Demonstration document processing method and device, electronic equipment and storage medium

Abstract

The invention provides a presentation document processing method and device, electronic equipment and a storage medium. The presentation document processing method comprises the steps that a presentation document to be processed is determined, and the presentation document comprises a plurality of pages; determining a starting page of each minimum chapter unit of the demonstration document and summary content of the starting page; inputting the presentation document or each start page into a pre-trained first multi-modal model to determine a chapter structure of the presentation document, the chapter structure comprising a chapter title and a corresponding page; processing the chapter structure according to the summary content of the starting page; according to the method, the chapter structure of the demonstration document can be generated, a user can clearly check the hierarchical relation of the whole demonstration document, and due to the fact that the first multi-mode model is adopted, front layout visual input is not needed, and dependence on a system is less.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method, device, electronic device and storage medium for processing a presentation document. Background Art

[0002] A presentation document can be a file of a presentation document or an image of a presentation document that appears in a video. The file of a presentation document is usually in the slide format (PPT, power point). The content of a presentation document usually implies a hierarchical structure, but the author of the presentation document does not necessarily write a hierarchical directory of the presentation document. Summary of the invention

[0003] The present disclosure provides a method, device, electronic device and storage medium for processing a presentation document.

[0004] The present disclosure adopts the following technical solutions.

[0005] In some embodiments, the present disclosure provides a method for processing a presentation document, comprising:

[0006] Determine a presentation document to be processed, wherein the presentation document includes a plurality of pages;

[0007] Determine the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page;

[0008] Inputting the presentation document or each of the start pages into a pre-trained first multimodal model to determine a chapter structure of the presentation document, wherein the chapter structure includes chapter titles and corresponding pages;

[0009] Processing the chapter structure according to the summary content of the start page;

[0010] The contents of each page in the minimum chapter unit have similar semantics, and the chapter structure includes one or more chapter titles.

[0011] In some embodiments, the present disclosure provides a device for processing a presentation document, comprising:

[0012] A determination unit, configured to determine a presentation document to be processed, wherein the presentation document includes a plurality of pages;

[0013] A control unit, used to determine the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page;

[0014] A control unit, configured to input the presentation document or each of the start pages into a pre-trained first multimodal model to determine a chapter structure of the presentation document, wherein the chapter structure includes chapter titles and corresponding pages;

[0015] A control unit, configured to process the chapter structure according to the summary content of the start page;

[0016] The contents of each page in the minimum chapter unit have similar semantics, and the chapter structure includes one or more chapter titles.

[0017] In some embodiments, the present disclosure provides an electronic device, comprising: at least one memory and at least one processor;

[0018] The memory is used to store program codes, and the processor is used to call the program codes stored in the memory to execute the above method.

[0019] In some embodiments, the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program code, and when the program code is executed by a processor, the processor is prompted to perform the above method.

[0020] The processing method of the presentation document provided by the embodiment of the present disclosure can generate the chapter structure of the presentation document, and the user can clearly view the hierarchical structure relationship of the entire presentation document, and can quickly understand the content and structure of the presentation document through the chapter title, the title and summary on the start page, etc. At the same time, because the first multimodal model is adopted, there is no need for pre-layout visual input, and the system is less dependent. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0022] Figure 1 It is a flowchart of a method for processing a presentation document according to an embodiment of the present disclosure.

[0023] Figure 2 It is a schematic diagram of the tree structure of an embodiment of the present disclosure.

[0024] Figure 3 It is a schematic diagram of the input and output of the first model of an embodiment of the present disclosure.

[0025] Figure 4 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0027] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0028] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0029] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0030] It is understandable that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and relevant provisions.

[0031] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0032] It should be understood that the various steps described in the method embodiments of the present disclosure can be performed in sequence and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0033] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0034] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0035] It should be noted that the modification of “one” mentioned in the present disclosure is illustrative rather than restrictive, and those skilled in the art should understand that it should be understood as “one or more” unless otherwise clearly indicated in the context.

[0036] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0037] The solution provided by the embodiment of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0038] like Figure 1 As shown, Figure 1 It is a flowchart of a method for processing a presentation document according to an embodiment of the present disclosure, and includes the following steps.

[0039] S11. Determine a presentation document to be processed, where the presentation document includes multiple pages.

[0040] In some embodiments, the method proposed in the present disclosure can be used in terminal devices such as mobile phones, computers, and servers. The presentation document to be processed can be a presentation document file (usually a ppt) or a presentation document image shown in a video. The presentation document has multiple pages, each of which can be a slide. The method of determining the presentation document can be to obtain a presentation document file or to obtain a presentation document image.

[0041] S12, determining the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page.

[0042] In some embodiments, the contents of each page in the minimum chapter unit have similar semantics, and a threshold of semantic similarity can be set. If the semantic similarity between the overall page contents of adjacent pages reaches the threshold, the adjacent pages are considered to belong to the same minimum chapter unit. A minimum chapter unit refers to a sequence of pages with similar semantics (for example, the semantic similarity is higher than the similarity threshold). The number of pages in each minimum chapter unit is not fixed. For example, the first three slides of a presentation document have similar semantics, and the next four slides have similar semantics. The first three are a minimum chapter unit, and the next four are another minimum chapter unit. Different minimum chapter units can be two chapters under the same parent chapter. The minimum chapter unit includes one or more pages, wherein a page of the minimum chapter unit as a start page can be marked as B, and a non-start page can be marked as i. For each start page, it is necessary to determine the summary content of the start page, and the summary content of the start page can be obtained by summarizing the content of the start page.

[0043] S13. Input the presentation document or each of the start pages into a pre-trained first multimodal model to determine the chapter structure of the presentation document.

[0044] In some embodiments, the first polymorphic model is used. The first polymorphic model excludes the front-end layout visual data, and the system relies less. The data input into the first polymorphic model can be a presentation document or a page sequence composed of various start pages. In some embodiments, such as Figure 2 As shown, it schematically shows how the chapter structure is represented by a tree diagram. Of course, the chapter structure can also be represented in other ways. The chapter structure includes chapter titles and corresponding pages. In some embodiments, the chapter structure includes one or more chapter titles, a hierarchical relationship between chapter titles, and pages corresponding to chapter titles. Chapter titles can have a hierarchy, and the lowest-level chapter title can be the chapter title of the smallest chapter unit. Chapter titles include, for example: a first-level title (title), a second-level title (e.g. Figure 2 1.section header), third-level headings (such as Figure 2 1.1 section header), Figure 2 This is just an example. The specific number of heading levels depends on the presentation document itself. There may be more or fewer heading levels. In addition, the chapter structure shows the pages corresponding to each chapter title. Specifically, the page can be displayed after the lowest level title, for example Figure 2In the example, page1 and page2 correspond to 1.1section and the higher-level section title, header. Page3 and page4 correspond to 1.2section header and the higher-level section title. In some embodiments, the determined section structure is not necessarily completely consistent with the section structure displayed in the page content of the presentation document. Similarly, the section title obtained by the first polymorphic model may or may not be the section title displayed on the page content of the presentation document. The first polymorphic model may be a combination of the section structure in the page content of the presentation document itself, which is usually written by the author in the page content (in some cases, the page content of the presentation document may not have the section structure written by the author in the page content) and the content in the presentation document, so as to output the section structure. The section title output by the first polymorphic model may be the same as or different from the title displayed in the presentation document page. In some cases, the page content of the presentation document may not have the section title written by the author in the page content, and the author of the presentation document may not write the section title in the page content of the presentation document. The first polymorphic model may divide the sections according to the summary of the meaning of the content of the presentation document, and may generate section titles according to the summary of the meaning of the page content corresponding to the divided sections.

[0045] S14, processing the chapter structure according to the summary content of the start page.

[0046] In some embodiments, S14 may include adding a page title to the start page in the chapter structure according to the summary content of the start page. In the chapter structure generated by the first polymorphic model, the page does not have a title, and the page title can be added to the start page in the chapter structure according to the summary content of the start page determined previously. Further, the method may also include generating summaries of each page, and the summary may be stored in the page in the chapter structure, and the summary may be displayed when the control mark (such as a mouse) is located on the page in the chapter structure.

[0047] In some embodiments of the present disclosure, the proposed method for processing a presentation document can generate a chapter structure of the presentation document, and the title will also be displayed on the start page of the smallest chapter unit in the chapter structure, so that the user can clearly view the hierarchical relationship of the entire presentation document, and can quickly understand the content and structure of the presentation document through the chapter title, the title and summary on the start page, etc. At the same time, because the first multimodal model is used, there is no need for pre-layout visual input, and there is less dependence on the system.

[0048] In some embodiments of the present disclosure, the presentation document or each of the starting pages is input into a pre-trained first multimodal model, including: determining a page sequence length of the presentation document; in response to the page sequence length of the presentation document being not greater than a preset threshold, inputting the presentation document into the first multimodal model, or, in response to the page sequence length of the presentation document being greater than a preset threshold, inputting the starting page of each of the minimum chapter units into the first multimodal model.

[0049] In some embodiments, the page sequence length of the presentation document can be expressed as the number of pages or the amount of page data. In this embodiment, when the page sequence length of the presentation document itself is not greater than a preset threshold, the presentation document is input into the first multimodal model, otherwise the sequence composed of the starting pages of each minimum chapter unit can be input into the first multimodal model. When the page sequence length of the presentation document is small, inputting it into the first multimodal model can more accurately identify the chapter structure of the presentation document, and the amount of calculation will not be too large. If the page sequence length of the presentation document is too large, then the amount of data that the first multimodal model needs to process is too large, and the calculation speed will be reduced too much. At this time, the sequence composed of the starting page is input, because the starting page is the homepage of the minimum chapter unit, and the content of the non-starting page behind it is similar to it, so the starting page and the page located in the same minimum chapter unit should be located under the same chapter title, therefore, both must be placed under the same minimum chapter title, so inputting the sequence composed of the starting page into the first multimodal model will not affect the generation of the chapter structure. By inputting the sequence composed of the starting page, the amount of data processing can also be reduced.

[0050] In some embodiments of the present disclosure, determining the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page includes: processing the presentation document into text information; inputting the text information into a pre-trained first model so that the first model recognizes the starting page in the presentation document; and inputting each of the starting pages into a pre-trained second model so that the second model generates the summary content of each of the starting pages.

[0051] In some embodiments, text in a presentation document may be extracted, and non-text information in the presentation document may be parsed to generate corresponding text descriptions, and the original text in the presentation document and the generated text description may be merged to obtain text information. A first model is set to classify page types of pages in the presentation document, and page types may be divided into start pages and non-start pages. A second model is used to generate summary content for the start page, which may be obtained by summarizing the content of the start page.

[0052] In some embodiments of the present disclosure, after determining the summary content of the start page, if the start page is input into the pre-trained first multimodal model, then before inputting the start page into the pre-trained first multimodal model, it also includes: determining whether there is a page mistakenly recognized as the start page based on the summary content of each of the start pages, and in response to the existence of a page mistakenly recognized as the start page, removing the page mistakenly recognized as the start page from the start page; or, after determining the summary content of the start page, if the presentation document is input into the pre-trained first multimodal model, then before processing the chapter structure based on the summary content of the start page, it also includes: determining whether there is a page mistakenly recognized as the start page based on the summary content of each of the start pages, and in response to the existence of a page mistakenly recognized as the start page, removing the page mistakenly recognized as the start page from the start page.

[0053] In some embodiments, after obtaining the summary content of the start page, the start page can be streamlined because there may be pages that are mistakenly identified as the start page in the determined start page. Before the start page or the summary content of the start page is used next time, the page that is mistakenly identified as the start page needs to be deleted, that is, before the start page is input into the first polymorphic model, or, if the start page is not input into the first polymorphic model, the page that is mistakenly identified as the start page can be deleted before the chapter structure is processed according to the summary content of the start page.

[0054] In some embodiments, the presentation document (images A) can be processed into text information (Texts A) by a preprocessing model through parsing, image recognition, etc., and then the text information is passed into the first model for segmentation, and the segmented start page (recorded as B) is retained, thereby obtaining a simplified page sequence (images B). At this time, there may be pages in the simplified page sequence that are mistakenly recognized as the start page. Then the simplified page sequence (images B) is input into the second model, so that each page is assigned a title (title B), and the simplified page sequence (images B) is corrected in combination with the assigned title to obtain a corrected page sequence (image C). Specifically, in response to the presence of an empty title in the summary content of the start page, it is determined that the page corresponding to the empty title is the page mistakenly recognized as the start page. In some embodiments, if there is a page with an empty title in the simplified page sequence, the page is mistakenly recognized as the start page. The start page can be corrected by the generated title, avoiding the problem of misrecognition that may exist in the first model.

[0055] In some embodiments of the present disclosure, the first model is a sequence labeling model, the text in the text information input to the first model is replaced by tags or tag identifiers, and the output of the first model is a classification of the tags, which is divided into: start page and non-start page.

[0056] In some embodiments, the module where the first model is located can be a minimum granularity segmentation module. The main task of this module is to screen the pages of the input data to reduce the number of pages processed by the subsequent model, especially to effectively eliminate the low information density pages such as duplication and animation of the presentation documents displayed in the video conference scene exhibition. For example, for a presentation document of 200 pages, after processing, 50 to 100 pages of high information density start pages can be obtained, which provides a better data basis for subsequent precise analysis. The first model can classify pages based on the xlm-roberta-base / large model. The input is a tag or tag identifier of the text information, which is also called a token, and the tag identifier can be a tokenid. This is to convert the text information into a digital encoding form that can be understood by the first model. The output is the classification of tokens, which are specifically divided into two categories: B (starting page) and I (non-starting page), which are used to mark the position of the page in the minimum chapter unit. B corresponds to the starting page of the minimum chapter unit, and I corresponds to the middle page of the minimum chapter unit. For example, the first model will analyze each token in the input text information to determine whether the page it is in belongs to the beginning or the middle part of the minimum chapter unit.

[0057] In some embodiments of the present disclosure, each page in the training document used to train the first model has a classification tag, and during pre-training, the first model only supervises the classification tags in the training document, but does not supervise other tags in the training document;

[0058] In some embodiments, Figure 3 The input and output of the first model are shown. During the training process, only the CLS token (classification mark) is supervised and trained, and other tokens are recorded as -100 and are not supervised. The CLS token plays the role of summarizing the entire sequence information in the first model. Through supervised training, the first model can learn how to classify pages based on the overall information. This training method helps to focus on key information and improve the classification accuracy of the model. Figure 3 The input of the first model in the lower middle part and the output of the first model in the upper middle part. It can be seen that it classifies the CLS token in the page into B or I, thereby determining whether the page type is a start page or a non-start page.

[0059] In some embodiments of the present disclosure, the text information input into the first model is first segmented as an input sequence according to the segmentation rules and then input into the first model. The segmentation rule is that the input sequence of the presentation document is segmented by page, each sequence has a maximum of 12 pages, and adjacent sequences overlap by 1 page. Each page has a maximum of 40 tokens, and this setting is intended to meet the requirement that the input of the first model is less than 480 sequence lengths. For example, for a longer presentation document, it will be divided into multiple sequences according to the page, and each sequence is strictly controlled in length and number of pages to avoid degradation of model performance or inability to process due to excessively long input.

[0060] In some embodiments of the present disclosure, during pre-training, the first model is evaluated by the accuracy, recall and F1 value of the start page and non-start page classification. In some embodiments, the performance of the first model is evaluated by the P (accuracy), R (recall) and F1 value of the B and I classifications. The accuracy rate indicates the proportion of the number of correctly classified pages to the total number of classified pages, the recall rate indicates the proportion of actually correct pages that are correctly classified, and the F1 value comprehensively considers the accuracy rate and the recall rate.

[0061] In some embodiments, determining the starting page of each minimum chapter unit of the presentation document includes: identifying a page without a title or content in the page content as a non-starting page; in some embodiments, in response to the same title being displayed in the page content of at least two consecutive pages in the presentation document, identifying the second page and subsequent pages of the at least two pages as non-starting pages. In some embodiments, in terms of edge case processing, pages without a title or content in the page content are uniformly marked as I; if the same title is displayed in the page content of multiple adjacent consecutive pages, it means that these pages are pages in the same minimum chapter unit, then at least it can be determined that the second and subsequent pages with the same title in the page content are not the starting page, so the second or more consecutive pages with repeated titles are also marked as I. This helps to maintain the consistency and rationality of classification in complex document structures.

[0062] In some embodiments of the present disclosure, the second model adopts a supervised fine-tuning method based on a multimodal large model. In some embodiments, during pre-training, the second model uses a method of matching the true value under the edit distance threshold of 0.5, 0.8, and 1.0 to evaluate the accuracy, recall, and F1 value of the second model.

[0063] In some embodiments, the second model adopts a supervised fine-tuning (SFT) method based on a multimodal large model. This method first uses the pre-training results of the multimodal large model on large-scale general data, and then uses annotated data (which can be data in the field of presentations and video conferencing) for fine-tuning for a specific title generation task, so that the model can better adapt to and accurately complete the task of generating page titles. The second model can use qwen2-vl-2b / 7b. This model has certain capabilities and advantages in processing multimodal information and performing tasks such as similar title generation. It can comprehensively consider multiple modal information such as input text and images to generate appropriate titles. Or the second model can use minicpm-v-2.6, which can also be used as a candidate model for the title generation task. With its own model architecture and pre-training parameters, it is expected to accurately extract appropriate titles from chapter pages after fine-tuning.

[0064] The module where the second model is located can be a page title module, and the training data used is mainly derived from data in the field of presentations and video conferencing. These data cover a wide variety of presentation content, including slide pages of different themes, styles, and structures, as well as the corresponding accurate title information. By training and learning on such targeted data, the model can better understand the intrinsic relationship between the presentation page content and the title, thereby improving the accuracy of title extraction.

[0065] In some embodiments, when evaluating the performance of the second model in generating titles, the method of matching GT (Ground Truth, i.e., true and accurate titles) titles at edit distance thresholds of 0.5 / 0.8 / 1.0 is adopted, and then the accuracy (P), recall (R), and F1 value of the title extraction task are evaluated. Edit distance is an indicator to measure the similarity between two strings. Three different edit distance thresholds of 0.5, 0.8, and 1.0 are set here. When the edit distance between the generated title and the real title is within a certain threshold range, they are considered to be matched. For example, at an edit distance threshold of 0.5, the difference between the generated title and the real title is relatively small, which is more stringent; while at an edit distance threshold of 1.0, the generated title is relatively allowed to have a certain difference from the real title, but it is still within an acceptable range. The accuracy (P) indicates the proportion of the number of correctly matched GT titles in the titles extracted by the model to the total number of titles generated by the model. It reflects the ability of the model to extract correct titles. The higher the value, the better the model performs in not generating wrong titles. Recall (R) refers to the ratio of the number of actually correct titles correctly generated by the model to the total number of actually correct titles. It reflects the model's ability to fully cover and generate correct titles. The higher the recall, the less the model misses correct titles. The F1 value is an evaluation indicator that comprehensively considers the precision and recall rates. Its calculation formula is: F1 = 2*(P*R) / (P+R). The higher the F1 value, the better the overall performance of the model in the title generation task, that is, it can accurately generate correct titles and cover all correct titles as comprehensively as possible.

[0066] In some embodiments, the content of the second model input and the output obtained may be as follows:

[0067] 1.[Prompt]

[0068] 2.

[0069] 3. What is the title of this slide?

[0070] 4.[Output]

[0071] 5. Section Header

[0072] [Prompt] is some guiding prompt information, which is used to guide the model to perform the task of extracting titles and inform the model of the key points or task requirements that need to be paid attention to. : Indicates that the input contains slide information, which reflects the characteristics of multimodality. The second model will comprehensively consider the slides and other relevant text information (such as subsequent query statements) to extract the title. "What is the title of this slide?" is a query statement that asks the second model to extract the title of the current page, so that the model knows the specific goal of this task. [Output]: Marks the beginning of the output result. "SectionHeader" is the title content of this slide generated by the model for the above input. It shows that the second model has analyzed and processed the input multimodal information and based on its pre-trained and fine-tuned knowledge, it has determined that the appropriate title for the current page is "Section Header".

[0073] In some embodiments of the present disclosure, before inputting the presentation document or each of the start pages into the pre-trained first multimodal model, it also includes: scaling the page to be input into the first multimodal model to a preset size so that the number of tags carried by each page does not exceed a preset number, and adding page numbers to pages without page numbers.

[0074] In some embodiments, the first multimodal model may be located in a chapter structure module, and the chapter structure module may be used to scale the page to be input into the first multimodal model to a preset size so that the number of tags carried by each page does not exceed the preset number, and to add page numbers to pages without page numbers. The input page will be scaled to a preset size (e.g., 448×252 pixels). In the scenario of presentation documents, it has been found through experiments that this scaling method will not lose important information, and can minimize the number of tokens per page, so that the number of tokens per page does not exceed the preset number (e.g., 144). For example, for a 50-page document, according to 144 tokens per page, there will be a total of 7,200 tokens input. Such an input scale can retain enough information for chapter structure analysis, and will not make the input data volume too large to exceed the model processing capacity. When scaling the page, a suitable image scaling algorithm will be used to ensure that the page content can still maintain the integrity of key information after the size is reduced, so that the model can accurately analyze it. Adding page numbers to the page, the page number information as additional auxiliary information can help the model accurately correspond to the corresponding page when outputting the chapter structure. For example, when the model outputs a chapter title, the page number can be used to clearly identify the location of the chapter in the document, making it easier to use and process it later. The way to add page numbers is simple and direct. In the preprocessing stage of input data, the page number information is integrated with the image information, so that the model can obtain both the page content and its location information at the same time.

[0075] In some embodiments of the present disclosure, the first multimodal model adopts a supervised fine-tuning method based on a long sequence multimodal large model; in some embodiments, during pre-training, the accuracy, recall rate and F1 value of the output results of the first multimodal model are used to evaluate the first multimodal model.

[0076] In some embodiments, the first multimodal model uses a supervised fine-tuning (SFT, Supervised Fine-Tuning) method based on a long sequence multimodal large model to construct a chapter structure for the demonstration question-answering problem. The method pre-trains the model based on large-scale multimodal data to enable it to process various types of information (such as images, text, etc.), and then uses annotated data for fine-tuning training on a specific chapter structure analysis task to optimize the performance of the model on the task. The sequence length of the first multimodal model can be cut off to 8192 tokens, which means that when the model processes the input data, it will limit the length of the input sequence to within 8192 tokens, avoiding the consumption of computing resources and performance degradation caused by too long sequences, and ensuring that the model can effectively process data within a reasonable resource range and generate accurate chapter structures. The first multimodal model can be qwen2-vl-2b / 7b and minicpm-v-2.6. The end-to-end first multimodal model excludes the pre-placed file layout visual input, which makes the system less dependent. Compared with some traditional models, no additional file layout information preprocessing module is required, which simplifies the system architecture. The first multimodal model can directly learn the characteristics of the chapter structure from the input page (after processing such as scaling) and other related information (such as page numbers), reducing the problem of error propagation and increased system complexity that may arise from the introduction of too many additional modules. For example, there is no need for a dedicated module to parse and process page layout information. The first multimodal model itself can mine clues to the chapter structure from information such as page content and page numbers.

[0077] In some embodiments, the first multimodal model is evaluated using the P (precision), R (recall), and F1 values ​​after hierarchical tiling. Hierarchical tiling is to unfold the multi-level title structure into a relatively flat structure according to certain rules, and then based on this flattened structure, the accuracy and recall between the output of the first multimodal model and the actual situation are counted, and the F1 value is calculated. The accuracy rate indicates the ratio of the number of correct chapter structures output by the first multimodal model to the total number of model outputs, reflecting the accuracy of the model output results; the recall rate indicates the ratio of the number of correct chapter structures correctly output by the model, reflecting the coverage of the correct results by the first multimodal model; the F1 value is an evaluation indicator that comprehensively considers the accuracy rate and the recall rate. These indicators can be used to comprehensively evaluate the accuracy and completeness of the first multimodal model's chapter structure analysis.

[0078] In some embodiments, the output of the model is a multi-level title with page numbers. For example, the output is:

[0079] [OUTPUT]

[0080] #1. Section Header#P1

[0081] ##1.1. Section Header#P3

[0082] #2. Section Header#P5 ....

[0084] "#1.Section Header#P1" means that the first-level title "Section Header" is on page 1, "##1.1.Section Header#P3" means that the second-level title "Section Header" is on page 3, and so on. This format clearly presents the hierarchical relationship of the chapter structure of the presentation, which is convenient for users to read and understand, and also convenient for interaction and integration with other document processing tools or systems. For example, in document editing software, you can quickly build a document outline based on the output of this format, or in presentation software, quickly locate and organize slide content based on chapter structure and page numbers.

[0085] In some embodiments of the present disclosure, the present invention further includes: displaying the hierarchical relationship of each of the chapter titles of the presentation document in a tree structure; wherein each of the chapter titles is a node in the tree structure. Figure 2 As shown, Figure 2 The tree structure is shown on the left. The root node of the tree structure can be the title of the entire presentation document. From the root node downward, it can be divided into multiple levels. Nodes at different levels correspond to chapter titles at different levels. Each chapter title is displayed as a node in the tree structure. The hierarchical relationship of the nodes is also the equal-level relationship of the chapter titles. In this way, the hierarchical relationship of the entire presentation document can be clearly viewed.

[0086] In some embodiments, the pages corresponding to the respective chapter titles in the tree structure are displayed in association. Figure 2 As shown, in Figure 2The right side of the display shows the associated pages, and the pages correspond to the chapter titles, for example, Page 1 corresponds to Title, Page 2 corresponds to 1.1 Section Header, and Page 3 and Page 4 correspond to 1.1 Section Header. The pages corresponding to the upper-level chapter title are all the pages corresponding to all the lower-level chapter titles under the upper-level chapter title. This makes it easy for users to clearly know which pages each chapter title corresponds to. In some embodiments, in the tree structure, if a chapter title is triggered, the page corresponding to the chapter title can be highlighted, and the presentation document can be opened to jump to the first page of the pages corresponding to the chapter title.

[0087] The present disclosure also provides a presentation document processing device, comprising:

[0088] A determination unit, configured to determine a presentation document to be processed, wherein the presentation document includes a plurality of pages;

[0089] A control unit, used to determine the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page;

[0090] A control unit, configured to input the presentation document or each of the start pages into a pre-trained first multimodal model to determine a chapter structure of the presentation document, wherein the chapter structure includes chapter titles and corresponding pages;

[0091] A control unit is used to add a page title to the start page in the chapter structure according to the summary content of the start page.

[0092] The contents of each page in the minimum chapter unit have similar semantics.

[0093] Inputting the presentation document or each of the start pages into a pre-trained first multimodal model comprises:

[0094] Determining the page sequence length of the presentation document;

[0095] In response to the page sequence length of the presentation document being no greater than a preset threshold, the presentation document is input into the first multimodal model; or, in response to the page sequence length of the presentation document being greater than a preset threshold, the starting page of each of the minimum chapter units is input into the first multimodal model.

[0096] In some embodiments, determining the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page includes:

[0097] Processing the presentation document into text information;

[0098] Inputting the text information into a pre-trained first model so that the first model recognizes a start page in the presentation document;

[0099] Each of the start pages is input into a pre-trained second model so that the second model generates summary content of each of the start pages.

[0100] In some embodiments, the control unit is further configured to: after determining the summary content of the start page, if the start page is input into the pre-trained first multimodal model, before inputting the start page into the pre-trained first multimodal model, determine whether there is a page mistakenly recognized as the start page according to the summary content of each of the start pages, and in response to the presence of a page mistakenly recognized as the start page, remove the page mistakenly recognized as the start page from the start page;

[0101] or,

[0102] In some embodiments, the control unit is also used to: after determining the summary content of the start page, if the presentation document is input into a pre-trained first multimodal model, before processing the chapter structure according to the summary content of the start page, determine whether there are pages mistakenly identified as start pages based on the summary content of each start page, and in response to the existence of pages mistakenly identified as start pages, remove the pages mistakenly identified as start pages from the start page.

[0103] In some embodiments, determining whether there is a page mistakenly identified as a start page according to the summary content of each start page includes:

[0104] In response to the presence of an empty title in the summary content of the start page, it is determined that a page corresponding to the empty title is a page mistakenly identified as the start page.

[0105] In some embodiments, determining the starting page of each minimum chapter unit of the presentation document includes: identifying a page without a title or without content in the page content as a non-starting page.

[0106] In some embodiments, determining the starting page of each minimum chapter unit of the presentation document includes: in response to the same title being displayed in the page content of at least two consecutive pages in the presentation document, identifying the second page and subsequent pages of the at least two pages as non-starting pages.

[0107] In some embodiments, the control unit is also used to: before inputting the presentation document or each of the start pages into the pre-trained first multimodal model, scale the page to be input into the first multimodal model to a preset size so that the number of tags carried by each page does not exceed a preset number, and add page numbers to pages without page numbers.

[0108] In some embodiments, the control unit is further used to: display the hierarchical relationship of each of the chapter titles of the presentation document in a tree structure; wherein each of the chapter titles is a node in the tree structure.

[0109] In some embodiments, the control unit is further configured to: associate and display pages corresponding to the respective chapter titles in the tree structure;

[0110] For the embodiments of the device, since they basically correspond to the method embodiments, the relevant parts refer to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the modules described as separation modules may or may not be separated. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present embodiment. Those of ordinary skill in the art can understand and implement without paying creative work.

[0111] The method and device of the present disclosure are described above based on the embodiments and application examples. In addition, the present disclosure also provides an electronic device and a computer-readable storage medium, which are described below.

[0112] Reference below Figure 4 , which shows a schematic diagram of the structure of an electronic device (such as a terminal device or a server) 800 suitable for implementing the embodiment of the present disclosure. The terminal device in the embodiment of the present disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. The electronic device shown in the figure is only an example and should not bring any limitation to the functions and scope of use of the embodiment of the present disclosure.

[0113] The electronic device 800 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage device 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the electronic device 800 are also stored. The processing device 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0114] Typically, the following devices may be connected to the I / O interface 805: input devices 806 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 808 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 809. The communication device 809 may allow the electronic device 800 to communicate with other devices wirelessly or by wire to exchange data. Although the electronic device 800 with various devices is shown in the figure, it should be understood that it is not required to implement or have all the devices shown. More or fewer devices may be implemented or have alternatively.

[0115] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 809, or installed from a storage device 808, or installed from a ROM 802. When the computer program is executed by the processing device 801, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0116] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer 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 of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0117] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0118] The computer-readable medium may be included in the electronic device, or may exist independently without being installed in the electronic device.

[0119] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device executes the method of the present disclosure.

[0120] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0121] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0122] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a unit does not, in some cases, limit the unit itself.

[0123] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0124] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer 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 of the foregoing.

[0125] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided, comprising:

[0126] Determine a presentation document to be processed, wherein the presentation document includes a plurality of pages;

[0127] Determine the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page;

[0128] Inputting the presentation document or each of the start pages into a pre-trained first multimodal model to determine a chapter structure of the presentation document, wherein the chapter structure includes chapter titles and corresponding pages;

[0129] Processing the chapter structure according to the summary content of the start page;

[0130] The contents of each page in the minimum chapter unit have similar semantics.

[0131] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided, wherein the presentation document or each of the start pages is input into a pre-trained first multimodal model, comprising:

[0132] Determining the page sequence length of the presentation document;

[0133] In response to the page sequence length of the presentation document being no greater than a preset threshold, the presentation document is input into the first multimodal model; or, in response to the page sequence length of the presentation document being greater than a preset threshold, the starting page of each of the minimum chapter units is input into the first multimodal model.

[0134] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided, which determines the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page, including:

[0135] Processing the presentation document into text information;

[0136] Inputting the text information into a pre-trained first model so that the first model recognizes a start page in the presentation document;

[0137] Each of the start pages is input into a pre-trained second model so that the second model generates summary content of each of the start pages.

[0138] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided. After determining the summary content of a start page, if the start page is input into a pre-trained first multimodal model, before inputting the start page into the pre-trained first multimodal model, the method further includes: determining whether there is a page mistakenly recognized as a start page according to the summary content of each start page, and in response to the presence of a page mistakenly recognized as a start page, removing the page mistakenly recognized as a start page from the start page;

[0139] or,

[0140] After determining the summary content of the start page, if the presentation document is input into a pre-trained first multimodal model, then before processing the chapter structure according to the summary content of the start page, it also includes: determining whether there is a page mistakenly identified as the start page based on the summary content of each of the start pages, and in response to the existence of a page mistakenly identified as the start page, removing the page mistakenly identified as the start page from the start page.

[0141] It is determined whether there is a page mistakenly recognized as a start page according to summary contents of each of the start pages, and in response to the presence of a page mistakenly recognized as a start page, the page mistakenly recognized as a start page is removed from the start pages.

[0142] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided, which determines whether there is a page mistakenly recognized as a start page according to the summary content of each start page, including:

[0143] In response to the presence of an empty title in the summary content of the start page, it is determined that a page corresponding to the empty title is a page mistakenly identified as the start page.

[0144] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided, which determines the starting page of each minimum chapter unit of the presentation document, including: identifying a page without a title or content in the page content as a non-starting page;

[0145] and / or,

[0146] Determining the starting page of each minimum chapter unit of the presentation document includes: in response to the same title being displayed in the page contents of at least two consecutive pages in the presentation document, identifying the second page and subsequent pages of the at least two pages as non-starting pages.

[0147] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided, which, before inputting the presentation document or each of the start pages into a pre-trained first multimodal model, further includes: scaling the page to be input into the first multimodal model to a preset size so that the number of tags carried by each page does not exceed a preset number, and adding page numbers to pages without page numbers.

[0148] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided, further comprising:

[0149] Displaying the hierarchical relationship of each of the chapter titles of the presentation document in a tree structure;

[0150] Each of the chapter titles is a node in the tree structure.

[0151] According to one or more embodiments of the present disclosure, a method for processing a presentation document is provided, further comprising:

[0152] The pages corresponding to the chapter titles in the tree structure are displayed in association,

[0153] According to one or more embodiments of the present disclosure, a device for processing a presentation document is provided, comprising:

[0154] A determination unit, configured to determine a presentation document to be processed, wherein the presentation document includes a plurality of pages;

[0155] A control unit, used to determine the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page;

[0156] A control unit, configured to input the presentation document or each of the start pages into a pre-trained first multimodal model to determine a chapter structure of the presentation document, wherein the chapter structure includes chapter titles and corresponding pages;

[0157] A control unit, configured to process the chapter structure according to the summary content of the start page;

[0158] The contents of each page in the minimum chapter unit have similar semantics.

[0159] According to one or more embodiments of the present disclosure, there is provided an electronic device, including: at least one memory and at least one processor;

[0160] The at least one memory is used to store program codes, and the at least one processor is used to call the program codes stored in the at least one memory to execute any one of the above methods.

[0161] According to one or more embodiments of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium is used to store program code, and when the program code is executed by a processor, the processor is prompted to execute the above method.

[0162] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0163] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0164] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A method for processing a presentation document, characterized in that: include: Determine a presentation document to be processed, wherein the presentation document includes a plurality of pages; Determine the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page; Inputting the presentation document or each of the start pages into a pre-trained first multimodal model to determine a chapter structure of the presentation document, wherein the chapter structure includes chapter titles and corresponding pages; Processing the chapter structure according to the summary content of the start page; The contents of each page in the minimum chapter unit have similar semantics.

2. The method according to claim 1, characterized in that Inputting the presentation document or each of the start pages into a pre-trained first multimodal model comprises: Determining the page sequence length of the presentation document; In response to the page sequence length of the presentation document being no greater than a preset threshold, the presentation document is input into the first multimodal model; or, in response to the page sequence length of the presentation document being greater than a preset threshold, the starting page of each of the minimum chapter units is input into the first multimodal model.

3. The method according to claim 1, characterized in that Determining the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page includes: Processing the presentation document into text information; Inputting the text information into a pre-trained first model so that the first model recognizes a start page in the presentation document; Each of the start pages is input into a pre-trained second model so that the second model generates summary content of each of the start pages.

4. The method according to claim 1, characterized in that After determining the summary content of the start page, if the start page is input into the pre-trained first multimodal model, before inputting the start page into the pre-trained first multimodal model, the method further includes: determining whether there is a page mistakenly recognized as the start page according to the summary content of each of the start pages, and in response to the presence of a page mistakenly recognized as the start page, removing the page mistakenly recognized as the start page from the start page; or, After determining the summary content of the start page, if the presentation document is input into a pre-trained first multimodal model, then before processing the chapter structure according to the summary content of the start page, it also includes: determining whether there is a page mistakenly identified as the start page based on the summary content of each of the start pages, and in response to the existence of a page mistakenly identified as the start page, removing the page mistakenly identified as the start page from the start page.

5. The method according to claim 4, characterized in that Determining whether there is a page mistakenly identified as a start page according to the summary content of each start page includes: In response to the presence of an empty title in the summary content of the start page, it is determined that a page corresponding to the empty title is a page mistakenly identified as the start page.

6. The method according to claim 1, characterized in that Determining the starting page of each minimum chapter unit of the presentation document includes: identifying a page without a title or content in the page content as a non-starting page; and / or, Determining the starting page of each minimum chapter unit of the presentation document includes: in response to the same title being displayed in the page contents of at least two consecutive pages in the presentation document, identifying the second page and subsequent pages of the at least two pages as non-starting pages.

7. The method according to claim 1, characterized in that Before inputting the presentation document or each of the start pages into the pre-trained first multimodal model, it also includes: scaling the page to be input into the first multimodal model to a preset size so that the number of tags carried by each page does not exceed a preset number, and adding page numbers to pages without page numbers.

8. The method according to claim 1, characterized in that Also includes: Displaying the hierarchical relationship of each of the chapter titles of the presentation document in a tree structure; Each of the chapter titles is a node in the tree structure.

9. The method according to claim 8, characterized in that Also includes: The pages corresponding to the respective chapter titles in the tree structure are displayed in association.

10. A device for processing a presentation document, characterized in that: include: A determination unit, configured to determine a presentation document to be processed, wherein the presentation document includes a plurality of pages; A control unit, used to determine the starting page of each minimum chapter unit of the presentation document and the summary content of the starting page; A control unit, configured to input the presentation document or each of the start pages into a pre-trained first multimodal model to determine a chapter structure of the presentation document, wherein the chapter structure includes chapter titles and corresponding pages; A control unit, configured to process the chapter structure according to the summary content of the start page; The contents of each page in the minimum chapter unit have similar semantics.

11. An electronic device, comprising: at least one memory and at least one processor; The at least one memory is used to store program code, and the at least one processor is used to call the program code stored in the at least one memory to execute the method according to any one of claims 1 to 9. 12 . A computer-readable storage medium, wherein the computer-readable storage medium is used to store program code, wherein when the program code is executed by a processor, the processor is prompted to execute the method according to claim 1 .

Images