Method, system and equipment for generating report lecture based on large model and medium

Abstract

The invention provides a report lecture generation method, system and device based on a large model and a medium, and belongs to the technical field of natural language process.The method comprises the steps that a data document and a discourse document input by a user are received; analyzing the data document and calling a large model to generate a first segment in combination with a first segment cue word template; integrating the discourse document and calling a large model to extract a central sentence in combination with a central sentence extraction cue word template; performing RAG retrieval on the discourse document based on the central sentence, determining a related fragment, combining with a paragraph to generate a cue word template, and calling a large model to generate a paragraph; the first segment and the paragraph are spliced, an ending prompt word template is combined, the large model is called to generate an ending, and the first segment, the paragraph and the ending are spliced to obtain a lecture; and performing fine tuning on the large model based on the historical conference text and the superior department problem data set, inputting the lecture into the fine-tuned large model, outputting a simulation problem and modifying the lecture. The method improves the writing efficiency of the report lecture, reduces the manual workload, and improves the accuracy and speciality of the lecture.

Description

Technical Field

[0001] The present application belongs to the field of natural language processing technology, and specifically relates to a method, system, device and medium for generating report manuscripts based on a large model. Background Art

[0002] Currently, text generation, such as writing presentations for professional fields, presents unique challenges and presents high demands. On the one hand, presentations must provide a high-level overview of the project's core points, including key data and progress information. On the other hand, presentations in different professional fields have specific formats and terminology. For example, academic research emphasizes innovation and rigor, while corporate project reports emphasize achievements and next steps. Traditional text generation methods, whether manual or through simple large-scale model generation, struggle to simultaneously meet these requirements. Manual writing is time-consuming and labor-intensive, and can easily omit key information, while direct generation through large-scale models is prone to content deviations and formatting discrepancies.

[0003] Furthermore, existing text generation technology has limitations in handling polysemy and semantic hierarchies. Complex polysemy and multi-layered semantic structures often confuse models, affecting the logical coherence and clarity of the text. For example, without considering the context, the word "bank" makes it difficult for the model to determine whether it refers to a financial institution or a riverbank, which can easily lead to confusion.

[0004] In summary, existing technologies for generating professional text suffer from issues such as poor accuracy, insufficient integration of specialized knowledge, and limitations in handling polysemy and semantic hierarchy. These issues severely impact the effectiveness of large models in generating professional presentations. There is an urgent need for a method that can integrate specialized domain knowledge and accurately generate high-quality presentations to improve both efficiency and quality. Summary of the Invention

[0005] In a first aspect, an embodiment of the present application provides a method for generating a report manuscript based on a large model, comprising the following steps: S1. Receive the data document in Excel format and the text document in Doc format input by the user; S2 parses the data document in Excel format, generates a JSON string, and fills the JSON string into the preset first paragraph prompt word template, and calls the general large model to generate the first paragraph of the lecture notes; S3. Integrate all Doc format text documents, fill in the preset central sentence extraction prompt word template, and call the general large model to extract a preset number of central sentences; S4. Using the extracted central sentence as input, perform RAG search on the text document to recall relevant segments of the central sentence. The central sentence and the corresponding recalled relevant segments are then inserted into a preset paragraph generation prompt word template. The general large model is then used to generate paragraphs of the lecture script. S5. After splicing the first paragraph of the lecture with the paragraphs of the lecture, input the preset ending prompt word template, call the universal large model according to the preset temperature parameters to generate the end of the lecture, and splice the first paragraph, paragraph and end to obtain a complete lecture; S6. Based on the historical meeting texts and the constructed dataset of questions from higher-level departments, fine-tune the general large model, input the complete lecture notes into the fine-tuned general large model, output simulated questions, and modify the complete lecture notes.

[0006] Furthermore, the specific steps of step S1 are as follows: S11 receives the document input by the user, identifies and distinguishes the data document in Excel format and the text document in Doc format by the file suffix; S12. Perform table file verification on the data document, and perform text file verification on the discourse document.

[0007] Furthermore, the specific steps of step S2 are as follows: S21. Use a Python script to call the pandas and openpyxl libraries to read Excel-formatted data documents and convert the table data into a DataFrame structure. S22. Convert the data in the DataFrame structure into a JSON string containing the header information and the corresponding data; S23. Fill the JSON string into the preset first paragraph prompt word template to generate a complete prompt word; S24. Call the universal large model Qwen-72B, input the complete prompt words, and generate the first paragraph of the speech.

[0008] Furthermore, the specific steps of step S3 are as follows: S31 reads all input Doc format text documents, and concatenates the name and content of each text document into a unified text text according to a preset format; S32. The concatenated text is filled into a preset central sentence extraction prompt word template, which contains instructions and material fields, wherein the instruction requires the output of a set number of central sentences covering all key points; S33. Call the general large model Qwen-72B, input the filled central sentence extraction prompt word template, and output the central sentence list.

[0009] Furthermore, the specific steps of step S4 are as follows: S41. Divide the text document into blocks and use the embedding model to convert each block of text into a vector of a set dimension, obtain several text block vectors and construct a vector index; S42. Convert each central sentence into a vector representation and use the bge_es_reranker retrieval algorithm to calculate the semantic similarity between the central sentence vector and the vectors of each text block in the text document; S43. Recall the top-K related text segments corresponding to each central sentence based on the similarity scores, where K is a preset integer; S44. Fill the central sentence and its corresponding Top-K related text fragments into the preset paragraph generation prompt word template to form a complete prompt word; S45. Call the universal large model Qwen-72B, input the complete prompt words, and generate the paragraphs of the lecture.

[0010] Furthermore, the specific steps of step S5 are as follows: S51. Sequentially concatenate the first paragraph of the speech and all generated paragraphs into an intermediate speech; S52. Fill in the preset ending prompt word template with the middle lecture notes and set the temperature parameters of the general large model; S53. Call the universal large model Qwen-72B and input the ending prompt word template according to the temperature parameter to generate the end of the speech; S54. Splice the middle speech manuscript with the end of the speech manuscript to obtain the complete speech manuscript, delete the colloquial expressions from the complete speech manuscript, compress it according to the preset ratio, and complete the polishing process.

[0011] Furthermore, the specific steps of step S6 are as follows: S61. Collect historical meeting texts and corresponding questions and instructions from higher authorities to form the original training dataset; S62. Desensitizing the original training data set to obtain a desensitized data set; S63. Use the desensitized dataset to perform LORA fine-tuning on the general large model Qwen-72B, set the rank parameter and scaling factor, and train the fine-tuned general large model Qwen-72B using the focal cross entropy loss function; S64. Input the complete lecture notes into the trained fine-tuned large model, and output simulated questions from higher-level authorities. S65. Modify the complete speech based on the simulated questions from the superior department.

[0012] In a second aspect, the present application also provides a large model-based report presentation system, comprising: The material document receiving module is used to receive the Excel format data document and Doc format text document input by the user; The first paragraph of the lecture script generation module is used to parse the data document in Excel format, generate a JSON string, fill the JSON string into the preset first paragraph prompt word template, and call the general large model to generate the first paragraph of the lecture script; The central sentence extraction module is used to integrate all Doc format text documents, fill in the preset central sentence extraction prompt word template, and call the general large model to extract a preset number of central sentences; The paragraph content improvement module is used to perform RAG search on the text document using the extracted central sentence as input, recall relevant fragments of the central sentence, and fill the central sentence and the corresponding recalled relevant fragments into the preset paragraph generation prompt word template. After that, the general large model is called to generate the paragraph of the speech script; The lecture ending generation module is used to splice the first paragraph of the lecture with the paragraphs of the lecture, input the preset ending prompt word template, call the universal large model to generate the ending of the lecture according to the preset temperature parameters, and splice the first paragraph, paragraphs and ending to obtain a complete lecture; The module for simulating questions from higher-level departments and modifying lecture notes is used to fine-tune the general large model based on historical meeting texts and the constructed dataset of questions from higher-level departments. The module then inputs the complete lecture notes into the fine-tuned general large model, outputs simulated questions, and modifies the complete lecture notes.

[0013] In a third aspect, an embodiment of the present application further provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method for generating a report script based on a large model as described in the first aspect are implemented.

[0014] In a fourth aspect, an embodiment of the present application further provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for generating a report manuscript based on a large model as described in the first aspect are implemented.

[0015] It can be seen from the above technical solutions that this application has the following advantages: The report manuscript generation method, system, equipment and medium based on the big model provided in this application realize the intelligent generation of report manuscripts, improve the efficiency and quality of report manuscript writing, and reduce the workload and error rate of manual writing; and combine the parsing and processing of data documents and discourse documents, and utilize the capabilities of general big models and fine-tuning models to realize the automated process from material input to complete manuscript output; at the same time, the simulation of questions from superior departments and manuscript modification links further enhances the pertinence and practicality of the manuscript, ensures that the report content can meet the requirements of high accuracy and high relevance in professional fields, and provide support for improving reporting efficiency and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for the description. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 The figure is a flow chart of the method for generating report manuscripts based on a large model of the present invention.

[0018] Figure 2 It is a schematic diagram of the report speech generation system based on the large model of the present invention. DETAILED DESCRIPTION

[0019] The specific steps of the large-scale model-based report presentation generation method will be described in detail below, and various embodiments of the present disclosure will be described more fully. The present disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of the present disclosure to the specific embodiments disclosed herein, but rather that the present disclosure should be understood to cover all adjustments, equivalents, and / or alternatives that fall within the spirit and scope of the various embodiments of the present disclosure.

[0020] For example, in the process of writing reports in today's professional fields, traditional text generation methods face many challenges due to their particularity and high requirements.

[0021] On the one hand, a report draft needs to provide a high-level summary of the project's key points, including key data and progress information. On the other hand, different professional fields have unique formats and terminology for report drafts. For example, academic research emphasizes innovation and rigor, while corporate project reports focus on the presentation of results and subsequent plans. Traditional text generation methods, whether manual or relying solely on large models, struggle to meet these diverse requirements. Manual writing is not only time-consuming and labor-intensive, but also prone to omissions of key information. Text generated directly by large models often suffers from content bias and formatting inconsistencies, making it difficult to directly apply. Furthermore, existing text generation technologies have significant limitations in handling polysemous words and complex semantic hierarchies.

[0022] Overall, existing technologies for generating professional texts present a range of challenges, including insufficient accuracy, insufficient integration of specialized knowledge, and limited processing of polysemy and semantic hierarchy. These issues severely restrict the widespread application of large models in generating professional presentations. Therefore, there is an urgent need for a method that can deeply integrate specialized domain knowledge and accurately generate high-quality, standardized presentations to effectively improve the efficiency and quality of presentations.

[0023] In response to the above problems, this embodiment provides a report script generation method based on a big model. By parsing Excel and Doc documents, the report script is generated using a big model, which improves writing efficiency, reduces human errors, ensures content accuracy and professionalism, and simulates questions from higher-level departments to enhance the pertinence of reports.

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] See also Figure 1 FIG. 1 is a flow chart of a method for generating a report script based on a large model in a specific embodiment, the method comprising the following steps: S1. Receive the data document in Excel format and the text document in Doc format input by the user; It should be noted that Excel and Doc documents are accepted as materials to provide basic data and corpus for subsequent speech generation, ensuring that the content of the speech is close to actual work scenarios and meeting the dual needs of professional field reports for data and text; S2 parses the data document in Excel format, generates a JSON string, and fills the JSON string into the preset first paragraph prompt word template, and calls the general large model to generate the first paragraph of the lecture notes; It should be noted that the Excel document is parsed to generate a JSON string and filled with the prompt word template to call the large model to generate the first paragraph, realizing data-driven automatic generation of the first paragraph of the lecture script, improving the efficiency and quality of the first paragraph generation, and ensuring that it contains key data and information; S3. Integrate all Doc format text documents, fill in the preset central sentence extraction prompt word template, and call the general large model to extract a preset number of central sentences, the central sentence can cover the main points of the text document; It should be noted that the Doc document is integrated with the prompt word template to extract the central sentence, build a framework for the main content of the speech, ensure that it covers the main points of the discourse, provide a clear theme and logical structure for the subsequent paragraph generation, and make the speech content clear and focused. S4. Using the extracted central sentence as input, perform RAG search on the text document to recall relevant segments of the central sentence. The central sentence and the corresponding recalled relevant segments are then inserted into a preset paragraph generation prompt word template. The general large model is then used to generate paragraphs of the lecture script. It should be noted that RAG search is performed with the central sentence as input to recall relevant fragments and generate paragraphs, ensuring that the paragraph content is highly relevant to the central sentence and accurate and detailed, thus improving the overall quality and logic of the speech and making the content more substantial and persuasive; S5. After splicing the first paragraph of the lecture with the paragraphs of the lecture, input the preset ending prompt word template, call the universal large model according to the preset temperature parameters to generate the end of the lecture, and splice the first paragraph, paragraph and end to obtain a complete lecture; It should be noted that the first paragraph, paragraphs and ending are spliced ​​together to generate a complete speech manuscript, and polished to improve the quality and professionalism of the speech manuscript, making it more suitable for the reporting scenario and enhancing the overall effect and professional image of the report; S6. Fine-tune the general model based on historical meeting texts and the constructed dataset of questions from higher-level departments. Input the complete speech notes into the fine-tuned general model, output simulated questions, and modify the complete speech notes. It should be noted that by fine-tuning the model based on historical data and simulating questions from higher-level departments to modify the speech, the speech can be made more targeted, potential problems can be discovered and improved in advance, the reporting effect can be enhanced, and the satisfaction of higher-level departments can be improved.

[0026] This embodiment parses Excel-formatted data documents and Doc-formatted discourse documents, and utilizes a general large model and a fine-tuning model to achieve an automated process from material input to complete speech output. This not only improves the efficiency of writing report manuscripts and reduces the workload and error rate of manual writing, but also further enhances the pertinence and practicality of the speech manuscripts by simulating questions from higher-level departments and the manuscript modification process, ensuring that the report content can meet the professional field's requirements for high accuracy and high relevance, and providing support for improving reporting efficiency and quality.

[0027] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process in this embodiment, another method for generating a report script based on a large model is provided, which includes the following steps: S1. Receive the data document in Excel format and the text document in Doc format input by the user; The specific steps of step S1 are as follows: S11 receives the document input by the user, identifies and distinguishes the data document in Excel format and the text document in Doc format by the file suffix; S12 performs table file verification on the data document and performs text file verification on the text document; Specifically, perform merged cell parsing on Excel format data documents: Detect the merged area and replace the values ​​of the merged cells with the values ​​of the first row and first column; To align the headers of a multi-tab Excel data document: Calculate the hash value H=hash(header) of each tab header. If (δ=0.15) then merge the data and add the page signature prefix; Among them, i and j are page tag codes; S2 parses the data document in Excel format, generates a JSON string, and fills the JSON string into the preset first paragraph prompt word template, and calls the general large model to generate the first paragraph of the lecture notes; The specific steps of step S2 are as follows: S21. Use a Python script to call the pandas and openpyxl libraries to read Excel-formatted data documents and convert the table data into a DataFrame structure. Specifically, call pandas.read_excel() and the openpyxl library to parse the Excel table into a DataFrame structure; S22. Convert the data in the DataFrame structure into a JSON string containing the header information and the corresponding data; Specifically, convert the DataFrame structure into a nested JSON string, retaining the header hierarchy, with the key name format being "main header@subheader"; S23. Fill the JSON string into the preset first paragraph prompt word template to generate a complete prompt word; For example, the first paragraph prompt word template includes: Instructions: Generate the first paragraph according to the example style, prohibit output of thinking process Example: Field-related demonstration text Data field: fill in the JSON string; S24. Call the universal large model Qwen-72B, input the complete prompt words, and generate the first paragraph of the speech script; S3. Integrate all Doc format text documents, fill in the preset central sentence extraction prompt word template, and call the general large model to extract a preset number of central sentences, the central sentence can cover the main points of the text document; The specific steps of step S3 are as follows: S31 reads all input Doc format text documents, and concatenates the name and content of each text document into a unified text text according to a preset format; S32. Fill the spliced ​​text into the preset central sentence extraction prompt word template, which contains instructions and material fields, where the instructions require the output of a set number (eg, 5-7) of central sentences covering all key points; Exemplarily, the central sentence extraction prompt word template includes: Instructions: Output 5-7 concise central sentences, covering all key points Material field: fill in the complete corpus; S33 calls the general large model Qwen-72B, inputs the filled central sentence extraction prompt word template, and outputs the central sentence list; For example, the length of each central sentence is limited to 15-25 words; S4. Using the extracted central sentence as input, perform RAG search on the text document to recall relevant segments of the central sentence. The central sentence and the corresponding recalled relevant segments are then inserted into a preset paragraph generation prompt word template. The general large model is then used to generate paragraphs of the lecture script. The specific steps of step S4 are as follows: S41. Divide the text document into chunks and use an embedding model to convert each chunk of text into a vector of a set dimension (e.g., a 768-dimensional vector). Obtain several chunk vectors and construct a vector index. It should be noted that the bge-large-zh model is used to encode the text documents into 768-dimensional vectors; S42. Convert each central sentence into a vector representation and use the bge_es_reranker retrieval algorithm to calculate the semantic similarity between the central sentence vector and the vectors of each text block in the text document; Specifically, the comprehensive similarity score of each central sentence is calculated using the following formula:

[0028] in, It is based on the semantic relevance of hfl / chinese-roberta-wwm-ext. is the Elasticsearch keyword weight, If the segment contains "budget / risk / schedule", it is assigned 1.0, otherwise it is 0; S43. Recall the top-K related text segments corresponding to each central sentence based on the similarity scores, where K is a preset integer; S44. Fill the central sentence and its corresponding Top-K related text fragments into the preset paragraph generation prompt word template to form a complete prompt word; For example, the paragraph generation prompt word template requires: Avoid point-by-point statements, avoid using colloquial expressions like "we," and avoid expanding into full paragraphs based on search snippets. S45. Call the universal large model Qwen-72B, input the complete prompt word, and generate the paragraph of the speech script; S5. After splicing the first paragraph of the lecture with the paragraphs of the lecture, input the preset ending prompt word template, call the universal large model according to the preset temperature parameters to generate the end of the lecture, and splice the first paragraph, paragraph and end to obtain a complete lecture; The specific steps of step S5 are as follows: S51. Sequentially concatenate the first paragraph of the speech and all generated paragraphs into an intermediate speech; S52. Fill in the preset ending prompt word template with the middle lecture notes and set the temperature parameter of the general large model (for example, it can be set to 0.5); S53. Call the universal large model Qwen-72B and input the ending prompt word template according to the temperature parameter to generate the end of the speech; S54. The middle speech manuscript is spliced ​​with the end of the speech manuscript to obtain a complete speech manuscript, and the colloquial expressions are deleted from the complete speech manuscript, and the manuscript is compressed according to a preset ratio to complete the polishing process; For example, text compression is performed according to the following formula: ; Remove colloquial expressions, such as replacing "we" with formal terms such as "this project team"; S6. Fine-tune the general model based on historical meeting texts and the constructed dataset of questions from higher-level departments. Input the complete speech notes into the fine-tuned general model, output simulated questions, and modify the complete speech notes. The specific steps of step S6 are as follows: S61. Collect historical meeting texts and corresponding questions and instructions from higher authorities to form the original training dataset; For example, in the original training dataset: Positive example: Historical lecture paragraph → Corresponding to the question from the superior department Negative samples: Randomly replace 30% of the keywords in the speech to generate interference problems; S62. Desensitizing the original training data set to obtain a desensitized data set; For example, regular expressions are used to replace phone numbers and amounts, and named entity recognition models are used to replace names, company names, and addresses. S63. Use the desensitized dataset to perform LORA fine-tuning on the general large model Qwen-72B, set the rank parameter and scaling factor, and train the fine-tuned general large model Qwen-72B using the focal cross entropy loss function; For example, the rank parameter r=8, the scaling factor α=32; S64. Input the complete lecture notes into the trained fine-tuned large model, and output simulated questions from higher-level authorities. S65. Modify the complete speech based on the simulated questions from the superior department.

[0029] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0030] like Figure 2 As shown, the following is an embodiment of the report manuscript generation system based on a big model provided by the embodiment of the present disclosure. This system and the report manuscript generation method based on a big model in the above-mentioned embodiments belong to the same inventive concept. For details not fully described in the embodiment of the report manuscript generation system based on a big model, please refer to the embodiment of the report manuscript generation method based on a big model.

[0031] The system includes: The material document receiving module is used to receive the Excel format data document and Doc format text document input by the user; The first paragraph of the lecture script generation module is used to parse the data document in Excel format, generate a JSON string, fill the JSON string into the preset first paragraph prompt word template, and call the general large model to generate the first paragraph of the lecture script; The core sentence extraction module is used to integrate all Doc format text documents, fill in the preset core sentence extraction prompt word template, and call the general large model to extract a preset number of core sentences that can cover the main points of the text document; The paragraph content improvement module is used to perform RAG search on the text document using the extracted central sentence as input, recall relevant fragments of the central sentence, and fill the central sentence and the corresponding recalled relevant fragments into the preset paragraph generation prompt word template. After that, the general large model is called to generate the paragraph of the speech script; The lecture ending generation module is used to splice the first paragraph of the lecture with the paragraphs of the lecture, input the preset ending prompt word template, call the universal large model to generate the ending of the lecture according to the preset temperature parameters, and splice the first paragraph, paragraphs and ending to obtain a complete lecture; The module for simulating questions from higher-level departments and modifying lecture notes is used to fine-tune the general large model based on historical meeting texts and the constructed dataset of questions from higher-level departments. The module then inputs the complete lecture notes into the fine-tuned general large model, outputs simulated questions, and modifies the complete lecture notes.

[0032] This embodiment realizes the automation and intelligence of the entire report speech generation process through the interactive collaboration of the material document receiving module, the speech first paragraph generation module, the central sentence extraction module, the paragraph content improvement module, the speech ending generation module, and the simulation of superior department questioning and speech modification module, thereby improving the overall performance and reliability of the system and providing users with a convenient report speech generation tool.

[0033] The large-scale model-based report manuscript generation method provided in the embodiment of the present application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiment of the present invention does not constitute a limitation on the electronic device, and the electronic device may include more or fewer components than shown, or combine certain components, or arrange components differently. In the embodiment of the present invention, the electronic device includes but is not limited to a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the present application described and / or required herein.

[0034] The electronic device may include a processor, an external memory interface, an internal memory, a universal serial bus (USB) interface, a charging management module, a power management module, a battery, a wireless communication module, an audio module, a speaker, a microphone, a sensor module, a button, a camera, a display, and a SIM card interface, etc.

[0035] It is understood that the structures illustrated in the embodiments of the present application do not constitute specific limitations on the electronic device. In other embodiments of the present application, the electronic device may include more or fewer components than shown, or combine or separate certain components, or arrange the components differently. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0036] A processor may include one or more processing units, such as a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0037] The processor can be the nerve center and command center of the electronic device. The controller can generate operation control signals based on the instruction opcode and timing signal to complete the control of instruction fetching and execution.

[0038] The processor may also include a memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or is reusing. If the processor needs to use the instruction or data again, it can directly call it from the memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0039] The electronic device implements the method of generating a report manuscript based on a large model of the present application, which receives a data document in Excel format and a discourse document in Doc format input by a user; parses the data document in Excel format, generates a JSON string, and fills the JSON string into a preset first paragraph prompt word template, and then calls the general large model to generate the first paragraph of the manuscript; integrates all discourse documents in Doc format, fills in a preset central sentence extraction prompt word template, and then calls the general large model to extract a preset number of central sentences, wherein the central sentences can cover the main points of the discourse document; uses the extracted central sentences as input, performs RAG retrieval on the discourse document, recalls relevant fragments of the central sentence, and fills the central sentence and the corresponding recalled relevant fragments into a preset paragraph generation prompt word template After that, the general large model is called to generate the paragraphs of the speech; after splicing the first paragraph of the speech with the paragraphs of the speech, after inputting the preset ending prompt word template, the general large model is called to generate the ending of the speech according to the preset temperature parameters, and the first paragraph, paragraphs and ending are spliced ​​to obtain a complete speech; based on the historical meeting text and the constructed superior department problem data set, the general large model is fine-tuned, and the complete speech is input into the fine-tuned general large model, the simulated problem is output and the complete speech is modified. The technical solution is achieved by parsing Excel and Doc documents and using the large model to generate report manuscripts, effectively improving writing efficiency, reducing human errors, ensuring content accuracy and professionalism, and at the same time simulating questions from superior departments to enhance the pertinence of reports.

[0040] The storage medium provided in this application stores a program product that can implement a method for generating report manuscripts based on a large model.

[0041] The method for generating report manuscripts based on the big model includes: receiving Excel format data documents and Doc format text documents input by users; parsing the Excel format data documents, generating JSON strings, and filling the JSON strings into a preset first paragraph prompt word template, and then calling the general big model to generate the first paragraph of the report manuscript; integrating all Doc format text documents, filling in a preset central sentence extraction prompt word template, and then calling the general big model to extract a preset number of central sentences, wherein the central sentences can cover the main points of the text documents; using the extracted central sentences as input, performing RAG retrieval on the text documents, and recalling the central sentences. The relevant fragments of the central sentence are collected, and the central sentence and the corresponding recalled relevant fragments are filled into the preset paragraph generation prompt word template, and then the general large model is called to generate the paragraph of the lecture; after splicing the first paragraph of the lecture with the paragraph of the lecture, after inputting the preset ending prompt word template, the general large model is called to generate the ending of the lecture according to the preset temperature parameters, and the first paragraph, paragraph and ending are spliced ​​to obtain a complete lecture; based on the historical meeting text and the constructed superior department problem data set, the general large model is fine-tuned, and the complete lecture is input into the fine-tuned general large model, the simulated problem is output and the complete lecture is modified.

[0042] In some possible implementations, the large model-based report script generation method disclosed herein can be implemented in the form of a program product, which includes program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps described in the above "Exemplary Method" section of this specification according to various exemplary implementations of the present disclosure.

[0043] The storage medium of the present disclosure can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0044] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating report manuscripts based on a large model, characterized in that: The steps include: S1. Receive the data document in Excel format and the text document in Doc format input by the user; S2 parses the data document in Excel format, generates a JSON string, and fills the JSON string into the preset first paragraph prompt word template, and calls the general large model to generate the first paragraph of the lecture notes; S3. Integrate all Doc format text documents, fill in the preset central sentence extraction prompt word template, and call the general large model to extract a preset number of central sentences; S4. Using the extracted central sentence as input, perform RAG search on the text document to recall relevant segments of the central sentence. The central sentence and the corresponding recalled relevant segments are then inserted into a preset paragraph generation prompt word template. The general large model is then used to generate paragraphs of the lecture script. S5. After splicing the first paragraph of the lecture with the paragraphs of the lecture, input the preset ending prompt word template, call the universal large model according to the preset temperature parameters to generate the end of the lecture, and splice the first paragraph, paragraph and end to obtain a complete lecture; S6. Based on the historical meeting texts and the constructed dataset of questions from higher-level departments, fine-tune the general large model, input the complete lecture notes into the fine-tuned general large model, output simulated questions, and modify the complete lecture notes.

2. The method for generating report manuscripts based on a large model according to claim 1, characterized in that: The specific steps of step S1 are as follows: S11 receives the document input by the user, identifies and distinguishes the data document in Excel format and the text document in Doc format by the file suffix; S12. Perform table file verification on the data document, and perform text file verification on the discourse document.

3. The method for generating report manuscripts based on a large model according to claim 1, characterized in that: The specific steps of step S2 are as follows: S21. Use a Python script to call the pandas and openpyxl libraries to read Excel-formatted data documents and convert the table data into a DataFrame structure. S22. Convert the data in the DataFrame structure into a JSON string containing the header information and the corresponding data; S23. Fill the JSON string into the preset first paragraph prompt word template to generate a complete prompt word; S24. Call the universal large model Qwen-72B, input the complete prompt words, and generate the first paragraph of the speech.

4. The method for generating report manuscripts based on a large model according to claim 3 is characterized in that: The specific steps of step S3 are as follows: S31 reads all input Doc format text documents, and concatenates the name and content of each text document into a unified text text according to a preset format; S32. The concatenated text is filled into a preset central sentence extraction prompt word template, which contains instructions and material fields, wherein the instruction requires the output of a set number of central sentences covering all key points; S33. Call the general large model Qwen-72B, input the filled central sentence extraction prompt word template, and output the central sentence list.

5. The method for generating report manuscripts based on a large model according to claim 4 is characterized in that: The specific steps of step S4 are as follows: S41. Divide the text document into blocks and use the embedding model to convert each block of text into a vector of a set dimension, obtain several text block vectors and construct a vector index; S42. Convert each central sentence into a vector representation and use the bge_es_reranker retrieval algorithm to calculate the semantic similarity between the central sentence vector and the vectors of each text block in the text document; S43. Recall the top-K related text segments corresponding to each central sentence based on the similarity scores, where K is a preset integer; S44. Fill the central sentence and its corresponding Top-K related text fragments into the preset paragraph generation prompt word template to form a complete prompt word; S45. Call the universal large model Qwen-72B, input the complete prompt words, and generate the paragraphs of the lecture.

6. The method for generating report manuscripts based on a large model according to claim 5, characterized in that: The specific steps of step S5 are as follows: S51. Sequentially concatenate the first paragraph of the speech and all generated paragraphs into an intermediate speech; S52. Fill in the preset ending prompt word template with the middle lecture notes and set the temperature parameters of the general large model; S53. Call the universal large model Qwen-72B and input the ending prompt word template according to the temperature parameter to generate the end of the speech; S54. Splice the middle speech manuscript with the end of the speech manuscript to obtain the complete speech manuscript, delete the colloquial expressions from the complete speech manuscript, compress it according to the preset ratio, and complete the polishing process.

7. The method for generating report manuscripts based on a large model according to claim 5, characterized in that: The specific steps of step S6 are as follows: S61. Collect historical meeting texts and corresponding questions and instructions from higher authorities to form the original training dataset; S62. Desensitizing the original training data set to obtain a desensitized data set; S63. Use the desensitized dataset to perform LORA fine-tuning on the general large model Qwen-72B, set the rank parameter and scaling factor, and train the fine-tuned general large model Qwen-72B using the focal cross entropy loss function; S64. Input the complete lecture notes into the trained fine-tuned large model, and output simulated questions from higher-level authorities. S65. Modify the complete speech based on the simulated questions from the superior department.

8. A report script generation system based on a large model, characterized in that: include: The material document receiving module is used to receive the Excel format data document and Doc format text document input by the user; The first paragraph of the lecture script generation module is used to parse the data document in Excel format, generate a JSON string, fill the JSON string into the preset first paragraph prompt word template, and call the general large model to generate the first paragraph of the lecture script; The central sentence extraction module is used to integrate all Doc format text documents, fill in the preset central sentence extraction prompt word template, and call the general large model to extract a preset number of central sentences; The paragraph content improvement module is used to perform RAG search on the text document using the extracted central sentence as input, recall relevant fragments of the central sentence, and fill the central sentence and the corresponding recalled relevant fragments into the preset paragraph generation prompt word template. After that, the general large model is called to generate the paragraph of the speech script; The lecture ending generation module is used to splice the first paragraph of the lecture with the paragraphs of the lecture, input the preset ending prompt word template, call the universal large model to generate the ending of the lecture according to the preset temperature parameters, and splice the first paragraph, paragraphs and ending to obtain a complete lecture; The module for simulating questions from higher-level departments and modifying lecture notes is used to fine-tune the general large model based on historical meeting texts and the constructed dataset of questions from higher-level departments. The module then inputs the complete lecture notes into the fine-tuned general large model, outputs simulated questions, and modifies the complete lecture notes.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the report speech generation method based on the large model as described in any one of claims 1 to 7 are implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for generating a report manuscript based on a large model as described in any one of claims 1 to 7 are implemented.

Images