Resume screening method and device based on large model prompt instruction and medium

Abstract

The invention discloses a resume screening method and device based on a large model prompt instruction and a medium, and relates to the field of data processing, and the method comprises the steps: obtaining resume information of a corresponding structure based on a plurality of sources; defining basic information items corresponding to the resume, and generating a large model prompt instruction; taking the resume information and the large model prompt instruction as input, and outputting and storing corresponding structured information through the trained large model; obtaining a query instruction input by a user, and extracting corresponding key information; and generating a corresponding structured query language according to the key information, and querying in a database storing the structured information through the structured query language. Unified structuring is carried out on resume information of different formats and sources through a large model prompt instruction, analysis failures caused by format differences in a traditional method are avoided, and the mass multi-source heterogeneous data processing efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and specifically to a resume screening method, device and medium based on large model prompt instructions. Background Art

[0002] At present, artificial intelligence and natural language processing technologies have been applied to the field of human resource recruitment to a certain extent, mainly in the intelligent screening and analysis of resumes.

[0003] Intelligent resume screening and parsing technology can extract key information from resumes, such as name, educational background and work experience. However, due to the diverse sources of resumes, the formats of different resumes are different, and considering the huge scale of resume data for some large companies during the peak recruitment season, there are problems of low screening efficiency and incomplete information capture.

[0004] In traditional solutions, the resume initial screening function is based on the information analysis of electronic resumes. It lacks the function of digitizing paper resumes and is difficult to process massive multi-source heterogeneous resume data in a short period of time, resulting in low resume screening efficiency.

[0005] In addition, resume screening methods based on scoring and ranking may ignore subtle but important information in the candidate's resume, such as soft skills such as communication skills and leadership skills, and career development trajectories that reflect career stability and development potential, resulting in incomplete information capture. Summary of the invention

[0006] In order to solve the above problems, this application proposes a resume screening method based on large model prompt instructions, including: Based on multiple sources, obtaining resume information of corresponding structure, and preprocessing the resume information; Defining basic information items corresponding to the resume, and generating large model prompt instructions based on the basic information items; Taking the resume information and the large model prompt instruction as input, extracting and structuring the resume information according to the large model prompt instruction through the trained large model, and outputting and storing the corresponding structured information; Obtaining the query instruction input by the user, and parsing the query instruction based on the natural language processing algorithm to extract the corresponding key information; A corresponding structured query language is generated according to the key information, and a query is performed in a database storing the structured information using the structured query language to return a corresponding query result.

[0007] In one example, based on multiple sources, resume information of corresponding structure is obtained, and the resume information is preprocessed, specifically including: Determine a plurality of preset sources; the sources include: a local database and a third-party system; For the local database, corresponding resume information is obtained by importing; and for the third-party system, resume information uploaded in the third-party system is obtained through a set system service interface; Determine the type of the resume information; the type includes: document type, image type; For the document type, the corresponding text data is parsed and extracted based on the corresponding document parsing tool; and for the image type, the corresponding text data is parsed and extracted through optical character recognition; The text data is preprocessed and formatted.

[0008] In one example, the resume information is extracted and structured according to the instructions of the large model through the trained large model, and the corresponding structured information is output and stored, specifically including: Determine the constraint conditions in the large model prompt instruction through the trained large model; the constraint conditions include information item selection constraint conditions, position constraint conditions, and format constraint conditions; Selecting at least some of the basic information items from all the basic information items as designated information items according to the information item selection constraint condition; According to the format constraint, in the resume information, converting the specified information item into intermediate information of a corresponding format; According to the position constraint condition, the intermediate information is combined at a corresponding position to generate and output corresponding structured information; The structured information is stored in a local database.

[0009] In one example, based on a natural language processing algorithm, the query instruction is parsed to extract corresponding key information, specifically including: Perform word segmentation and part-of-speech tagging on the query instruction, and filter the words to obtain key information; The key information is subjected to entity recognition through named entity recognition to obtain corresponding entity information.

[0010] In one example, generating a corresponding structured query language according to the key information specifically includes: determining a database for storing the structured information; Mapping the identified entity information to the table structure of the database so as to map each entity information to a corresponding field; Based on a preset filling template or a pre-trained deep learning model, a corresponding structured query language is generated according to the fields and the entity information.

[0011] In one example, querying a database storing the structured information using the structured query language and returning corresponding query results specifically includes: Performing a query in a database storing the structured information through fields included in the structured query language; Generate corresponding query results; If the query result is empty, synonym mapping and / or quantified fuzzy expression are performed on the entity information according to a pre-established domain dictionary to generate corresponding replacement information; The corresponding query result is regenerated according to the replacement information and returned.

[0012] In one example, synonym mapping and / or quantitative fuzzy expression are performed on entity information according to a pre-established domain dictionary to generate corresponding replacement information, specifically including: Query each field contained in the structured query language separately to obtain the corresponding sub-query results; If there is no field whose subquery result is empty, the query result is returned as empty; If there is a first specified field whose subquery result is empty, then the entity information corresponding to the field is obtained, and synonym mapping and / or quantitative fuzzy expression are performed on the entity information according to a pre-established domain dictionary to generate corresponding replacement information; According to the second designated field corresponding to the replacement information, query through each replacement information to obtain the corresponding designated sub-query results respectively; If all the specified sub-query results are empty, the query result is returned as empty; If there is a specified subquery result that is not empty, the entity information corresponding to the first specified field is replaced with the replacement information of the second specified field.

[0013] In one example, the corresponding query results are returned, including: Obtaining access rights of the user; Determining, according to a third designated field in the query result, designated resume information in the query result for which the user does not have access rights; In the fields mapped to the entity information, determine the corresponding privacy field and the remaining fields except the privacy field; In the specified resume information, hiding the fields except the remaining fields and the sequence field; Returns the resume information that the user has access rights to, as well as the specified resume information after hiding.

[0014] On the other hand, the present application also proposes a resume screening device based on a large model prompt instruction, comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the resume screening method based on the large model prompt instructions as described in any of the above examples.

[0015] On the other hand, the present application also proposes a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured as: a resume screening method based on large model prompt instructions as described in any of the above examples.

[0016] The resume screening method based on large model prompt instructions proposed in this application can bring the following beneficial effects: 1. Through the large model prompt instructions, resume information of different formats and sources is uniformly structured to avoid the parsing failure caused by format differences in traditional methods, improve the efficiency of processing massive multi-source heterogeneous data, and improve the efficiency of resume screening and the accuracy of resume data.

[0017] 2. Based on the ability to understand context, the big model can extract implicit information such as soft skills and career development trajectories that are difficult to cover with traditional rules, deeply capture subtle information, and improve the comprehensiveness of information capture.

[0018] 3. The entire process from information extraction to query generation is automated, which reduces the cost of manual labeling and rule configuration, reduces manual intervention, and can adapt to high-concurrency scenarios during the peak recruitment season.

[0019] 4. Generate SQL based on natural language query instructions, flexibly support complex combination condition retrieval, realize dynamic adaptive screening, and avoid the static rule limitations of traditional scoring models. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flowchart of a resume screening method based on a large model prompt instruction in an embodiment of the present application; Figure 2 This is a flow chart of a resume screening method based on a large model prompt instruction in one scenario in an embodiment of the present application; Figure 3 This is a schematic diagram of a resume screening device based on a large model prompt instruction in an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0022] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0023] like Figure 1 and Figure 2 As shown, the embodiment of the present application provides a resume screening method based on a large model prompt instruction, including: S101: Based on multiple sources, obtain resume information of corresponding structure, and pre-process the resume information.

[0024] Here, the platform for executing the resume screening method based on the large model prompt instruction can be called the resume intelligent screening platform (hereinafter referred to as the platform).

[0025] The platform determines multiple pre-set sources, including local databases and third-party systems. For the local database, the corresponding resume information (also called resume data) is obtained through import; and for the third-party system, the resume information uploaded in the third-party system is obtained through the set system service interface.

[0026] The resume information in the local database can be imported by scanning paper resumes or by local import. The management node can select single or multiple files for upload through the graphical user interface (GUI), or use drag and drop operations to add files to the platform. The third-party system can be the corresponding recruitment system, campus system, etc. When uploading through a third-party system, the system service interface (Restful API interface) can be used to connect the third-party system to the platform. The platform provides a unified Java SDK and JS SDK, which provide encapsulated functions and methods for Java developers and JavaScript developers respectively, to facilitate data transmission through the Java SDK or JS SDK.

[0027] Determine the type of resume information; the types include: document type and image type. The type of resume information can be determined by its corresponding file suffix. For example, jpeg and PNG suffixes are images, and txt and Word suffixes are documents. For pdf suffixes, you can first treat it as a document. For the document type, parse and extract the corresponding text data based on the corresponding document parsing tool. Document parsing tools refer to existing parsing tools that can extract text content. For example, for pdf documents, parsing tools include PyPDF2 and PyPDF4; for Word documents, parsing tools include python-docx and Apache POI; for Excel documents, parsing tools include pandas, openpyxl, etc. And for the image type, parse and extract the corresponding text data through optical character recognition (OCR).

[0028] For document types (such as text documents, PDF documents, etc.), the corresponding parsing tools can be used to parse and extract text data. For scanned image types and some PDF documents that cannot be parsed and recognized, OCR can be used to recognize and obtain text data. The text data obtained here is the multiple dimensions in the basic information items below, which will be explained below.

[0029] Perform text preprocessing on text data and perform format standardization. Text preprocessing can include document segmentation, data cleaning, stop word replacement, etc., while formatting standardization is to encode text data in a standard format.

[0030] S102: Define basic information items corresponding to the resume, and generate large model prompt instructions based on the basic information items.

[0031] Basic information items may include multiple dimensions, and each dimension may further include multiple sub-dimensions. For example, dimensions may include basic information, contact information, education experience, work experience, skill list, etc. Sub-dimensions in basic information may include name, gender, age, etc., and sub-dimensions in work experience may include start time, company, company size, position, etc.

[0032] For dimensions (sub-dimensions) that can be directly quantified in the basic information items, they are directly quantified. If the dimensions are difficult to quantify, they are classified to obtain all categories corresponding to the dimension (sub-dimension). Data quantization refers to the process of converting data into discrete, measurable numerical forms. Common quantification types include: continuous data binning, categorical data encoding, and feature engineering quantification. Among them, continuous data binning corresponds to dimensions that can be directly quantified. For example, for the age dimension, it is segmented, 20 to 30 years old is young, 31 to 50 years old is middle-aged, and 51 years old and above is old. And categorical data encoding corresponds to dimensions that are difficult to quantify. For example, for the gender dimension, males are classified and coded as 1, and females are classified and coded as 0. For example, in the educational experience dimension, junior college or below is coded as 0, junior college is coded as 1, undergraduate is coded as 2, and graduate students and above are coded as 3, etc. Feature engineering quantification converts abstract features into computable indicators. For example, for the position dimension, a position conversion table is established, including interns corresponding to 1, assistants corresponding to 2, specialists corresponding to 3, supervisors corresponding to 4, and so on. The higher the number in the table, the higher the level, and the corresponding quantification is performed. For the company dimension, similar quantification can be used, or no quantification is required, so as to facilitate accurate query.

[0033] Among them, the algorithm for continuous data binning can adopt equal width binning (Equal Width Binning), equal frequency binning (Equal Frequency Binning), clustering-based binning (Clustering-Based Binning), etc., and the algorithm for categorical data encoding can adopt label encoding (Label Encoding), one-hot encoding (One-Hot Encoding) and other algorithms.

[0034] The following explains how to generate large model prompt instructions based on basic information items. Its essence is to fill in the required basic information items into the corresponding positions in the pre-set text format (examples are given below) to generate large model prompt instructions.

[0035] Specifically, the text format corresponding to the large model prompt instruction can include indication information, relevant context, examples helpful for the large model to understand, clear input, description of the expected output format, and set constraints. A complete prompt instruction can include the following parts: Instructions are used to clearly describe the task, such as "Please extract personal information, educational background, and work experience from the given resume text"; Context is used to provide background information related to the task, especially when dealing with multi-round interactions; Examples are used to give specific examples to help the model understand the task requirements and output format; Input is used for specific resume information; Output describes the expected output format, such as "Please output the extracted information in JSON format"; Constraints are used for various constraints, which can be listed separately or combined with other parts.

[0036] For example, the following is a text format with JSON format as an example, through which the large model prompt instruction can be generated: { "prompt": "Please extract the name, contact information, educational background, and work experience from the following resume text and output it in JSON format.\nInput text:\n[Insert resume text here]\nOutput format example:\n{\"name\": \"Zhang San\", \"contact\": {\"phone\": \"123456789\"}, \"education\": [{\"degree\":\"Master\", \"institution\": \"University A\"}], \"experience\": [{\"company\": \"Company B\", \"position\": \"Position C\"}]}", "model": "haiyue-2.0" } S103: Use the resume information and the large model prompt instruction as input, and through the trained large model, extract and structure the information from the resume information according to the large model prompt instruction, and output and store the corresponding structured information.

[0037] The large model is a Large Language Model (LLM). The resume information is the information to be processed as input, and the large model prompt instruction is used as an instruction to process the resume information, so as to extract the corresponding structured information and store it in the local database.

[0038] Specifically, through the trained large model, determine the constraint conditions in the large model's prompt instructions; the constraint conditions include information item selection constraint conditions, position constraint conditions, and format constraint conditions. The information item selection constraint conditions are used to only select some of the basic information items in the resume for outputting structured information. The format constraint conditions are used to constrain the output format of the structured information. The position constraint conditions are used to locate the positions of the basic information items in the output structured information. Among them, each constraint condition can be added based on requirements.

[0039] Based on this, through the information item selection constraint conditions, select at least some of the basic information items from all the basic information items as the specified information items. According to the format constraint conditions, convert the specified information items into intermediate information in the corresponding format (such as JSON format) in the resume information. According to the position constraint conditions, combine the intermediate information at the corresponding positions to generate and output the corresponding structured information.

[0040] Store the structured information in the local database to form the storage of the structured information.

[0041] S104: Obtain the query instruction input by the user, and based on the natural language processing algorithm, parse the query instruction to extract the corresponding key information.

[0042] The query instruction refers to the instruction entered by the user on the platform with the goal of querying for corresponding conditions in the structured information. It is usually the natural language input by the user and is difficult to be directly used for querying. Therefore, it is necessary to parse the query instruction based on the natural language processing algorithm to extract the corresponding key information.

[0043] Among them, the specific algorithms of the natural language processing algorithm include word segmentation processing, part-of-speech tagging, word filtering, and named entity recognition, which will be explained in detail below.

[0044] Use the corresponding tools to perform word segmentation processing and part-of-speech tagging (including nouns, verbs, numerical values, etc.) on the query instruction. For example, use Jieba for word segmentation processing and support custom dictionaries and part-of-speech tagging. You can also use the LTP tool, which supports word segmentation, part-of-speech tagging, and dependency parsing.

[0045] Perform word filtering, filter out meaningless words (such as "de", "all"), etc., and remove stop words, and use the remaining words as key information. For example, by establishing a blacklist in advance and then filtering through the blacklist.

[0046] Perform entity recognition on the key information through named entity recognition (NER) to obtain the corresponding entity information.

[0047] The process of named entity recognition algorithm mainly includes: data preprocessing, including text cleaning: removing noise (including special symbols, HTML tags, etc.), unifying the encoding format. Word segmentation: word segmentation for non-space-delimited languages ​​such as Chinese (for example, Jieba word segmentation). Annotation system, for example, using BIO (Begin, Inside, Outside) or BIOES (add Single, End) format to annotate entities.

[0048] Feature extraction is performed, where word-level features include: word vectors (e.g., Word2Vec, GloVe, BERT embedding), part-of-speech tags (e.g., nouns, verbs), word form features (e.g., starting with capital letters, containing numbers, etc.); character-level features include character combination patterns; context features include window words (that is, feature combinations of the previous and next n words).

[0049] Build a model, for example, using conditional random fields (CRF) to build a model. Alternatively, use deep learning models such as the BiLSTM-CRF architecture and the BERT architecture to build a model.

[0050] Perform model training, use the extracted features as training samples, and use cross entropy loss or negative log-likelihood of CRF as the loss function to train the model.

[0051] The trained model uses the Viterbi algorithm to achieve the optimal label sequence for CRF layer decoding, and uses beam search to retain the Top-K candidate paths to balance efficiency and accuracy, thereby achieving decoding prediction.

[0052] Merge consecutive B / I tags into complete entities to achieve entity merging, complete the date format through regular expressions (for example, complete "2023 August" to "2023-08") to achieve rule correction, and use professional terminology libraries to correct model errors (for example, special job titles) to achieve post-processing of output results.

[0053] For example, if the query command input is "find the resumes of all software development engineers with more than 5 years of work experience", after word segmentation and word filtering, the key information obtained includes ["5 years", "work experience", "software development engineer", "resume"]. After NER entity recognition, the entity information obtained is: Position: Software Development Engineer, Years: More than 5 years, Target object: Resume.

[0054] S105: Generate a corresponding structured query language according to the key information, and query the database storing the structured information using the structured query language to return a corresponding query result.

[0055] Structured Query Language (SQL query statement) is a standard language specifically used to manage relational databases. The generated SQL query statement will be sent to the database for execution to retrieve data that meets the user's conditions and return the data set that meets the conditions.

[0056] Specifically, when generating the structured query language, determine the database that stores the structured information. Map the identified entity information to the table structure of the database, so that each entity information is mapped to the corresponding field. For example, the fields include: position, experience_years, skills, and the mapping field can be software development engineer mapped to position = 'Software Development Engineer'.

[0057] Based on the pre-set filling template or the pre-trained deep learning model, the corresponding structured query language is generated according to the field and entity information.

[0058] For simple structured query language, it can be achieved by filling in the template. For example, the template is: SELECT * FROM resumes WHERE [condition], and after filling in: SELECT * FROM resumes WHERE position = 'Software Development Engineer' AND experience_years>= 5.

[0059] Complex structured query languages ​​can be generated using pre-trained deep learning models (e.g., trained large language models).

[0060] Based on this, when querying through structured query language, the fields contained in the structured query language are used to query in the database storing structured information (wherein corresponding indexes can be established for high-frequency fields to speed up the query), and the corresponding query results are generated, which can then be converted into the corresponding format for output.

[0061] After obtaining the final query results, they can be returned and displayed to the user. Based on user needs, they can choose from a variety of forms such as lists, tables, and classified displays. They can also add a corresponding resume summary.

[0062] In addition, if Figure 2 As shown, the user can also further input query instructions based on the query results based on the needs to continue the screening query.

[0063] 1. Through the large model prompt instructions, resume information of different formats and sources is uniformly structured to avoid the parsing failure caused by format differences in traditional methods, improve the efficiency of processing massive multi-source heterogeneous data, and improve the efficiency of resume screening and the accuracy of resume data.

[0064] 2. Based on the ability to understand context, the big model can extract implicit information such as soft skills and career development trajectories that are difficult to cover with traditional rules, deeply capture subtle information, and improve the comprehensiveness of information capture.

[0065] 3. The entire process from information extraction to query generation is automated, which reduces the cost of manual labeling and rule configuration, reduces manual intervention, and can adapt to high-concurrency scenarios during the peak recruitment season.

[0066] 4. Generate SQL based on natural language query instructions, flexibly support complex combination condition retrieval, realize dynamic adaptive screening, and avoid the static rule limitations of traditional scoring models.

[0067] In one embodiment, the query result may be empty. In this case, there are two possibilities. One is that there is no information that meets the requirements in the structured information, and the other is that the description of some keywords in the user's query instruction is inaccurate.

[0068] Based on this, synonym mapping and / or quantitative fuzzy expression are performed on the entity information according to the pre-established domain dictionary to generate corresponding replacement information. This process can be directly generated and replaced during the query, so as to perform retrieval based on the replacement information, or it can be directly retrieved based on the user's input, and then replaced when the corresponding resume cannot be retrieved.

[0069] At this time, the original field information is replaced according to the replacement information, so as to regenerate the corresponding query result and return it.

[0070] Furthermore, when generating replacement information, if the user's query instruction contains many conditions, one condition may have multiple replaceable replacement information. If all fields are replaced and all possibilities are queried through permutations and combinations, when the amount of resume information in the structured information is large, a large amount of computing resources will be consumed during the query process, and the process will be slow.

[0071] Based on this, queries are performed on each field included in the structured query language to obtain corresponding sub-query results, that is, structured information is queried for each resource separately.

[0072] If there is no field with an empty subquery result, it is considered that each query condition in the user's query instruction can be matched to the corresponding resume, but after being combined together, no resume corresponding to the condition can be found. At this time, the returned query result is empty.

[0073] If there is a first specified field whose subquery result is empty, it is generally believed that the probability of not retrieving information under certain conditions through only a single field in a large amount of structured information is extremely low. Therefore, it is believed that this may be caused by the inconsistency between the words used in the user's query instruction and the words used in the structured information.

[0074] At this time, the entity information corresponding to the field is obtained, and the entity information is synonymously mapped and / or quantified fuzzily expressed according to the pre-established domain dictionary to generate the corresponding replacement information. For a single entity information, there may be multiple replacement information. For example, if the user enters "senior engineer", the fuzzy description of "senior" can be quantified into multiple replacement information such as "having certificate A" or "having title B" or "having more than 5 years of work experience" according to the domain dictionary. For the "resume" entered by the user, its synonyms can be mapped to "CV" and "resume".

[0075] At this time, since the replacement information may belong to a different field from the original entity information, for example, the field corresponding to "senior engineer" is the position, the field corresponding to "having certificate A" or "having professional title B" is the certificate or professional title, and the field corresponding to "having more than 5 years of work experience" is work experience.

[0076] Therefore, for each replacement information, the corresponding second designated field is determined, and according to the second designated field corresponding to the replacement information, only the second designated field is searched, and each replacement information is searched separately to obtain the corresponding designated sub-query results.

[0077] If all the specified subquery results of all the replacement information corresponding to a single entity information are empty, it is considered that the query condition corresponding to the single entity information queried by the user is not because of vague description, but because the corresponding resume is indeed not included in the structured information, so the query result returned is empty.

[0078] If there is a specified subquery result that is not empty, it is considered that it is caused by the user's ambiguous description. At this time, the replacement information of the second specified field replaces the entity information corresponding to the first specified field. When replacing, if there are multiple specified subquery results, the "or" form can be used in the replacement information corresponding to these multiple specified subquery results to replace the original entity information.

[0079] For example, the original entity information is "Senior Engineer", and the corresponding first designated field "Position" is empty. Multiple replacement information "Has Certificate A", "Has Professional Title B", and "Has more than 5 years of work experience" are generated, and the corresponding second designated fields are "Certificate", "Professional Title", and "Years of Work" respectively.

[0080] At this time, a separate query is performed in each second specified field to determine that the sub-query results corresponding to "having certificate A" and "having more than 5 years of work experience" are not empty, then "having certificate A or having more than 5 years of work experience" is used to replace the "senior" in "senior engineer".

[0081] Of course, when the search results are not empty but contain only a very small amount of information, a similar solution can be used to perform a corresponding re-search.

[0082] In one embodiment, after the final query result is determined, the user's access rights may be obtained. Different users may have different rights for the structured information in the database.

[0083] According to the third designated field in the query result, the designated resume information in the query result that the user does not have access rights is determined. The third designated field refers to a field used to set access rights, for example, it can be a field such as department or level, and different users have corresponding access rights for resumes of different levels and departments.

[0084] In the fields mapped to entity information, determine the corresponding privacy fields and the remaining fields except the privacy fields. Privacy fields can be set based on actual needs, for example, including name, contact information, education background, etc.

[0085] At this time, in the specified resume information, all fields except the remaining fields and the sequence field are hidden, and the resume information to which the user has access rights and the specified resume information after hiding are returned. For the resume information to which the user does not have access rights, the set privacy fields are hidden and returned to the user, so that the user can know the approximate number corresponding to his query instruction while protecting the privacy of the job seeker and preventing the user from seeing the content that he does not have access rights to.

[0086] like Figure 3 As shown, the embodiment of the present application also proposes a resume screening device based on a large model prompt instruction, including: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the resume screening method based on the large model prompt instructions as described in any of the above embodiments.

[0087] The embodiment of the present application also proposes a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured as: a resume screening method based on large model prompt instructions as described in any of the above embodiments.

[0088] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0089] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects as the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0090] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A resume screening method based on large model prompt instructions, characterized in that: include: Based on multiple sources, obtaining resume information of corresponding structure, and preprocessing the resume information; Defining basic information items corresponding to the resume, and generating large model prompt instructions based on the basic information items; Taking the resume information and the large model prompt instruction as input, extracting and structuring the resume information according to the large model prompt instruction through the trained large model, and outputting and storing the corresponding structured information; Obtaining the query instruction input by the user, and parsing the query instruction based on the natural language processing algorithm to extract the corresponding key information; A corresponding structured query language is generated according to the key information, and a query is performed in a database storing the structured information using the structured query language to return a corresponding query result.

2. The resume screening method based on large model prompt instructions according to claim 1 is characterized in that: Based on multiple sources, resume information of corresponding structure is obtained, and the resume information is preprocessed, specifically including: Determine a plurality of preset sources; the sources include: a local database and a third-party system; For the local database, corresponding resume information is obtained by importing; and for the third-party system, the resume information uploaded in the third-party system is obtained through the set system service interface; Determine the type of the resume information; the type includes: document type, image type; For the document type, the corresponding text data is parsed and extracted based on the corresponding document parsing tool; and for the image type, the corresponding text data is parsed and extracted through optical character recognition; The text data is preprocessed and formatted.

3. The resume screening method based on large model prompt instructions according to claim 1 is characterized in that: Through the trained big model, the resume information is extracted and structured according to the big model prompt instructions, and the corresponding structured information is output and stored, specifically including: Determine the constraint conditions in the large model prompt instruction through the trained large model; the constraint conditions include information item selection constraint conditions, position constraint conditions, and format constraint conditions; Selecting at least some of the basic information items from all the basic information items as designated information items according to the information item selection constraint condition; According to the format constraint, in the resume information, converting the specified information item into intermediate information of a corresponding format; According to the position constraint condition, the intermediate information is combined at a corresponding position to generate and output corresponding structured information; The structured information is stored in a local database.

4. The resume screening method based on large model prompt instructions according to claim 1 is characterized in that: Based on the natural language processing algorithm, the query instruction is parsed to extract the corresponding key information, including: Perform word segmentation and part-of-speech tagging on the query instruction, and filter the words to obtain key information; The key information is subjected to entity recognition through named entity recognition to obtain corresponding entity information.

5. The resume screening method based on large model prompt instructions according to claim 4 is characterized in that: Generate a corresponding structured query language according to the key information, specifically including: determining a database for storing the structured information; Mapping the identified entity information to the table structure of the database so as to map each entity information to a corresponding field; Based on a preset filling template or a pre-trained deep learning model, a corresponding structured query language is generated according to the fields and the entity information.

6. The resume screening method based on large model prompt instructions according to claim 4 is characterized in that: Performing a query in the database storing the structured information by using the structured query language and returning a corresponding query result specifically includes: Performing a query in a database storing the structured information through fields included in the structured query language; Generate corresponding query results; If the query result is empty, synonym mapping and / or quantified fuzzy expression are performed on the entity information according to a pre-established domain dictionary to generate corresponding replacement information; The corresponding query result is regenerated according to the replacement information and returned.

7. The resume screening method based on large model prompt instructions according to claim 6 is characterized in that: According to the pre-established domain dictionary, synonym mapping and / or quantitative fuzzy expression are performed on the entity information to generate corresponding replacement information, including: Query each field contained in the structured query language separately to obtain the corresponding sub-query results; If there is no field whose subquery result is empty, the query result is returned as empty; If there is a first specified field whose subquery result is empty, then the entity information corresponding to the field is obtained, and synonym mapping and / or quantitative fuzzy expression are performed on the entity information according to a pre-established domain dictionary to generate corresponding replacement information; According to the second designated field corresponding to the replacement information, query through each replacement information to obtain the corresponding designated sub-query results respectively; If all the specified sub-query results are empty, the query result is returned as empty; If there is a specified subquery result that is not empty, the entity information corresponding to the first specified field is replaced with the replacement information of the second specified field.

8. The resume screening method based on large model prompt instructions according to claim 5 is characterized in that: Returns the corresponding query results, including: Obtaining access rights of the user; Determining, according to a third designated field in the query result, designated resume information in the query result for which the user does not have access rights; In the fields mapped to the entity information, determine the corresponding privacy field and the remaining fields except the privacy field; In the specified resume information, hiding the fields except the remaining fields and the sequence field; Returns the resume information that the user has access rights to, as well as the specified resume information after hiding.

9. A resume screening device based on large model prompt instructions, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the resume screening method based on large model prompt instructions as described in any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured as: a resume screening method based on large model prompt instructions as described in any one of claims 1 to 8.

Images