. This implementation is similar to the above method of retrieving resumes in ascending order of similarity, only the order is reversed. For specific implementation details, please refer to the relevant descriptions above; this embodiment will not repeat them here.

[0203] Furthermore, when identifying target resumes using the first method described above, the order in which the resumes are retrieved is positively correlated with the degree to which they match the search criteria. Therefore, the resumes can be sorted according to the order in which they were retrieved. Conversely, when identifying target resumes using the second method, the order in which they are retrieved is negatively correlated with the degree to which they match the search criteria. Therefore, the resumes can be sorted in the reverse order of the order in which they were retrieved. In this way, recruiting companies can see the target resumes that are most relevant to their search criteria and best meet their recruitment needs first, thus facilitating their recruitment process.

[0204] In addition to sorting target resumes from highest to lowest relevance to the search criteria as described above, they can also be sorted from lowest to highest relevance to the search criteria. In this case, target resumes can be filtered in reverse order, or a portion of the target resumes can be extracted for filtering, or resumes can be filtered sequentially. This application does not limit the sorting method of target resumes or the filtering method used by recruiting companies.

[0205] It should be noted that step 804 is a specific implementation of step 208.

[0206] In this embodiment, after determining the target search keywords associated with the content of multiple child-level entities, they are sorted from high to low according to their similarity to the content of the multiple child-level entities, and the sorting results are stored in the target search keyword list. Then, when searching for resumes in the future, the professional aspects can be examined according to the order of the target search keywords in the target search keyword list, so as to quickly find the target resumes that meet the professional requirements, thereby improving the efficiency and accuracy of resume retrieval. Furthermore, sorting the target resumes according to the order of the retrieved target resumes provides convenience for users processing the target resumes.

[0207] Figure 9 This is a flowchart of another resume retrieval method provided according to an embodiment of this application, including steps 902 to 910.

[0208] Step 902: Obtain the resume search text.

[0209] It should be noted that the specific implementation of step 902 can be found in the relevant description of step 202 above, and will not be repeated here in this embodiment.

[0210] Step 904: Standardize the resume search text to obtain a standard format resume search text.

[0211] In some embodiments, since the resume search text may contain some erroneous characters, spaces, hyphens and other meaningless characters, these characters are not helpful for resume retrieval and may affect the subsequent processing of the resume search text. Therefore, the resume search text can be standardized by deleting or modifying the meaningless characters to obtain a standard format resume search text.

[0212] As an example, standardizing resume search text can involve correcting typos and removing meaningless characters such as spaces. For instance, if the resume search text is "2020 Tsinghua University Computer Science graduate who knows Java", then the hyphens after 2020 and the spaces after Java can be removed to obtain the standard format resume search text "2020 Tsinghua University Computer Science graduate who knows Java".

[0213] Step 906: Perform entity recognition on the standard format resume search text to identify the content of each entity in the resume search text.

[0214] Step 908: For each entity content, use preset association rules to determine the target search keywords associated with that entity content.

[0215] Step 910: Retrieve target resumes from the resume database based on the target search keywords.

[0216] It should be noted that the specific implementation of steps 906-910 can be found in the relevant descriptions of steps 206-208 above, and will not be repeated here in this embodiment.

[0217] In this embodiment, after obtaining the resume retrieval text, standardization processing is performed first, which can filter out factors in the resume retrieval text that are irrelevant to resume retrieval or affect resume retrieval, avoid the influence of irrelevant factors on the subsequently determined target search keywords, and thus improve the accuracy of resume retrieval.

[0218] Figure 10 This is a flowchart of another resume retrieval method provided according to an embodiment of this application. Figure 11It is a schematic diagram of a resume retrieval method provided according to an embodiment of the present application. Next, in combination with Figure 10 and Figure 11 , taking the resume retrieval text as "students who graduated from the computer major of Peking University before 2020 and know Java and Python" as an example, the specific process of the resume retrieval method will be described.

[0219] Step 1002: Obtain the resume retrieval text.

[0220] Referring to Figure 11 , first obtain the resume retrieval text "students who graduated from the computer major of Peking University before 2020 and know Java and Python".

[0221] Step 1004: Perform standardization processing on the resume retrieval text to obtain the resume retrieval text in standard format.

[0222] As an example, the standardization processing can also be called preprocessing. Referring to Figure 11 , perform preprocessing on the resume retrieval text to obtain the standardized resume retrieval text "students who graduated from the computer major of Peking University before 2020 and know Java and Python".

[0223] Step 1006: Perform entity recognition on the resume retrieval text to identify each entity content in the resume retrieval text.

[0224] Referring to Figure 11 , the entity content obtained after entity recognition includes [(computer, major), (Peking University, school), (Java, skill), (before 2020, time), (Python, skill)].

[0225] Step 1008: For the school name in the entity content, query the school name from the keyword library. If the school name is queried, execute step 1010. If the school name is not queried, execute step 1012.

[0226] Step 1010: Determine the school name as the target retrieval keyword of the school category.

[0227] Step 1012: Determine the full name of the school through the abbreviation mapping rule, and then determine the target retrieval keyword of the school category through any one of the methods of word vector similarity, crawler, and university ranking.

[0228] For example, through the abbreviation mapping rule, the full names of "Peking University" include "Peking University", "Northeastern University", and "University of Science and Technology Beijing". Through word vector similarity, it is determined that the similarity of "Peking University" is the highest, then it is determined that the target retrieval keyword of the school category is "Peking University". Referring to Figure 11 , Figure 11The target search keyword for the school category, determined by four methods—school abbreviation, abbreviation mapping rules, university ranking, and word vector similarity—is "Peking University," which can be represented as ['Peking University', school].

[0229] Step 1014: For the professional name in the entity content, search for the professional name in the keyword database. If the professional name is found, proceed to step 1016. If the professional name is not found, proceed to step 1018.

[0230] Step 1016: Determine the name of the major as the target search keyword for the major category.

[0231] See Figure 11 Determine if the major name exists in an SQL database (relational database). If it does, directly identify the major name as the target search keyword for the major category.

[0232] Step 1018: Determine the target search keywords for the professional category using the first similarity model, the second similarity model, and the ranking algorithm.

[0233] As an example, the first similarity model could be the BERT model, the second similarity model could be the BiLSTM model, and the ranking algorithm could be the w2v similarity re-ranking algorithm.

[0234] See Figure 11 The system retrieves first-level disciplines related to the major name using the BERT model, determines second-level disciplines using the BiLSTM model, and sorts the union of first-level and second-level disciplines using the w2v similarity re-ranking algorithm to obtain target search keywords for the major category, including ['Computer', 'Computer and Applications', 'Computer Science and Engineering', 'Computer Engineering Department', ...].

[0235] Step 1020: For skills in entity content, directly identify the skill as the target search keyword for the skill category.

[0236] See Figure 11 The target search keywords for skills include "java" and "python", which can be represented as ['java', 'python', skills].

[0237] Step 1022: For time in the entity content, directly determine time as the target search keyword for the time category.

[0238] See Figure 11 Time-related target search keywords include "before 2020", which can be represented as ['before 2020', time].

[0239] Step 1024: For target search keywords related to schools, majors, and skills, use Elastic Search for retrieval; for target search keywords related to time, use MySQL for retrieval.

[0240] See Figure 11 The system uses Elastic Search to search for resumes based on the keywords "Peking University", "Computer Science", "Computer Science and Applications", "Computer Science and Engineering", "Computer Engineering", "Java", and "Python". It also uses MySQL to search for resumes based on the keyword "before 2020".

[0241] Step 1026: Take the intersection of the resumes retrieved by the two search methods to obtain the target resume.

[0242] See Figure 11 The target resume can be obtained by taking the intersection of the resumes retrieved using Elasticsearch and the resumes retrieved using MySQL.

[0243] It should be noted that the specific implementation of steps 1002-1026 in this embodiment can be found in the relevant descriptions of the above embodiments, and will not be repeated here.

[0244] In this embodiment, the problem of abbreviated school names is solved by methods such as abbreviation rules. The problems of different major abbreviations, first and second-level disciplines, and undergraduate and graduate major names are solved by using two models, BERT and BiLSTM. More target search keywords are identified, making the search more comprehensive. This can avoid missing resumes that meet the conditions and eliminate the need for multiple searches, thereby improving the accuracy and efficiency of resume retrieval.

[0245] Corresponding to the above method embodiments, this application also provides embodiments of a resume retrieval device. Figure 12 This is a schematic diagram of a resume retrieval device according to an embodiment of this application. Figure 12 As shown, the device 1200 includes:

[0246] Module 1202 is configured to retrieve resume search text.

[0247] The recognition module 1204 is configured to perform entity recognition on the resume search text and recognize the content of each entity in the resume search text.

[0248] The determination module 1206 is configured to determine the target search keywords associated with each entity content using preset association rules;

[0249] The retrieval module 1208 is configured to retrieve target resumes from the resume database based on various target search keywords.

[0250] Optionally, the content of each entity includes a first type of entity content related to additional experience information;

[0251] The determining module 1206 is further configured as follows:

[0252] The content of the first type of entity is determined as the target search keyword.

[0253] Optionally, the first type of entity content includes time-related entity content;

[0254] The determining module 1206 is further configured as follows:

[0255] If the time-related entity content is determined to be a time point based on a preset template, then the time-related entity content is determined as the target search keyword;

[0256] If the time-related entity content is determined to be a time period based on a preset template, the time-related entity content is converted into a standard time search expression, and the standard time search expression is determined as the target search keyword for the time-related entity content.

[0257] Optionally, the content of each entity includes a second type of entity content related to educational experience;

[0258] The determining module 1206 is further configured as follows:

[0259] Search the keyword database for the content of the second type of entity;

[0260] If the second type of entity content is found, the second type of entity content will be identified as the target search keyword;

[0261] If no content of the second type of entity is found, the target search keywords associated with the second type of entity are determined using preset matching rules.

[0262] Optionally, the second type of entity content includes abbreviation-type entity content;

[0263] The determining module 1206 is further configured as follows:

[0264] Based on the abbreviation mapping rules, candidate search keywords corresponding to the content of the abbreviation-type entity are determined;

[0265] Determine the similarity between each candidate search keyword and the content of the abbreviation-type entity;

[0266] The candidate search keywords with the highest similarity are determined as the target search keywords associated with the content of the abbreviation-type entity.

[0267] Optionally, the second type of entity content includes multi-subclass entity content;

[0268] The determining module 1206 is further configured as follows:

[0269] Input the content of the multi-sub-level entity class into the first similarity model to determine the first-level candidate search keywords related to the content of the multi-sub-level entity class;

[0270] The first-level candidate search keywords are input into the second similarity model to determine the second-level candidate search keywords corresponding to the first-level candidate search keywords;

[0271] The union of the first-level candidate search keywords and the second-level candidate search keywords is determined as the target search keyword associated with the content of the multi-sub-level entity.

[0272] Optionally, the determining module 1206 is further configured to:

[0273] Determine the similarity between the target search keywords and the content of the multi-level class entities;

[0274] Accordingly, the retrieval module 1208 is further configured as follows:

[0275] Based on the similarity ranking, the resumes in the resume database are searched sequentially using the target search keywords to obtain target resumes containing the target search keywords.

[0276] Optionally, the device further includes:

[0277] The standardization processing module is configured to standardize the resume search text to obtain a standard format resume search text.

[0278] The resume retrieval device provided in this application can obtain more retrieval keywords by determining the target retrieval keywords corresponding to the entity content in the resume retrieval text. This allows for the identification of more resumes that meet the criteria, making the retrieval more comprehensive. It can avoid missing resumes that meet the criteria and eliminates the need for multiple searches, thereby improving the accuracy and efficiency of resume retrieval.

[0279] The above is a schematic scheme of a resume retrieval device according to this embodiment. It should be noted that the technical solution of this resume retrieval device and the technical solution of the resume retrieval method described above belong to the same concept. For details not described in detail in the technical solution of the resume retrieval device, please refer to the description of the technical solution of the resume retrieval method described above.

[0280] It should be noted that each component in the device claim should be understood as a functional module necessary to implement each step of the program flow or method, and the functional modules are not actual functional divisions or separations. A device claim defined by such a set of functional modules should be understood as a functional module architecture that implements the solution primarily through the computer program described in the specification, and not as a physical device that implements the solution primarily through hardware.

[0281] Figure 13 A structural block diagram of a computing device 1300 according to an embodiment of this application is shown. The components of the computing device 1300 include, but are not limited to, a memory 1310 and a processor 1320. The processor 1320 is connected to the memory 1310 via a bus 1330, and a database 1350 is used to store data.

[0282] The computing device 1300 also includes an access device 1340, which enables the computing device 1300 to communicate via one or more networks 1360. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 1340 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0283] In one embodiment of this application, the aforementioned components of the computing device 1300 and Figure 13 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 13 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this application. Those skilled in the art can add or replace other components as needed.

[0284] The computing device 1300 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. The computing device 1300 can also be a mobile or stationary server.

[0285] The processor 1320 is used to execute the following computer-executable instructions:

[0286] Obtain the resume search text;

[0287] Entity recognition is performed on the resume search text to identify the content of each entity in the resume search text;

[0288] For each entity content, target search keywords associated with that entity content are determined using preset association rules;

[0289] Based on the target search keywords, the system retrieves target resumes from the resume database.

[0290] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the resume retrieval method described above belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the resume retrieval method described above.

[0291] An embodiment of this application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of the resume retrieval method as described above.

[0292] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solution of the resume retrieval method described above. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the resume retrieval method described above.

[0293] This application discloses a chip that stores computer instructions, which, when executed by a processor, implement the steps of the resume retrieval method described above.

[0294] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0295] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0296] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0297] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0298] The preferred embodiments disclosed above are merely illustrative of this application. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this application. These embodiments are selected and specifically described in this application to better explain the principles and practical applications of this application, thereby enabling those skilled in the art to better understand and utilize this application. This application is limited only by the claims and their full scope and equivalents.

Claims

1. A resume retrieval method, characterized in that, The method includes: Obtain the resume search text; Entity recognition is performed on the resume search text to identify the content of each entity in the resume search text; For each entity content, target search keywords associated with the entity content are determined using preset association rules. The entity content includes at least one of the following: time-related entity content related to additional experience information, abbreviation-related entity content related to educational experience, and multi-level entity content related to educational experience. Different preset association rules are used for different types of entity content. The preset association rules include at least one of the following: abbreviation mapping rules, rules on the association between first-level disciplines and multi-level disciplines, and rules on determining whether the time-related entity content is the graduation time in educational experience or the working time in work experience based on the characters in the context of the time-related entity content. Based on the target search keywords, the system retrieves target resumes from the resume database.

2. The method as described in claim 1, characterized in that, The content of each entity includes the first type of entity content related to additional experience information; For each entity content, using preset association rules, the target search keywords associated with that entity content are determined, including: The content of the first type of entity is determined as the target search keyword.

3. The method as described in claim 2, characterized in that, The first type of entity content includes time-related entity content; The step of determining the content of the first type of entity as the target search keyword includes: If the time-related entity content is determined to be a time point based on a preset template, then the time-related entity content is determined as the target search keyword; If the time-related entity content is determined to be a time period based on a preset template, the time-related entity content is converted into a standard time search expression, and the standard time search expression is determined as the target search keyword for the time-related entity content.

4. The method as described in claim 1, characterized in that, The content of each entity includes a second category of entity content related to educational experience; For each entity content, using preset association rules, the target search keywords associated with that entity content are determined, including: Search the keyword database for the content of the second type of entity; If the second type of entity content is found, the second type of entity content will be identified as the target search keyword; If no content of the second type of entity is found, the target search keywords associated with the second type of entity are determined using preset matching rules.

5. The method as described in claim 4, characterized in that, The second type of entity content includes abbreviation-based entity content; The step of determining the target search keywords associated with the second type of entity content using preset matching rules includes: Based on the abbreviation mapping rules, candidate search keywords corresponding to the content of the abbreviation-type entity are determined; Determine the similarity between each candidate search keyword and the content of the abbreviation-type entity; The candidate search keywords with the highest similarity are determined as the target search keywords associated with the content of the abbreviation-type entity.

6. The method as described in claim 4, characterized in that, The second type of entity content includes multi-level subclass entity content; The step of determining the target search keywords associated with the second type of entity content using preset matching rules includes: Input the content of the multi-sub-level entity class into the first similarity model to determine the first-level candidate search keywords related to the content of the multi-sub-level entity class; The first-level candidate search keywords are input into the second similarity model to determine the second-level candidate search keywords corresponding to the first-level candidate search keywords; The union of the first-level candidate search keywords and the second-level candidate search keywords is determined as the target search keyword associated with the content of the multi-sub-level entity.

7. The method as described in claim 6, characterized in that, After determining the union of the first-level candidate search keywords and the second-level candidate search keywords as the target search keywords associated with the content of the multi-child class entity, the method further includes: Determine the similarity between the target search keywords and the content of the multi-level class entities; The process of retrieving target resumes from the resume database based on each target search keyword includes: Based on the similarity ranking, the resumes in the resume database are searched sequentially using the target search keywords to obtain target resumes containing the target search keywords.

8. The method according to any one of claims 1-7, characterized in that, Before performing entity recognition on the resume search text, the method further includes: The resume search text is standardized to obtain a standard format resume search text.

9. A resume retrieval device, characterized in that, The device includes: The acquisition module is configured to retrieve resume search text; The recognition module is configured to perform entity recognition on the resume search text and recognize the content of each entity in the resume search text. The determination module is configured to determine the target search keywords associated with each entity content using preset association rules. Each entity content includes at least one of the following: time-related entity content related to additional experience information, abbreviation-related entity content related to educational experience, and multi-level entity content related to educational experience. Different preset association rules are used for different types of entity content. When the entity content is abbreviation-related entity content related to educational experience, the preset association rule is an abbreviation mapping rule, which associates abbreviations or alternative names with full names. When the entity content is time-related entity content related to additional experience information, the time-related entity content is determined to be either graduation date in educational experience or work date in work experience based on the characters preceding and following it. The search module is configured to retrieve target resumes from the resume database based on various target search keywords.

10. A computing device, comprising a memory, a processor, and computer instructions stored in the memory and executable on the processor, characterized in that, When the processor executes the instructions, it implements the steps of the method according to any one of claims 1-8.

11. A computer-readable storage medium storing computer instructions, characterized in that, When executed by the processor, this instruction implements the steps of the method according to any one of claims 1-8.

12. A computer program product, characterized in that, Includes computer instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-8.

Images