Project automatic auditing method based on AI large model

Abstract

The invention relates to an automatic project auditing method based on an AI large model, which comprises the following steps: step 1, storing and analyzing project materials in different formats, including word, pdf and scanning document formats, step 2, performing industry category auditing on a project by using analyzed contents, step 3, performing industrial added value energy consumption auditing on the project, and step 4, performing industrial added value energy consumption auditing on the project. The method comprises the following steps of: (1) performing energy consumption index implementation, demonstration and review on a project, (2) summarizing project form review results, (6) performing automatic screening and distribution on project materials of the summarized form review results in the step (5) by an entrusting unit, and (7) performing in-mechanism review and uploading review opinions after the entrusting unit receives project information. Step 8, constructing a large model context of a final review result for the project passing the review of the mechanism in the step 7, and step 9, generating a final review meaning through the large model context and the cue word set in the step 8; according to the scheme, the final review result is automatically generated, and the review efficiency is comprehensively improved.

Description

Technical Field

[0001] This invention relates to an automatic review method, specifically an automatic project review method based on an AI large model, belonging to the fields of artificial intelligence and data processing technology. Background Technology

[0002] Currently, relevant departments need to conduct strict manual review of the project materials submitted by various cities. The review process mainly covers four stages: project form review, project entrustment and distribution, third-party institution review, and the formation of final review opinions.

[0003] The project formal review includes industry category review, industrial added value energy consumption (equivalent value) review, and energy consumption indicator implementation demonstration review. After the project formal review is completed, the relevant materials will be distributed to designated third-party institutions, entering the project commissioning and distribution stage. Upon receiving the project commission, the third-party institution will conduct an offline review and upload its preliminary review opinions to the system after the review. Finally, the final review opinion will be formed by combining the project formal review results with the preliminary review opinions of the third-party institution.

[0004] Currently, the materials for such projects are diverse in form, complex in content, and inconsistent in format. The material formats include Word, PDF, scanned documents (images), and many others. This means that the review process requires a significant amount of manpower to read each project material individually, and writing the final review comments also requires a substantial investment of time. Summary of the Invention

[0005] This invention addresses the technical problems existing in current technologies by designing an automated project review method based on an AI large-scale model. This invention integrates deep learning and large-scale model technology to construct an automated project review method and system. Its core process is as follows: First, all text content in scanned documents is extracted using image recognition and OCR technologies, while key seal information in project materials is captured using seal recognition technology; second, a balanced algorithm is designed to intelligently distribute project assignments, automatically allocating project materials to corresponding third-party organizations; finally, using large-scale model technology and combining the review opinions of third-party organizations, the final review result is automatically generated.

[0006] The project materials are diverse in format, complex in content, and inconsistent in format. The materials include various types such as Word, PDF, and scanned documents (images), which were traditionally printed manually and processed offline. This invention designs filtering rules to handle materials of different formats. For Word or PDF text, all content is directly extracted from the Word document; for image-based text, OCR technology is used to obtain the text content.

[0007] Project formal review encompasses industry category review, industrial added value energy consumption (equivalent value) review, and energy consumption indicator implementation demonstration review. Among these, industry category review verifies whether the construction content described in the project materials matches the declared industry category. However, since construction content is often lengthy, manual review is not only time-consuming and labor-intensive but also prone to judgment errors. This invention addresses this problem by first constructing an industry category knowledge base based on the National Economic Classification System; then, using the construction content in the project materials to search this knowledge base and obtain matching industry classifications; simultaneously, by setting large model prompts, the declared industry classification is directly extracted from the materials; finally, by comparing the industry classifications obtained through the two methods, the industry category review is automated.

[0008] In the project formal review, the core of the industrial added value energy consumption (equivalent value) review is to compare the energy consumption level and intensity information recorded in the project materials with the energy consumption (equivalent value) of the corresponding city. The traditional manual review mode is time-consuming and inefficient. This invention addresses this pain point by using OCR technology to identify the issuing unit of the project materials, and by using a large model prompt word design to accurately extract the industrial added value energy consumption (equivalent value) data from the materials. Finally, based on the issuing unit, the lowest industrial added value energy consumption (equivalent value) value of the corresponding city is matched and automatically compared with the data extracted by the large model, thereby achieving full automation of the review process for this indicator.

[0009] In the project formal review, the core of the "energy consumption indicator implementation demonstration" review lies in comparing the project's energy consumption recorded in the project materials with the corresponding city's energy consumption. Traditional manual review methods are time-consuming and inefficient. This invention addresses this pain point by automating the entire review process for this indicator through two methods: first, using a large model to extract the title text of the project materials; and second, using OCR technology to identify the issuing unit in the project materials' seals, thereby completing the automatic review and judgment.

[0010] The traditional project assignment process involves manually selecting a candidate list of agencies based on the industry classifications listed in the project materials, and then randomly selecting one from this list as the agency to review the project materials. This random allocation method easily leads to an uneven distribution of projects among agencies, affecting overall review efficiency. This invention addresses this problem by designing a dedicated balancing algorithm to automate and evenly distribute materials to relevant agencies, effectively solving the problem of imbalanced allocation.

[0011] Traditionally, the final review results are written manually, combining third-party review opinions and project formal review results. The content covers multiple aspects, including the review process, basic project information, overall energy consumption, and its impact. Due to the length of the document, manual writing is often time-consuming. This invention addresses this issue by using third-party review opinions and project formal review results as input context for a large model. By setting targeted prompts, it achieves automatic generation of the final project review results.

[0012] In summary, this invention addresses the problems of complex material formats, low efficiency of manual review, uneven distribution, and time-consuming result compilation in project review. It proposes an automated review solution integrating deep learning and large-scale model technology. The solution extracts text from scanned documents through OCR and image recognition, and obtains key information by combining seal recognition; it constructs a knowledge base based on the National Economic Classification, and achieves automated industry category review by comparing industry classifications extracted through retrieval and large-scale model; it uses OCR to identify the issuing unit and large-scale model to extract energy consumption data, comparing it with corresponding municipal indicators to complete energy consumption review; it designs a balancing algorithm to solve the problem of unbalanced project delegation distribution; and it uses third-party review opinions and formal review results as input to the large-scale model to automatically generate the final review result, comprehensively improving review efficiency.

[0013] To achieve the above objectives, the technical solution of the present invention is as follows: a method for automatic project review based on an AI large model, the method comprising the following steps:

[0014] Step 1: Store and parse project materials in different formats, including Word, PDF, scanned documents (images), etc.

[0015] Step Two: Using the information analyzed in Step One, conduct an industry category review of the project.

[0016] Step 3: Using the information analyzed in Step 1, conduct an energy consumption review of the project's industrial added value.

[0017] Step Four: Using the information analyzed in Step One, conduct a feasibility study and review of the project's energy consumption indicators.

[0018] Step 5: Summarize the results of the project formal review.

[0019] Step Six: The project materials based on the summarized review results from Step Five are automatically filtered and distributed by the client.

[0020] Step Seven: After receiving the project information, the client organization conducts an internal review and uploads its review comments.

[0021] Step 8: For projects that passed the institutional review in Step 7, construct a large model context for the final review results.

[0022] Step Nine: Generate final review comments using the large model layout and prompts set in Step Eight.

[0023] Step one specifically includes the following:

[0024] First, the project format is determined by the suffix of the project materials submitted to the system. If the suffix is ​​doc, docx, or pdf, all project material information is directly parsed through code and persistently stored.

[0025] If the file extension is png / jpg or other image format, then OCR technology is used to extract all project material information and store it persistently.

[0026] Step two is detailed as follows:

[0027] 1) First, using the National Economic Classification, construct an industry category knowledge base. The specific results of this knowledge base are as follows:

[0028] 2) Process the structured material information extracted in step one to extract the project construction content.

[0029] 3) The extracted construction content is retrieved from the existing industry category knowledge base to obtain the predicted industry category information.

[0030] 4) Use the project construction content as input to the large model and set specific prompts to obtain industry category information.

[0031] 5) Compare whether the industry categories retrieved through the knowledge base are consistent with the extracted industry category information.

[0032] Step three is detailed as follows:

[0033] 1) Process the structured material information extracted in step one to extract energy efficiency level and energy consumption intensity.

[0034] 2) Input energy efficiency levels and energy intensity into a large model and set specific prompts to obtain energy consumption data for industrial added value.

[0035] 3) Input the project construction content extracted in step two into the large model and set specific prompts to obtain the city information where the project is located.

[0036] 4) Construct a knowledge base of the lowest energy consumption (equivalent value) per unit of industrial added value in prefecture-level cities.

[0037] 5) Compare the extracted energy consumption data of prefecture-level cities and industrial added value with the knowledge base of the lowest energy consumption of industrial added value of prefecture-level cities.

[0038] In step four, the energy consumption indicators of the structured project material information extracted in step one are reviewed and verified, as follows:

[0039] 1) Process the structured material information extracted in step one to extract the project's energy consumption information.

[0040] 2) Input energy consumption data into the large model and set specific prompts to obtain energy consumption data.

[0041] 3) Using the project materials collected in step one, label over 1000 stamp samples with LabelMe to pre-train the YOLOv8 algorithm, thus obtaining a stamp recognition engine.

[0042] 4) Using a seal recognition engine, extract the seals from the project materials, and use PaddleOCR to extract the issuing unit information from the seals.

[0043] 5) Determine the issuing authority based on the project's energy consumption;

[0044] 6) Compare the issuing unit obtained in the previous step with the issuing unit extracted in the fourth step to complete the verification and review of the implementation of energy consumption indicators.

[0045] In step six, a dedicated balancing algorithm was designed to automate and evenly distribute the materials to be reviewed to the relevant commissioning agencies. The specific calculation process is as follows:

[0046] 1) Symbol Definition: Let A be the set of all industry types, A = ,in This represents the j-th industry type, where n is the total number of industry types.

[0047] 2) For each industry type ,set up This refers to the collection of institutions corresponding to this industry. = ,in Indicates industry, The corresponding Kth institution, For the industry The corresponding number of institutions

[0048] 3) Let For the industry The selection counter is initialized to 0.

[0049] 4) Initialization: Obtain all industry types A and the set of institutions corresponding to each industry. and select counters for all industries Initialize to 0,

[0050] 5) Select Institution: When a given industry type is specified... When selecting the appropriate mechanism, follow the formula below: SA= Where mod is the modulo operation. The meaning of this formula is that, through industry... Selection counter Perform a modulo operation to obtain a set of mechanisms. The values ​​within the index range are used to select the corresponding institution, for example... =5, =3, then 5mod3=2, so choose As a selected institution.

[0051] 6) Update the counter: After selecting an institution, the selection counter for that industry needs to be updated: C is =C is +1.

[0052] In step seven, after selecting the commissioning unit using the balancing algorithm designed in step six, the commissioned unit reviews the application and uploads its review comments.

[0053] In step eight, the review opinions from the commissioning agency in step seven and the project formal review results in step five are summarized to construct a large model context for the final review results.

[0054] In step nine, the results summarized in step eight are used as the context of the large model, and specific prompts are set to generate the final review comments.

[0055] An automated project review system based on an AI-powered large-scale model, comprising the following modules:

[0056] The content parsing module's core function is to accurately extract the document content from the uploaded project materials by formulating corresponding rules based on the uploaded project material's format using a rule engine. The project format review module specifically includes sub-modules for industry category review, industrial added value energy consumption (equivalent value) review, and energy consumption indicator implementation demonstration review, each responsible for reviewing its respective indicator.

[0057] The project delegation module primarily utilizes a designed balancing algorithm to automate and evenly distribute materials awaiting review to relevant delegation agencies, ensuring both the rationality and efficiency of the allocation.

[0058] The third-party review module stores the review comments submitted by third-party organizations, providing data support for subsequent processes.

[0059] The key to the final review results module lies in using the content stored in the third-party review module and the content stored in the project form review module as the input context of the large model. By setting targeted prompts, the final review results of the project are automatically generated.

[0060] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for automatic project review based on an AI large model.

[0061] A computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the method for automatic project review based on an AI large model.

[0062] Compared to existing technologies, this invention has the following advantages: 1. It defines an automated project review process, including project formal review, automatic project delegation, third-party institution review, and final project review opinions. This standardized automated process system can quickly generate final project review opinions, improving overall review efficiency. 2. It defines a balanced algorithm to achieve automated and balanced distribution of materials to be reviewed to relevant delegation institutions. Using the labelme tool, over 1000 seal samples were manually labeled, and the YOLOv8 algorithm was pre-trained to obtain a seal recognition engine. 3. It constructs an industry classification knowledge base and a knowledge base of minimum energy consumption (equivalent value) for industrial added value in prefecture-level cities. 4. This invention uses the labelme tool to manually label over 1000 seal samples, pre-trains the YOLOv8 algorithm to obtain a seal recognition engine, extracts seal images from project materials of various formats, and uses PaddleOCR to extract the issuing unit information from the seal. This invention also constructs an industry classification knowledge base and a knowledge base of minimum energy consumption (equivalent value) for industrial added value in prefecture-level cities. This invention achieves automated and balanced distribution of materials to be reviewed to relevant commissioning agencies through a specially designed balancing algorithm. It automatically extracts industry classification and industrial added value energy consumption values ​​from project materials using a large model by setting specific prompts such as industry classification and industrial added value energy consumption. Furthermore, it automatically generates the final project review results by setting specific project formal review results and third-party review opinions as the context of the large model and setting specific prompts. Attached Figure Description

[0063] Figure 1 This is a schematic diagram of the overall system framework of the present invention. Detailed Implementation

[0064] To enhance understanding of the present invention, the embodiments will be described in detail below with reference to the accompanying drawings.

[0065] Example 1: A method for automatic project review based on AI large model, the method includes the following steps:

[0066] Step 1: Store and parse project materials in different formats, including Word, PDF, scanned documents (images), etc.

[0067] Step Two: Using the information analyzed in Step One, conduct an industry category review of the project.

[0068] Step 3: Using the information analyzed in Step 1, conduct an energy consumption review per unit of industrial added value for the project.

[0069] Step Four: Using the information analyzed in Step One, conduct a feasibility study and review of the project's energy consumption indicators.

[0070] Step 5: Summarize the results of the project formal review.

[0071] Step Six: The project materials based on the summarized review results from Step Five are automatically filtered and distributed by the client.

[0072] Step Seven: After receiving the project information, the client organization conducts an internal review and uploads its review comments.

[0073] Step 8: For projects that passed the institutional review in Step 7, construct a large model context for the final review results.

[0074] Step Nine: Generate final review comments using the large model layout and prompts set in Step Eight.

[0075] Step one specifically includes the following:

[0076] First, the project format is determined by the suffix of the project materials submitted to the system. If the suffix is ​​doc, docx, or pdf, all project material information is directly parsed through code and persistently stored.

[0077] If the file extension is png / jpg or other image format, then OCR technology is used to extract all project material information and store it persistently.

[0078] Step two is detailed as follows:

[0079] 1) First, the national economic classification was used to construct an industry category knowledge base. The specific knowledge base results are shown in Table 1.

[0080] 2) Process the structured material information extracted in step one to extract the project construction content.

[0081] 3) The extracted construction content is retrieved from the existing industry category knowledge base to obtain the predicted industry category information.

[0082] 4) Use the project construction content as input to the large model and set specific prompts to obtain industry category information.

[0083] 5) Compare whether the industry categories retrieved through the knowledge base are consistent with the extracted industry category information.

[0084] Table 1:

[0085]

[0086]

[0087]

[0088]

[0089] Step three is detailed as follows: 1. Process the structured material information extracted in step one to extract energy efficiency level and energy intensity. 2. By inputting energy efficiency levels and energy intensity into a large model and setting specific prompts, energy consumption data for industrial added value can be obtained.

[0090] 3. Input the project construction content extracted in step two into the large model and set specific prompts to obtain the city information where the project is located.

[0091] 4. Construct a knowledge base for the minimum energy consumption (equivalent value) of industrial added value in prefecture-level cities, as shown in Table 2.

[0092] Table 2:

[0093] area Energy consumption per ton of standard coal equivalent in industrial added value / 10,000 yuan (equivalent value) Jiangsu Province 0.37 Nanjing 0.72 Wuxi 0.57 Xuzhou 0.94 Changzhou 0.67 Suzhou 0.64 Nantong 0.58 Lianyungang 0.63 Huai'an 0.66 Yancheng 0.56 Yangzhou 0.51 Zhenjiang 0.83 Taizhou 0.69 Suqian 0.71

[0094] 5. Compare the extracted energy consumption data of prefecture-level cities and industrial added value with the knowledge base of the lowest energy consumption of industrial added value of prefecture-level cities.

[0095] In step four, the energy consumption indicators of the structured project material information extracted in step one are reviewed and verified, as follows:

[0096] 1) Process the structured material information extracted in step one to extract the project's energy consumption information.

[0097] 2) Input energy consumption data into the large model and set specific prompts to obtain energy consumption data.

[0098] 3) Using the project materials collected in step one, label over 1000 stamp samples with LabelMe to pre-train the YOLOv8 algorithm, thus obtaining a stamp recognition engine.

[0099] 4) Using a seal recognition engine, extract the seals from the project materials, and use PaddleOCR to extract the issuing unit information from the seals.

[0100] 5) Determine the issuing authority based on the project's energy consumption;

[0101] 6) Compare the issuing unit obtained in the previous step with the issuing unit extracted in the fourth step to complete the verification and review of the implementation of energy consumption indicators.

[0102] In step six, a dedicated balancing algorithm was designed to automate and evenly distribute the materials to be reviewed to the relevant commissioning agencies. The specific calculation process is as follows:

[0103] 1) Symbol Definition: Let A be the set of all industry types, A = ,in This represents the j-th industry type, where n is the total number of industry types.

[0104] 2) For each industry type ,set up This refers to the collection of institutions corresponding to this industry. = ,in Indicates industry, The corresponding Kth institution, For the industry The corresponding number of institutions

[0105] 3) Let For the industry The selection counter is initialized to 0.

[0106] 4) Initialization: Obtain all industry types A and the set of institutions corresponding to each industry. and select counters for all industries Initialize to 0,

[0107] 5) Select Institution: When a given industry type is specified... When selecting the appropriate mechanism, follow the formula below: SA= Where mod is the modulo operation. The meaning of this formula is that, through industry... Selection counter Perform a modulo operation to obtain a set of mechanisms. The values ​​within the index range are used to select the corresponding institution, for example... =5, =3, then 5mod3=2, so choose As a selected institution.

[0108] 6) Update the counter: After selecting an institution, the selection counter for that industry needs to be updated: C is =C is +1.

[0109] In step seven, after selecting the commissioning unit using the balancing algorithm designed in step six, the commissioned unit reviews the application and uploads its review comments.

[0110] In step eight, the review opinions from the commissioning agency in step seven and the project formal review results in step five are summarized to construct a large model context for the final review results.

[0111] In step nine, the results summarized in step eight are used as the context of the large model, and specific prompts are set to generate the final review comments.

[0112] Example 2: See Figure 1 An automated project review system based on an AI-powered large-scale model, comprising the following modules:

[0113] The content parsing module's core function is to accurately extract the file content from the uploaded project materials by using a rule engine to formulate corresponding rules based on the format of the uploaded project materials.

[0114] The project formal review module specifically includes sub-modules for industry category review, industrial added value energy consumption (equivalent value) review, and energy consumption indicator implementation demonstration review, each responsible for reviewing the corresponding indicators.

[0115] The project delegation module primarily utilizes a designed balancing algorithm to automate and evenly distribute materials awaiting review to relevant delegation agencies, ensuring both the rationality and efficiency of the allocation.

[0116] The third-party review module stores the review comments submitted by third-party organizations, providing data support for subsequent processes.

[0117] The key to the final review results module lies in using the content stored in the third-party review module and the content stored in the project form review module as the input context of the large model. By setting targeted prompts, the final review results of the project are automatically generated.

[0118] It should be noted that the above embodiments are not intended to limit the scope of protection of the present invention. Equivalent transformations or substitutions made based on the above technical solutions all fall within the scope of protection of the claims of the present invention.

Claims

1. A method for automatic project review based on AI large-scale models, characterized in that, The method includes the following steps: Step 1: Store and parse project materials in different formats, including Word, PDF, and scanned documents. Step Two: Using the information analyzed in Step One, conduct an industry category review of the project. Step 3: Using the information analyzed in Step 1, conduct an energy consumption review of the project's industrial added value. Step Four: Using the information analyzed in Step One, conduct a feasibility study and review of the project's energy consumption indicators. Step 5: Summarize the results of the project formal review. Step Six: The project materials based on the summarized review results from Step Five are automatically filtered and distributed by the client. Step Seven: After receiving the project information, the client organization conducts an internal review and uploads its review comments. Step 8: For projects that passed the institutional review in Step 7, construct a large model context for the final review results. Step Nine: Generate final review comments using the large model layout and prompts set in Step Eight.

2. The method for automatic project review based on an AI large model according to claim 1, characterized in that, Step one, specifically as follows: First, the project format is determined by the suffix of the project materials submitted to the system. If the suffix is ​​doc, docx, or pdf, all project material information is directly parsed through code and persistently stored. If the file extension is png / jpg or other image format, then OCR technology is used to extract all project material information and store it persistently.

3. The method for automatic project review based on an AI large model according to claim 1, characterized in that, Step two is detailed as follows: 1) First, using the National Economic Classification, construct an industry category knowledge base. The specific results of this knowledge base are as follows: 2) Process the structured material information extracted in step one to extract the project construction content. 3) The extracted construction content is retrieved from the existing industry category knowledge base to obtain the predicted industry category information. 4) Use the project construction content as input to the large model and set specific prompts to obtain industry category information. 5) Compare whether the industry categories retrieved through the knowledge base are consistent with the extracted industry category information.

4. The method for automatic project review based on an AI large model according to claim 1, characterized in that, Step three is detailed as follows: 1) Process the structured material information extracted in step one to extract energy efficiency level and energy consumption intensity. 2) Input energy efficiency levels and energy intensity into a large model and set specific prompts to obtain energy consumption data for industrial added value. 3) Input the project construction content extracted in step two into the large model and set specific prompts to obtain the city information where the project is located. 4) Construct a knowledge base of the lowest energy consumption (equivalent value) per unit of industrial added value in prefecture-level cities. 5) Compare the extracted energy consumption data of prefecture-level cities and industrial added value with the knowledge base of the lowest energy consumption of industrial added value of prefecture-level cities.

5. The method for automatic project review based on an AI large model according to claim 1, characterized in that, In step four, the energy consumption indicators of the structured project material information extracted in step one are reviewed and verified, as follows: 1) Process the structured material information extracted in step one to extract the project's energy consumption information. 2) Input energy consumption data into the large model and set specific prompts to obtain energy consumption data. 3) Using the project materials collected in step one, label over 1000 stamp samples with LabelMe to pre-train the YOLOv8 algorithm, thus obtaining a stamp recognition engine. 4) Using a seal recognition engine, extract the seals from the project materials, and use PaddleOCR to extract the issuing unit information from the seals. 5) The issuing authority is determined based on the project's energy consumption. 6) Compare the issuing unit obtained in the previous step with the issuing unit extracted in the fourth step to complete the verification and review of the implementation of energy consumption indicators.

6. The method for automatic project review based on an AI large model according to claim 1, characterized in that, In step six, a dedicated balancing algorithm was designed to automate and evenly distribute the materials to be reviewed to the relevant commissioning agencies. The specific calculation process is as follows: 1) Symbol Definition: Let A be the set of all industry types, A = {a1, a2, ..., a...} n }, where a j This represents the j-th industry type, where n is the total number of industry types. 2) For each industry type a j Let A ∈ A. ij This refers to the collection of institutions corresponding to this industry. = ,in Indicates industry, a j The corresponding Kth institution, m j For industry a j The corresponding number of institutions 3) Let C ij For industry a j The selection counter is initialized to 0. 4) Initialization: Obtain all industry types A and the set of institutions A corresponding to each industry. ij And select counter C for all industries ij Initialize to 0, 5) Select Institution: When an industry type a is given s When selecting the appropriate mechanism, follow the formula below: SA= Where mod stands for modulo operation, the meaning of this formula is that by taking industry a... s Select Counter C is Perform a modulo operation to obtain a set of mechanisms. 6) Update the counter: After selecting the institution, the selection counter for that industry needs to be updated: C is =C is +1.

7. The method for automatic project review based on an AI large model according to claim 6, characterized in that, In step seven, after selecting the commissioning unit using the balancing algorithm designed in step six, the commissioned unit reviews the application and uploads its review comments. In step eight, the review comments from the commissioning agency in step seven and the project formal review results from step five are summarized to construct a large-scale model context for the final review results. In step nine, the results summarized in step eight are used as the context of the large model, and specific prompts are set to generate the final review comments.

8. An automated project review system based on an AI large-scale model, characterized in that, The system includes the following modules: The content parsing module's core function is to accurately extract the file content from the uploaded project materials by using a rule engine to formulate corresponding rules based on the format of the uploaded project materials. The project formal review module specifically includes sub-modules for industry category review, industrial added value energy consumption (equivalent value) review, and energy consumption indicator implementation demonstration review, each responsible for reviewing the corresponding indicators. The project delegation module primarily utilizes a designed balancing algorithm to automate and evenly distribute materials awaiting review to relevant delegation agencies, ensuring both the rationality and efficiency of the allocation. The third-party review module stores the review comments submitted by third-party organizations, providing data support for subsequent processes. The key to the final review results module lies in using the content stored in the third-party review module and the content stored in the project form review module as the input context of the large model. By setting targeted prompts, the final review results of the project are automatically generated.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a method for automatic project review based on an AI large model as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instruction is executed by the processor, it implements a method for automatic project review based on an AI large model as described in any one of claims 1-7.

Images