AI-based large-scale model-based intelligent generation system for resume documents and talent profile reports

Abstract

The application relates to the technical field of named entity recognition, in particular to a resume file talent portrait report intelligent generation system based on an AI large model, which comprises the following steps: a resume file analysis module is used for converting a Chinese resume file format; a resume information analysis module is used for performing public word replacement on each Chinese resume text, analyzing word group nesting probabilities of each noun corresponding word group in the public words, obtaining a supplementary resume set of each Chinese resume, determining entity nesting confidence based on the word group nesting probabilities of each noun corresponding word group in the supplementary resume set, determining a sparse resume sample, counting word group nesting probabilities of each noun corresponding word group in a webpage text, determining a search auxiliary discrimination value, determining a nested entity screening value, and obtaining a nested long entity; a resume information extraction module is used for obtaining a Chinese resume named entity recognition result; and a talent portrait report generation module is used for generating a talent portrait report. The application aims to improve the resume entity recognition accuracy.

Description

Technical Field

[0001] This application relates to the field of named entity recognition technology, specifically to an intelligent generation system for talent profile reports based on AI large-scale models. Background Technology

[0002] Resumes, being textual information, not only contain key information such as the job seeker's basic information, educational background, work experience, and project experience, but also reflect the job seeker's personalized characteristics, such as self-evaluation and career goals. They serve as an important bridge between individual job seekers and corporate recruiters, while companies face the challenge of sifting through massive amounts of resumes. Resumes are characterized by semi-structured format, large quantity, personalization, rich entity types, wide domain span, and high information density, often requiring Named Entity Recognition (NAME) for information extraction to better perform feature extraction and analysis.

[0003] Deep learning-based entity recognition can autonomously learn the characteristics and contextual relationships of entities, reducing reliance on manually constructed features. It is highly adaptable to personalized electronic resumes with diverse entity types and broad domain scope. The invention published under "CN115169349A" proposes an "ALBERT-based named entity recognition method for Chinese electronic resumes." It utilizes a trained ALBERT model for character embedding to obtain dynamic word vectors, addressing the problem of polysemy. However, the entity lengths of job titles and project names in resumes are often too long, with unclear grammatical structures, and even entity nesting and blurred entity boundaries, affecting the accuracy of resume entity recognition and reducing the effectiveness of talent profile report generation. Summary of the Invention

[0004] In light of the above, it is necessary to provide an intelligent talent profile report generation system for resume documents based on AI large-scale models to solve the above problems.

[0005] One embodiment of this application provides an intelligent generation system for talent profile reports from resume files based on AI large-scale models, the system comprising:

[0006] Resume Information Analysis Module: Based on the frequency of occurrence of each word segment and its synonyms in all Chinese resumes, each common word sequence is obtained; based on the frequency of occurrence of nouns in each common word sequence and word groups composed of adjacent common words in each Chinese resume, the probability of word group nesting is determined;

[0007] Based on the similarity between each Chinese resume and other Chinese resumes, a supplementary resume set for each Chinese resume is obtained; the probability of word phrases corresponding to each noun in each Chinese resume being nested in each Chinese resume in its supplementary resume set is analyzed, and the entity nesting confidence is determined in combination with the similarity.

[0008] Based on the distribution characteristics of the probability of word nesting of each noun corresponding to each common word sequence in Chinese resumes in the web page text, a search auxiliary discrimination value is determined;

[0009] Based on the probability of phrase nesting, the confidence of entity nesting, and the search-aided discrimination value in Chinese resumes, the nested entity filtering value of each noun corresponding to the phrase is determined, and nested long entities are obtained.

[0010] Resume Information Extraction Module: Based on a neural network model, this module extracts Chinese resume named entity recognition results by analyzing the character sequences of nested long entities and common words.

[0011] The talent profile report generation module is used to match entities based on the named entity recognition results of Chinese resumes and generate talent profile reports.

[0012] Preferably, obtaining each public word sequence specifically involves:

[0013] For each segmented sequence obtained from the word segmentation process, the set of all synonyms of each segment in the Chinese general concept knowledge graph is used as the extended word set of each segment;

[0014] The frequency of each word in the extended word set of each segment is counted in all Chinese resumes. The word with the highest frequency is taken as the common word of the extended word set, and the corresponding segment is replaced to obtain the common word sequence.

[0015] Preferably, the specific process for determining the probability of phrase nesting is as follows:

[0016] The frequency of each noun in each public word sequence in all public word sequences of the corresponding Chinese resume is obtained and recorded as the first frequency; the frequency of each phrase of each noun in each public word sequence in all public word sequences of the corresponding Chinese resume is obtained and recorded as the second frequency; the ratio between the second frequency and the first frequency is used as the phrase nesting probability of each noun in each public word sequence.

[0017] Preferably, the supplementary resume set for each Chinese resume specifically includes:

[0018] We use regular expressions to extract the target job entities from each Chinese resume and compare them with the target job entities from other Chinese resumes to obtain the job semantic similarity between each Chinese resume and other Chinese resumes.

[0019] Threshold segmentation is performed on the semantic similarity of each Chinese resume to all other Chinese resumes for job postings;

[0020] The set of all Chinese resumes whose job title semantic similarity to each Chinese resume is greater than the segmentation threshold is used as a supplementary resume set for each Chinese resume.

[0021] Preferably, the semantic similarity of the job titles is obtained by the cosine similarity between each Chinese resume and the intended job title entity in another Chinese resume.

[0022] Preferably, the formula for determining the confidence level of entity nesting is: In the formula, Phrases representing the Chinese resume (k) , Nested formulas for entities; This refers to the i-th noun in the common word sequence S of Chinese resumes; This represents the next common word in the sequence S of common words in Chinese resumes, following the i-th noun. This is the Chinese resume k and its supplementary resume collection. semantic similarity of job titles among Chinese resumes It is a supplementary resume collection for Chinese resume k, phrases ( , ) in the The probability of phrase nesting in a Chinese resume. It represents the total number of all Chinese resumes in the supplementary resume set of Chinese resume k.

[0023] Preferably, determining the search-aid discriminant value specifically involves:

[0024] Based on the number of elements in the supplementary resume set for each resume, a sparse resume sample is determined.

[0025] The average probability of word nesting for each noun phrase in the sparse resume sample across all web page texts is used as the search auxiliary discriminant value for each noun phrase.

[0026] Preferably, the sparse resume sample specifically refers to Chinese resumes in the supplementary resume set whose number of Chinese resumes is less than a preset value.

[0027] Preferably, the step of determining the nested entity filtering value of each noun corresponding to each word group in each word segmentation sequence specifically involves:

[0028] If the Chinese resumes are sparse resume samples, the average of the probability of phrase nesting, the confidence of entity nesting, and the search auxiliary discrimination value will be used as the nested entity filtering value of each noun in the corresponding Chinese resume.

[0029] Otherwise, the mean of the probability of phrase nesting and the confidence level of entity nesting will be used as the nested entity screening value for each noun in the corresponding Chinese resume.

[0030] Preferably, obtaining nested long entities specifically involves merging phrases whose nested long entity filtering value is greater than a preset nesting filtering threshold to form nested long entities.

[0031] This application has at least the following beneficial effects:

[0032] This application obtains synonyms from word segmentation through knowledge graphs and replaces them with common words, which solves the problem of cross-document semantic drift caused by the mixed use of abbreviations and synonyms for the same entity in resumes, reduces the false judgment rate of entity recognition, and improves the consistency of named entity recognition.

[0033] Furthermore, to address the issues of entity nesting and ambiguous boundaries in resumes, this study calculates the probability of phrase nesting and the confidence level of entity nesting, and combines supplementary resume sets and search engine-assisted discrimination to effectively distinguish between nested entities that are common in job seekers' personal habits and industry fields. This improves the boundary recognition accuracy of long nested entities, making it particularly suitable for niche or emerging positions.

[0034] Finally, an ALBERT+BiLSTM+CRF architecture was adopted, in which the word embedding layer was concatenated with ALBERT dynamic vectors, word segmentation feature vectors and boundary feature vectors, which fully integrated character semantics, word segmentation information and word boundary features, improved the model's ability to recognize complex entities in resumes, and improved the overall accuracy of named entity recognition and the accuracy of talent profile reports. Attached Figure Description

[0035] Figure 1 A flowchart of the AI-based large-scale model-based intelligent generation system for resume document talent profile reports provided for this application;

[0036] Figure 2 A schematic diagram of the Chinese resume named entity recognition model provided in this application. Detailed Implementation

[0037] In the description of the embodiments in this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0039] It should also be noted that the terms "first" and "second" in this application and its accompanying drawings are used to distinguish similar objects, rather than to describe a specific order or sequence. The methods disclosed in the embodiments of this application or the methods shown in the flowcharts include one or more steps for implementing the method. Without departing from the scope of protection of this application, the execution order of multiple steps can be interchanged, and some steps can also be deleted.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0041] The following section, in conjunction with the accompanying drawings, details the specific solution of the AI-based large-scale model-based intelligent generation system for resume files and talent profile reports provided in this application.

[0042] Please see Figure 1 The diagram illustrates a block diagram of an AI-based large model-based intelligent generation system for resume files and talent profile reports, provided in one embodiment of this application. The system includes: a resume file parsing module, a resume information analysis module, a resume information extraction module, and a talent profile report generation module.

[0043] This application first proposes an intelligent generation system for talent profile reports based on AI large-scale models, applied in the field of named entity recognition technology. The system includes:

[0044] Resume file parsing module:

[0045] For image-format resume files, OCR technology is used to process scanned copies or photographs of the resume, converting them into PDF files. For Word format resume files, a Word-to-PDF converter can be used to convert them into PDF files. The resume parsing module takes the resume PDF file as input and calls the API interface of the large language model to divide the text content of the resume PDF file according to the resume module. In this embodiment, the large language model used is deepseek, which divides the resume structure into: basic information, education background, work experience, project experience, skills and abilities, and self-evaluation.

[0046] In another embodiment, a resume structure partitioning pattern can be independently constructed to realize the resume structure partitioning of PDF format electronic resumes, specifically:

[0047] Obtain Chinese resume text in PDF format, and use the pdfplumber tool to divide the electronic resume into several single-line texts. Then, import these texts into the Label Studio annotation platform to annotate each single-line text with a corresponding module label: basic information, educational background, work experience, project experience, skills and abilities, and self-evaluation. Use the annotated resume texts as the dataset, with a training, validation, and test set ratio of 8:1:1. Employ PyTorch as the deep learning platform, importing the torch, numpy, and transformer libraries. Use TextRCNN as the neural network training model, setting the batch size to 50, the maximum sentence length to 150, the dropout size to 0.5, and the learning rate to 0.001. Train the resume structure segmentation model. Convert the PDF electronic resume into single-line text data using pdfplumber and use this as input to the resume structure segmentation model to achieve resume structure segmentation.

[0048] This application inputs N Chinese resumes into the resume parsing module. Each Chinese resume can be divided into 6 resume structure texts. The larger the value of N, the higher the accuracy of resume named entity recognition. The value must be at least greater than 200. In this embodiment, it is 500. Implementers can adjust the value according to the actual situation.

[0049] Resume Information Analysis Module:

[0050] Current resume information extraction modules primarily focus on the highly structured parts of resumes, such as basic information, name, contact information, and educational background. However, less structured content, such as skills and abilities, project experience, and career goals, often receives insufficient attention. Yet, this less structured information is crucial for building talent profiles and helping employers assess candidates' overall capabilities. Therefore, the core task of a resume information extraction module is to extract valuable unstructured information from resumes using Named Entity Recognition (NER) technology, providing robust data support for subsequent resume analysis and decision-making.

[0051] For each Chinese resume, the LAC (Lexical Analysis of Chinese) tool was used to segment each resume structure text, resulting in several segmented word sequences for each resume structure text. Each segmented word sequence consists of several words, and each segmented word sequence corresponds to one sentence. Then, the Harbin Institute of Technology stop word list was used to process each segmented word sequence and delete stop words that have no impact on the semantics of the sentences or have no practical meaning.

[0052] Resume information is usually highly condensed, and there are various ways to express entities. Abbreviations and synonyms are often mixed in resumes, lacking consistency. For example, "risk control" is often abbreviated as "risk management", and "Risk Management" may also be represented by "RM". This diverse way of expression makes information extraction and understanding more complex. Named Entity Recognition (NER) models are usually trained based on the mapping relationship between surface words and entity labels. When the same entity appears in multiple forms, the samples of each variant in the training set are sparse, and it is difficult for the NER model to learn the semantic features shared by these variants. Some variants have too large literal differences, resulting in cross-document semantic drift of the same entity and increasing the misjudgment rate of entity recognition.

[0053] Therefore, based on the information extraction method of the Chinese general concept knowledge graph CN-DBpedia, this application obtains all synonyms of any segmented word, and jointly forms an extended word set with the segmented word. Among them, the synonyms are specifically the full name, abbreviation, Chinese name, and foreign name. The information extraction based on the knowledge graph is specifically: extracting entities in the CN-DBpedia knowledge graph whose relationship with the segmented word is the full name, abbreviation, Chinese name, and foreign name. Count the occurrence frequency of each word in the extended word set of each segmented word in all Chinese resumes, and take the word with the maximum occurrence frequency as the common word of the extended word set, and replace each segmented word. The segmented word sequence after replacement is recorded as the common word sequence.

[0054] Due to the complexity of named entity recognition in Chinese resumes, Chinese sentences are usually written continuously, and there is no clear separator between words, which makes it difficult to extract word boundary information. Since most named entity terms are nouns, taking the common word sequence of each sentence in each Chinese resume as an example, this application uses the Baidu Chinese lexical analysis tool LAC to obtain the词性 of any common word in the common word sequence S. The i-th noun in the common word sequence S is denoted as , and the noun The next common word after is denoted as , and and form the phrase of the i-th noun ( , ). Among them, nouns are specifically common nouns, locative nouns, place nouns, work names, other proper names, nominal verbs, nominal adjectives, personal names, place names, organization names, and time in the LAC词性 annotation.

[0055] The frequency of each noun in each public word sequence in all public word sequences of the corresponding Chinese resume is obtained and recorded as the first frequency; the frequency of each phrase of each noun in each public word sequence in all public word sequences of the corresponding Chinese resume is obtained and recorded as the second frequency; the ratio between the second frequency and the first frequency is used as the phrase nesting probability of each noun in each public word sequence.

[0056] Specifically, in this embodiment, the phrase is obtained using the following formula ( , Probability of phrase nesting in Chinese resume k Its formula is: In the formula, It is a public word sequence The frequency of the i-th noun in the corresponding Chinese resume among all common word sequences. It is a public word sequence The i-th noun in the corresponding Chinese resume is in all common word sequences (word groups) The frequency of occurrence of ).

[0057] An entity may completely contain one or more other entities, exhibiting nested behavior. For example, in a Chinese resume, the entity "Senior Software Development Engineer" contains nested entities for "software development" and "engineer," making entity recognition relying on Chinese word segmentation technology prone to errors. A higher proportion of phrase co-occurrence increases the probability of phrases acting as nested entities, thus increasing the probability of phrase nesting. The larger.

[0058] Nested long entities consist of multiple short entities. Resumes, as a crucial bridge between job seekers and recruiters, are characterized by concise text, with the same entity appearing infrequently within a single resume. Due to the diverse formats and strong individuality of Chinese resumes, the use of consecutive or nested phrases may stem from personal habits and not necessarily constitute nested long entities. Since the intended position is structured and thematically focused in the resume (e.g., "Intended Position: Backend Development Engineer"), subsequent work experience and project experience are highly likely to be relevant to this intended position.

[0059] This application obtains basic information from Chinese resumes and uses a regular expression-based information extraction method to extract the intended job entity from the resumes. Specifically, it uses regular expressions to match common word combinations after "intended job" as the intended job entity in the Chinese resume. The intended job entity is then used as input to a Word2Vec model to obtain the corresponding word vectors, where the word vector output by the Word2Vec model has a length of 100. The cosine similarity between the intended job entity word vectors of each Chinese resume and those of another Chinese resume is used as the job semantic similarity between the two resumes.

[0060] The meanings of job titles in Chinese resumes vary greatly across different industries; the smaller the semantic differences in the intended job, the better. The larger the similarity, the more similar the job responsibilities and skill requirements, and the more likely Chinese resumes with similar job aspirations belong to the same industry and have the same named entities. Using the job semantic similarity between each Chinese resume and all other Chinese resumes as input, the Otsu's method is used to obtain the segmentation threshold. Other Chinese resumes with job semantic similarity greater than the segmentation threshold and a value greater than 0 are then used to form a supplementary resume set for each Chinese resume.

[0061] This application uses the following formula to obtain the word group of the i-th noun in each common word sequence within each Chinese resume ( , Entity nesting confidence Its formula is: In the formula, This is the Chinese resume k and its supplementary resume collection. semantic similarity of job titles among Chinese resumes It is a supplementary resume collection for Chinese resume k, phrases ( , ) in the The probability of phrase nesting in a Chinese resume. It represents the total number of all Chinese resumes in the supplementary resume set of Chinese resume k.

[0062] It should be understood that, The larger the number of resumes, the more similar they are to the intended job positions, and the more similar the job application information they may have. Therefore, supplementing the resume collection with Chinese resumes is more valuable for reference. The larger the value, the more frequent the nested phrases appear, indicating that the nested phrases are less likely to be a personal habit and more likely to be nested entities within an industry domain. (Entity nesting confidence level) The larger.

[0063] The industry is constantly evolving, and new technical terms and job titles are appearing in Chinese resumes. Furthermore, for less popular positions and majors, the number of relevant Chinese resumes is scarce, and the screening value for nested long entities is generally small, increasing the difficulty of nested entity recognition. If the number of Chinese resumes in the supplementary resume set is less than a preset value, then that Chinese resume is recorded as a sparse resume sample; in this embodiment, the preset value is 10.

[0064] For sparse samples, this application utilizes a search engine to identify whether phrases are nested entities based on the knowledge of the retrieved web pages. Specifically, the intended job entity for each Chinese resume is input into the search engine, and the first M web pages are retrieved based on the search results. Within each web page, phrases are statistically analyzed (…). , ), words The frequency of occurrence, M, is set to 10 in this embodiment. Treating the webpage as a resume sample, the probability of nesting with the phrase is calculated using the same formula. Similarly, the phrase ( , The probability of phrase nesting on each webpage is calculated. The search engine can be Google or Baidu; this application uses Baidu, but the implementer can choose a suitable search engine. (The calculation of phrases...) , The mean probability of phrase nesting across all web pages is used as the sparse resume sample k with respect to phrases ( , Search auxiliary discriminant value .

[0065] Based on the above steps, we can obtain the phrase ( , Probability of phrase nesting in Chinese resume k Entity nesting confidence and search auxiliary discriminant value This application obtains the phrases within k of a Chinese resume using the following formula ( , Nested long entity filter values Its formula is:

[0066] ; where Z represents sparse resume samples.

[0067] The probability of phrase nesting and the confidence of entity nesting are used to eliminate the individual habits of job seekers in Chinese resumes. Synonymous common words are used to replace the original words, and cross-document resume association analysis is performed based on the intentional job entities with significant structural and thematic features. This solves the problems of difficulty in extracting effective summaries from Chinese resume text and difficulty in cross-document resume association analysis. It also eliminates differences in resume description habits, improves the recognition accuracy of nested long entity boundaries in resumes, and obtains external knowledge for sparse samples in Chinese resumes, improving the recognition accuracy of nested long entity boundaries for niche or emerging jobs.

[0068] It is a common word for nouns. In the public word sequence The last common word, It is a common word for nouns. In the public word sequence The first common word in the middle, therefore, similarly, we can obtain the word groups within k in the Chinese resume ( The nested long entity filtering value is used to merge phrases with nested long entity filtering values ​​greater than a preset nested filtering threshold to form new words, which are then denoted as nested long entities. In this embodiment, the nested filtering threshold is set to 0.7. For example, if the original common word sequence is: ("person", "at", "former company", "serving as", "software", "development", "engineer"), the nested long entity filtering value of the phrase ("software", "development") is 0.84, and the nested long entity filtering value of the phrase ("development", "engineer") is 0.76, the merged nested long entity is "software development engineer".

[0069] Resume information extraction module:

[0070] This application adopts a Chinese resume named entity recognition model based on the ALBERT+BiLSTM+CRF architecture. The Chinese resume named entity recognition model has a three-layer structure: a character embedding layer, an encoding layer, and a decoding layer. The dynamic vector output by ALBERT is used as the character embedding representation, and word segmentation and boundary feature vectors are concatenated on top of this, resulting in the concatenated character vectors. As input to the encoding layer, BiLSTM is used for encoding, and CRF is used for decoding. A schematic diagram of the Chinese resume named entity recognition model is shown below. Figure 2 As shown.

[0071] (1) ALBERT character vector: is the character vector output by the ALBERT model.

[0072] The ALBERT model extracts rich semantic features from Chinese resume text. This application splits each nested long entity and common word into several characters, obtaining the character sequence corresponding to the common word sequence. This character sequence is then used as input to the ALBERT model. After feature extraction via a bidirectional Transformer encoder, a dynamic character vector with specific information is output for each character in the character sequence. This solves the problem of polysemy.

[0073] (2) Word segmentation feature character vector:

[0074] For any nested long entity containing characters, the nested long entity is used as input to the Word2vec model to obtain the word segmentation vector of the nested long entity. This word segmentation vector is then used as the word segmentation feature character vector for each character within the nested long entity. .

[0075] For common word characters of non-nested long entities, the common word is used as input to the Word2vec model to obtain the word segmentation vector. The word segmentation vector of the common word is then used as the word segmentation feature character vector for each character within the common word. .

[0076] (3) Boundary feature character vector:

[0077] For any nested long entity, the word boundary of the first character of the nested long entity is marked as B (word boundary vector is [1,1,0,0,0]), the word boundary of the middle character of the nested long entity is marked as I (word boundary vector is [1,0,1,0,0]), and the word boundary of the last character of the nested long entity is marked as E (word boundary vector is [1,0,0,1,0]). The boundary feature character vector of each character in the nested long entity is obtained. .

[0078] For any common word, the word boundary of the first character of the common word is marked as B (word boundary vector is [0,1,0,0,0]), the word boundary of the middle character of the common word is marked as I (word boundary vector is [0,0,1,0,0]), and the word boundary of the last character of the common word is marked as E (word boundary vector is [0,0,0,1,0]). If the common word has only one character, then the single character is marked as S (word boundary vector is [0,0,0,0,1]). Obtain the boundary feature character vector of each character within the nested long entity. The boundary feature character vector has a length of 5 and can only take the value BIES.

[0079] The main parameters of the Chinese resume named entity recognition model include the ALBERT output vector dimension, batch size, dropout, number of BiLSTM hidden units, optimizer, and initial learning rate, as shown in the table below:

[0080] Table 1: List of Model Parameter Settings

[0081]

[0082] Talent profile report generation module:

[0083] The talent profile report comprises two parts: a template-based resume introduction and a data-driven competency assessment. The template-based resume introduction visually displays the job seeker's information and requires a pre-defined structured template. Then, based on Chinese resume naming recognition results, named entities are matched to the corresponding positions in the template. The data-driven competency assessment visually displays the job seeker's evaluation results and automatically generates a skills and competency assessment and career development potential based on the Chinese resume, helping HR quickly understand the job seeker's overall qualities.

[0084] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0085] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An AI large model-based resume file talent portrait report intelligent generation system, characterized in that, The system comprises: The resume information analysis module: for each segmented sequence obtained by the segmentation processing, a set of all synonyms of each segmented word in the Chinese general concept knowledge graph is taken as an extended word set of each segmented word; the occurrence frequency of each word in the extended word set of each segmented word in all Chinese resumes is counted, the word with the maximum occurrence frequency is taken as the public word of the extended word set, and the corresponding segmented word is replaced to obtain a public word sequence; the occurrence frequency of each noun of each public word sequence in all public word sequences of the corresponding Chinese resume is obtained, denoted as a first frequency; the occurrence frequency of each noun phrase of each public word sequence in all public word sequences of the corresponding Chinese resume is obtained, denoted as a second frequency; the ratio between the second frequency and the first frequency is taken as the noun phrase nesting probability of each noun of each public word sequence; The intended post entity of each Chinese resume is extracted by using a regular expression, and compared with the intended post entity of other Chinese resumes to obtain the post word sense similarity between each Chinese resume and other Chinese resumes; the post word sense similarity of each Chinese resume and all other Chinese resumes is threshold segmented; a set of all Chinese resumes with a post word sense similarity greater than the segmentation threshold to each Chinese resume is taken as the supplementary resume set of each Chinese resume; the word group nesting probability of each noun in each Chinese resume in the supplementary resume set of the Chinese resume is analyzed, and the entity nesting confidence is determined in combination with the similarity; the formula for determining the entity nesting confidence is: ; in the formula, represents the entity nesting formula of the word group ( , ) of the Chinese resume k; represents the i th noun of the Chinese resume common word sequence S; represents the next common word of the i th noun in the Chinese resume common word sequence S; is the post semantic similarity of the Chinese resume k and the th Chinese resume in the supplementary resume set of the Chinese resume k, is the word group nesting probability of the word group ( , ) in the th Chinese resume in the supplementary resume set of the Chinese resume k, is the total number of all Chinese resumes in the supplementary resume set of the Chinese resume k; Based on the number of elements of the supplementary resume set of each resume, a sparse resume sample is determined; the average value of the noun phrase nesting probability of each noun phrase in all web texts in the sparse resume sample is taken as the search auxiliary discrimination value of each noun phrase; Based on the Chinese resume noun phrase nesting probability, the entity nesting confidence, and the search auxiliary discrimination value, the nested long entity screening value of each noun phrase is determined, and the noun phrases with the nested long entity screening value greater than the preset nested screening threshold are combined to form a nested long entity. The resume information extraction module is configured to obtain a Chinese resume named entity recognition result based on a neural network model according to the nested long entity and the character sequence of the public word. The talent portrait report generation module is configured to match the entity based on the Chinese resume named entity recognition result, and generate a talent portrait report. 2.The AI large model-based resume file talent portrait report intelligent generation system of claim 1, wherein, The post semantic similarity is obtained by the cosine similarity between the intended post entity of each Chinese resume and another Chinese resume. 3.The AI large model-based resume file talent portrait report intelligent generation system of claim 1, wherein, The sparse resume sample is specifically a Chinese resume whose number in the supplementary resume set is less than a preset number. 4.The AI large model-based resume file talent portrait report intelligent generation system of claim 1, wherein, The determination of the nested long entity screening value of each noun phrase of each segmented sequence is specifically: If the Chinese resume is a sparse resume sample, the average value of the noun phrase nesting probability, the entity nesting confidence, and the search auxiliary discrimination value is taken as the nested long entity screening value of each noun phrase of the corresponding Chinese resume; Otherwise, the average value of the noun phrase nesting probability and the entity nesting confidence is taken as the nested long entity screening value of each noun phrase of the corresponding Chinese resume.

Images