Enterprise project intelligent declaration method based on deep learning large model

Abstract

The invention relates to the field of deep learning large models, in particular to an enterprise project intelligent declaration method based on a deep learning large model. According to the method, through a data labeling technology, an interpretation algorithm, an intelligent recommendation model, a process monitoring mechanism, a material pre-auditing system and intelligent project declaration, comprehensive automation and intelligentization of a project declaration process are realized. The declaration efficiency and accuracy can be improved, and personalized and precise declaration service can be provided for the user.

Description

Technical Field

[0001] The present invention relates to the field of deep learning big models, and in particular to an enterprise project intelligent declaration method based on deep learning big models. Background Art

[0002] As global scientific and technological competition becomes increasingly fierce, enterprises' demand for technological innovation and R&D investment continues to increase. As an important means for enterprises to obtain government support and promote technological innovation, project application is becoming increasingly important. Project application not only involves the application of funds and the use of documents, but also relates to the strategic development and market competitiveness of enterprises. Traditional project application methods mainly rely on manual operations, including the collection of project information, interpretation of documents, preparation and review of materials. These links are not only time-consuming and labor-intensive, but also easily interfered by human factors, resulting in low application efficiency and high error rate, and sometimes even missed application opportunities due to information asymmetry.

[0003] In order to solve these problems, some automated tools and information systems have been introduced into the project application process. These tools and systems can improve the efficiency of application to a certain extent, but due to the lack of intelligent processing capabilities, they can only handle some simple and standardized tasks, and their adaptability and decision-making support capabilities for complex situations are limited.

[0004] In recent years, with the rise of big data and artificial intelligence technologies, people have begun to explore the application of these advanced technologies in the field of project application. Big data analysis can process and analyze massive, multi-source heterogeneous data, while artificial intelligence technology can simulate the human decision-making process and provide intelligent recommendations and services. The combination of these technologies provides new possibilities for the automation and intelligence of the project application process.

[0005] Nevertheless, the existing project application methods based on big data and artificial intelligence still have some problems and challenges. For example, how to effectively integrate and utilize multi-dimensional enterprise information, how to accurately interpret and match project documents, how to achieve automated pre-examination of application materials, and how to intelligently fill out forms. The existence of these problems limits the improvement of the intelligent level of project application, and also affects the success rate of application and the satisfaction of enterprise users. Summary of the invention

[0006] The present invention discloses a method for intelligent application of enterprise projects based on a deep learning big model. The method realizes the comprehensive automation and intelligence of the project application process through innovative data annotation technology, file interpretation algorithm, intelligent recommendation model, process monitoring mechanism, material pre-examination system and project intelligent application. It can not only improve the application efficiency and accuracy, but also provide users with personalized and precise application services.

[0007] The following technical solution is used to solve the technical problem: an enterprise project intelligent declaration method based on a deep learning large model, the method comprising the following steps: S1. Use data annotation technology to annotate enterprise information in multiple dimensions, use large models to extract key elements of documents, and achieve intelligent project recommendations through feature vector fusion; S2. Build a project application process model and intelligent reminder system to ensure that enterprises can respond to the application process in a timely manner and avoid delays; S3. Set up an intelligent pre-examination mechanism for application materials, automatically analyze documents using document quality assessment, and provide optimization suggestions through review problem prompts; S4. Intelligent project declaration: Based on OCR document recognition and matching templates, intelligent form filling can be realized and project declaration content can be automatically submitted.

[0008] Multi-dimensional data labeling uses the Doccano data labeling tool to label enterprise data based on dimensions such as enterprise size, industry field, region, and business scope.

[0009] The feature vector fusion described is to fuse the key application elements extracted from the project document with the enterprise's own annotation information after feature vector quantization, use the machine learning model random forest for intelligent recommendation, and output the project application document that best meets the enterprise's conditions.

[0010] The intelligent reminder system identifies key node information through a large model, automatically generates notification content through a preset rule engine, and automatically pushes reminders through multiple channels such as emails, text messages, robots, etc.

[0011] The document quality assessment uses NLP technology to perform syntax verification, logical consistency check and key information integrity analysis on the uploaded documents.

[0012] The audit problem prompts described above use the deep learning large model BERT for intelligent prompts to guide companies to make modifications and improvements. The prompts include missing materials, format errors, and untrue content.

[0013] The intelligent form filler uses the OCR recognition results to locate the position of the form by template matching, performs data processing on the extracted content, including spaces, special symbols, etc., and then matches the fields with the form in the database, and retrieves and queries the corresponding data from the database according to the matching results. Use file resources to improve the success of project application.

[0014] The beneficial effects of the present invention are: a method for intelligent application for enterprise projects based on a deep learning big model, which simplifies the enterprise project application process through intelligent recommendation and process assistance, material pre-examination, intelligent filling and other functions, and realizes the intelligence and customization of the application process through the deep integration of deep learning and artificial intelligence technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of the declaration method; Figure 2 This is a schematic diagram of the detailed process of project application recommendation in step S1 of the embodiment; Figure 3 This is a schematic diagram of the detailed process of material pre-examination in step S3 of the embodiment. DETAILED DESCRIPTION

[0016] A method for intelligently declaring enterprise projects based on a deep learning big model, comprising the following steps: S1. Use data annotation technology to annotate enterprise information in multiple dimensions, use large models to extract key elements of documents, and achieve intelligent project recommendations through feature vector fusion; S2. Build a project application process model and intelligent reminder system to ensure that enterprises can respond to the application process in a timely manner and avoid delays; S3. Set up an intelligent pre-examination mechanism for application materials, automatically analyze documents using document quality assessment, and provide optimization suggestions through review problem prompts; S4. Intelligent project declaration: Based on OCR document recognition and matching templates, intelligent form filling can be realized and project declaration content can be automatically submitted.

[0017] Multi-dimensional data labeling uses the Doccano data labeling tool to label enterprise data based on dimensions such as enterprise size, industry field, region, and business scope.

[0018] The feature vector fusion described is to fuse the key application elements extracted from the project document with the enterprise's own annotation information after feature vector quantization, use the machine learning model random forest for intelligent recommendation, and output the project application document that best meets the enterprise's conditions.

[0019] The intelligent reminder system identifies key node information through a large model, automatically generates notification content through a preset rule engine, and automatically pushes reminders through multiple channels such as emails, text messages, robots, etc.

[0020] The document quality assessment uses NLP technology to perform syntax verification, logical consistency check and key information integrity analysis on the uploaded documents.

[0021] The audit problem prompts are modified and improved through intelligent prompts using the deep learning large model BERT, and the prompts include missing materials, incorrect format, and untrue content.

[0022] The intelligent form fills in the results recognized by OCR, locates the position of the form by template matching, performs data processing on the extracted content, including spaces, special symbols, etc., and then matches the fields with the form in the database, and retrieves and queries the corresponding data from the database according to the matching results. Specific embodiments

[0023] like Figures 1 to 3 As shown, according to the first aspect of the present disclosure, the present disclosure provides an enterprise project intelligent declaration method based on a deep learning big model, comprising the following steps: A method for intelligently declaring enterprise projects based on a deep learning big model, comprising the following steps: S1. Multi-dimensional data annotation of enterprise information is carried out to realize intelligent project application recommendation.

[0024] The deep learning model deeply interprets the content of the project application documents, the key elements of the enterprise project application, and integrates the key element information with the enterprise annotation information. The multi-dimensional data labeling of the enterprise information is performed by using the Doccano data labeling tool to label the enterprise data according to dimensions such as enterprise size, industry field, location, and business scope.

[0025] The key point refers to using large-model natural language processing technology to interpret document content, automatically identify and accurately extract key elements of enterprise project applications from labeled data.

[0026] Furthermore, the key application elements extracted from the project documents are fused with the company's own annotation information after feature vector quantization, and the machine learning model random forest is used for intelligent recommendation to output the project application documents that best meet the company's conditions.

[0027] The feature extraction and vectorization are integrated as follows: The key elements extracted from the file form a vector ,in Representative Numerical representation of the key elements.

[0028] The enterprise's own labeling information constitutes a vector ,in The corresponding enterprise’s own labeled characteristic values.

[0029] Combine these two vectors to form the feature vector , that is, a sample point in the feature space The random forest model intelligent recommendation trains the model through the random forest algorithm, automatically performs feature selection and model construction, obtains the predicted score of the matching degree or applicability of each file to the enterprise, sorts all the files according to the predicted score, and selects the files with the highest scores as the recommendation results.

[0030] In this implementation, the random forest model intelligently recommends files, and the specific implementation method is as follows: S11. Filter the features in the file that have a greater impact on the enterprise, such as enterprise scale and industry, and perform feature vectorization on the enterprise data and file data to form fused feature vector data; S12. Use the random forest algorithm to train the model, divide the data into training data (75%) and test data (25%), set the parameters of the random forest to train the model, including the number of trees and the maximum depth, and evaluate the performance of the model through indicators such as accuracy and recall rate; S13, using the trained random forest model to predict the matching score of each file; S14. Sort all the documents and select the documents with the highest scores as the recommended results of project application documents that best meet the enterprise's conditions.

[0031] Specific examples are as follows: File data: File A: Type = Technological innovation, Funding scale = 1 million, Application qualification = Small and medium-sized enterprises; File B: Type = Environmental protection, Funding scale = 500,000, Application qualification = All enterprises.

[0032] Company data: Company X: Size = Small, Industry Type = IT, Years Established = 5 years.

[0033] After vectorizing the two feature data, the following vectors can be obtained: File A: [1, 0, 1 million, 1] (technological innovation = 1, environmental protection = 0, capital scale, small and medium-sized enterprises = 1) File B: [0, 1, 500,000, 0] (Technological innovation = 0, Environmental protection = 1, Funding scale, Small and medium-sized enterprises = 0) Company X: [1, 0, 5] (IT=1, Manufacturing=0, Years of Establishment) The fused feature vector: File A and Company X: [1, 0, 1 million, 1, 1, 0, 5] File B and Enterprise X: [0, 1, 500,000, 0, 1, 0, 5] After training the random forest model, the predicted output is that the matching degree of file A to the enterprise is 0.8, and the matching degree of file B to the enterprise is 0.75, so file A is more in line with the conditions of enterprise X.

[0034] S2. Assist in the key declaration process, using deep learning big models to monitor each node of the declaration process and realize intelligent management of process nodes.

[0035] Furthermore, in step S2, the big model automatically sends process node notifications, and identifies key nodes, such as the deadline for material submission, the review announcement time, etc. When the big model identifies key nodes, the notification content is automatically generated through the preset rule engine, and reminders are automatically pushed through multiple channels such as emails, SMS, robots, etc., to ensure the smooth progress of the application process.

[0036] S3. Pre-examine the application materials, conduct document quality assessment, check whether the materials are complete and the format is correct, etc., feedback problem prompts through large model reasoning, promptly discover and correct problems in the materials, rectify them according to the problem prompts, and resubmit for pre-examination after rectification.

[0037] The document quality assessment uses NLP technology to perform syntax verification, logical consistency check and key information integrity analysis on the uploaded documents, and provide instant feedback and modification suggestions.

[0038] The document quality score adopts a weighted comprehensive scoring method, combining multiple evaluation indicators (such as the number of grammatical errors, logical consistency score, and key information completeness score).

[0039] The weighted comprehensive scoring method is calculated as follows: in, Indicates the syntax, Represents logic, Indicates key information. is the weight of each indicator.

[0040] For example, if the full score for each item in the document submitted for review is 100 points and 10 grammatical errors are found, the grammar score is approximately 90 points; the internal logic of the document is checked using NLP technology, and after evaluation, it is scored 85 points; check whether the report contains all the necessary information, such as research objectives, methods, data analysis, conclusions, etc., and the score is assumed to be 95 points.

[0041] The weights are set as follows: , , Calculating Quality Score Substitute the above scores and weights into the formula to calculate: Quality Score=0.3×90+0.4×85+0.3×95=27+34+28.5=89.5Quality Score=0.3×90+0.4×85+0.3×95=27+34+28.5=89.5 Therefore, the overall quality score of the submitted document is 89.5 points. If the threshold is 60 points, the document can be used as a submission material, but the issues raised in the pre-examination must be revised.

[0042] The audit problem prompts are to find problems in the pre-examination, use the deep learning big model to provide intelligent prompts, and guide enterprises to make modifications and improvements. These prompts include specific suggestions on missing materials, format errors, untrue content, etc.

[0043] The deep learning model provides intelligent prompts, and uses tools such as Jieba to clean and segment text. The pre-trained deep learning model BERT extracts the semantic features of the text, and further trains the model to identify whether the material is complete, the format is correct, and the content is true. Specific guidance prompts are generated according to the question type.

[0044] The generation of specific guidance prompts mainly includes the following processes: By converting the text into an integer sequence, using the Transformer structure, for a given input sequence, the encoder converts it into a context vector, and the decoder begins to generate an output sequence, starting with a special start token, predicting the probability distribution of the next word at each time step, and selecting the word with the highest probability as the output; Convert the generated integer sequence back to text and provide output through the API interface.

[0045] S4, intelligent project declaration, based on the preliminary review materials in step S3, the key fields of the form are identified through OCR for the data that needs to be filled in, and the field data in the database is queried through the interface to fill in the form intelligently. The filled-in information is manually checked. After verification, the project declaration content can be automatically submitted through the program.

[0046] The smart form filling is specifically implemented as follows: S41, the result recognized by OCR; S42, locating the position of the form by using template matching; S43, performing data processing on the extracted content, including spaces, special symbols, etc.; S44, using the interface method to match the fields with the form in the database, and retrieving and querying the corresponding data from the database according to the matching results.

[0047] Through the above specific implementation methods, the present invention realizes an intelligent enterprise project declaration method based on a deep learning big model, which integrates enterprise multi-dimensional data processing, deep file analysis, intelligent strategy recommendation, process automation monitoring, document quality intelligent pre-examination, and automatic form filling and other functions, fundamentally improving the efficiency and accuracy of declaration, ensuring that enterprises can make full use of file resources, accurately formulate declaration strategies, and effectively promote the success rate of projects.

[0048] Glossary Doccano Tool: An open source text annotation tool widely used in natural language processing (NLP) projects. It supports a variety of annotation needs including text classification, sequence annotation, and sequence-to-sequence tasks, and is a key step in data preprocessing to train machine learning or deep learning models.

[0049] NLP technology: the full name is Natural Language Processing, which is a branch of artificial intelligence (AI) that aims to enable computers to understand, parse, generate and manipulate natural language (the language used by humans in daily life). NLP combines knowledge from multiple fields such as computer science, artificial intelligence, linguistics and cognitive science, with the goal of narrowing the gap between human language and the language that computers understand and generate.

[0050] OCR recognition: Optical Character Recognition (OCR) is a technology designed to convert text content in an image into an editable and searchable digital text format. This technology uses image processing, pattern recognition, and machine learning algorithms to recognize and extract printed or handwritten text so that it can be processed in a computer without human input.

Claims

1. A method for intelligent enterprise project declaration based on deep learning big model, characterized by The following steps are involved: S1. Use data annotation technology to annotate enterprise information in multiple dimensions, use large models to extract key elements of documents, and achieve intelligent project recommendations through feature vector fusion; S2. Build a project application process model and intelligent reminder system to ensure that enterprises can respond to the application process in a timely manner and avoid delays; S3. Set up an intelligent pre-examination mechanism for application materials, automatically analyze documents using document quality assessment, and provide optimization suggestions through review problem prompts; S4. Intelligent project declaration: Based on OCR document recognition and matching templates, intelligent form filling can be realized and project declaration content can be automatically submitted.

2. The enterprise project intelligent declaration method based on deep learning big model according to claim 1 is characterized in that: Multi-dimensional data labeling uses the Doccano data labeling tool to label enterprise data based on dimensions such as enterprise size, industry field, region, and business scope.

3. The enterprise project intelligent declaration method based on deep learning big model according to claim 1 is characterized in that: The feature vector fusion described is to fuse the key application elements extracted from the project document with the enterprise's own annotation information after feature vector quantization, use the machine learning model random forest for intelligent recommendation, and output the project application document that best meets the enterprise's conditions.

4. The enterprise project intelligent declaration method based on deep learning big model according to claim 1 is characterized in that: The intelligent reminder system identifies key node information through a large model, automatically generates notification content through a preset rule engine, and automatically pushes reminders through multiple channels such as emails, text messages, robots, etc.

5. The enterprise project intelligent declaration method based on deep learning big model according to claim 1 is characterized in that: The document quality assessment uses NLP technology to perform syntax verification, logical consistency check and key information integrity analysis on the uploaded documents.

6. The enterprise project intelligent declaration method based on deep learning big model according to claim 1 is characterized in that: The audit problem prompts are modified and improved through intelligent prompts using the deep learning large model BERT, and the prompts include missing materials, format errors, and untrue content.

7. The enterprise project intelligent declaration method based on deep learning big model according to claim 1 is characterized in that: The intelligent form fills in the results recognized by OCR, locates the position of the form by template matching, performs data processing on the extracted content, including spaces, special symbols, etc., and then matches the fields with the form in the database, and retrieves and queries the corresponding data from the database according to the matching results.

Images