Intelligent auxiliary method, medium and equipment for evaluation of constructional engineering title

Abstract

The invention discloses an intelligent auxiliary method, a medium and equipment for constructional engineering title review. According to the method, key feature information of unstructured text data in a declaration material is extracted by using a natural language processing technology, image data features are analyzed by using a ResNet-50 deep learning model, and a comprehensive quantitative scoring mechanism of a construction engineering performance material is established. And integrating education background, working experience, scientific research achievements and multi-modal performance data, constructing a constructional engineering title review rule base, setting a standard score threshold based on historical declaration data, and generating a title simulation review score and a review prediction result. And finally, aiming at the model, designing a feature weight analysis layer, and automatically outputting improvement suggestions of the declaration material defects according to the simulation review score after being associated with the associated review rule base. According to the method, automatic feature extraction and quantitative evaluation of the unstructured material are realized, the problems of low manual auditing efficiency and fuzzy standard are solved, and the objectivity of title evaluation is improved through the rule base and the evaluation prediction model.

Description

Technical Field

[0001] This invention relates to the field of decision support technology, specifically to an intelligent auxiliary method, medium, and device for professional title evaluation in the field of construction engineering. Background Technology

[0002] Traditional professional title evaluation involves two core stages: an internal review by the company and a final review by the government's human resources and social security department. Applicants must submit documents such as academic certificates, professional qualification certificates, research achievements, papers, performance materials, and award certificates. During the internal review stage, the human resources department focuses on the format, completeness, and internal compliance of the materials. If the materials are not up to standard, the applicant must be asked to supplement and revise them. After the review is passed, the results will be publicized for no less than five working days, and complaints will be handled and information verified. The government's final review stage focuses on verifying the authenticity of the materials and conducting a substantive assessment of academic level and work performance. Materials that do not meet the application requirements or have procedural flaws will be returned. After the review is passed, the results will be publicized and professional title certificates will be issued.

[0003] In existing technologies, using simple algorithms to assist in professional title evaluation has certain limitations. First, simple quantitative models such as linear weighted scoring cannot accurately reflect the complexity of professional and technical capabilities, easily leading to evaluation results that deviate from reality. Second, the algorithms lack the ability to process unstructured materials such as technical summary reports and engineering design drawings. They cannot automatically associate implicit indicators in the evaluation rules, requiring manual feature extraction. Furthermore, simple algorithms are susceptible to interference from falsified application materials; for example, acts such as ghostwriting of papers and patent plagiarism are difficult to identify through basic numerical verification mechanisms, and the algorithms themselves lack the ability to cross-verify the authenticity of data. Finally, simple algorithms are difficult to adapt to the ambiguity and subjectivity of evaluation standards. For example, the difference in the value weight of achievements in different professional fields requires expert judgment, and the applicant's professional and technical capabilities may conflict with the rigid settings of the rule base. Summary of the Invention

[0004] The present invention proposes an intelligent auxiliary method, medium and equipment for the evaluation of professional titles in construction engineering, which can at least solve one of the technical problems in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A smart auxiliary method for professional title evaluation in construction engineering involves performing the following steps using computer equipment: S100. Collect historical materials for professional title applications in construction engineering, establish a database of application materials, distinguish between structured and unstructured data, and establish a data management model. S200. Extract data features from unstructured text and image performance data, and perform comprehensive quantitative analysis of unstructured data based on the extracted data features. S300. Based on the comprehensive quantitative analysis results, a rule library for the evaluation of professional titles in construction engineering is constructed. The evaluation rules are embedded into the professional title evaluation auxiliary prediction model in the form of characteristic constraints, and the model is trained using historical application materials. S400. Based on the trained professional title evaluation auxiliary prediction model, construct an interpretability module, combine the interpretability module with the construction engineering professional title rule base, and output personalized suggestions for improving professional title materials.

[0006] Furthermore, the historical materials for the application for professional titles in construction engineering in step S100 of this invention include: Structured data, unstructured text data, and unstructured image data; Structured data includes basic information that can be quantified, such as the applicant's years of work experience, educational level, continuing education hours, professional qualifications, award level, and number of papers. Unstructured text data includes technical summary reports, special achievement documents, engineering performance materials, and research project application forms, reflecting professional and technical capabilities. Unstructured image data is crucial for verifying the authenticity and evaluating the richness of performance materials, including award certificates, official seals on contracts, engineering design drawings, construction process images, and attached diagrams in acceptance reports.

[0007] Furthermore, the data feature extraction method in step S200 of the present invention includes: Methods for feature extraction from unstructured text-based performance data and unstructured image-based performance data: The feature extraction method for unstructured text performance data is as follows: The text data of the technical summary report and engineering performance materials submitted by the applicant are retrieved from the feature library, and the continuous text is segmented into sentences to generate a set of sentences. ; Then, noise reduction processing is performed on stop words and irrelevant words to generate word vector sequences. This achieves the standardization of input data; A language database for professional title evaluation in the construction engineering field is constructed. This database uses specialized terminology within the construction engineering domain to generate domain-enhanced word vectors, eliminating ambiguity in model judgments caused by industry-specific terminology.

[0008] in, For general word vectors, For domain knowledge vectors, These are the weighting coefficients.

[0009] By introducing the hidden state vector of the BiLSTM model, the contextual information of the text material can be captured; Calculate the probability of the label sequence using a probability model. : ; Then, the project complexity score is calculated based on the probability of the label sequence. : ,

[0010] in, The hidden state vector contains forward and backward text information; It is the conditional probability distribution function; Here is the complexity label at time step t; Statistical tags The number of times it appears in the text; The label weighting function increases the weight of projects with higher levels, more participants, and higher amounts of money involved. The entity frequency decay coefficient transforms the text description into a calculated indicator. Based on the applicant's responsibilities in the performance projects With contribution Construct a contribution weight function ; The individual contribution metric function is set as follows:

[0011]

[0012] in, This is the technical object coefficient; the higher the technical difficulty involved in the performance project, the larger this coefficient will be. This is a moderating factor for project difficulty; the higher the difficulty of completing a project, the larger this coefficient becomes. The number of performance projects submitted by the applicant; The score is based on the applicant's written performance materials. The weighting coefficient for the text material score; Furthermore, the method for feature extraction of unstructured image performance material data according to the present invention includes: Extracting engineering drawing data from a graph database Converted into a three-channel normalized image matrix This linearly transforms the pixel value to interval; Among them, spatial dimension This represents the image height and width; channel dimension 3 corresponds to the RGB color space. Use the ResNet-50 deep learning model to extract features from image data; Spatial downsampling and feature enhancement are performed through convolutional layers and residual modules to achieve multi-level residual feature extraction; Initial feature extraction of convolutional layers:

[0013] Establish residual block groups to implement feature transformation:

[0014]

[0015]

[0016]

[0017] Establish a global average pooling layer Reduce the number of model parameters to prevent overfitting:

[0018] Establish a fully connected layer to retain and flexibly combine the feature information extracted by global average pooling:

[0019] Among them, the weight matrix Bias term ; Features of related image professional title data:

[0020] in, The corresponding architectural engineering attribute in the image-based professional title data; Among them, the output 2048-dimensional feature vector Through a fully connected layer Feature map compression preserves spatial semantic information and meets the input specifications of the review model; Calculate the scale of performance projects based on extracted data features. With project complexity :

[0021]

[0022]

[0023] Among them, the maximum pooling layer Defined as: ; in, It is the Sigmoid activation function. ; and These are trainable weight vectors; The score is based on the applicant's image and performance materials. , which represents the weighting coefficient for the image material score.

[0024] Furthermore, the comprehensive quantitative analysis method in step S200 of the present invention is as follows: S231. Combining the project complexity and individual contribution of the textual performance materials, construct characteristic dimension indicators for the textual performance materials. ; By combining the project scale and complexity of image-based performance materials, feature dimension indicators for text-based performance materials are constructed. ; Then, a text-image association matrix is ​​established to achieve cross-modal data alignment:

[0025] Among them, the row direction represents the single-modal feature integrity of the text line and the image line, respectively, and the column direction represents the cross-modal correlation of scale correlation and technical depth correlation, respectively. S232. Calculate the fused feature vector based on the text-image correlation matrix:

[0026] in, As text complexity weights, Compensation weights for text contribution. Weights based on image project size. This represents the complexity weight of the image project.

[0027] Establish a comprehensive scoring formula for unstructured performance materials: ,

[0028] in, For cross-modal weight vectors, Corresponding text feature weights, Corresponding image feature weights.

[0029] Furthermore, the method for constructing the rule base for professional title evaluation in construction engineering in step S300 of the present invention includes: Based on the key elements of professional title evaluation, including academic qualifications, work experience, work achievements, written performance materials, and image performance materials, a scoring method library is defined for applicants. for:

[0030]

[0031] in, For academic education scoring functions; For work experience scoring function; A scoring function to quantify work results; This is a comprehensive scoring function for performance materials; in, For the number of published papers, This is the paper's grade coefficient, with values ​​divided into core journal papers and ordinary journal papers; The number of patents granted. This is the patent level coefficient, and its value is divided into invention patents and utility model patents. For the number of science and technology awards, This is the award level coefficient, with values ​​divided into national-level awards, provincial-level awards, and municipal-level awards. The database is used to filter historical application materials that have passed the professional title evaluation, and the scores of the application materials are calculated. The minimum score for passing the evaluation at different professional title levels is set as the standard score.

[0032] in, Standard scores for different professional title levels, The standard score represents the 5th percentile. It is calculated based on the scores of historically approved application materials. The 5th percentile of the score distribution of this group is taken as the review benchmark to ensure that the standard score represents the lowest historical passing requirement. in, These are the weighting coefficients. , and The weighting coefficients correspond to the junior, intermediate, and senior professional titles, respectively. in, For the collection of applicants, , and These are the sets of applicants for junior, intermediate, and senior professional titles, respectively.

[0033] Furthermore, the method for constructing an interpretability module in step S400 of the present invention includes: Based on the professional title evaluation auxiliary prediction model, a feature weight parsing layer is designed:

[0034] in, This is the cross-membrane state weight vector; For text-image fusion features; The feature contribution is used as the influence coefficient of each feature dimension on the total score.

[0035] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0036] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0037] As can be seen from the above technical solutions, this solution uses NLP technology to extract and quantify key feature information in text data, and uses a deep learning ResNet-50 model to extract and quantify image data features, eliminating information omissions and improving the objectivity of the review. Based on the professional title review auxiliary prediction model, a feature weight parsing layer is designed, and an interpretability module is constructed. This interpretability module is combined with the construction engineering professional title rule base, and personalized suggestions for improving professional title materials are output based on the review score. This achieves personalized improvement suggestions for applicants, enhancing the model's credibility and user acceptance. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the intelligent auxiliary method for professional title evaluation in construction engineering according to the present invention. Figure 2 This is a diagram of the ResNet-50 algorithm structure of this invention; Figure 3 This is a flowchart illustrating the process of constructing an auxiliary prediction model for professional title evaluation according to the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0040] like Figure 1 As shown in this embodiment, the intelligent auxiliary method for professional title evaluation in construction engineering performs the following steps using computer equipment: S100. Collect historical materials for professional title applications in construction engineering, establish a database of application materials, distinguish between structured and unstructured data, and establish a data management model. S200. Extract data features from unstructured text and image performance data, and perform comprehensive quantitative analysis of unstructured data based on the extracted data features. S300. Based on the comprehensive quantitative analysis results, a rule library for the evaluation of professional titles in construction engineering is constructed. The evaluation rules are embedded into the professional title evaluation auxiliary prediction model in the form of characteristic constraints, and the model is trained using historical application materials. S400. Based on the trained professional title evaluation auxiliary prediction model, construct an interpretability module, combine the interpretability module with the construction engineering professional title rule base, and output personalized suggestions for improving professional title materials.

[0041] The following provides a detailed explanation of each step: S100. Collect historical materials for professional title applications in construction engineering, establish a database of application materials, distinguish between structured and unstructured data, and establish a data management model.

[0042] S110. In view of the professional characteristics of the evaluation of professional titles in construction engineering, the application materials are divided into three categories according to the data type: structured data, unstructured text data, and unstructured image data.

[0043] The structured data includes basic, quantifiable information such as the applicant's years of work experience, educational level, continuing education hours, professional qualifications, award level, and number of papers.

[0044] The unstructured text data includes technical summary reports, special achievement documents, engineering performance materials, and scientific research project proposals, reflecting professional and technical capabilities. Among them, engineering performance materials are diverse in form, and it is necessary to extract key performance information from these text materials, such as project level, project scale, technical positions, duration, project results, and technological innovation measures.

[0045] The unstructured image data is crucial for verifying the authenticity and evaluating the richness of performance materials, and mainly includes award certificates, official seals on contracts, engineering design drawings, construction process images, and attached drawings in acceptance reports.

[0046] S120. Based on the data characteristics of various materials, different types of databases are used for storage, and a data update and maintenance mechanism is established. For structured data: In terms of storage, a relational database is used to store structured data. A core data table is established with the applicant's ID as the primary key, and redundancy in numerical fields is eliminated through normalization. Multi-valued attributes use related sub-tables to achieve one-to-many mapping, ensuring data atomicity and integrity. Regarding the update mechanism, applicants submit incremental data through a standardized interface, and the system automatically performs numerical logic verification and updates the core tables in real time. Administrators review requests for changes to related sub-tables, and upon approval, perform batch updates and retain version snapshots to ensure data atomicity and historical traceability.

[0047] For unstructured text data: In terms of storage, a document database is used to store the raw text data, with each record bound to metadata tags. Simultaneously, an independent feature library is established to store key structured features extracted from the text. Regarding the update mechanism, a unique file identifier is automatically assigned and persistently stored after text is uploaded. When the text content is revised, a feature re-extraction process is triggered, the updated features are written into the feature library, and a version mapping relationship is established, supporting the tracing of historical features.

[0048] For unstructured image data: In terms of storage, an object storage service is used to implement hierarchical image management, with frequently accessed data stored in the hot storage layer and large files stored in the cold storage layer; image hash values, technical tags and associations are stored through a graph database. Regarding the update mechanism, image technology tags are automatically generated during the upload phase. When the associated data changes, image consistency verification is triggered, and the verification results are updated to the graph database in real time. Modification records of key images are stored through chained hashing to ensure the traceability of data sources.

[0049] S200. Extract data features from unstructured text and image performance data, and perform comprehensive quantitative analysis of unstructured data based on the extracted data features. For unstructured text-based performance materials, the data feature extraction method is as follows: S211. Retrieve text data such as technical summary reports and engineering performance materials submitted by the applicant from the feature library, perform sentence segmentation on the continuous text, and generate a set of sentences. Then, noise reduction is performed on stop words and irrelevant words to generate a word vector sequence. This achieves the standardization of input data; S212. Construct a language database for professional title evaluation in the construction engineering field, and use professional terms in the construction engineering field to generate domain-enhanced word vectors to eliminate ambiguity in model judgment caused by industry terminology:

[0050] in, For general word vectors, For domain knowledge vectors, These are the weighting coefficients.

[0051] S213. Introduce the hidden state vector of the BiLSTM model to capture the contextual information of the text material, and calculate the probability of the label sequence through a probabilistic model. : ; Then, the project complexity score is calculated based on the probability of the label sequence. : ,

[0052] in, The hidden state vector contains forward and backward text information; It is the conditional probability distribution function; Here is the complexity label at time step t; Statistical tags The number of times it appears in the text; The label weighting function increases the weight of projects with higher levels, more participants, and higher amounts of money involved. The entity frequency attenuation coefficient is used to convert text descriptions into computational metrics.

[0053] S214. Based on the applicant's responsibilities in the performance projects. With contribution Construct a contribution weight function Then, the individual contribution metric function is set as follows:

[0054]

[0055] in, This is the technical object coefficient; the higher the technical difficulty involved in the performance project, the larger this coefficient will be. This is a moderating factor for project difficulty; the higher the difficulty of completing a project, the larger this coefficient becomes. The number of performance projects submitted by the applicant; The score is based on the applicant's written performance materials. This represents the weighting coefficient for the text material's score.

[0056] For unstructured image performance materials, the data feature extraction method is as follows: S221. Extracting engineering drawing data from the drawing database. Converted into a three-channel normalized image matrix This linearly transforms the pixel value to Interval. Wherein, spatial dimension... This represents the image height and width; channel dimension 3 corresponds to the RGB color space. S222. Use the ResNet-50 deep learning model to extract features from image data; Spatial downsampling and feature enhancement are performed through convolutional layers and residual modules to achieve multi-level residual feature extraction; Initial feature extraction of convolutional layers:

[0057] Establish residual block groups to implement feature transformation:

[0058]

[0059]

[0060]

[0061] Establish a global average pooling layer Reduce the number of model parameters to prevent overfitting:

[0062] Establish a fully connected layer to retain and flexibly combine the feature information extracted by global average pooling:

[0063] Among them, the weight matrix Bias term ; Features of related image professional title data:

[0064] in, The corresponding architectural engineering attribute in the image-based professional title data; Among them, the output 2048-dimensional feature vector Through a fully connected layer Feature map compression preserves spatial semantic information and meets the input specifications of the review model. An explicit mapping relationship is established between the feature dimensions and the building engineering attribute database to ensure interpretability.

[0065] S223. Calculate the performance project scale based on the extracted data characteristics. With project complexity :

[0066]

[0067]

[0068] Among them, the maximum pooling layer Defined as: ; in, It is the Sigmoid activation function. ; and These are trainable weight vectors; The score is based on the applicant's image and performance materials. , which represents the weighting coefficient for the image material score.

[0069] Among them, the global max pooling layer Extract the most prominent features from the engineering drawings and compare them with the global average pooling layer. They form complementary features, respectively representing the complexity and scale of the project.

[0070] Combining feature extraction methods for unstructured text and image data, a comprehensive quantitative analysis method for construction engineering performance materials is constructed as follows: S231. Combining the project complexity and individual contribution of the textual performance materials, construct characteristic dimension indicators for the textual performance materials. ; By combining the project scale and complexity of image-based performance materials, feature dimension indicators for text-based performance materials are constructed. ; Then, a text-image association matrix is ​​established to achieve cross-modal data alignment:

[0071] Among them, the row direction represents the integrity of the single-modal features of the text line and the image line, respectively, and the column direction represents the cross-modal correlation of scale correlation and technology depth correlation, respectively.

[0072] S232. Calculate the fused feature vector based on the text-image correlation matrix:

[0073] in, As text complexity weights, Compensation weights for text contribution. Weights based on image project size. This represents the complexity weight of the image project.

[0074] Establish a comprehensive scoring formula for unstructured performance materials: ,

[0075] in, For cross-modal weight vectors, Corresponding text feature weights, Corresponding image feature weights.

[0076] S300. Based on the comprehensive quantitative analysis results, a rule library for the evaluation of professional titles in construction engineering is constructed. The evaluation rules are embedded into the professional title evaluation auxiliary prediction model in the form of characteristic constraints, and the model is trained using historical application materials.

[0077] S310, Key elements in comprehensive professional title evaluation, and construct a rule base for professional title evaluation in construction engineering; A comprehensive scoring method library is defined for applicants, taking into account key elements such as academic education, work experience, work achievements, written performance materials, and image performance materials. for:

[0078]

[0079] in, This is a scoring function for formal education, used to map formal education to scores; This is a work experience scoring function used to map work experience into scores; A scoring function to quantify work results; This is a comprehensive score function for performance materials.

[0080] in, For the number of published papers, This is the paper's grade coefficient, with values ​​divided into core journal papers and ordinary journal papers; The number of patents granted. This is the patent level coefficient, and its value is divided into invention patents and utility model patents. For the number of science and technology awards, This is the award level coefficient, with values ​​divided into national-level awards, provincial (ministerial)-level awards, and municipal (departmental)-level awards; S320. Based on the scores of historical application materials, set a standard score for passing the professional title evaluation. Calculate the applicant's evaluation score and output the evaluation analysis results; S321. Filter historical application materials that have passed the professional title evaluation from the database, calculate the score of the application materials, and set the minimum score for passing the evaluation under different professional title levels as the standard score:

[0081] in, Standard scores for different professional title levels, The standard score represents the 5th percentile. It is calculated based on the scores of application materials that have passed the review in the past. The 5th percentile of the score distribution of this group is taken as the review benchmark to ensure that the standard score represents the lowest passing requirement in history.

[0082] in, These are the weighting coefficients. , and These correspond to the weighting coefficients for junior, intermediate, and senior professional titles, respectively.

[0083] in, For the collection of applicants, , and These are the sets of applicants for junior, intermediate, and senior professional titles, respectively.

[0084] S322. Calculate the applicant's review score, compare it with the standard score, and output the review result:

[0085]

[0086] in, The applicant's review score For the review results; In particular, to avoid directly classifying applicants whose scores are close to the historical lowest passing score as having passed, a coefficient... Adjustments can be made based on the error rate of historical review data: For review scores greater than or equal to Applicants whose scores exceed the standard score will be deemed to have submitted materials that will pass the review. For review scores less than For applicants whose scores are multiples of the standard score and greater than or equal to the standard score, the result of the review is pending and they need to supplement materials based on the improvement suggestions for re-review. Applicants whose review scores are lower than the standard score will be deemed unable to pass the review and will need to supplement their materials and be reviewed again based on the improvement suggestions.

[0087] S400. Based on the trained professional title evaluation auxiliary prediction model, construct an interpretability module, combine the interpretability module with the construction engineering professional title rule base, and output personalized suggestions for improving professional title materials.

[0088] S410. Construct an interpretability module based on the professional title evaluation auxiliary prediction model. Design a feature weight parsing layer based on the professional title evaluation auxiliary prediction model:

[0089] in, This refers to the transmembrane state weight vector in S232; For text-image fusion features in S232; The feature contribution is used as the influence coefficient of each feature dimension on the total score.

[0090] The gradient term is used to characterize the effect of small increases in each feature on the change in the total score; the Hadamard product combines global weights with local sensitivity gradients to eliminate bias caused by modal differences.

[0091] S420. For applicants whose review results are pending, the interpretability module will be integrated with the architectural engineering professional title rule base. Based on the review scores, personalized suggestions for improving professional title application materials will be provided. S421. Based on the applicant's score, select the weakest indicators and calculate the missing scores for each weakest indicator: First, calculate the applicant's score rate in each dimension:

[0092] in, This represents the k-th dimension score of the applicant's rule base. These correspond to academic education, work experience, technological achievements, and work performance, respectively. This represents the 5th percentile of the historical score in the k-th dimension for the current professional title level; Then, the dimensions in the applicant's materials that did not meet the standards but had a relatively high weight are designated as weak points, and improvement suggestions need to be generated. The formula for identifying the weakest link indicator is set as follows:

[0093] A quantitative formula is set to address the applicant's weaknesses:

[0094] S422. For the four indicators—academic education, work experience, technological achievements, and work performance—matched with the professional title evaluation rule base, the following method is set for generating improvement suggestions:

[0095] In summary, this solution utilizes NLP technology to extract and quantify key feature information from text data, and employs a deep learning ResNet-50 model to extract and quantify features from image data, thus eliminating information omissions and improving the objectivity of the review process. Based on the professional title review auxiliary prediction model, a feature weight parsing layer is designed, and an interpretability module is constructed. This interpretability module is integrated with the construction engineering professional title rule base, and personalized suggestions for improving professional title application materials are output based on the review score. This achieves personalized improvement suggestions for applicants, enhancing the model's credibility and user acceptance.

[0096] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0097] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0098] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the intelligent auxiliary methods for professional title evaluation in construction engineering described in the above embodiments.

[0099] It is understood that the systems, devices, and storage media provided in the embodiments of the present invention correspond to the methods provided in the embodiments of the present invention, and the explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0100] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0102] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0103] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention 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 spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent auxiliary method for professional title evaluation in the field of architectural engineering, characterized in that, Perform the following steps using a computer device: S100. Collect historical materials for professional title applications in construction engineering, establish a database of application materials, distinguish between structured and unstructured data, and establish a data management model. S200. Extract data features from unstructured text and image performance data, and perform comprehensive quantitative analysis of unstructured data based on the extracted data features. S300. Based on the comprehensive quantitative analysis results, a rule library for the evaluation of professional titles in construction engineering is constructed. The evaluation rules are embedded into the professional title evaluation auxiliary prediction model in the form of characteristic constraints, and the model is trained using historical application materials. S400. Based on the trained professional title evaluation auxiliary prediction model, construct an interpretability module, combine the interpretability module with the construction engineering professional title rule base, and output personalized suggestions for improving professional title materials.

2. The intelligent auxiliary method for professional title evaluation in construction engineering according to claim 1, characterized in that, The historical materials for the application for professional titles in construction engineering in step S100 include: Structured data, unstructured text data, and unstructured image data; Structured data includes basic information that can be quantified, such as the applicant's years of work experience, educational level, continuing education hours, professional qualifications, award level, and number of papers. Unstructured text data includes technical summary reports, special achievement documents, engineering performance materials, and research project application forms, reflecting professional and technical capabilities. Unstructured image data is crucial for verifying the authenticity and evaluating the richness of performance materials, including award certificates, official seals on contracts, engineering design drawings, construction process images, and attached diagrams in acceptance reports.

3. The intelligent auxiliary method for professional title evaluation in construction engineering according to claim 1, characterized in that, The data feature extraction methods in step S200 include: Methods for feature extraction from unstructured text-based performance data and unstructured image-based performance data: The feature extraction method for unstructured text performance data is as follows: The text data of the technical summary report and engineering performance materials submitted by the applicant are retrieved from the feature library, and the continuous text is segmented into sentences to generate a set of sentences. ; Then, noise reduction processing is performed on stop words and irrelevant words to generate word vector sequences. This achieves the standardization of input data; A language database for professional title evaluation in the construction engineering field is constructed. This database uses specialized terminology within the construction engineering domain to generate domain-enhanced word vectors, eliminating ambiguity in model judgments caused by industry-specific terminology. in, For general word vectors, For domain knowledge vectors, These are the weighting coefficients; By introducing the hidden state vector of the BiLSTM model, the contextual information of the text material can be captured; Calculate the probability of the label sequence using a probability model. : ； Then, the project complexity score is calculated based on the probability of the label sequence. : , The hidden state vector contains forward and backward text information; It is the conditional probability distribution function; Here is the complexity label at time step t; Statistical tags The number of times it appears in the text; The label weighting function increases the weight of projects with higher levels, more participants, and higher amounts of money involved. The entity frequency attenuation coefficient is used to transform the textual description into a calculated indicator; based on the applicant's responsibilities in the performance projects. With contribution Construct a contribution weight function ; The individual contribution metric function is set as follows: in, This is the technical object coefficient; the higher the technical difficulty involved in the performance project, the larger this coefficient will be. This is a moderating factor for project difficulty; the higher the difficulty of completing a project, the larger this coefficient becomes. The number of performance projects submitted by the applicant; The score is based on the applicant's written performance materials. This represents the weighting coefficient for the text material's score.

4. The intelligent auxiliary method for professional title evaluation in construction engineering according to claim 3, characterized in that, Methods for feature extraction from unstructured image performance data include: Extracting engineering drawing data from a graph database Converted into a three-channel normalized image matrix This linearly transforms the pixel value to interval; Among them, spatial dimension This represents the image height and width; channel dimension 3 corresponds to the RGB color space. Use the ResNet-50 deep learning model to extract features from image data; Spatial downsampling and feature enhancement are performed through convolutional layers and residual modules to achieve multi-level residual feature extraction; Initial feature extraction of convolutional layers: Establish residual block groups to implement feature transformation: Establish a global average pooling layer Reduce the number of model parameters to prevent overfitting: Establish a fully connected layer to retain and flexibly combine the feature information extracted by global average pooling: Among them, the weight matrix Bias term ; Features of related image professional title data: in, The corresponding architectural engineering attribute in the image-based professional title data; Among them, the output 2048-dimensional feature vector Through a fully connected layer Feature map compression preserves spatial semantic information and meets the input specifications of the review model; Calculate the scale of performance projects based on extracted data features. With project complexity : Among them, the maximum pooling layer Defined as: ; in, It is the Sigmoid activation function. ; and These are trainable weight vectors; The score is based on the applicant's image and performance materials. , which represents the weighting coefficient for the image material score.

5. The intelligent auxiliary method for professional title evaluation in construction engineering according to claim 1, characterized in that, The comprehensive quantitative analysis method in step S200 is as follows: S231. Combining the project complexity and individual contribution of the textual performance materials, construct characteristic dimension indicators for the textual performance materials. ; By combining the project scale and complexity of image-based performance materials, feature dimension indicators for text-based performance materials are constructed. ; Then, a text-image association matrix is ​​established to achieve cross-modal data alignment: Among them, the row direction represents the single-modal feature integrity of the text line and the image line, respectively, and the column direction represents the cross-modal correlation of scale correlation and technical depth correlation, respectively. S232. Calculate the fused feature vector based on the text-image correlation matrix: in, As text complexity weights, Compensation weights for text contribution. Weights based on image project size. Weights for image item complexity; Establish a comprehensive scoring formula for unstructured performance materials: ， in, For cross-modal weight vectors, Corresponding text feature weights, Corresponding image feature weights.

6. The intelligent auxiliary method for professional title evaluation in construction engineering according to claim 1, characterized in that, The method for constructing the rule base for professional title evaluation in construction engineering in step S300 includes: Based on the key elements of professional title evaluation, including academic qualifications, work experience, work achievements, written performance materials, and image performance materials, a scoring method library is defined for applicants. for: in, For academic education scoring functions; For work experience scoring function; A scoring function to quantify work results; This is a comprehensive scoring function for performance materials; in, For the number of published papers, This is the paper's grade coefficient, with values ​​divided into core journal papers and ordinary journal papers; The number of patents granted. This is the patent level coefficient, and its value is divided into invention patents and utility model patents. For the number of science and technology awards, This is the award level coefficient, with values ​​divided into national-level awards, provincial-level awards, and municipal-level awards. The database is used to filter historical application materials that have passed the professional title evaluation, and the scores of the application materials are calculated. The minimum score for passing the evaluation at different professional title levels is set as the standard score. in, Standard scores for different professional title levels, The standard score represents the 5th percentile. It is calculated based on the scores of historically approved application materials. The 5th percentile of the score distribution of this group is taken as the review benchmark to ensure that the standard score represents the lowest historical passing requirement. in, These are the weighting coefficients. , and The weighting coefficients correspond to the junior, intermediate, and senior professional titles, respectively. in, For the collection of applicants, , and These are the sets of applicants for junior, intermediate, and senior professional titles, respectively.

7. The intelligent auxiliary method for professional title evaluation in construction engineering according to claim 1, characterized in that, The method for constructing an interpretable module in step S400 includes: Based on the professional title evaluation auxiliary prediction model, a feature weight parsing layer is designed: in, This is the cross-membrane state weight vector; For text-image fusion features; The feature contribution is used as the influence coefficient of each feature dimension on the total score.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, the processor performs the method as described in any one of claims 1 to 7.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program that, when executed by the processor, causes the processor to perform the method as described in any one of claims 1 to 7.

Images