Intelligent resume evaluation method, system and equipment based on large model and storage medium

Abstract

The invention provides an intelligent resume evaluation method, system and equipment based on a large model and a storage medium, belongs to the technical field of natural language processing, and aims to clean and label real resume data and supervise and finely adjust the large language model, so that the model has the capabilities of understanding Chinese resume contents and performing structured scoring. The system receives resume information input by a user, automatically extracts key features, performs multi-dimensional scoring in combination with post requirements and industry standards, outputs detailed rule scores including experience matching degree, education background, skill level, employment flexibility and the like, and gives a total score and enrollment suggestions. The system supports personalized scoring rule configuration, is suitable for talent screening of professional technical posts such as IT and Java development, can dynamically optimize evaluation results, and improves recruitment efficiency and decision scientificity.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of natural language processing, more specifically, particularly relates to a resume intelligent evaluation method, system, device and storage medium based on a large model. BACKGROUND

[0002] With the rapid development of Internet recruitment platforms, the number of resumes received by enterprises and recruitment agencies is growing exponentially. Traditional resume screening work mainly relies on manual reading and judgment, which is time-consuming and labor-intensive and is easily affected by subjective factors, and the evaluation results lack consistency and objectivity.

[0003] In recent years, some studies have attempted to use traditional natural language processing techniques (such as TF-IDF, word vectors, LSTM, etc.) to model the semantics of resumes, but due to limitations in model expression ability and training data size, they perform poorly in understanding complex semantics, context associations, and job matching, resulting in low accuracy of evaluation results.

[0004] Therefore, how to build an intelligent evaluation system that can efficiently understand the semantics of resumes, automatically identify key capabilities, and accurately evaluate the overall quality of candidates has become a key technical problem in the current intelligent transformation of human resources. SUMMARY

[0005] To solve the above technical problems, the present application provides a resume intelligent evaluation method, system, device and storage medium based on a large model to solve the technical problem that the traditional resume evaluation method in the prior art performs poorly in understanding complex semantics, context associations, and job matching, resulting in low accuracy of evaluation results.

[0006] The purpose and effect of the resume intelligent evaluation method, system, device and storage medium based on a large model of the present application are achieved by the following specific technical means: The resume intelligent evaluation method based on a large model comprises the following steps: S1: Obtain user resume information, including but not limited to the age, experience, and education of the job seeker; S2: Perform semantic recognition and entity extraction based on the text of the user resume information to extract key dimension features of the resume; S3: Combine a pre-set rule base and a semantic understanding model to perform implicit knowledge reasoning and generate a preliminary evaluation conclusion; S4: Perform multi-dimensional semantic scoring based on the key dimension features of the resume to obtain resume scoring details; S5: Perform interactive enhanced scoring based on multi-dimensional semantic features to determine the semantic consistency and job matching degree of the resume scoring details; S6: Obtain the comprehensive score of the job seeker's resume after fusing the key features and semantic interaction features.

[0007] According to a preferred embodiment, the obtaining of the user resume information comprises: Receiving a resume description in the form of unstructured text, and extracting key semantic information through a pre-trained language model.

[0008] According to a preferred embodiment, the semantic recognition and entity extraction comprises: The semantic recognition and entity extraction process adopts a method that combines named entity recognition and context modeling, uses a BERT-type model to identify entity boundaries, and captures the implicit ability information in the context through an Attention mechanism, where the ability information represents soft skills and tacit abilities.

[0009] According to a preferred embodiment, the generating of an enhanced score based on multi-dimensional semantic features for interaction, and determining the semantic consistency and job matching degree of the resume scoring rules comprise: Cross-verification of the relevance between the skills and certificates mentioned in the resume, and if there is a significant mismatch, a penalty item that prompts the attention of the recruiter is introduced into the total score.

[0010] According to a preferred embodiment, the semantic understanding model comprises: The training phase of the semantic understanding model uses a large amount of annotated resume data to supervise the fine-tuning of the basic large language model, with the original resume score and evaluation analysis instance dataset as the training target; LoRA low-rank adaptation technology is used in the fine-tuning process.

[0011] According to a preferred embodiment, the implicit knowledge reasoning based on the preset rule base and the semantic understanding model, and the generating of the preliminary evaluation conclusion comprise: inputting the structured semantic information extracted from the resume into the rule base, and performing standardized matching on the candidate's educational background, industry experience, and job adaptation degree.

[0012] The resume intelligent evaluation system based on a large model comprises: A resume information acquisition and semantic analysis module for receiving a user-input unstructured resume text; extracting key semantic information and converting it into a unified semantic vector representation for subsequent evaluation.

[0013] A rule reasoning and preliminary evaluation module that combines a rule base to perform standardized matching on the extracted resume information; a DeepSeekR1 model that has been fine-tuned to model and reason the non-explicitly described ability information in the context; recognition of soft skill keywords; output of a preliminary evaluation conclusion containing advantage items and potential risk items as a basis for multi-dimensional scoring.

[0014] A multi-dimensional score generation module sets multiple evaluation dimensions and their corresponding weight distribution strategies, including education matching degree, experience length, skill relevance, industry adaptation degree, in-service state, and certificate gold content. Each dimension uses a knowledge base-based scoring rule and a semantic understanding model to score cooperatively. Finally, the total score value is calculated by weighted calculation, and a score rule with strong interpretability is formed.

[0015] An interaction enhancement scoring module introduces a semantic consistency verification mechanism to cross-verify the relevance between the skills and certificates mentioned in the resume. If there is obvious mismatch, a penalty term is introduced in the corresponding dimension, and a risk item is marked. At the same time, the semantic similarity between the keywords in the job description and the resume content is calculated to identify exaggerated or ambiguous description behaviors.

[0016] An evaluation result generation and output module uses a natural language generation mechanism to integrate the multi-dimensional score results and evaluation basis into a natural language paragraph with strong interpretability. The output content includes: score rule items, advantage items, disadvantage items, potential risk points, comprehensive suggestions, and final scores.

[0017] An electronic device includes a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to implement the above-mentioned large model-based resume intelligent evaluation method.

[0018] A storage medium stores a computer program, and the computer program executes the above-mentioned large model-based resume intelligent evaluation method when running.

[0019] Compared with the prior art, the present application has the following beneficial effects: 1. By fusing named entity recognition and context modeling technology, combining BERT model for entity boundary recognition, and using Attention mechanism to capture the implicit ability information in the context, at the same time, using pre-training language model to extract key semantic information, converting unstructured text into unified semantic vector, and cooperating with semantic understanding model for implicit knowledge reasoning, avoiding information omission caused by shallow semantic understanding in traditional methods, covering explicit information and implicit ability, reflecting the comprehensive quality of job seekers more comprehensively, solving the problem of insufficient understanding of complex semantics and context association in traditional evaluation methods.

[0020] 2. The multi-dimensional score generation module sets multiple evaluation dimensions such as education matching degree, experience length, skill relevance, industry adaptation degree, in-service status, and certificate gold content. Each dimension is scored in combination with the knowledge base scoring rules and the semantic understanding model, and cross-validation is performed through the interactive enhancement scoring module. If a mismatch is found, a penalty term will be introduced in the corresponding dimension and a risk point will be marked. At the same time, exaggerated or ambiguous descriptions are identified to avoid misjudgment caused by the lack of verification in traditional evaluation, making the scoring result more consistent with the real ability of the job seeker and reducing evaluation bias.

[0021] 3. By combining the keywords in the job requirement description with the resume content to calculate the semantic similarity, the adaptation difference is judged to avoid misjudgment caused by surface matching. At the same time, the evaluation result generation and output module uses natural language generation mechanism to integrate the multi-dimensional score results into natural language paragraphs containing score rule items, advantage items, disadvantage items, potential risk points and comprehensive suggestions. The employer can clearly understand the scoring basis and the strengths and weaknesses of the job seeker, solving the problem of low job adaptation and difficult to understand the evaluation results in traditional methods. BRIEF DESCRIPTION OF DRAWINGS

[0022] Fig. 1 is the step flow chart of the resume intelligent evaluation method based on the large model of the present application; Fig. 2 is the technical roadmap of the resume intelligent evaluation system based on the large model of the present application; Fig. 3 is the workflow chart of the resume intelligent evaluation method based on the large model of the present application. DETAILED DESCRIPTION

[0023] The embodiments of the present application will be further described in detail below in combination with the drawings and examples. The following examples are used to illustrate the technical solutions of the present application, but cannot be used to limit the protection scope of the present application.

[0024] Embodiment:

[0025] As shown in the accompanying Figs. 1 to 3 : The present application provides a resume intelligent evaluation method based on a large model, comprising the following steps: S1: Obtain user resume information, including but not limited to the age, experience length, and education of the job seeker.

[0026] Specifically, the resume description in the form of unstructured text is received, and the key semantic information is extracted through a pre-trained language model; first, the original resume text is collected from multiple channels, including recruitment platforms, enterprise databases, and user direct input. The resume information specifically covers: age, gender, educational background, work experience, skill labels, certificate qualifications, project experience, job intention, and other key fields. Since the original resume is in the form of unstructured text, it needs to be standardized and cleaned: remove format disorder content, redundant information, and invalid information. After cleaning, a unified structured resume dataset is formed, with each line corresponding to a field type and specific content; the cleaned resume data is normalized: experience is converted to "X years X months" format, education is standardized to "primary school / junior high school / high school / college / undergraduate / master's / doctorate", skill labels are classified according to industry standards, certificate qualifications are marked, to ensure data format uniformity and reduce ambiguity in subsequent model processing. At the same time, a training corpus is constructed by combining manual annotation, and the annotation personnel are composed of senior HR and industry business experts. When annotating, each resume sample is matched with the corresponding professional score result. The scoring dimensions include educational match degree, experience, skill relevance, industry adaptability, and certificate gold content. This training corpus is used for subsequent supervised fine-tuning of the pre-trained language model, providing professional evaluation standard reference for the model, ensuring that the model can accurately learn to extract key semantic information from the resume text and correspond to the evaluation dimensions, laying an accurate data foundation for subsequent steps of semantic recognition, entity extraction, and evaluation conclusion generation.

[0027] S2: Semantic recognition and entity extraction based on the text of the user's resume information, extracting key dimension features of the resume.

[0028] Specifically, first, the cleaned and normalized resume text in S1 is preprocessed, including using the WordPiece segmentation method to divide the text into tokens recognizable by the model, and adding special markers for subsequent model input preparation; the semantic recognition and entity extraction process uses a method that combines named entity recognition and context modeling, and selects a pre-trained BERT model for entity boundary recognition. The model learns the semantic features of different fields in the resume text and outputs pre-set entity labels, covering explicit entity types. At the same time, the model's built-in Attention mechanism captures the implicit ability information in the context. This mechanism calculates the correlation weight of each token with other tokens, focusing on key words and their context that reflect abilities. Soft skills and implicit abilities cannot be directly obtained through explicit entities, but they are crucial for evaluating the comprehensive competence of job seekers. Through the Attention mechanism, such information can be effectively mined, ensuring that the extracted key dimension features not only include explicit basic information, but also cover implicit ability competence.

[0029] The large model fine-tuning based on LoRA technology first selects a basic large language model as a base model, and fine-tunes the instructions on the resume dataset constructed in S1. To improve training efficiency and reduce parameter size, the Low-Rank Adaptation technology is introduced. Specifically, a low-rank adaptation matrix is inserted into the Transformer layer of the base model, and only the parameters of the two low-rank matrices are updated during training. The original weights of the base model are completely frozen and not modified. During fine-tuning, the resume dataset is structured into training samples according to the format of instructions, inputs, and outputs. The error between the model prediction results and the manually labeled results is calculated using the cross-entropy loss function. The LoRA matrix parameters are iteratively updated using the AdamW optimizer until the feature extraction accuracy of the model on the validation set reaches a stable level. In this way, the model can accurately adapt to the resume evaluation task while maintaining the original general language understanding ability, avoiding the problem of declining generalization ability caused by full fine-tuning, reducing the required computing resources, shortening the training period, and ensuring that the model can efficiently and stably process large-scale resume text semantic recognition and entity extraction tasks, providing comprehensive and accurate key dimension feature input for subsequent rule reasoning and multi-dimensional scoring.

[0030] S3: Combine the preset rule library and the semantic understanding model to perform implicit knowledge reasoning and generate preliminary evaluation conclusions.

[0031] Specifically, the training phase of the semantic understanding model first constructs an annotated resume dataset, which contains original resume texts of multiple industries and positions, manually annotated dimension scores, and evaluation analysis instances. These data are used as training targets to supervise the fine-tuning of the basic large language model. During fine-tuning, the LoRA low-rank adaptation technology is used to update only the low-rank matrix parameters in the model's Transformer layer, and the original model weights are frozen. This approach can reduce the size of the training parameters, reduce the consumption of computing resources, and enable the model to learn the semantic features and evaluation logic of different industry resumes, adapt to diverse resume evaluation scenarios, improve the model's generalization ability, and speed up the model's reasoning speed. The preset rule library contains multi-dimensional standardized rules, including education matching rules, industry experience rules, and position adaptation degree rules. After inputting the structured semantic information extracted in S2 into the rule library, the rule library automatically calls the corresponding rules and compares them with the current matching position requirements to output standardized matching results.

[0032] In the model inference stage, specific prompt templates are designed to guide the model to process according to the flow, so that the model can accurately understand the resume text input by the user and generate structured output. For example, the prompt template guides the model to complete tasks such as entity extraction, ability identification, rule matching, and score suggestion in sequence. At the same time, the prompt words are optimized in combination with user feedback mechanism, and the content of the prompt words is dynamically adjusted to improve the accuracy and individuality of the generated results.

[0033] The model that has been fine-tuned and optimized for prompt words is deployed to the backend inference module of the evaluation system, which realizes data intercommunication with the semantic recognition and entity extraction module of S2 and the preset rule library module. When receiving the user input resume information, the model first extracts the explicit information in the resume, and then identifies the implicit ability through implicit knowledge reasoning; then it calls the preset rule library to perform rule matching between the explicit information and the job requirements; subsequently, it calculates the preliminary score in combination with the multi-dimensional scoring mechanism and generates a preliminary evaluation conclusion; at the same time, it introduces a semantic consistency verification mechanism to detect the matching between skills and certificates; finally, the model calls the natural language generation module to integrate the above content into an evaluation report in natural language form, including strengths, weaknesses, risk warnings, and total score.

[0034] S4: Perform multi-dimensional semantic scoring based on the key dimensional characteristics of the resume to obtain resume scoring details; During the model construction process, quality evaluation of the model output evaluation results is performed, and the evaluation indicators include scoring accuracy, explainability, and job matching degree. If there is a deviation between the evaluation results and the professional scores, a feedback mechanism is introduced to iteratively optimize the model parameters, prompt word templates, and scoring rules, ensuring continuous improvement of the model and adaptation to the needs of different job types and recruitment scenarios.

[0035] The construction of the entire system aims to achieve efficient and automated processing of the resume evaluation task and improve the objectivity, consistency, and explainability of the evaluation process. The tool module provides natural language processing, knowledge retrieval, semantic understanding, and other functions; the memory module stores general knowledge and intermediate results; the planning module is responsible for task decomposition and execution sequence scheduling. The modules work together to jointly promote the entire process of resume evaluation.

[0036] S5: Perform interactive enhanced scoring based on multi-dimensional semantic features to judge the semantic consistency and job matching degree of the resume scoring details.

[0037] Specifically, first, the semantic understanding model fine-tuned by LoRA low-rank adaptation technology is called to load the multi-dimensional semantic features extracted by S2, and the demand information of the current matching post is called from the system database; then the multi-module interactive verification process is started, the first step is to cross verify the skills and certificates: call the skill and certificate association rules in the preset rule library, compare the resume features and rules to find that the skilled Python / Postman and software test engineer intermediate certificate meet the association requirements, and there is no obvious mismatch, and no penalty term is introduced; if the resume features are skilled Python and certificate English level 6, it is determined that the skill and certificate are not associated and lack Python related auxiliary certificate, a prompt that the skill Python and certificate are not associated is generated, and it is suggested that the subsequent Python skill mastery degree is verified through practical operation test, which is marked as a recruitment party attention item.

[0038] S6: Obtain the comprehensive score of the job seeker's resume after fusing the key features and semantic interaction features.

[0039] Specifically, first, the specific content of the key features and the semantic interaction features is determined, the key features refer to the multi-dimensional basic scores corresponding to the explicit and implicit features extracted by S2, and the semantic interaction features refer to the adjustment scores generated by the interactive verification in S5; then the system fusion algorithm is called, the key feature basic total score is calculated according to the post weight first; then the semantic interaction feature adjustment score is added, and the total score after fusion is obtained. In the fusion process, the calculation basis of each link needs to be retained; at the same time, a comprehensive score report is generated, which needs to distinguish the key feature score and the semantic interaction adjustment item, and list the specific adjustment reasons; the role of this fusion process is to make the comprehensive score reflect the basic quality of the job seeker and embody the adaptation details of the job seeker and the target post, avoid the one-sidedness of single-dimensional evaluation, and at the same time ensure the traceability of the score through clear calculation basis, help the recruitment party to fully understand the matching degree of the job seeker and the post, and improve the recruitment decision-making efficiency.

[0040] The resume intelligent evaluation system based on a large model comprises: A resume information acquisition and semantic analysis module is used to receive a user inputted unstructured resume text; extract key semantic information and convert it into a unified semantic vector representation for subsequent evaluation.

[0041] A rule reasoning and preliminary evaluation module is used to standardize matching of the extracted resume information in combination with a rule library; a DeepSeekR1 model fine-tuned by a context modeling and reasoning is used to model and reason the non-explicit description of the ability information; soft skill keywords are identified; and a preliminary evaluation conclusion containing advantage items and potential risk items is outputted as a basis for multi-dimensional scoring.

[0042] Combining the pre-set rule base and the semantic understanding model, the preliminary evaluation conclusion is generated. This module aims to quickly locate whether the candidate meets the basic job requirements and provides basic support for subsequent multi-dimensional scoring. The module first matches the key features extracted from the resume (such as education, experience, skills, etc.) with the standards in the pre-set rule base. For example, for a "Java Senior Development Engineer" position, the rule base will define the following conditions: "Education requirement: Bachelor's degree or above; Experience: 5 years or more; Skill requirement: proficient in Java, Spring framework, etc."

[0043] At the same time, the module uses the semantic understanding model to further analyze the candidate's implicit abilities. For example, if the resume mentions "responsible for multiple cross-department collaboration projects", the module will consider it as having strong communication and coordination skills and give extra points in the preliminary evaluation.

[0044] Finally, the module outputs a preliminary evaluation report, including the candidate's strengths, potential risks, and preliminary scores, providing a reference for subsequent multi-dimensional scoring.

[0045] The multi-dimensional scoring generation module sets multiple evaluation dimensions and calculates the comprehensive score according to the weight distribution, ensuring that the evaluation results are comprehensive and objective.

[0046] Education level: Determine whether the candidate's educational background meets the requirements of the job.

[0047] Experience: Evaluate whether the candidate's actual work experience meets the requirements of the job.

[0048] Professional relevance: Analyze the matching degree of the candidate's skill tags and job requirements.

[0049] Arrival time: Consider whether the candidate's arrival time meets the needs of the recruiter.

[0050] Experience and certificate value: Evaluate whether the candidate's certificates have actual value.

[0051] The scoring of each dimension is completed by the rule base and the semantic understanding model. For example, when evaluating "professional relevance", the module not only checks the skill tags mentioned in the resume, but also combines the job requirements for semantic similarity calculation to ensure accurate scoring.

[0052] The module calculates the subdivision scores of each dimension by weighting and forms detailed scoring rules to provide basic data for subsequent interactive enhanced scoring.

[0053] The interactive enhanced scoring module introduces a semantic consistency verification mechanism to cross-verify the relevance between the skills and certificates mentioned in the resume.

[0054] The module first compares the skill tags in the resume with the certificate qualifications to determine their relevance. For example, if the candidate claims to have "Java development experience" but only holds an "intermediate PLC certificate", the module will flag this inconsistency and introduce a penalty term in the corresponding dimension. At the same time, the module also analyzes the semantic consistency between the resume content and the job requirements, for example, if the job requires "team management experience", but the resume does not explicitly mention similar experience, the module will reduce the score in the "team leadership" dimension.

[0055] In addition, the module supports user feedback mechanisms, allowing the recruiter to adjust the scoring weights or supplement the evaluation rules according to actual situations. For example, the recruiter can manually increase the emphasis on "cross-department collaboration ability", thereby affecting the final score.

[0056] The evaluation result generation and output module is responsible for integrating the multi-dimensional score results into a natural language report with strong interpretability, making it easy for recruiters to quickly understand the strengths and weaknesses of candidates.

[0057] The module first summarizes the dimension scores and scoring details to generate a structured evaluation report, including: Strengths: the candidate's outstanding abilities or experiences; Weaknesses: areas where the candidate needs to improve; Risk warnings: information that may be controversial or needs further verification; Total score: the comprehensive score of all dimensions.

[0058] Finally, the module outputs a detailed evaluation report for the recruiter's decision-making reference.

[0059] The above modules together constitute the complete process of the large model-based job seeker resume intelligent evaluation system. The resume information acquisition and semantic analysis module is responsible for extracting key features; the rule reasoning and preliminary evaluation module generates a basic score; the multi-dimensional score generation module implements comprehensive evaluation; the interactive enhanced scoring module improves the accuracy of the score; the evaluation result generation and output module provides a report with strong interpretability. Each module works together to ensure that the evaluation process is efficient, accurate, and transparent.

[0060] An electronic device includes a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to implement the large model-based resume intelligent evaluation method described above.

[0061] A storage medium, the storage medium stores a computer program, and the computer program runs to execute the large model-based resume intelligent evaluation method described above.

[0062] The above-described embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A resume intelligent evaluation method based on a large model, characterized in that, Comprise the following steps: S1: Obtain user resume information, resume information includes but is not limited to the age of the job seeker, experience, education; S2: Based on the text of user resume information, semantic recognition and entity extraction, extract resume key dimension characteristics; S3: Combine the preset rule base and semantic understanding model, implicit knowledge reasoning, generate preliminary evaluation conclusion; S4: According to the key dimension characteristics of resume, multi-dimensional semantic scoring, get resume scoring details; S5: Based on multi-dimensional semantic features, interactive enhanced scoring is generated, and the semantic consistency and post matching degree of resume scoring details are judged; S6: Obtain the comprehensive score of the job seeker's resume after the fusion of key features and semantic interaction features.

2. The large model-based resume intelligent evaluation method according to claim 1, characterized in that, The acquisition of user resume information includes: Receive the resume description in the form of unstructured text, and extract the key semantic information through the pre-trained language model.

3. The large model-based resume intelligent evaluation method according to claim 1, characterized in that, The semantic recognition and entity extraction include: The semantic recognition and entity extraction process adopts the method of fusing named entity recognition and context modeling, combines BERT model for entity boundary recognition, and captures the implicit ability information in the context through Attention mechanism, and the ability information is represented as soft skills and implicit ability.

4. The large model-based resume intelligent evaluation method according to claim 1, characterized in that, The interactive enhanced scoring based on multi-dimensional semantic features includes: Cross verification is carried out on the relevance between the skills and certificates mentioned in the resume, and if there is obvious mismatch, a penalty item that attracts the attention of the employer is introduced in the total score.

5. The large model-based resume intelligent evaluation method according to claim 1, characterized in that, The semantic understanding model includes: The training stage of semantic understanding model adopts a large number of annotated resume data to supervise the fine-tuning of the basic large language model, and uses the original resume score and evaluation analysis instance data set as the training target; LoRA low rank adaptation technology is used in the fine-tuning process.

6. The large model-based resume intelligent evaluation method according to claim 1, characterized in that, The combination of preset rule base and semantic understanding model includes: input the structured semantic information extracted from the resume into the rule base, and standardize the matching of candidate's education background, industry experience and post adaptation degree.

7. A large model-based resume intelligent evaluation system for implementing the method of any one of claims 1 to 6, characterized in that, It includes: Resume information acquisition and semantic analysis module, used for receiving user input unstructured resume text; extract key semantic information and convert it into unified semantic vector representation for subsequent evaluation; Rule reasoning and preliminary evaluation module, combined with rule base, the extracted resume information is standardized matched; the non-explicit description ability information is modeled and reasoned by the fine-tuned DeepSeekR1 model; Identify soft skill keywords; output the preliminary evaluation conclusion containing advantage items and potential risk items as the basis for multi-dimensional scoring; Multi-dimensional scoring generation module, set multiple evaluation dimensions and their corresponding weight distribution strategy, including education matching degree, experience, skill relevance, industry adaptation degree, in-service state, certificate gold content. Each dimension adopts knowledge base based scoring rule and semantic understanding model collaborative scoring; finally, the total score value is calculated by weighted calculation, and the scoring details with strong interpretability are formed; An interaction enhancement scoring module introduces a semantic consistency verification mechanism to cross-verify the relevance between the skills and certificates mentioned in the resume; If there is obvious mismatch, a penalty term is introduced in the corresponding dimension, and the risk item is marked; at the same time, the semantic similarity calculation is combined with the keywords in the job description and the content of the resume to identify the exaggeration or vague description behavior; An evaluation result generation and output module uses natural language generation mechanism to integrate the multi-dimensional scoring results and evaluation basis into natural language paragraphs with strong interpretability. The output content includes: scoring details, advantages, disadvantages, potential risk points, comprehensive suggestions and final score.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to implement the large model-based intelligent resume evaluation method of any one of claims 1-6.

9. A storage medium, characterized by The storage medium stores a computer program, and the computer program runs to execute the large model-based intelligent resume evaluation method of any one of claims 1-6.

Images