Planning development report compiling method and system based on RAG technology

Abstract

The invention relates to the technical field of artificial intelligence application software development, and discloses a planning development report compiling method and system based on an RAG technology. The method comprises the following steps: traversing a root directory, recursively decompressing a compressed file, and carrying out word segmentation and classification on a file name to a target directory based on keyword mapping; performing multi-modal text extraction on the classified files, and recombining text data according to an original file logic sequence; slicing the text data according to natural segments, converting the text data into high-dimensional vectors, and storing the high-dimensional vectors into a vector database; obtaining a report outline containing preset chapter keywords and user preference characteristics; based on user preference and chapter keyword combination, constructing a retrieval statement, and retrieving related text fragments in a vector library; organizing the retrieval fragments into structured cue words, and inputting the structured cue words into a large language model to generate chapter content; and synthesizing and outputting all the chapter contents and the report outline as a report document. According to the method, automatic generation from a multi-source heterogeneous file to a personalized report is realized, and the normalization and pertinence of report generation are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence application software development technology, and in particular to an intelligent report generation method and system that integrates multimodal information perception and understanding. Specifically, this invention relates to an implementation scheme for computer software based on retrieval-enhanced generation (RAG) technology, capable of processing and analyzing multimodal data such as text, images, and audio, and automatically generating structured reports. Background Technology

[0002] As enterprises deepen their digital transformation, formulating scientific and standardized strategic plans and industry development reports has become a core decision-making basis for various industries to clarify their development direction and enhance their core competitiveness. However, existing solutions have significant limitations: First, while traditional manual compilation methods can ensure professionalism, they are time-consuming, labor-intensive, and difficult to standardize in terms of quality. Second, most common automated generation tools can only process text data and lack effective automatic perception and understanding capabilities for non-text multimedia content such as charts, images, video screenshots, and interview recordings that are crucial to reports. This results in insufficient dimensionality and depth of information in the generated content, severe template-based content, and still requires extensive manual modification in the later stages. In particular, when planning reports need to integrate and analyze visual data such as product design drawings and site photos, as well as auditory data such as meeting minutes and recordings, existing technologies cannot simultaneously meet the requirements of standardization and personalization, seriously restricting the efficiency and quality of planning and development report compilation. Summary of the Invention

[0003] To address the technical problem that existing technologies struggle to balance standardized and personalized needs in report preparation, severely limiting the efficiency and quality of report generation, this invention provides a planning and development report preparation method and system based on RAG technology.

[0004] The first aspect of this invention discloses a method for compiling a planning and development report based on RAG technology, the method comprising: The program iterates through the files in the specified root directory, identifies and decompresses the compressed files encountered, and recursively traverses the decompressed directories to obtain all files to be processed; the files to be processed include text files, image files, and audio files. Based on a preset keyword-directory mapping relationship, the filenames of the obtained files to be processed are segmented to obtain filename keywords, and the files are moved to the corresponding target directories by matching them. Multiple files in the preprocessed target directory are obtained, and multimodal content extraction is performed on the files. Specifically, text content is extracted from text files, optical character recognition (OCR) and / or image description information is extracted from image files, and speech recognition (ASR) is performed on audio files. All extracted content is then reassembled according to its logical order in the original files to form coherent text data. The text data is divided into multiple text segments according to natural paragraphs, and each text segment is converted into a high-dimensional vector through a text embedding model and then stored in a vector database. Obtain a pre-set report outline, which contains multiple chapters and each chapter has preset chapter keywords, and simultaneously obtain user preference characteristics; For each chapter, a search query statement constructed based on the user preference features and chapter keywords is retrieved from the vector database to obtain multiple text fragments related to each chapter. The retrieved text fragments are dynamically organized into structured prompt words, and the prompt words are input into a large language model to request the generation of the text content for that chapter; The text content generated in each chapter is combined with the report outline to output a report document.

[0005] As an optional implementation, in the first aspect of the present invention, after outputting a report document, the method further includes: Record user actions on specific chapters of the report document, analyze user preferences for content depth, writing style, or data density based on these actions, and dynamically update user personalized profiles. In the subsequent report generation process, the updated user personalization profile will be used as input to adaptively adjust the content of the generated text.

[0006] As an optional implementation, in the first aspect of the present invention, for each chapter, a search query statement constructed based on the combination of user preference features and chapter keywords is performed in the vector database to obtain multiple text fragments related to each chapter, including: The preset chapter keywords for each chapter are subjected to limited semantic expansion; By combining preset prompt word templates, user preference features, and expanded chapter keywords, a search query is constructed to perform a similarity search in the vector database and obtain the top N most relevant text fragments.

[0007] As an optional implementation, in the first aspect of the present invention, the step of dynamically organizing the retrieved text fragments into structured prompt words specifically includes: Based on the position and level of the current chapter in the report outline, select the corresponding context organization strategy from the preset prompt word strategy library; The retrieved text fragments are sorted and truncated according to their order of appearance in the original file or their relevance score to the chapter keywords; The sorted and truncated text fragments are then filled into the corresponding fields of the prompt word template according to the selected context organization strategy, forming the final prompt word for this request's large language model.

[0008] A second aspect of this invention discloses a planning and development report preparation system based on RAG technology, comprising: The file traversal module is used to traverse the files in the specified root directory, identify and decompress the encountered compressed files, and recursively traverse the decompressed directory to obtain all files to be processed; the files to be processed include text files, image files and audio files. The word segmentation and matching module is used to perform word segmentation on the filenames of the acquired files to be processed based on a preset keyword-directory mapping relationship to obtain filename keywords, and then move the files to the corresponding target directories through matching. The file acquisition module is used to acquire multiple files in the target directory after preprocessing. The text processing module is used to perform multimodal content extraction on the file, including extracting text content from text files, performing optical character recognition (OCR) and / or image description information extraction from image files, performing speech recognition (ASR) conversion on audio files, and reorganizing all the extracted content according to its logical order in the original file to form coherent text data. The vectorization processing module is used to divide the text data into multiple text segments according to natural paragraphs, and to convert each text segment into a high-dimensional vector through a text embedding model and then store it in a vector database. The outline and preference acquisition module is used to acquire a preset report outline, which contains multiple chapters and each chapter has preset chapter keywords, while also acquiring user preference features; The data retrieval module is used to retrieve multiple text fragments related to each chapter by performing a retrieval query statement constructed based on the user preference features and chapter keywords in the vector database. The content generation module is used to dynamically organize the retrieved text fragments into structured prompt words, and input the prompt words into the large language model to request the generation of the text content of the chapter; The report output module is used to combine the text content generated in each chapter with the report outline and output it as a report document.

[0009] As an optional implementation, in a second aspect of the present invention, the system further includes: The configuration update module is used to record the user's operation behavior on specific chapters of the report document, analyze the user's preference characteristics in terms of content depth, writing style or data density based on the behavior, and dynamically update the user's personalized configuration file. The adaptive adjustment module is used to adaptively adjust the content of the generated text by taking the updated user personalization profile as input during the subsequent report generation process.

[0010] As an optional implementation, in a second aspect of the present invention, the data retrieval module includes: The semantic expansion submodule is used to perform limited semantic expansion on the preset chapter keywords for each chapter; The retrieval submodule is used to combine preset prompt word templates, user preference features, and expanded chapter keywords to construct a retrieval query statement, so as to perform a similarity retrieval in the vector database and obtain the top N most relevant text fragments.

[0011] As an optional implementation, in a second aspect of the present invention, the content generation module specifically includes: The selection submodule is used to select the corresponding context organization strategy from the preset prompt word strategy library based on the position and level of the current chapter in the report outline; The processing submodule is used to sort and truncate the retrieved text fragments according to their order of appearance in the original file or their relevance score to the chapter keywords; The prompt word acquisition submodule is used to fill the corresponding fields of the prompt word template with sorted and truncated text fragments according to the selected context organization strategy to form structured prompt words; A generation submodule is used to input the structured prompt words into a large language model to request the generation of the text content for that chapter.

[0012] A third aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute a planning and development report preparation method based on RAG technology disclosed in the first aspect of the present invention.

[0013] A fourth aspect of this invention discloses an electronic device, the electronic device comprising: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method described above.

[0014] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: The method provided by this invention includes the following steps: obtaining multiple files in a target directory and extracting multimodal content; reorganizing the extracted content according to the logical order of the original files to form coherent text data; segmenting the text data into text fragments according to natural paragraphs, and converting them into high-dimensional vectors through a text embedding model and storing them in a vector database; obtaining a pre-set report outline containing multiple chapters and pre-setting keywords for each chapter; retrieving relevant text fragments from the vector database based on the chapter keywords; dynamically organizing the retrieved text fragments into structured prompt words and inputting them into a large language model to generate the content of each chapter; and finally synthesizing the generated content of each chapter with the report outline and outputting it as a report document.

[0015] This method enables the automated generation of personalized reports from multi-source heterogeneous files, solving the problems of low report generation efficiency, insufficient content accuracy, and lack of personalization in existing technologies, and effectively improving the accuracy, standardization, and relevance of report generation.

[0016] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit the invention. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0018] Figure 1 This is a flowchart illustrating a planning and development report preparation method based on RAG technology according to an exemplary embodiment; Figure 2 This is a block diagram illustrating a planning and development report preparation system based on RAG technology, according to another exemplary embodiment. Figure 3 This is a block diagram illustrating a planning and development report preparation system based on RAG technology, according to another exemplary embodiment. Detailed Implementation

[0019] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0020] Figure 1 This is a flowchart illustrating a planning and development report preparation method based on RAG technology, according to an exemplary embodiment. Figure 1 As shown, this method includes the following steps.

[0021] Step 101: Traverse the files in the specified root directory, identify and decompress the compressed files encountered, and recursively traverse the decompressed directory to obtain all files to be processed, including text files, image files and audio files.

[0022] In this embodiment of the invention, optionally, a recursive traversal algorithm is used to automatically decompress and perform depth-first traversal of compressed files (such as ZIP and TAR) in the root directory, ensuring that the contents of the nested file structure are obtained without omission. This step can improve the problems of manual processing of compressed packages or incomplete traversal in the prior art.

[0023] In this embodiment of the invention, optionally, the files in the specified root directory can be massive amounts of data files that are updated in real time based on the user's report compilation needs (including report type, report object, report occasion, etc.).

[0024] Step 102: Based on the preset keyword-directory mapping relationship, the filenames of the obtained files to be processed are segmented to obtain filename keywords, and the files are moved to the corresponding target directories by matching.

[0025] In this embodiment of the invention, preferably, Chinese word segmentation technology (such as Jieba) can be used to extract semantic features from filenames, and the files can be automatically classified and placed into predefined target directories according to the mapping relationship. This initially achieves the sorting and classification of massive amounts of files, laying the foundation for subsequent processing.

[0026] Step 103: Obtain multiple files in the preprocessed target directory, extract multimodal content from the files, and reassemble all the extracted text content according to their logical order in the original files to form coherent text data.

[0027] In this embodiment of the invention, a unified format parsing and content extraction can be performed on various file formats (DOC, PDF, images, audio, etc.). Text content extraction is performed on text files, optical character recognition (OCR) and / or image description information extraction is performed on image files, and automatic speech recognition (ASR) conversion is performed on audio files. Furthermore, the original text and the OCR-recognized text are sequentially reassembled according to the original logical order in the original file to generate a complete set of content data that maintains the original document logic. This embodiment ensures the integrity and fidelity of the document information. Furthermore, the aforementioned image files may include on-site survey photos or ID photos of people; the extraction of image description information from the image files may specifically include: The face recognition model is invoked to identify the person in the image, and the identified identity information is stored in the vector database as part of the image's description information.

[0028] Based on the above embodiments, preferably, predefined entity keywords are identified and extracted from the extracted image description information or speech recognition text. These entity keywords include geographical location, device number, and personnel name. The entity keywords are then added to the file name keyword set to optimize the matching accuracy of the keyword-directory mapping relationship.

[0029] Step 104: Divide the text data into multiple text segments according to natural paragraphs, and then convert each text segment into a high-dimensional vector through a text embedding model and store it in a vector database.

[0030] In this embodiment of the invention, adaptive segmentation can be performed using natural segment boundaries, and an upper limit for segment length (e.g., 256 characters) can be set. Furthermore, a text embedding model is used to transform each text segment into a high-dimensional space vector, which is then stored in a vector database.

[0031] Step 105: Obtain the pre-set report outline, which contains multiple chapters and pre-set chapter keywords for each chapter, while also obtaining user preference characteristics.

[0032] Step 106: For each chapter, a search query statement constructed based on user preference features and chapter keywords is retrieved from the vector database to obtain multiple text fragments related to each chapter.

[0033] In this embodiment of the invention, user preference characteristics can optionally be determined from multiple perspectives, such as user historical behavior data and industry role data.

[0034] As an optional implementation, for each chapter, the search query constructed based on user preference features and chapter keywords is retrieved from a vector database to obtain multiple text fragments related to each chapter, including: Limited semantic expansion is performed on the pre-set chapter keywords for each chapter; By combining pre-set prompt word templates, user preference features, and expanded chapter keywords, a search query is constructed to perform a similarity search in a vector database and obtain the top N most relevant text fragments.

[0035] Step 107: Dynamically organize the retrieved text fragments into structured prompt words, and input the prompt words into the large language model to request the generation of the text content for that chapter.

[0036] As an optional implementation, the above-mentioned dynamic organization of retrieved text fragments into structured prompt words may specifically include: Based on the position and level of the current chapter in the report outline, select the corresponding context organization strategy from the preset prompt word strategy library; The retrieved text fragments are sorted and truncated according to their order of appearance in the original file or their relevance score to the chapter keywords; The sorted and truncated text fragments are then populated into the corresponding fields of the prompt word template according to the selected context organization strategy, forming the final prompt words for this request's large language model. Step 108: Combine the text content generated in each chapter with the report outline and output the report document.

[0037] As an optional implementation, after the above output is a report document, it may also include: Record user actions on specific chapters of the report document, analyze user preferences for content depth, writing style, or data density based on these actions, and dynamically update user personalized profiles. In the subsequent report generation process, the updated user personalization profile will be used as input to adaptively adjust the content of the generated text.

[0038] It is evident that implementation Figure 1 The described RAG-based planning and development report compilation method ensures the integrity and accuracy of original information through multimodal text extraction and logical sequence reorganization; it then constructs search queries based on user preference features and chapter keywords to accurately locate materials from a vector database; finally, it generates a personalized report deeply integrated with the outline structure through a large language model. This method automates the entire process of planning and development report compilation, from data preparation to content generation, significantly improving the report's standardization and relevance.

[0039] Figure 2 This is a block diagram illustrating a report generation system using RAG technology according to an exemplary embodiment. Figure 2 As shown, the system includes: The file traversal module 201 is used to traverse the files in the specified root directory, identify and decompress the encountered compressed files, and recursively traverse the decompressed directory to obtain all files to be processed, including text files, image files and audio files.

[0040] The word segmentation and matching module 202 is used to segment the filenames of the obtained files to be processed based on the preset keyword-directory mapping relationship to obtain filename keywords, and move the files to the corresponding target directories through matching.

[0041] The file acquisition module 203 is used to acquire multiple files in the preprocessed target directory.

[0042] The text processing module 204 is used to extract multimodal content from the file and reorganize all the extracted text content according to its logical order in the original file to form coherent text data.

[0043] The vectorization processing module 205 is used to divide the text data into multiple text segments according to natural paragraphs, and then convert each text segment into a high-dimensional vector through a text embedding model and store it in the vector database.

[0044] The outline and preference acquisition module 206 is used to acquire a preset report outline, which contains multiple chapters and each chapter has preset chapter keywords, while also acquiring user preference features.

[0045] The data retrieval module 207 is used to retrieve multiple text fragments related to each chapter by constructing a retrieval query statement based on user preference features and chapter keywords in the vector database.

[0046] The content generation module 208 is used to dynamically organize the retrieved text fragments into structured prompt words, and input the prompt words into the large language model to request the generation of the text content of the chapter.

[0047] The report output module 209 is used to combine the text content generated in each chapter with the report outline and output it as a report document.

[0048] It is evident that implementation Figure 2 The described planning and development report compilation system based on RAG technology ensures the integrity and accuracy of original information through multimodal text extraction and logical sequence reorganization; it then constructs search queries based on user preference features and chapter keywords to accurately locate materials from a vector database; finally, it generates personalized reports deeply integrated with the outline structure through a large language model. This method automates the entire process of planning and development report compilation, from data preparation to content generation, significantly improving the standardization and relevance of the reports.

[0049] Figure 3 This is a block diagram illustrating another planning and development report preparation system based on RAG technology, according to an exemplary embodiment. Figure 3 The planning and development report preparation system based on RAG technology shown is... Figure 2 The planning and development report preparation system based on RAG technology shown is obtained through further optimization. Figure 2 The report generation pen shown is a comparison with the IoT-based smart signature RAG technology. Figure 3The planning and development report preparation system based on RAG technology shown may also include: The configuration update module 210 is used to record the user's operation behavior on specific chapters of the report document, analyze the user's preference characteristics in terms of content depth, writing style or data density based on behavior analysis, and dynamically update the user's personalized configuration file.

[0050] The adaptive adjustment module 211 is used to adaptively adjust the content of the generated text by taking the updated user personalization profile as input during the subsequent report generation process.

[0051] As an optional implementation, the data retrieval module 207 described above may include: The semantic extension submodule 2071 is used to perform limited semantic extensions on the preset chapter keywords for each chapter.

[0052] The retrieval submodule 2072 is used to combine preset prompt word templates, user preference features and expanded chapter keywords to construct a retrieval query statement, so as to perform similarity retrieval in the vector database and obtain the top N most relevant text fragments.

[0053] As another optional implementation, the content generation module 208 described above may specifically include: Select submodule 2081 to select the corresponding context organization strategy from the preset prompt word strategy library based on the position and level of the current chapter in the report outline.

[0054] Processing submodule 2082 is used to sort and truncate the retrieved text fragments according to their order of appearance in the original file or their relevance score to the chapter keywords.

[0055] The prompt word acquisition submodule 2083 is used to fill the sorted and truncated text fragments into the corresponding fields of the prompt word template according to the selected context organization strategy, forming structured prompt words.

[0056] The generation submodule 2084 is used to input structured prompt words into the large language model to request the generation of the text content for this chapter.

[0057] It is evident that implementation Figure 3 The described planning and development report compilation system based on RAG technology ensures the integrity and accuracy of original information through multimodal text extraction and logical sequence reorganization; it then constructs search queries based on user preference features and chapter keywords to accurately locate materials from a vector database; finally, it generates personalized reports deeply integrated with the outline structure through a large language model. This method automates the entire process of planning and development report compilation, from data preparation to content generation, significantly improving the standardization and relevance of the reports.

[0058] In one exemplary embodiment, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the planning and development report preparation method based on RAG technology as described above.

[0059] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for compiling planning and development reports based on RAG technology, characterized in that, include: Iterate through the files in the specified root directory, identify and decompress the compressed files encountered, and recursively traverse the decompressed directory to obtain all files to be processed. The files to be processed include text files, image files, and audio files; Based on a preset keyword-directory mapping relationship, the filenames of the obtained files to be processed are segmented to obtain filename keywords, and the files are moved to the corresponding target directories by matching them. Multiple files in the preprocessed target directory are obtained, and multimodal content extraction is performed on the files. Specifically, text content is extracted from text files, optical character recognition (OCR) and / or image description information is extracted from image files, and speech recognition (ASR) is performed on audio files. All extracted content is then reassembled according to its logical order in the original files to form coherent text data. The text data is divided into multiple text segments according to natural paragraphs, and each text segment is converted into a high-dimensional vector through a text embedding model and then stored in a vector database. Obtain a pre-set report outline, which contains multiple chapters and each chapter has preset chapter keywords, and simultaneously obtain user preference characteristics; For each chapter, a search query statement constructed based on the user preference features and chapter keywords is retrieved from the vector database to obtain multiple text fragments related to each chapter. The retrieved text fragments are dynamically organized into structured prompt words, and the prompt words are input into a large language model to request the generation of the text content for that chapter; The text content generated in each chapter is combined with the report outline to output a report document.

2. The method according to claim 1, characterized in that, After the output is a report document, the method further includes: Record user actions on specific chapters of the report document, analyze user preferences for content depth, writing style, or data density based on these actions, and dynamically update user personalized profiles. In the subsequent report generation process, the updated user personalization profile will be used as input to adaptively adjust the content of the generated text.

3. The method according to claim 1, characterized in that, For each chapter, a search query constructed based on the user preference features and chapter keywords is retrieved from the vector database to obtain multiple text fragments related to each chapter, including: The preset chapter keywords for each chapter are subjected to limited semantic expansion; By combining preset prompt word templates, user preference features, and expanded chapter keywords, a search query is constructed to perform a similarity search in the vector database and obtain the top N most relevant text fragments.

4. The method according to any one of claims 1 to 3, characterized in that, The process of dynamically organizing the retrieved text fragments into structured prompt words specifically includes: Based on the position and level of the current chapter in the report outline, select the corresponding context organization strategy from the preset prompt word strategy library; The retrieved text fragments are sorted and truncated according to their order of appearance in the original file or their relevance score to the chapter keywords; The sorted and truncated text fragments are then filled into the corresponding fields of the prompt word template according to the selected context organization strategy, forming the final prompt word for this request's large language model.

5. A planning and development report preparation system based on RAG technology, characterized in that, include: The file traversal module is used to traverse the files in the specified root directory, identify and decompress the encountered compressed files, and recursively traverse the decompressed directory to obtain all files to be processed; the files to be processed include text files, image files and audio files. The word segmentation and matching module is used to perform word segmentation on the filenames of the acquired files to be processed based on a preset keyword-directory mapping relationship to obtain filename keywords, and then move the files to the corresponding target directories through matching. The file acquisition module is used to acquire multiple files in the target directory after preprocessing. The text processing module is used to perform multimodal content extraction on the file, including extracting text content from text files, performing optical character recognition (OCR) and / or image description information extraction from image files, performing speech recognition (ASR) conversion on audio files, and reorganizing all the extracted content according to its logical order in the original file to form coherent text data. The vectorization processing module is used to divide the text data into multiple text segments according to natural paragraphs, and to convert each text segment into a high-dimensional vector through a text embedding model and then store it in a vector database. The outline and preference acquisition module is used to acquire a preset report outline, which contains multiple chapters and each chapter has preset chapter keywords, while also acquiring user preference features; The data retrieval module is used to retrieve multiple text fragments related to each chapter by performing a retrieval query statement constructed based on the user preference features and chapter keywords in the vector database. The content generation module is used to dynamically organize the retrieved text fragments into structured prompt words, and input the prompt words into the large language model to request the generation of the text content of the chapter; The report output module is used to combine the text content generated in each chapter with the report outline and output it as a report document.

6. The system according to claim 5, characterized in that, The system also includes: The configuration update module is used to record the user's operation behavior on specific chapters of the report document, analyze the user's preference characteristics in terms of content depth, writing style or data density based on the behavior, and dynamically update the user's personalized configuration file. The adaptive adjustment module is used to adaptively adjust the content of the generated text by taking the updated user personalization profile as input during the subsequent report generation process.

7. The system according to claim 5, characterized in that, The data retrieval module includes: The semantic expansion submodule is used to perform limited semantic expansion on the preset chapter keywords for each chapter; The retrieval submodule is used to combine preset prompt word templates, user preference features, and expanded chapter keywords to construct a retrieval query statement, so as to perform a similarity retrieval in the vector database and obtain the top N most relevant text fragments.

8. The system according to any one of claims 5 to 7, characterized in that, The content generation module specifically includes: The selection submodule is used to select the corresponding context organization strategy from the preset prompt word strategy library based on the position and level of the current chapter in the report outline; The processing submodule is used to sort and truncate the retrieved text fragments according to their order of appearance in the original file or their relevance score to the chapter keywords; The prompt word acquisition submodule is used to fill the corresponding fields of the prompt word template with sorted and truncated text fragments according to the selected context organization strategy to form structured prompt words; The generation submodule is used to input the structured prompt words into the large language model to request the generation of the text content for this chapter.

Images