Image-based model training method and device, question and answer method and device and electronic equipment

By determining the style parameters of the table image and training the table question and answer model, the problems of insufficient text modal processing and insufficient image modal training in the table image question and answer task are solved, and the accuracy and adaptation of the model are improved sex.

CN118966353APending Publication Date: 2024-11-15BEIJING BAIDU NETCOM SCI & TECH CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410964262.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The multimodal large model has insufficient text modal processing, insufficient image modal training, and insufficient Q&A sample classification in the table image Q&A task, resulting in poor performance when processing table images.

Method used

By determining the style parameters of the table image and training the preset model based on these style parameters and corresponding question and answer information, the table question and answer model is obtained to improve the model's understanding and processing ability of the table data.

Benefits of technology

It improves the accuracy and adaptability of the table question and answer questions related to table images more accurately, and enhances the model's ability to handle table question and answer tasks in different scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118966353A_ABST
    Figure CN118966353A_ABST
Patent Text Reader

Abstract

The invention provides an image-based model training method and device, a question and answer method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, in particular to the technical fields of computer vision, deep learning, large models and the like. According to the specific implementation scheme, style parameters of a first table image are determined; wherein the style parameter is used for representing a display style of the first table image; and based on the style parameter of the first table image and the first question and answer information corresponding to the first table image, training a preset model to obtain a table question and answer model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to technical fields such as computer vision, deep learning, and large models. Background Art

[0002] With the development of artificial intelligence technology, human-computer interaction based on multimodal large models has gradually emerged. At present, multimodal large models are widely used to process natural language understanding and image recognition tasks. Question answering tasks based on table images are one of the common tasks. Summary of the invention

[0003] The present disclosure provides an image-based model training method, a question-answering method, a device, an electronic device, and a storage medium.

[0004] According to one aspect of the present disclosure, there is provided an image-based model training method, comprising:

[0005] Determining style parameters of the first table image; wherein the style parameters are used to characterize the display style of the first table image;

[0006] Based on the style parameters of the first table image and the first question and answer information corresponding to the first table image, the preset model is trained to obtain a table question and answer model.

[0007] According to another aspect of the present disclosure, a question-answering method is provided, comprising:

[0008] Obtain a second table image and a question corresponding to the second table image;

[0009] Based on the table question and answer model, the second table image and the question corresponding to the second table image are processed to obtain the answer corresponding to the second table image; wherein the table question and answer model is trained based on the style parameters of the first table image and the first question and answer information corresponding to the first table image.

[0010] According to another aspect of the present disclosure, there is provided an image-based model training device, comprising:

[0011] A style determination module, used to determine style parameters of the first table image; wherein the style parameters are used to characterize the display style of the first table image;

[0012] The model training module is used to train the preset model based on the style parameters of the first table image and the first question and answer information corresponding to the first table image to obtain a table question and answer model.

[0013] According to another aspect of the present disclosure, there is provided a question-answering device, comprising:

[0014] A question acquisition module, used for acquiring the second table image and the question corresponding to the second table image;

[0015] A model processing module is used to process the second table image and the question corresponding to the second table image based on the table question and answer model to obtain the answer corresponding to the second table image; wherein the table question and answer model is trained based on the style parameters of the first table image and the first question and answer information corresponding to the first table image.

[0016] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0017] at least one processor; and

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute any method in the embodiments of the present disclosure.

[0020] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any method according to the embodiments of the present disclosure.

[0021] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements any method according to the embodiments of the present disclosure when executed by a processor.

[0022] The technical solution of the disclosed embodiment can train a preset model through the style parameters of the first table image and the first question and answer information corresponding to the first table image to obtain a table question and answer model, which can improve the accuracy of the table question and answer model in processing table question and answer tasks.

[0023] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0025] Figure 1 is a flowchart of an image-based model training method provided in an embodiment of the present disclosure;

[0026] Figure 2 is a flowchart of a question-and-answer method provided by an embodiment of the present disclosure;

[0027] Figure 3 It is a schematic diagram of the overall process of the image-based model training method and the question-answering method in an application example of the present disclosure;

[0028] Figure 4 is a schematic block diagram of an image-based model training device provided by an embodiment of the present disclosure;

[0029] Figure 5 is a schematic block diagram of an image-based model training device provided by another embodiment of the present disclosure;

[0030] Figure 6 is a schematic block diagram of an image-based model training device provided by another embodiment of the present disclosure;

[0031] Figure 7 is a schematic block diagram of a question-answering device provided by an embodiment of the present disclosure;

[0032] Figure 8 It is a block diagram of an electronic device used to implement the image-based model training method and / or question-answering method of the embodiments of the present disclosure. DETAILED DESCRIPTION

[0033] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0034] In order to facilitate understanding of the image-based model training method and question-answering method provided by the embodiments of the present disclosure, the relevant technologies of the embodiments of the present disclosure are explained below. The following related technologies can be arbitrarily combined with the technical solutions of the embodiments of the present disclosure as optional solutions, and they all belong to the protection scope of the embodiments of the present disclosure.

[0035] At present, multimodal large models perform well in natural language understanding and image recognition tasks, but they are still insufficient in table image question answering tasks. These models mainly rely on large-scale general data sets for training and lack training and optimization specifically for table data. Specifically, multimodal large models face the following limitations in table image question answering tasks:

[0036] 1. Inadequate processing of text modality: Tabular data exhibits unique structural characteristics in text modality. Unlike ordinary natural language text, they usually contain wired and wireless table formats. At the same time, tabular data also has a variety of hierarchical structures, such as single-column, multi-column, and multi-level complex tables. Traditional text processing models lack a special processing mechanism for tabular data, resulting in poor performance in table parsing and question-answering tasks.

[0037] 2. Insufficient training of image modalities: Table image data is diverse and has different styles, including different fonts, colors, layouts, and formats. Existing multimodal large models lack training specifically for these different types of table image data, resulting in unstable performance of the model when processing table images of different styles. For example, handwritten tables and printed tables, horizontal tables and vertical tables are different in form, and existing models often find it difficult to fully cover these diverse table image types.

[0038] 3. Insufficient classification of question-answering samples: The table question-answering task involves image-text alignment, that is, effectively matching and aligning the information in the table image with the text. The related technology lacks systematic classification of table question-answering sample pairs and formulates specific prompts as instructions, which affects the accuracy of the model's question-answering. For example, in the table question-answering task, users may ask questions such as "What is the total revenue this quarter?" or "What is the highest value shown in this table?" Existing models often lack clear guidance and prompts when dealing with these specific questions, resulting in inaccurate or incomplete answers.

[0039] The technical solution of the embodiment of the present disclosure can solve at least one of the above technical problems.

[0040] The following is an introduction to the application scenarios of the method provided by the embodiment of the present disclosure. The application scenarios may include but are not limited to the following aspects:

[0041] 1. Financial statement analysis: Automatically parse and analyze financial statements, answer users' questions about financial indicators such as income, expenditure, and profit, and help companies and individuals with financial management and decision-making.

[0042] 2. Data statistics and analysis: Process various data statistical tables, answer users' questions about data trends, comparative analysis, statistical results, etc., and support data-driven decision-making processes.

[0043] 3. Medical data management: Parse and analyze medical record forms, such as patient medical records, test reports, diagnosis results, etc., provide answers about patient health status, treatment plans, medical statistics, etc., and assist doctors and medical institutions in clinical decision-making and management.

[0044] 4. Educational performance analysis: Process student performance tables, answer questions about student performance distribution, class ranking, subject performance comparison, etc., to help educational institutions and teachers conduct teaching evaluation and improve strategies.

[0045] 5. Market research and study: Analyze market survey questionnaire results, answer questions about consumer preferences, market trends, product feedback, etc., and help companies conduct market analysis and strategic planning.

[0046] Figure 1 An image-based model training method provided by an embodiment of the present disclosure is shown. The method can be applied to an image-based model training device, which can be deployed in an electronic device. The electronic device is, for example, a single or multi-machine terminal, server or other processing device. Among them, the terminal can be a user equipment (User Equipment, UE) such as a mobile device, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc.; the server can be a stand-alone server or a server cluster. In some possible implementations, the method can also be implemented by a processor calling computer-readable instructions stored in a memory. As Figure 1 As shown, the method may include the following steps S110 to S120.

[0047] Step S110: determining style parameters of the first table image; wherein the style parameters are used to characterize the display style of the first table image.

[0048] In the disclosed embodiment, the first table image may be an image of a table in any field, for example, the first table image may be a table image in various forms such as a handwritten table, a printed table, a horizontal table, a vertical table, a multi-level table, etc., and the first table image may include fonts and borders of any style and color. Exemplarily, the first table image may be a financial report, a medical record table, a student's grade table, etc.

[0049] In actual applications, the first table image can be obtained from a variety of sources (such as PDF files, scanned documents, spreadsheets). Of course, the first table image can also be an image of a table drawn manually by a user, or can be generated by an electronic device calling AI (Artificial Intelligence).

[0050] In the disclosed embodiment, the style parameters of the first table image may characterize the display style of the first table image, and the display style may include at least one type of information such as structural information (such as the arrangement of rows and columns), font type, color, and border style.

[0051] Optionally, an image processing model may be deployed in the electronic device, or the electronic device may be connected to other devices or call the image processing model through an API (Application Programming Interface), thereby obtaining the style parameters of the first table image through the first table image. Alternatively, the style image of the first table image may also be obtained through manual annotation.

[0052] Step S120: Based on the style parameters of the first table image and the first question and answer information corresponding to the first table image, the preset model is trained to obtain a table question and answer model.

[0053] In the embodiments of the present disclosure, the preset model can be understood as a model with a preset network structure. Exemplarily, the preset model can include a multimodal model, that is, a model that can be used to process multimodal data. In other words, the preset model can process input data containing any one or more of multiple modalities. Multiple modalities include, for example, images, text, voice, etc.

[0054] Exemplarily, the preset model may include a large model, also referred to as a large language model (LLM), having extremely large-scale parameters, complex network structure, and high computational requirements, capable of capturing complex patterns and relationships in the input data.

[0055] Exemplarily, the preset model may include a pre-trained model, such as a pre-trained large model. The pre-trained model is trained on a large-scale general data set and has learned some general knowledge. In the above step S120, the training of the pre-trained model may be fine-tuning, that is, a training process for adjusting the performance of the pre-trained model in the table question and answer task or the table question and answer data set. Specifically, further adjustments can be made through a small number of samples of the table question and answer task to make the model output more in line with expectations.

[0056] Optionally, the training of the preset model may be supervised training, such as supervised fine-tuning (SFT).

[0057] In the embodiment of the present disclosure, the first question and answer information corresponding to the first table image may include questions and answers. Generally speaking, the table question and answer task is to give corresponding answers based on the input table image and question. Therefore, in the embodiment of the present disclosure, the data used to train the preset model to obtain the table question and answer model includes the first question and answer information to provide samples in the form of question-answer pairs to achieve supervised training. Exemplarily, the question can be any type of question, for example, the type of question can be information extraction type, numerical statistics type, comparison and sorting type, etc.

[0058] In an embodiment of the present disclosure, the data used to train the preset model to obtain a table question-answering model also includes style parameters. Exemplarily, the style parameters can be used to input the text branch of the preset model. Optionally, the preset model may include a text branch and a graph branch, and the style parameters and the question in the first question-answering information are input into the text branch, and the table image is input into the graph branch; or, the text branch and the graph branch may not be distinguished, and the style parameters, the question in the first question-answering information, and the table image are input into the preset model. The preset model outputs a predicted answer, and the loss can be calculated based on the predicted answer and the answer in the first question information, and the parameters of the preset model are updated based on the loss, and the training / fine-tuning of the preset model is completed through multiple rounds of iterations.

[0059] In practical applications, the number of first table images can be greater than or equal to 2, the fields and styles involved in each first table image can be different, and thus the style parameters of each first table image can be different. The preset model can be trained based on multiple different style parameters and the first question and answer information corresponding to each first table image, and a more generalized table question and answer model can be obtained. In addition, since the style parameters characterize the display style of the table image, the style parameters can help the model capture the impact of the display style on the table text, enhance the model's understanding and processing capabilities of the table text, and enable it to more accurately parse the data in the table. For example, financial statements and questionnaires often have different display styles, and the same text may have different meanings in financial statements and questionnaires. By introducing style parameters, the model's processing capabilities for table question and answer tasks in different scenarios can be enhanced.

[0060] According to the technical solution of the embodiment of the present disclosure, the preset model can be trained using the style parameters of the first table image and the first question and answer information corresponding to the first table image to obtain a table question and answer model that can more accurately understand and answer questions related to the table image, which not only improves the accuracy of the question and answer, but also improves the adaptability of the model to table images.

[0061] In some embodiments, step S110: determining the style parameters of the first table image includes:

[0062] Generate HTML (Hyper Text Markup Language) data corresponding to the first table image according to the structured data of the first table image;

[0063] Based on the style information in the HTML data, style parameters of the first table image are determined.

[0064] In the embodiment of the present disclosure, the structured data may be structured text data obtained by parsing the first table image, for example, each element in the text data has row and column information.

[0065] Exemplarily, the first table image may be analyzed using OCR (Optical Character Recognition) technology to convert the content in the first table image into structured text data, that is, to obtain structured data of the first table image.

[0066] In the disclosed embodiment, HTML data corresponding to the first table image can be generated by analyzing the row and column structure of the table in the structured data. HTML is a standard markup language used to create web pages, and is used to tell the browser how to display web page content, such as text, pictures, links, tables, etc. HTML data can also be called an HTML document, which includes a series of elements, which are defined by tags.

[0067] The HTML data may include style information. For example, the style information may include <style>元素,用于描述HTML数据中各元素的样式,如颜色、字体、间距、布局等。<style>元素通常放在头部(<head>部分)内。

[0068] 在本公开实施例中,基于HTML数据中的样式信息,确定第一表格图像的风格参数,例如可以在HTML数据中的<style>元素中进行选取,将选取到的元素作为风格参数。

[0069] 根据上述实施例,对第一表格图像解析后生成HTML数据,再基于HTML数据确定风格参数。由于HTML数据包含有表征显示样式的元素,因此可以直接提取到风格参数,提高获取风格参数的效率和准确性。

[0070] 在一些实施例中,步骤S120:基于第一表格图像的风格参数以及第一表格图像对应的第一问答信息,对预设模型进行训练,得到表格问答模型,包括:

[0071] 基于第一表格图像对应的第一问答信息,确定第一表格图像对应的任务类型;

[0072] 基于第一表格图像、任务类型、风格参数以及第一问答信息中的问题,得到预设模型的输入信息;

[0073] 基于输入信息,得到预设模型的输出信息;

[0074] 基于输出信息以及第一问答信息中的答案,对预设模型进行微调,得到表格问答模型。

[0075] 在本公开实施例中,第一表格图像对应的任务类型可以理解为第一问答信息中问题的类型,每个第一问答信息对应有一个任务类型。例如,任务类型可以包括信息抽取类型、比较和排序类型、数值统计类型、以及其他类型。

[0076] 示例性地,可以根据第一问答信息中的问题和 / 或答案,确定第一表格图像对应的任务类型。

[0077] 示例性地,可以基于第一表格图像对应的任务类型,确定任务类型对应的提示符,在SFT可以给预设模型输入对应的提示符作为引导信息,用来学习对应任务类型的训练数据的特征,以提升对应任务类型的问题的回答准确率。

[0078] 示例性地,第一表格图像可以对应有多个第一问答信息,从而第一表格图像可以对应有多个任务类型,不同的任务类型对应的提示符不同。

[0079] 在一个应用示例中,可以分别设置每个任务类型对应的提示符<Info Extraction>、<Compare>、<Count>以及<Others>用来区分信息抽取类型、比较和排序类型、数值统计类型以及其他类型,从而在SFT阶段可显式地给预设模型输入引导信息,用来学习不同任务类型的训练数据的特征,以提升表格问答模型对不同任务类型的问题的回答准确率,提高了表格问答模型的泛化能力。

[0080] 示例性地,在模型的微调阶段,可以基于任务类型和风格参数构造引导信息,并与第一表格图像、第一问答信息中的问题拼接,得到用于输入预设模型的输入信息。将该输入信息输入预设模型,可以得到预设模型的输出信息,利用输出信息第一问答信息中的答案之间的损失,对预设模型进行参数调整,通过多轮迭代,预设模型的参数收敛,从而该预设模型可以作为表格问答模型。

[0081] 根据上述实施例,通过利用第一表格图像、任务类型、风格参数以及第一问答信息对预设模型进行有监督的训练,可以得到适用于表格问答的模型,从而提高了表格问答的准确性。

[0082] 在一些实施例中,基于第一表格图像对应的第一问答信息,确定第一表格图像对应的任务类型,包括:

[0083] 在第一表格图像的结构化数据中检索第一问答信息中的答案;

[0084] 响应于在结构化数据中检索到答案,确定第一表格图像对应的任务类型为信息提取类型。

[0085] 第一表格图像的结构化数据中可以包括多个元素。可选地,结构化数据中各个元素可以与第一表格图像对应的表格的各个单元格一一对应。或者,结构化数据中的元素与第一表格图像中的文本的各个分词结果一一对应。在本公开实施例中,可以在结构化数据的多个元素中检索第一问答信息中的答案,当检索到多个元素中的其中一个元素的数据与答案相同时,可以确定任务类型为信息提取类型,当检索到每个元素与答案都不相同时,可以确定任务类型不为信息提取类型。

[0086] 信息提取类型可以理解为,第一问答信息中的问题为在第一表格图像中查找并提取指定的信息,在这种情况下,第一表格图像中可以包括第一问答信息中的答案,因此可以通过在第一表格图像的结构化数据中检索第一问答信息中的答案,来确定任务类型是否信息提取类型。

[0087] 根据本公开实施例的技术方案,可以利用第一表格图像的结构化数据以及第一问答信息中的答案,准确地确定信息提取类型的问题,从而基于准确的样本进行有监督微调,提升模型的准确性。

[0088] 在一些实施例中,基于第一表格图像对应的第一问答信息,确定第一表格图像对应的任务类型,还包括:

[0089] 响应于在结构化数据中检索不到答案,基于第一问答信息中的问题中的关键词,确定第一表格图像对应的任务类型。

[0090] 在本公开实施例中,当检索到结构化数据的每个元素与答案都不相同时,可以基于第一问答信息中的问题中的关键词,确定第一表格图像对应的任务类型。示例性地,可以解析关键词得到关键词的语义特征,可以基于关键词的语义特征,确定任务类型。

[0091] 实际应用中,预设模型可以具备自然语言处理(Natural Language Processing,NLP)的能力,通过预设模型可以解析第一问答信息中的问题,并提取问题中的关键词,进而基于关键词,确定第一表格图像对应的任务类型。当然,也可以通过利用其他模型或其他能力来提取问题中的关键词。

[0092] 根据本公开实施例的技术方案,在结构化数据中检索不到答案的情况下,利用第一问答信息中问题的关键词,准确地确定除了信息提取类型之外的任务类型,从而基于准确的样本进行有监督微调,提升模型的准确性。

[0093] 在一些实施例中,基于第一问答信息中的问题中的关键词,确定第一表格图像对应的任务类型,包括:

[0094] 在与多个任务类型一一对应的多个预设词集合中,检索第一问答信息中的问题中的关键词;

[0095] 响应于在多个预设词集合中的第一预设词集合中检索到关键词,确定第一表格图像对应的任务类型为第一预设词集合对应的任务类型。

[0096] 在本公开实施例中,可以预先配置多个预设词集合,每个预设词集合中可以包括至少一个词汇,并且每个预设词集合对应有一个任务类型,可以在多个预设词集合中检索关键词,当在某个预设词集合中检索到关键词时,可以确定第一表格图像对应的任务类型为该预设词集合对应的任务类型。

[0097] 示例性地,多个预设词集合中,与比较和排序类型对应的预设词集合可以包括:most(最)、more(更)、rank(排序)、between(之间)、difference(不同)、less(较少的)、compare(比较)、next(下一个)、after(在……之后)、first(第一位的、最重要的)、top(高于)等,当在该预设词集合中检索到关键词的情况下,可以确定第一表格图像对应的任务类型为比较和排序类型。

[0098] 示例性地,多个预设词集合中,与数值统计类型对应的预设词集合可以包括:howmany(多少)、number of(数量)、total(总数)、count(计算)等,当在该预设词集合中检索到关键词的情况下,可以确定第一表格图像对应的任务类型为数值统计类型。当在上述两个预设词集合中都没有检索到关键词的情况下,可以确定第一表格图像对应的任务类型为其他类型。

[0099] 根据本公开实施例的技术方案,可以利用多个预设词集合和问题中的关键词,通过关键词匹配的方式快速确定任务类型,从而可以提升模型训练的效率。

[0100] 在一些实施例中,第一表格图像的获取方式包括:

[0101] 通过网络收集或AI(Artificial Intelligence,人工智能)生成的方式,获取HTML数据;

[0102] 基于HTML数据进行渲染,得到第一表格图像。

[0103] 在本公开实施例中,可以通过网络收集或AI生成的方式,获取到多个HTML数据,每个HTML数据中可以包括的字体、颜色、布局和格式可以不同,进而可以渲染HTML数据,得到第一表格图像。可选地,第一表格图像为样本图像,实际应用中可以有多个第一表格图像,其中,一部分第一表格图像是通过基于HTML数据进行渲染得到的,该部分第一表格图像可以直接基于HTML数据获得风格参数;另一部分表格图像可以是收集得到的真实表格图像,可以通过生成对应的HTLM数据得到风格参数。

[0104] 示例性地,电子设备可以通过收集网络平台中的数据,获取HTML数据,也可以通过API调用AI,由AI生成HTML数据,通过上述两种渠道可以获取到不同风格和类型的HTML数据,HTML数据的数量可以达到百万数量以上。

[0105] 示例性地,可以使用渲染引擎将HTML数据渲染为第一表格图像,第一表格图像可以包括不同的字体、颜色、布局和格式,确保覆盖了手写表格和打印表格、横向表格和纵向表格、多级复杂表格等各种形式,从而构建了多样化的第一表格图像。

[0106] 根据本公开实施例的技术方案,通过渲染HTML数据,可以得到多样化的第一表格图像,克服实际应用中图像样本相比文本样本更难收集获取的问题,利用扩充的多样化的第一表格图像,对预设模型进行训练,提高模型对不同类型表格图像的泛化能力,确保其在处理各种表格图像时表现稳定。

[0107] 图2示出了本公开一实施例提供的问答方法。该方法可以应用于问答装置,该装置可以部署于电子设备中。电子设备例如是单机或多机的终端、服务器或其他处理设备。其中,终端可以为移动设备、PDA、手持设备、计算设备、车载设备、可穿戴设备等用户设备。在一些可能的实现方式中,该方法还可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。如图2所示,该方法可以包括:

[0108] 步骤S210:获取第二表格图像以及第二表格图像对应的问题;

[0109] 步骤S220:基于表格问答模型,对第二表格图像以及第二表格图像对应的问题进行处理,得到第二表格图像对应的答案;其中,表格问答模型是基于第一表格图像的风格参数以及第一表格图像对应的第一问答信息训练得到的。

[0110] 在本公开实施例中,第二表格图像可以为待处理的表格图像。示例性地,第二表格图像可以为任意领域中的表格图像,例如第二表格图像可以为财务报表、医疗记录表格、以及学生成绩表格等,第二表格图像也可以为任意形式的表格图像,例如第二表格图像可以为手写表格、打印表格、横向表格、纵向表格、多级表格等。

[0111] 在本公开实施例中,第二表格图像对应的问题可以理解为基于第二表格图像提出的问题。示例性地,问题可以为任意类型的问题,例如问题的类型可以为信息抽取类型、数值统计类型、以及比较和排序类型等。

[0112] 在本公开实施例中,可以利用前述任一实施例中的基于图像的模型训练方法训练得到的表格问答模型,对第二表格图像以及第二表格图像对应的问题进行处理,得到第二表格图像对应的答案。

[0113] 示例性地,在上述问答方法中所采用的表格问答模型,是基于第一表格图像的风格参数得到的,该风格参数可以是根据第一表格图像对应的HTML数据中的样式信息确定的,该HTML数据的获得方式可以参考前述实施例实现。

[0114] 示例性地,上述问答方法中所采用的表格问答模型,是基于第一表格图像、第一表格图像对应的任务类型、风格参数以及第一问答信息,对预设模型进行微调得到的。例如,基于第一表格图像、任务类型、风格参数以及第一问答信息中的问题,得到预设模型的输入信息;基于该输入信息,得到预设模型的输出信息,基于输出信息以及第一问答信息中的答案,对预设模型进行微调,得到表格问答模型。可选地,该任务类型的确定方式可以参考前述任一实施例实现,例如通过在结构化数据中检索第一问答信息中的答案、问题中的关键词等方式确定。

[0115] 示例性地,上述问答方法中所采用的表格问答模型,是基于第一表格图像训练得到的,该第一表格图像可以包括基于HTML数据进行渲染得到的图像,该HTML数据可以通过网络收集或AI生成的方式得到。

[0116] 可以理解,上述问答方法中的各技术细节,均可以参考前述任一实施例中对应的内容实现,或者说与前述任一实施例的基于图像的模型训练方法进行结合,所得到的技术方案均属于本公开保护范围内。为了简洁,在此不一一列举。

[0117] 根据本公开实施例的技术方案,可以利用基于第一表格图像的风格参数以及第一表格图像对应的第一问答信息训练得到的表格问答模型,对第二表格图像以及第二表格图像对应的问题进行处理,得到第二表格图像对应的答案,提高了问答的准确性,进而提高了用户体验。

[0118] 在一些实施例中,基于表格问答模型,对第二表格图像以及第二表格图像对应的问题进行处理,得到第二表格图像对应的答案,包括:

[0119] 在第二表格图像对应的问题为信息提取任务的问题的情况下,基于第二表格图像以及第二表格图像对应的问题,得到第一提示信息;其中,第一提示信息用于引导表格问答模型将第二表格图像处理为HTML数据,并基于HTML数据输出针对问题的答案;

[0120] 将第一提示信息输入表格问答模型,得到表格问答模型输出的第二表格图像对应的答案。

[0121] 在本公开实施例中,针对于信息提取任务的问题,采用多轮问答的方式,第一提示信息可以先引导表格问答模型解析第二表格图像,得到第二表格图像对应的HTML数据,然后将HTML数据作为下一轮的输入,表格问答模型可以基于第二表格图像对应的HTML数据以及第二表格图像对应的问题,输出第二表格图像对应的答案。

[0122] 示例性地,可以预先配置信息提取任务对应的预设词集合,通过在信息提取任务对应的预设词集合中,检索第二表格图像对应的问题中的关键词,响应于在信息提取任务对应的预设词集合中检索到关键词,可以确定问题对应的任务类型为信息提取任务。

[0123] 可以看到,由于信息提取类型的问题是要去精确的匹配到表格里面的某一个单元格的内容,即信息提取类型的问题对第二表格图像的准确性要求较高,因此第一提示信息会引导表格问答模型,先解析第二表格图像得到更准确的HTML数据,然后表格问答模型可以基于HTML数据以及第二表格图像对应的问题,输出的第二表格图像对应的答案,这种定制化的引导设计可以使得表格问答模型在回答问题时更加准确,显著提升了用户体验。

[0124] 在一些实施例中,基于表格问答模型,对第二表格图像以及第二表格图像对应的问题进行处理,得到第二表格图像对应的答案,包括:

[0125] 基于第二表格图像以及第二表格图像对应的问题,得到第二提示信息;其中,第二提示信息用于引导表格问答模型根据第二表格图像以及问题,输出针对问题的答案以及获得答案的逻辑推理过程;

[0126] 将第二提示信息输入表格问答模型,得到表格问答模型输出的第二表格图像对应的答案以及获得答案的逻辑推理过程。

[0127] 在本公开实施例中,采用思维链路(Chain of Though,CoT)方法,可以针对于任意任务类型的问题,得到第二提示信息,第二提示信息可以引导表格问答模型根据第二表格图像以及问题,输出针对问题的答案以及获得答案的逻辑推理过程,并可以逐步显示答案的逻辑推理过程以及针对问题的答案。

[0128] 示例性,可以根据问题对应的任务类型,确定答案的逻辑推理过程对应的固定格式,每个任务类型可以对应有不同的固定格式,可以按照固定格式显示对应的答案的逻辑推理过程。

[0129] 具体而言,可以在对表格问答模型的prompt(提示)中添加推理过程的指令(即第二提示信息):"你是一个功能强大的表格信息提取器,可以根据输入的CSV(Comma-Separated Values,逗号分隔值)表格内容和相应的问题来生成固定格式的精准的答案及逻辑推理过程。请根据问题给出逻辑推理过程以及对应的答案,按照|问题|答案|推理过程|的格式返回。”

[0130] 可以看到,本公开实施例中采用CoT方法,通过分步骤推理和回答问题,使用户在复杂问题的解答中获得更清晰和准确的答案。

[0131] 为了便于理解上述技术方案,图3提供了基于图像的模型训练方法和问答方法的整体流程的示意图,可以执行以下处理:

[0132] 步骤S310:将第一表格图像、任务类型、风格参数以及第一问答信息中的问题作为输入信息,输入至预设模型;

[0133] 步骤S320:利用预设模型对上述输入信息进行处理,得到预设模型的输出信息;

[0134] 步骤S330:基于输出信息以及第一问答信息中的答案,对预设模型进行微调,得到表格问答模型;

[0135] 步骤S340:将第二表格图像以及第二表格图像对应的问题作为输入信息,输入至表格问答模型;

[0136] 步骤S350:利用表格问答模型,对第二表格图像以及第二表格图像对应的问题进行处理,得到第二表格图像对应的答案。

[0137] 根据本公开的实施例,本公开还提供了一种基于图像的模型训练装置,图4示出了本公开一实施例提供的基于图像的模型训练装置的示意性框图,如图4所示,该装置包括:

[0138] 风格确定模块410,用于确定第一表格图像的风格参数;其中,风格参数用于表征第一表格图像的显示样式;

[0139] 模型训练模块420,用于基于第一表格图像的风格参数以及第一表格图像对应的第一问答信息,对预设模型进行训练,得到表格问答模型。

[0140] 在一些实施例中,风格确定模块410,还用于:

[0141] 根据第一表格图像的结构化数据,生成第一表格图像对应的HTML数据;

[0142] 基于HTML数据中的样式信息,确定第一表格图像的风格参数。

[0143] 在一些实施例中,如图4所示,模型训练模块420,包括:

[0144] 类型确定模块510,用于基于第一表格图像对应的第一问答信息,确定第一表格图像对应的任务类型;

[0145] 输入模块520,用于基于第一表格图像、任务类型、风格参数以及第一问答信息中的问题,得到预设模型的输入信息;

[0146] 输出模块530,用于基于输入信息,得到预设模型的输出信息;

[0147] 模型微调模块540,用于基于输出信息以及第一问答信息中的答案,对预设模型进行微调,得到表格问答模型。

[0148] 在一些实施例中,类型确定模块510,还用于:

[0149] 在第一表格图像的结构化数据中检索第一问答信息中的答案;

[0150] 响应于在结构化数据中检索到答案,确定第一表格图像对应的任务类型为信息提取类型。

[0151] 在一些实施例中,类型确定模块510,还用于:

[0152] 响应于在结构化数据中检索不到答案,基于第一问答信息中的问题中的关键词,确定第一表格图像对应的任务类型。

[0153] 在一些实施例中,类型确定模块510,还用于:

[0154] 在与多个任务类型一一对应的多个预设词集合中,检索第一问答信息中的问题中的关键词;

[0155] 响应于在多个预设词集合中的第一预设词集合中检索到关键词,确定第一表格图像对应的任务类型为第一预设词集合对应的任务类型。

[0156] 在一些实施例中,如图6所示,还包括表格获取模块610,表格获取模块610用于:

[0157] 通过网络收集或AI生成的方式,获取HTML数据;

[0158] 基于HTML数据进行渲染,得到第一表格图像。

[0159] 根据本公开的实施例,本公开还提供了一种问答装置,图7示出了本公开一实施例提供的问答装置的示意性框图,如图7所示,该装置包括:

[0160] 问题获取模块710,用于获取第二表格图像以及第二表格图像对应的问题;

[0161] 模型处理模块720,用于基于表格问答模型,对第二表格图像以及第二表格图像对应的问题进行处理,得到第二表格图像对应的答案;其中,表格问答模型是基于第一表格图像的风格参数以及第一表格图像对应的第一问答信息训练得到的。

[0162] 在一些实施例中,模型处理模块720,还用于:

[0163] 在第二表格图像对应的问题为信息提取任务的问题的情况下,基于第二表格图像以及第二表格图像对应的问题,得到第一提示信息;其中,第一提示信息用于引导表格问答模型将第二表格图像处理为HTML数据,并基于HTML数据输出针对问题的答案;

[0164] 将第一提示信息输入表格问答模型,得到表格问答模型输出的第二表格图像对应的答案。

[0165] 在一些实施例中,模型处理模块720,还用于:

[0166] 基于第二表格图像以及第二表格图像对应的问题,得到第二提示信息;其中,第二提示信息用于引导表格问答模型根据第二表格图像以及问题,输出针对问题的答案以及获得答案的逻辑推理过程;

[0167] 将第二提示信息输入表格问答模型,得到表格问答模型输出的第二表格图像对应的答案以及获得答案的逻辑推理过程。

[0168] 本公开实施例的装置的各模块、子模块的具体功能和示例的描述,可以参见上述方法实施例中对应步骤的相关描述,在此不再赘述。

[0169] 本公开的技术方案中,所涉及的用户个人信息的获取,存储和应用等,均符合相关法律法规的规定,且不违背公序良俗。

[0170] 根据本公开的实施例,本公开还提供了一种电子设备、一种可读存储介质和一种计算机程序产品。

[0171] 图8示出了可以用来实施本公开的实施例的示例电子设备800的示意性框图。电子设备旨在表示各种形式的数字计算机,诸如,膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置,诸如,个人数字助理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例,并且不意在限制本文中描述的和 / 或者要求的本公开的实现。

[0172] 如图8所示,设备800包括计算单元801,其可以根据存储在只读存储器(Read-OnlyMemory,ROM)802中的计算机程序或者从存储单元808加载到随机访问存储器(RandomAccess Memory,RAM)803中的计算机程序,来执行各种适当的动作和处理。在RAM 803中,还可存储设备800操作所需的各种程序和数据。计算单元801、ROM 802以及RAM 803通过总线804彼此相连。输入 / 输出(Input / Output,I / O)接口805也连接至总线804。

[0173] 设备800中的多个部件连接至I / O接口805,包括:输入单元806,例如键盘、鼠标等;输出单元807,例如各种类型的显示器、扬声器等;存储单元808,例如磁盘、光盘等;以及通信单元809,例如网卡、调制解调器、无线通信收发机等。通信单元809允许设备800通过诸如因特网的计算机网络和 / 或各种电信网络与其他设备交换信息 / 数据。

[0174] 计算单元801可以是各种具有处理和计算能力的通用和 / 或专用处理组件。计算单元801的一些示例包括但不限于中央处理单元(Central Processing Unit,CPU)、图形处理单元(Graphics Processing Unit,GPU)、各种专用的人工智能(ArtificialIntelligence,AI)计算芯片、各种运行机器学习模型算法的计算单元、数字信号处理器(Digital Signal Processor,DSP)、以及任何适当的处理器、控制器、微控制器等。计算单元801执行上文所描述的各个方法和处理,例如基于图像的模型训练方法和 / 或问答方法。例如,在一些实施例中,基于图像的模型训练方法和 / 或问答方法可被实现为计算机软件程序,其被有形地包含于机器可读介质,例如存储单元808。在一些实施例中,计算机程序的部分或者全部可以经由ROM 802和 / 或通信单元809而被载入和 / 或安装到设备800上。当计算机程序加载到RAM 803并由计算单元801执行时,可以执行上文描述的基于图像的模型训练方法和 / 或问答方法的一个或多个步骤。备选地,在其他实施例中,计算单元801可以通过其他任何适当的方式(例如,借助于固件)而被配置为执行基于图像的模型训练方法和 / 或问答方法。

[0175] 本文中以上描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、现场可编程门阵列(Field Programmable Gate Array,FPGA)、专用集成电路(Application Specific Integrated Circuit,ASIC)、专用标准产品(Application-Specific Standard Products,ASSP)、芯片上系统的系统(System on Chip,SOC)、负载可编程逻辑设备(Complex Programmable Logic Device,CPLD)、计算机硬件、固件、软件、和 / 或它们的组合中实现。这些各种实施方式可以包括:实施在一个或者多个计算机程序中,该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和 / 或解释,该可编程处理器可以是专用或者通用可编程处理器,可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令,并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。

[0176] 用于实施本公开的方法的程序代码可以采用一个或多个编程语言的任何组合来编写。这些程序代码可以提供给通用计算机、专用计算机或其他可编程数据处理装置的处理器或控制器,使得程序代码当由处理器或控制器执行时使流程图和 / 或框图中所规定的功能 / 操作被实施。程序代码可以完全在机器上执行、部分地在机器上执行,作为独立软件包部分地在机器上执行且部分地在远程机器上执行或完全在远程机器或服务器上执行。

[0177] 在本公开的上下文中,机器可读介质可以是有形的介质,其可以包含或存储以供指令执行系统、装置或设备使用或与指令执行系统、装置或设备结合地使用的程序。机器可读介质可以是机器可读信号介质或机器可读储存介质。机器可读介质可以包括但不限于电子的、磁性的、光学的、电磁的、红外的、或半导体系统、装置或设备,或者上述内容的任何合适组合。机器可读存储介质的更具体示例会包括基于一个或多个线的电气连接、便携式计算机盘、硬盘、随机存取存储器、只读存储器、可擦除可编程只读存储器(ErasableProgrammable Read-Only Memory,EPROM)、快闪存储器、光纤、便捷式紧凑盘只读存储器(Compact Disk Read Only Memory,CD-ROM)、光学储存设备、磁储存设备、或上述内容的任何合适组合。

[0178] 为了提供与用户的交互,可以在计算机上实施此处描述的系统和技术,该计算机具有:用于向用户显示信息的显示装置(例如,阴极射线管(Cathode Ray Tube,CRT)或者液晶显示器(Liquid Crystal Display,LCD)监视器);以及键盘和指向装置(例如,鼠标或者轨迹球),用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互;例如,提供给用户的反馈可以是任何形式的传感反馈(例如,视觉反馈、听觉反馈、或者触觉反馈);并且可以用任何形式(包括声输入、语音输入、或者触觉输入)来接收来自用户的输入。

[0179] 可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如,作为数据服务器)、或者包括中间件部件的计算系统(例如,应用服务器)、或者包括前端部件的计算系统(例如,具有图形用户界面或者网络浏览器的用户计算机,用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如,通信网络)来将系统的部件相互连接。通信网络的示例包括:局域网(Local Area Network,LAN)、广域网(Wide Area Network,WAN)和互联网。

[0180] 计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。服务器可以是云服务器,也可以为分布式系统的服务器,或者是结合了区块链的服务器。

[0181] 应该理解,可以使用上面所示的各种形式的流程,重新排序、增加或删除步骤。例如,本公开中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行,只要能够实现本公开公开的技术方案所期望的结果,本文在此不进行限制。

[0182] 上述具体实施方式,并不构成对本公开保护范围的限制。本领域技术人员应该明白的是,根据设计要求和其他因素,可以进行各种修改、组合、子组合和替代。任何在本公开的原则之内所作的修改、等同替换和改进等,均应包含在本公开保护范围之内。< / style>

Claims

1. An image-based model training method, comprising: Determining style parameters of the first table image; wherein the style parameters are used to characterize the display style of the first table image; Based on the style parameters of the first table image and the first question and answer information corresponding to the first table image, the preset model is trained to obtain a table question and answer model.

2. The method according to claim 1, wherein: The step of determining the style parameters of the first table image includes: Generate hypertext markup language HTML data corresponding to the first table image according to the structured data of the first table image; Based on the style information in the HTML data, style parameters of the first table image are determined.

3. The method according to claim 1 or 2, wherein: The step of training a preset model based on the style parameter of the first table image and the first question-answering information corresponding to the first table image to obtain a table question-answering model includes: Determining a task type corresponding to the first table image based on the first question and answer information corresponding to the first table image; Based on the first table image, the task type, the style parameter, and the question in the first question-and-answer information, obtaining input information of the preset model; Based on the input information, obtaining output information of the preset model; Based on the output information and the answers in the first question and answer information, the preset model is fine-tuned to obtain the table question and answer model.

4. The method according to claim 3, wherein: The determining, based on the first question and answer information corresponding to the first table image, the task type corresponding to the first table image includes: Retrieving an answer in the first question-and-answer information from the structured data of the first table image; In response to retrieving the answer from the structured data, it is determined that the task type corresponding to the first table image is an information extraction type.

5. The method according to claim 4, wherein: The determining, based on the first question and answer information corresponding to the first table image, the task type corresponding to the first table image further includes: In response to failing to retrieve the answer from the structured data, determining a task type corresponding to the first table image based on keywords in the question in the first question-and-answer information.

6. The method according to claim 5, wherein: The determining, based on the keywords in the question in the first question-and-answer information, the task type corresponding to the first table image includes: Retrieving keywords in the question in the first question-and-answer information from a plurality of preset word sets corresponding to a plurality of task types one by one; In response to retrieving the keyword in a first preset word set among the plurality of preset word sets, determining that the task type corresponding to the first table image is the task type corresponding to the first preset word set.

7. The method according to any one of claims 1 to 6, wherein: The method of acquiring the first table image includes: Obtain HTML data through network collection or AI generation; Rendering is performed based on the HTML data to obtain the first table image.

8. A question-answering method, comprising: Obtain a second table image and a question corresponding to the second table image; Based on the table question and answer model, the second table image and the question corresponding to the second table image are processed to obtain the answer corresponding to the second table image; wherein the table question and answer model is trained based on the style parameters of the first table image and the first question and answer information corresponding to the first table image.

9. The method according to claim 8, wherein: The table question-answering model based on the second table image and the question corresponding to the second table image are processed to obtain the answer corresponding to the second table image, including: In the case where the question corresponding to the second table image is a question of the information extraction task, first prompt information is obtained based on the second table image and the question corresponding to the second table image; wherein the first prompt information is used to guide the table question-answering model to process the second table image into HTML data, and output an answer to the question based on the HTML data; The first prompt information is input into the table question and answer model to obtain an answer corresponding to the second table image output by the table question and answer model.

10. The method according to claim 8, wherein: The table-based question-answering model processes the second table image and the question corresponding to the second table image to obtain an answer corresponding to the second table image, including: Based on the second table image and the question corresponding to the second table image, second prompt information is obtained; wherein the second prompt information is used to guide the table question-answering model to output an answer to the question and a logical reasoning process for obtaining the answer according to the second table image and the question; The second prompt information is input into the table question and answer model to obtain the answer corresponding to the second table image output by the table question and answer model and the logical reasoning process for obtaining the answer.

11. An image-based model training device, comprising: A style determination module, used to determine style parameters of the first table image; wherein the style parameters are used to characterize the display style of the first table image; The model training module is used to train the preset model based on the style parameters of the first table image and the first question and answer information corresponding to the first table image to obtain a table question and answer model.

12. The device according to claim 11, wherein The style determination module is further used for: generating HTML data corresponding to the first table image according to the structured data of the first table image; Based on the style information in the HTML data, style parameters of the first table image are determined.

13. The device according to claim 11 or 12, wherein: The model training module includes: a type determination module, configured to determine a task type corresponding to the first table image based on the first question and answer information corresponding to the first table image; An input module, configured to obtain input information of the preset model based on the first table image, the task type, the style parameter, and the question in the first question-and-answer information; An output module, used to obtain output information of the preset model based on the input information; A model fine-tuning module is used to fine-tune the preset model based on the output information and the answers in the first question and answer information to obtain the tabular question and answer model.

14. The device according to claim 13, wherein: The type determination module is further used for: Retrieving an answer in the first question-and-answer information from the structured data of the first table image; In response to retrieving the answer from the structured data, it is determined that the task type corresponding to the first table image is an information extraction type.

15. The device according to claim 14, wherein: The type determination module is further used for: In response to failing to retrieve the answer from the structured data, determining a task type corresponding to the first table image based on keywords in the question in the first question-and-answer information.

16. The device according to claim 15, wherein: The type determination module is further used for: Retrieving keywords in the question in the first question-and-answer information from a plurality of preset word sets corresponding to a plurality of task types one by one; In response to retrieving the keyword in a first preset word set among the plurality of preset word sets, determining that the task type corresponding to the first table image is the task type corresponding to the first preset word set.

17. The device according to any one of claims 11 to 16, wherein: It also includes a table acquisition module, which is used to: Obtain HTML data through network collection or AI generation; Rendering is performed based on the HTML data to obtain the first table image.

18. A question-answering device, comprising: A question acquisition module, used for acquiring a second table image and a question corresponding to the second table image; A model processing module is used to process the second table image and the question corresponding to the second table image based on a table question and answer model to obtain an answer corresponding to the second table image; wherein the table question and answer model is trained based on the style parameters of the first table image and the first question and answer information corresponding to the first table image.

19. The device according to claim 18, wherein The model processing module is also used for: In the case where the question corresponding to the second table image is a question of the information extraction task, first prompt information is obtained based on the second table image and the question corresponding to the second table image; wherein the first prompt information is used to guide the table question-answering model to process the second table image into HTML data, and output an answer to the question based on the HTML data; The first prompt information is input into the table question and answer model to obtain an answer corresponding to the second table image output by the table question and answer model.

20. The device according to claim 18, wherein The model processing module is also used for: Based on the second table image and the question corresponding to the second table image, second prompt information is obtained; wherein the second prompt information is used to guide the table question-answering model to output an answer to the question and a logical reasoning process for obtaining the answer according to the second table image and the question; The second prompt information is input into the table question and answer model to obtain the answer corresponding to the second table image output by the table question and answer model and the logical reasoning process for obtaining the answer.

21. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 10.

22. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-10.

23. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 10.

Citation Information

Cited By

  • MTU data decoupling formation and model training method, system and device and medium

    CN119832579A

  • Answer support system, answer support method, and answer support program

    JP7742475B1