Enterprise transformation diagnosis report generation method based on large model

Abstract

The invention relates to the technical field of natural language processing, in particular to an enterprise transformation diagnosis report generation method based on a large model. Collecting corpus data by using an enterprise diagnosis questionnaire; constructing a digital transformation diagnosis knowledge base, and storing the knowledge base into a vector database; training a sample model, and constructing a diagnosis report template through fine tuning of Prompt cue words; and taking the fine-tuned cue word as input, automatically generating a diagnosis report by utilizing the trained large model, selecting different file formats by utilizing an agent, generating a diagnosis report and outputting the diagnosis report. Through questionnaires and industry characteristic analysis, diagnosis suggestions are intelligently generated, enterprise transformation pain points are identified, and the efficiency and accuracy of enterprise transformation diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a method for generating an enterprise transformation diagnosis report based on a large model. Background Art

[0002] With the development of the global economy, enterprises are facing increasingly fierce market competition and a constantly changing market environment. In order to adapt to this environment, enterprises need to continuously transform and upgrade. In the process of enterprise transformation, comprehensive diagnosis and analysis are required to determine the direction and strategy of transformation. Traditional enterprise transformation diagnosis methods mainly rely on manual analysis and empirical judgment, which are highly subjective and uncertain, and inefficient. Therefore, a method for generating enterprise transformation diagnosis reports based on large models is needed to improve the accuracy and efficiency of diagnosis.

[0003] Traditional diagnostic reports and diagnostic methods lack personalization. Different enterprises have different business models, market positioning and cultural backgrounds, so personalized transformation strategies are needed. However, existing diagnostic methods often use standardized templates, which are difficult to meet the personalized needs of enterprises. The enterprise transformation diagnostic report generation method based on big models is to use big model technology to conduct comprehensive, rapid and accurate diagnosis and analysis of enterprises, generate diagnostic reports, and provide decision support for enterprise transformation. Compared with traditional enterprise transformation diagnostic methods, the enterprise transformation diagnostic report generation based on big models has strong feature extraction and pattern recognition capabilities, which can more accurately identify the advantages and disadvantages of enterprises and provide more targeted suggestions for enterprise transformation. The enterprise transformation diagnostic report generation method based on big models mainly relies on data and models, reduces manual intervention and subjective judgment, and improves the objectivity of diagnosis. Big model technology can continuously learn and optimize. With the accumulation of data and technology, its diagnostic ability will continue to improve. The enterprise transformation diagnostic report generation method based on big models is a technology with broad application prospects, which can help enterprises quickly and accurately diagnose and analyze transformation and improve their competitiveness and sustainable development capabilities. Summary of the invention

[0004] The present invention discloses a method for generating enterprise transformation diagnostic reports based on a large model, which aims to use advanced intelligent computing technology and large model text parsing capabilities to provide enterprises with accurate and comprehensive digital transformation assessment and strategic guidance. The method comprehensively uses questionnaires, diagnostic indicator scoring algorithms and diagnostic level research and judgment, and applies industry large models, prompt words, agent intelligent bodies and other technologies to automatically generate diagnostic reports and help enterprises make digital transformation decisions.

[0005] The present invention solves the deficiencies of the prior art and provides a method for generating an enterprise transformation diagnosis report based on a large model, comprising the following steps: S1. Collect corpus data using the enterprise diagnostic questionnaire; S2. Build a digital transformation diagnosis knowledge base and store it in the vector database; S3, sample model training, constructing a diagnosis report template through fine-tuning of prompt words; S4. Use the fine-tuned prompt words as input, use the trained large model to automatically generate a diagnosis report, use the agent to select different file formats, and generate a diagnosis report output.

[0006] The digital transformation diagnosis knowledge base constructed in S2 is stored in the vector database, which specifically includes the following contents: S21. Use the tool OpenRefine to clean, sample, and deduplicate the data and convert it into a format that the model can understand. The conversion methods include: text encoding, standardizing numerical features, and handling missing values. S22, build a knowledge base for the cleaned and standardized data, and segment the text; load and segment a large number of basic documents constituting the knowledge base into text blocks for subsequent indexing; S23, enterprise knowledge base embedding index generation, after dividing the knowledge fragments, using the embedding model to convert them into high-dimensional embedding vectors that can be used for semantic search, and the vectors are associated with the text fragments and stored in the vector database; S24. Vectorize the query question and use the retriever to sort and retrieve the knowledge fragments that best match the question.

[0007] The sample model training in step S3, the Prompt word fine-tuning, and the construction of the diagnosis report template are specifically implemented as follows: S31. Use the constructed digital transformation knowledge base to train sample models based on the ChatGLM3-6B large model; S32, performing sample learning on the sample metadata by retrieving similar instances in the vector database; S33, constructing a prompt word template for enterprise diagnosis report, using P-Tuning v2 to form a continuous vector of the prompt word template and search content as input, forming a new output after model training, and outputting prompt words that meet the diagnosis template after repeated iterations; The tips for fine-tuning P-Tuning v2 are as follows: Suppose the pre-trained model is M, a function that accepts input text And output text , optimize the prompt words and introduce the prompt vector , where each They are all a set of continuous prompt vectors; Model input: original Combined with the prompt vector P to form a new input text ,Right now: Model output: Model M uses the modified input Produces output text : During training, the prompt vector P is updated, the parameters of the pre-trained model M remain unchanged, and the gradient descent algorithm is used to update P: - ; in, represents the prompt vector at iteration t, Table is the updated vector, is the learning rate, is the gradient of the loss function with respect to P.

[0008] The automatic writing of the diagnostic report in step S4 takes the fine-tuned prompt words as input, uses the trained large model to automatically generate the diagnostic report, and uses the Agent intelligent body to select different file formats for output. The specific steps are as follows: S41, using Agent to match templates of different industries, using the constructed diagnostic report prompt words, conversation history records, and search results as inputs for the large model; The Agent matches different industry models by predefining the industry type and presentation form, and using a matching tool to select the Prompt words that meet the requirements after training as input; S42. Use the trained industry big model to generate reports in various formats and text descriptions of enterprise digital transformation diagnosis calls, write them into files, and form a diagnosis report; S43, using Agent to select different file formats, providing Word, PDF, Markdown and other file types for diagnosis report viewing and downloading; S44. Label the data of industry solutions and SaaS applications in the database, analyze the industry classification and digital diagnosis and evaluation level through industry big model reasoning, match the labeled data, and screen out solutions and SaaS applications that meet the characteristics of the industry.

[0009] The beneficial effects of the present invention are: A method for generating enterprise transformation diagnosis reports based on a large model. Through questionnaires, industry characteristics analysis, and intelligent generation of diagnostic suggestions, it identifies the pain points of enterprise transformation, improves the efficiency and accuracy of enterprise transformation diagnosis, and can provide personalized and customized transformation suggestions and recommend SaaS applications that meet the current status of the enterprise. Through prompt word engineering and agent intelligent technology, the generation process of the diagnostic report is simplified and the user experience is improved. It helps enterprises better understand and grasp industry development trends and formulate effective transformation strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the report generation process of the present invention. DETAILED DESCRIPTION

[0012] The following is a detailed description of a method for generating an enterprise transformation diagnosis report based on a large model, with reference to the accompanying drawings and implementation cases, including the following steps: S1. Collect corpus data, use the dynamic configuration of the enterprise diagnostic questionnaire, and obtain corpus data through data collection, score calculation, and grade judgment; S2. Build a digital transformation diagnosis knowledge base and store it in the vector database; S3: Small sample model training, prompt word fine-tuning, and construction of diagnosis report template; S4, automatic writing of diagnostic reports, using the fine-tuned prompt word engineering as input, using the trained large model to automatically generate diagnostic reports, using the agent to select different file formats, and generate diagnostic report output; Furthermore, in step S1, questionnaire collection, diagnosis score calculation, and diagnosis level determination specifically include the following steps: S11. Construct a dynamic enterprise maturity questionnaire for manufacturing and service industries based on the four dimensions of enterprise digital foundation, operation, management, and effectiveness, collect classified data, and store the scores and answers in a structured database as JSON strings; The dynamic diagnostic maturity questionnaire is flexibly configured through low-code tools according to the different indicators collected by each industry; S12. According to the collected diagnostic questionnaire data, set the quantifiable indicator scores and weights for the indicator questions under each dimension, build a maturity score algorithm, and obtain the enterprise diagnostic score; The existing enterprise digital diagnosis and evaluation scoring algorithm is as follows: Indicator scoring: For each specific indicator i, let its score be ,generally , indicating a rating range from low to high.

[0013] Dimension weights: The weights of the four dimensions of digital foundation, operation, management, and effectiveness are , and full , the weight reflects the importance of each dimension to the digital transformation of the enterprise.

[0014] Dimension score calculation: The score of each dimension is the weighted average of all its subordinate indicator scores multiplied by the corresponding weight. For example, the score of the digital foundation dimension is: in is the relative weight of indicator i in the digital capability dimension, to ensure normalization.

[0015] The total diagnostic score TotalScore is the sum of the scores of each dimension multiplied by their weights: in, They represent the weights of the four dimensions of digital foundation, operation, management, and effectiveness. It represents the weighted average score of the four dimensions of digital foundation, operation, management, and effectiveness; The technical effect of the above solution is that the collection of enterprise questionnaires based on multi-dimensional and dynamically configured can formulate more professional and industry-specific diagnostic needs, use dimension and indicator combination algorithms to improve the objectivity and accuracy of scores, judge the diagnostic level, clarify the digitalization gap of the enterprise, make the corpus data more in line with the current situation of the enterprise, and generate more accurate diagnostic reports.

[0016] S13. According to the enterprise diagnosis score and the diagnostic assessment level range, match the enterprise diagnosis level. For example, if the diagnosis score of enterprise A is 36 points, the enterprise maturity level corresponds to level 1, and the diagnosis result is that the enterprise lacks digital transformation methods and tools.

[0017] S14, storing the obtained enterprise questionnaire information, enterprise information, industry, diagnosis score, diagnosis level and other structured data into a database; Furthermore, the step S2 of constructing a digital transformation diagnosis knowledge base and storing it in a vector database is specifically implemented as follows: S21. Use the tool OpenRefine to clean, sample, and deduplicate the enterprise diagnostics and assessment surveys, industry solutions, and one-company-one-file data, and convert them into a format that the model can understand. The conversion methods include: text encoding, standardizing numerical features, and processing missing values; S22. Build a knowledge base for the cleaned and standardized data and segment the text. Load and segment the large number of basic documents that constitute the knowledge base into text blocks of appropriate size for subsequent indexing; S23, enterprise knowledge base embedding index generation. After dividing the knowledge fragments, the embedding model is used to convert them into high-dimensional embedding vectors that can be used for semantic search, and the vectors are mapped and associated with the text fragments and stored in the vector database.

[0018] S24, vectorize the query question, and use the retriever to sort and retrieve the knowledge fragments that best match the question; Furthermore, the small sample model training in step S3, the Prompt word fine-tuning, and the construction of the diagnosis report template are specifically implemented as follows: S31. Use the constructed digital transformation knowledge base to train a few-sample model based on the ChatGLM3-6B large model; S32, performing few-sample learning on sample metadata by retrieving similar instances in a vector database; The retrieval vector database sample similar instances are measured using cosine similarity. For each vector in the database , calculate its cosine similarity with the query vector q: Sort all calculated similarities and find the k cosine similarity values ​​closest to 1; Select the vectors corresponding to the k largest cosine similarities from the vector database The data is used as a training set for model training; For example: There are three industry vectors in the vector database Company A: [0.6, 0.3, 0.1] Company B: [0.2, 0.7, 0.1] It may represent the weights of different characteristics of the industry to which the enterprise belongs, such as the proportion of technology, manufacturing and services.

[0019] If you want to query companies in the technology industry, create a query vector Q: [0.8, 0.1, 0.1] Calculated by cosine similarity: The similarity of Enterprise A is approximately 0.94, and the similarity of Enterprise B is approximately 0.45. According to cosine similarity, among the top K technology-related enterprises, Enterprise A is the first choice.

[0020] S33, constructing a prompt word template for enterprise diagnosis report, using P-Tuning v2 to form a continuous vector of the prompt word template and search content as input, forming a new output after model training, and outputting prompt words that meet the diagnosis template after repeated iterations; The enterprise diagnosis report template mainly consists of a structure, examples, and outputs, wherein the structure consists of role objectives and specific task requirements; the variable data in the task requirements are mainly formed by enterprise information, diagnosis scores, diagnosis levels, enterprise status and industry characteristics, transformation suggestions and other enterprise diagnosis contents; The tips for fine-tuning P-Tuning v2 are as follows: Let the pre-trained model be M, a function that accepts And output In order to optimize the prompt words, the prompt vector is introduced , where each They are all a set of continuous prompt vectors; Model input: original Combined with the prompt vector P to form a new input ,Right now: Model output: Model M uses the modified input Producing Output : During training, only the prompt vector P is updated, the parameters of the pre-trained model M remain unchanged, and the gradient descent algorithm is used to update P: - in, represents the prompt vector at iteration t, Table is the updated vector, is the learning rate, is the gradient of the loss function with respect to P.

[0021] For example, if the input is "[CLS]", the current status of the enterprise is [MASK]. Among them, MASk is the token to be predicted; Add a prompt vector P to represent the industry to which the enterprise belongs. After training, the output forms a new input prompt word: "[CLS]" The current status of the industry described by the enterprise is [MASK]; Furthermore, the automatic writing of the diagnostic report in step S4 takes the fine-tuned prompt word project as input, uses the trained large model to automatically generate the diagnostic report, and uses the Agent intelligent body to select different file formats for output. The specific steps are as follows: S41. Use Agent to match templates of different industries, and use the constructed diagnostic report prompt words, conversation history records, and search results as inputs for the large model.

[0022] The Agent matches different industry models by predefining the industry type and presentation form, and using a matching tool to select the Prompt words that meet the requirements after training as input; S42. Use the trained industry big model to generate reports in various formats and text descriptions of enterprise digital transformation diagnosis calls, write them into files, and form a diagnosis report; S43, using Agent to select different file formats, providing Word, PDF, Markdown and other file types for diagnosis report viewing and downloading; The Agent selects different file formats, mainly including: Develop file download plug-ins, including Word, PDF, Markdown and image formats. The plug-ins contain file format prompt information; The Prompt project contains key information for selecting the file type. The agent intelligently matches the prompt containing the key information with the developed file format plug-in. After generating the required files, the system will provide the document's download name and URL link for users to download and review. S44. Label the data of industry solutions and SaaS applications in the database, analyze the industry classification and digital diagnosis and evaluation level through industry big model reasoning, match the labeled data, and screen out solutions and SaaS applications that meet the characteristics of the industry.

[0023] Through the above-mentioned specific implementation methods, the present invention realizes an end-to-end enterprise digital transformation diagnosis and strategy generation system, collects and analyzes enterprise data in multiple dimensions, accurately evaluates its digitalization level, deeply analyzes the internal and external conditions and industry characteristics of the enterprise, and finally generates a customized transformation recommendation report, which greatly enhances the pertinence and timeliness of enterprise transformation and is a powerful tool to promote enterprise digital transformation.

[0024] Summary of the invention Technical points: A method for generating an enterprise transformation diagnosis report based on a large model, the method comprising the following steps: S1. Collect corpus data, use the dynamic configuration of the enterprise diagnostic questionnaire, and obtain corpus data through data collection, score calculation, and grade judgment; S2. Build a digital transformation diagnosis knowledge base and store it in a vector database; S3. Train a small sample model, fine-tune the prompt words, and build a diagnosis report template; S4, automatic writing of diagnostic reports, using the fine-tuned prompt word engineering as input, using the trained large model to automatically generate diagnostic reports, using Agent intelligence to select different file formats, and generate diagnostic report output; The digital maturity assessment model is constructed from four dimensions: digital foundation, operation, management, and effectiveness. Specific questions and corresponding quantifiable or evaluable indicators are set under each dimension. Each indicator is set with a corresponding weight. The collected questionnaire data is sorted and analyzed to calculate the weighted average sum of each dimension of the maturity assessment model, reflecting the comprehensive diagnostic assessment score of the digital maturity of the enterprise. The dynamic configuration of the questionnaire is flexibly configured using low-code tools according to factors such as the types of indicators collected by each industry and the score weights. The industry big model cleans, samples, removes duplicates, and converts the industry data of the diagnostic report, one-enterprise-one-file, and research report into a format that the model can understand. For text data, it performs word segmentation, indexing, and word vectorization. Based on the ChatGLM big model, the model is pre-trained using enterprise knowledge base data. The industry big model cleans, samples, removes duplicates, and converts the industry data of diagnostic reports, enterprise-specific files, and research reports into a format that the model can understand. For text data, it performs word segmentation, indexing, and word vectorization. Based on the ChatGLM big model, it uses enterprise knowledge base data for model pre-training. The Prompt word fine-tuning described above uses the prompt fine-tuning technology to form a continuous vector based on the knowledge base vector data and the fixed template as input, and uses the gradient descent algorithm to update the prompt word engineering; only the prompt vector P is updated, and the parameters of the pre-trained model M remain unchanged, and the gradient descent algorithm is used to update P, as shown below: The enterprise knowledge base mainly includes: industry solutions, diagnostic reports, one file for each enterprise, enterprise research status, and also includes historical data in the above-mentioned diagnostic process, enterprise status and industry characteristics, and transformation suggestions; the prompt word template includes the following contents: enterprise overview, overall development status, enterprise digital status analysis, and digital transformation suggestions, and forms a PDF diagnostic report through the Agent intelligent body; the Agent intelligent body filters the industry diagnosis prompt template and report display method by industry type to form a digital transformation diagnostic report in PDF carrier.

[0025] Glossary Prompt: A piece of information or instruction given to an AI system to guide it to generate a specific output or perform a task. For example, when using a large language model, a well-designed prompt can help the model understand user needs and generate high-quality text content.

[0026] Agent: Agent is a core concept in the field of artificial intelligence (AI), referring to an entity that can autonomously perceive, make decisions, and perform actions in a specific environment to achieve specific goals or complete specific tasks.

[0027] SAAS application: SaaS (Software as a Service) application is a cloud computing-based software delivery model, in which software applications are provided to customers via the Internet in a subscription form, without the need for users to install or maintain the software locally. In the SaaS model, the software provider is responsible for hosting the application and ensuring its normal operation through the cloud infrastructure, including data storage, security, software updates and maintenance.

[0028] Transformer: A breakthrough neural network architecture proposed by Vaswani et al. in 2017. This model uses a self-attention mechanism to process sequence data without the need for recurrent or convolutional networks, greatly improving processing speed and parallel capabilities. It is particularly suitable for tasks such as machine translation, text generation, and question-answering systems. The Transformer model, through its encoder-decoder structure, can understand and generate complex language structures and is a cornerstone of current NLP research and applications.

Claims

1. A method for generating enterprise transformation diagnosis report based on a large model, characterized in that The steps include: S1. Collect corpus data using the enterprise diagnostic questionnaire; S2. Build a digital transformation diagnosis knowledge base and store it in the vector database; S3, sample model training, constructing a diagnosis report template through fine-tuning of prompt words; S4. Use the fine-tuned prompt words as input, use the trained large model to automatically generate a diagnosis report, use the agent to select different file formats, and generate a diagnosis report output.

2. According to the method for generating enterprise transformation diagnosis report based on big model in claim 1, it is characterized in that The digital transformation diagnosis knowledge base constructed in S2 is stored in the vector database, which specifically includes the following contents: S21. Use the tool OpenRefine to clean, sample, and deduplicate the data and convert it into a format that the model can understand. The conversion methods include: text encoding, standardizing numerical features, and handling missing values. S22, build a knowledge base for the cleaned and standardized data, and segment the text; load and segment a large number of basic documents constituting the knowledge base into text blocks for subsequent indexing; S23, enterprise knowledge base embedding index generation, after dividing the knowledge fragments, using the embedding model to convert them into high-dimensional embedding vectors that can be used for semantic search, and the vectors are associated with the text fragments and stored in the vector database; S24. Vectorize the query question and use the retriever to sort and retrieve the knowledge fragments that best match the question.

3. According to the method for generating enterprise transformation diagnosis report based on big model in claim 1, it is characterized in that The sample model training in step S3, the Prompt word fine-tuning, and the construction of the diagnosis report template are specifically implemented as follows: S31. Use the constructed digital transformation knowledge base to train sample models based on the ChatGLM3-6B large model; S32, performing sample learning on the sample metadata by retrieving similar instances in the vector database; S33, constructing a prompt word template for enterprise diagnosis report, using P-Tuning v2 to form a continuous vector of the prompt word template and search content as input, forming a new output after model training, and outputting prompt words that meet the diagnosis template after repeated iterations; The tips for fine-tuning P-Tuning v2 are as follows: Suppose the pre-trained model is M, a function that accepts input text And output text , optimize the prompt words and introduce the prompt vector , each of which They are all a set of continuous prompt vectors; Model input: original Combined with the prompt vector P to form the new input text x ,Right now: ; Model output: Model M uses the modified input x to produce output text y : ; During training, the prompt vector P is updated, the parameters of the pre-trained model M remain unchanged, and the gradient descent algorithm is used to update P: ; in, represents the prompt vector at iteration t, Table is the updated vector, is the learning rate, is the gradient of the loss function with respect to P.

4. According to the method for generating enterprise transformation diagnosis report based on large model in claim 1, it is characterized in that The automatic writing of the diagnostic report in step S4 takes the fine-tuned prompt words as input, uses the trained large model to automatically generate the diagnostic report, and uses the Agent intelligent body to select different file formats for output. The specific steps are as follows: S41, using Agent to match templates of different industries, using the constructed diagnostic report prompt words, conversation history records, and search results as inputs for the large model; The Agent matches different industry models by predefining the industry type and presentation form, and using a matching tool to select the Prompt words that meet the requirements after training as input; S42. Use the trained industry big model to generate reports in various formats and text descriptions of enterprise digital transformation diagnosis calls, write them into files, and form a diagnosis report; S43, using Agent to select different file formats, providing Word, PDF, Markdown and other file types for diagnosis report viewing and downloading; S44. Label the data of industry solutions and SaaS applications in the database, analyze the industry classification and digital diagnosis and evaluation level through industry big model reasoning, match the labeled data, and screen out solutions and SaaS applications that meet the characteristics of the industry.

Images