Query task processing method, document question answering method, and information processing method based on processing model

Abstract

The present disclosure provides a query task processing method, a document question answering method, and an information processing method based on a processing model. The query task processing method is applied to the technical field of computers, and comprises: acquiring document query information of a target query task; on the basis of the document query information, screening out a target text block among a plurality of text blocks comprised in a target document; and using a text processing model and a visual processing model to process the document query information and the target text block, and obtaining a document query result of the target query task.

Description

Query task processing, document question answering, and information processing methods based on processing models

[0001] Cross-reference

[0002] This disclosure claims priority to Chinese Patent Application No. 2025100205300, filed on January 6, 2025, entitled "Query Task Processing, Document Question Answering and Information Processing Method Based on Processing Model", the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure relates to the field of computer technology, and in particular to query task processing, document question answering, and information processing methods based on processing models. Background Technology

[0004] With the development of computer technology, retrieval-enhanced generation has gradually become a research focus. Retrieval-enhanced generation combines information retrieval techniques (such as vector similarity search methods) to retrieve relevant information from external databases. It combines the original input query with the retrieved relevant information as input to the generative model, enabling the model to obtain more reference knowledge or real-time information. This overcomes the limitations of the model itself in terms of training time and allows the model to provide more accurate and evidence-based responses during the generation phase.

[0005] However, since generative models are trained on one-dimensional text sequences, their comprehension capabilities are typically poor. When faced with inputs containing complex document structures or graphical tables, generative models struggle to correctly answer user queries. Therefore, a comprehensive and accurate query task processing solution is urgently needed. Summary of the Invention

[0006] In view of this, the present disclosure provides a query task processing method. One or more embodiments of this specification also relate to a document question-and-answer method, an information processing method based on a processing model, a task platform, a query task processing device, a document question-and-answer device, an information processing device based on a processing model, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0007] According to a first aspect of this disclosure, a query task processing method is provided, comprising:

[0008] Retrieve document query information for the target query task;

[0009] Based on the document query information, the target text block is selected from multiple text blocks included in the target document;

[0010] By using text processing and visual processing models, document query information and target text blocks are processed to obtain document query results for the target query task.

[0011] According to a second aspect of this disclosure, a document question-and-answer method is provided, comprising:

[0012] The receiving terminal device sends a document related to the target document;

[0013] Based on the document question, filter out the target text block from multiple text blocks included in the target document;

[0014] Using text processing and visual processing models, the document question and target text block are processed to obtain the answer to the document question, and the answer is sent to the terminal device.

[0015] According to a third aspect of this disclosure, an information processing method based on a processing model is provided, applied to a task platform, comprising:

[0016] Receive model requests sent by terminal devices;

[0017] Based on the model request, a text processing model and a visual processing model are determined from multiple processing models. The text processing model and the visual processing model are used to query the execution process of the task processing method.

[0018] According to a fourth aspect of this disclosure, a task platform is provided, including a request interface and a response component;

[0019] The request interface is configured to receive model requests sent by terminal devices. The model request includes at least one of the following: the scene identifier of the target scene, the scene input data of the target scene, and the model specification parameters.

[0020] The response component is configured to determine the text processing model and the visual processing model from multiple processing models based on the model request. The text processing model and the visual processing model are used to query the execution process of the task processing method.

[0021] According to a fifth aspect of this disclosure, a query task processing apparatus is provided, comprising:

[0022] The first acquisition component is configured to acquire document query information for the target query task;

[0023] The first filtering component is configured to filter out target text blocks from multiple text blocks included in the target document based on document query information;

[0024] The first processing component is configured to use text processing models and visual processing models to process document query information and target text blocks to obtain document query results for the target query task.

[0025] According to a sixth aspect of this disclosure, a document question-answering apparatus is provided, comprising:

[0026] The first receiving component is configured to receive document questions sent by the terminal device in response to the target document;

[0027] The second filtering component is configured to filter out target text blocks from multiple text blocks included in the target document based on the document question.

[0028] The second processing component is configured to use text processing models and visual processing models to process the document question and the target text block, obtain the answer to the document question, and send the answer to the terminal device.

[0029] According to the seventh aspect of this disclosure, an information processing apparatus based on a processing model is provided, applied to a task platform, comprising:

[0030] The second receiving component is configured to receive model requests sent by the terminal device;

[0031] The component is configured to determine a text processing model and a visual processing model from multiple processing models based on model requests. The text processing model and the visual processing model are used to query the execution process of the task processing method.

[0032] According to an eighth aspect of this disclosure, a computing device is provided, comprising:

[0033] Memory and processor;

[0034] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the methods provided in the first, second, or third aspects described above.

[0035] According to a ninth aspect of this disclosure, an electronic device is provided, comprising:

[0036] The memory and processor are connected via a bus;

[0037] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the methods provided in the first, second, or third aspects described above.

[0038] According to a tenth aspect of this disclosure, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the methods provided in the first, second, or third aspects described above.

[0039] According to the eleventh aspect of this disclosure, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the methods provided in the first, second, or third aspects described above.

[0040] This specification provides a query task processing method according to one embodiment, comprising: obtaining document query information for a target query task; filtering target text blocks from multiple text blocks included in the target document based on the document query information; and processing the document query information and target text blocks using a text processing model and a visual processing model to obtain the document query results for the target query task. By parsing the text blocks of the target document and locating the target text blocks corresponding to the document query information, the accuracy of query task processing is improved. Furthermore, while using the text processing model to understand the text content in the target text blocks, the visual processing model also parses the complex document structures or non-text elements such as images and tables contained in the target text blocks, providing a more comprehensive query task processing service and further improving the comprehensiveness and accuracy of the document query results. Attached Figure Description

[0041] Figure 1 is a flowchart of a query task processing method provided in an embodiment of this specification;

[0042] Figure 2 is an architecture diagram of a query task processing system provided in one embodiment of this specification;

[0043] Figure 3 is a flowchart of a query task processing method provided in an embodiment of this specification;

[0044] Figure 4 is a flowchart of a query task processing method provided in one embodiment of this specification;

[0045] Figure 5 is a flowchart of another query task processing method provided in one embodiment of this specification;

[0046] Figure 6 is a flowchart of a document question-and-answer method provided in one embodiment of this specification;

[0047] Figure 7 is a flowchart of an information processing method based on a processing model provided in one embodiment of this specification;

[0048] Figure 8 is a schematic diagram of the structure of a task platform provided in one embodiment of this specification;

[0049] Figure 9 is a schematic diagram of a query task processing device provided in one embodiment of this specification;

[0050] Figure 10 is a schematic diagram of a document question-and-answer device provided in one embodiment of this specification;

[0051] Figure 11 is a schematic diagram of the structure of an information processing device based on a processing model provided in one embodiment of this specification;

[0052] Figure 12 is a structural block diagram of a computing device provided in one embodiment of this specification;

[0053] Figure 13 is a structural block diagram of an electronic device provided in one embodiment of this specification. Detailed Implementation

[0054] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0055] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0056] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0057] In addition, optionally, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data shall comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0058] In one or more embodiments of this specification, a large model refers to a deep learning model with a large number of model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even tens of trillions of model parameters. A large model can also be called a foundation model. It is pre-trained on a large-scale unlabeled corpus to produce a pre-trained model with hundreds of millions of parameters. Such models can adapt to a wide range of downstream tasks and have good generalization ability; examples include large-scale language models and multi-modal pre-training models.

[0059] In practical applications, large models only require a small number of samples to fine-tune the pre-trained model before they can be applied to different tasks. Large models can be widely used in fields such as Natural Language Processing (NLP) and Computer Vision. Specifically, they can be applied to computer vision tasks such as Visual Question Answering (VQA), Image Captioning (IC), and Image Generation, as well as NLP tasks such as text-based sentiment classification, text summarization, and machine translation. The main application scenarios for large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design.

[0060] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0061] Retrieval-Augmented Generation (RAG) refers to the process of combining information retrieval techniques (such as vector similarity search methods) to retrieve relevant information from external databases, combining the original input query with the retrieved relevant text fragments, and using this combination as input to the generative model. This allows the generative model to obtain more reference knowledge or real-time information, thereby overcoming the limitations of the model itself in terms of information during training time, and enabling the model to provide more accurate and evidence-based responses during the generation stage.

[0062] Document parsing refers to the process of converting electronic documents of different formats into processable plain text, thereby making it easier to perform operations such as searching, statistical analysis, and information extraction.

[0063] Text slicing refers to the process of dividing long text into smaller paragraphs or segments. Text slicing allows each segment to serve as an independent query unit. This enables more efficient searching and matching of text segments relevant to the input query during the retrieval phase.

[0064] Vectorized storage refers to the process of converting text slices into vectors and storing them in a vector database. By converting text content into vectors, vector similarity search methods (such as cosine similarity, Euclidean distance, etc.) can be used to improve retrieval accuracy. This method is better at capturing semantic similarity than traditional keyword search.

[0065] Vector retrieval refers to the process of using vector similarity search methods to find several documents or fragments that are most relevant to the input query from a pre-built vector database.

[0066] Optical Character Recognition (OCR) is a technology that extracts and converts printed or handwritten text from paper documents, images, or other physical media into digital text. Its purpose is to enable computers to read and understand text content by recognizing characters in images.

[0067] Large Language Models (LLMs) are artificial intelligence models trained on massive amounts of text data, designed to generate and understand human language. They typically have billions or even hundreds of billions of parameters and can generate high-quality text output, understand contextual relationships, and perform natural language processing tasks.

[0068] Visual Language Models (VLMs) are models trained using both visual and linguistic modalities to understand the relationship between the two. They can process multimodal data, such as images and text, to perform various joint visual and linguistic tasks.

[0069] Chain of Thought (CoT): Its basic concept is to generate a series of step-by-step reasoning steps through a model to arrive at the final answer. This approach helps improve the performance of models on complex tasks because it allows the model to "think" or "reason" about the process when generating the answer, rather than just outputting the result.

[0070] Markup languages ​​are programming languages ​​used for text markup. Typical markup languages ​​include HyperText Markup Language (HTML), LaTeX, eXtensible Markup Language (XML), and Markdown. They define the structure and format of text by inserting specific symbols or tags. These tags are typically used to control how text is displayed, its formatting, and its functionality.

[0071] Markdown is a lightweight markup language that allows people to write documents in an easy-to-read and easy-to-write plain text format, and then convert them into structured HTML web pages.

[0072] Deep self-attention (Transformer) models are network structures based on multi-head self-attention mechanisms, primarily used for processing sequential data. A Transformer model consists of repeatedly stacked encoder and decoder units. This design allows the Transformer to efficiently learn long-term dependencies, making it suitable for various natural language processing tasks, including machine translation, text summarization, and question answering systems.

[0073] Bidirectional Encoder Representations from Transformers (BERT) is a pre-trained NLP model. By learning from large amounts of unlabeled text data, this model can capture deep semantic information from text and achieves significant performance improvements on numerous NLP tasks.

[0074] The Text-to-Text Transfer Transformer (T5) model is a type of NLP model. The main characteristic of the T5 model is that it unifies all NLP tasks into a text-to-text format, meaning both the input and output are text sequences. This design makes the model more adaptable to various tasks, such as translation, question answering, and summarization.

[0075] Contrastive Language–Image Pre-training (CLIP) models are multimodal deep learning models that are pre-trained using a large number of text-image pairs. They employ a contrastive learning framework to map images and text into the same high-dimensional embedding space. This allows CLIP to understand image content and generate corresponding text descriptions, or to find the most relevant images based on text queries. The power of CLIP lies in its zero-shot transfer learning capability, meaning it can be applied to unseen tasks or categories without additional fine-tuning. It is widely applicable to various cross-modal tasks such as image classification, image retrieval, and image caption generation.

[0076] The Vision-and-Language Transformer (ViLT) model is a multimodal deep learning model. Based on the Transformer architecture, ViLT directly inputs images and text at the pixel level, eliminating the need for traditional region feature extraction steps and significantly reducing computational complexity. By jointly training image and text data, it achieves excellent performance on various vision-language tasks, such as visual question answering and image-text matching. ViLT's design simplifies the vision-language processing flow, improves model efficiency and performance, making it a powerful tool for handling cross-modal tasks.

[0077] Referring to Figure 1, which is a flowchart of a query task processing method provided in one embodiment of this specification, the query task processing method is applied to a query system, which includes a document parsing component, a text slicing and vectorization storage component, a vector retrieval component, and a generation component. The processing procedures of these components will be described below. A user uploads a document, which is parsed by the document parsing component to obtain the parsing results. The text slicing and vectorization storage component performs text slicing and vectorization storage on the parsed text results. Next, the vector retrieval component performs vector retrieval based on the user's query. Finally, the generation model generates a response based on the retrieval results from the vector retrieval component, obtaining the query results. The generation component often relies on the LLM's ability to understand the original query and the retrieved relevant text fragments. Although LLMs possess strong understanding and reasoning capabilities when faced with text input, because they are trained on one-dimensional text sequences, their understanding capabilities are often insufficient when faced with inputs containing complex document structures or graphical tables, resulting in the LLM failing to fully extract the effective information from the retrieval results, thus making it difficult to correctly answer the user's query.

[0078] To address the aforementioned issues, this disclosure proposes a query task processing scheme combining LLM and VLM. The process involves: acquiring document query information for the target query task; filtering target text blocks from multiple text blocks included in the target document based on the document query information; and processing the document query information and target text blocks using text processing and visual processing models to obtain the document query results for the target query task. By combining the powerful understanding and reasoning capabilities of LLM when faced with text input with the ability of VLM to understand complex visual and layout information, this scheme comprehensively covers complex query tasks and improves the end-to-end RAG (Research and Development Environment) link performance.

[0079] This specification provides a query task processing method, and also relates to a document question-and-answer method, an information processing method based on a processing model, a task platform, a query task processing device, a document question-and-answer device, an information processing device based on a processing model, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0080] Considering the large number of model parameters in text processing and visual processing models, and the limited computing resources of terminal devices, the query task processing proposed in this disclosure can be applied to the query task processing system shown in Figure 1, but is not limited thereto. Referring to Figure 2, Figure 2 shows an architecture diagram of a query task processing system provided in one embodiment of this specification. The query task processing system may include a terminal device 202 and a server 204.

[0081] Terminal device 202 is configured to send document query information for the target query task to server 204;

[0082] Server 204 is configured to filter out the target text block from multiple text blocks included in the target document based on the document query information; process the document query information and the target text block using a text processing model and a visual processing model to obtain the document query result for the target query task; and send the document query result to terminal device 202.

[0083] Terminal device 202 is also used to receive document query results sent by server 204.

[0084] As shown in Figure 2, the text processing model and the visual processing model are deployed in the server 204. The server 204 can connect to one or more terminal devices 202 via a local area network (LAN), a wide area network (WAN), the Internet, or other types of data networks. Terminal devices 202 may include, but are not limited to, smartphones, tablets, laptops, PDAs, personal computers, smart home devices, and in-vehicle devices. Terminal devices 202 can also interact with users through a graphical user interface to invoke the text processing model and the visual processing model, thereby implementing the query task processing method provided in this disclosure.

[0085] Optionally, the query task processing method provided in this disclosure is generally executed by the server. However, in other embodiments of this specification, if the terminal device's operating resources can meet the deployment and operating conditions of the text processing model and the visual processing model, the terminal device may also have similar functions to the server, thereby executing the query task processing method provided in this disclosure. In other embodiments, the query task processing method provided in this disclosure may also be executed jointly by the terminal device and the server. Next, taking the server executing the query task processing method proposed in this disclosure as an example, the query task processing method will be described in detail.

[0086] Referring to Figure 3, Figure 3 shows a flowchart of a query task processing method provided in an embodiment of this specification, which specifically includes the following steps:

[0087] Step 302: Obtain the document query information for the target query task.

[0088] Optionally, a target query task refers to a task set by the user or system to find specific information from a target document based on document query information. The target query task can be a task in different scenarios, such as a legal information query task in a legal scenario, a knowledge question-and-answer task in an academic scenario, etc. Document query information refers to the set of information associated with the target query task. Document query information represents the query intent of the target query task. Document query information can be a complete document question or a set of parameters used to guide and limit the query process, such as query conditions, query scope, and expected result format, etc., which are selected according to the actual situation, and this disclosure does not impose any limitations on them. Document query information can be in text format or in markup language format. In practical applications, there are multiple ways to obtain the document query information of the target query task, and the specific method is selected according to the actual situation, and this disclosure does not impose any limitations on them. In one possible implementation of this specification, the document query information of the target query task sent by the user through a terminal device can be received. In another possible implementation of this specification, the document query information of the target query task can be read from other databases or data acquisition devices.

[0089] Step 304: Based on the document query information, filter out the target text block from the multiple text blocks included in the target document.

[0090] Optionally, the target document refers to the document that the target query task seeks to retrieve, or the document to which the document query information is targeted. A target document can include various types of elements, such as text, images, and tables. Text refers to the plain text content in the target document, such as natural language text, code, and text data in tables. Images refer to visual content embedded in the target document, such as photographs, illustrations, charts, and icons. Tables are a way to organize and display data, consisting of rows and columns. Tables can contain various types of data, such as text, numbers, and formulas. Table structures help express complex relationships, such as financial statements and experimental results. A target document typically includes multiple text blocks. A text block refers to a relatively independent part of the content in the target document. A text block can be a paragraph, chapter, cell in a table, or other text units with logical integrity. In the query task processing, text blocks are the basic units for analysis and retrieval. A target text block refers to at least one text block selected from multiple text blocks based on the document query information that is most likely to contain the information the user needs; that is, a text block with a high degree of matching with the document query information.

[0091] In practical applications, there are various ways to filter target text blocks from multiple text blocks included in a target document based on document query information. The specific method should be selected according to the actual situation, and this disclosure does not impose any limitations on it. In one possible implementation of this specification, the document query information and each text block can be matched by text recall to obtain multiple text matching results, wherein each text matching result corresponds one-to-one with each text block; and the target text block is filtered out from the multiple text blocks based on the text matching results.

[0092] In another possible implementation of this specification, target text blocks can be selected from multiple text blocks included in the target document based on document query information using vector recall. That is, the above-mentioned selection of target text blocks from multiple text blocks included in the target document based on document query information may include the following steps:

[0093] Multiple text block features are obtained, where each text block feature corresponds one-to-one with a text block, and the text block features are obtained based on feature extraction of the text blocks;

[0094] Feature extraction is performed on document query information to obtain document query features;

[0095] The document query features are matched with multiple text block features to obtain multiple matching results, where each matching result corresponds one-to-one with a text block.

[0096] Based on multiple matching results, the target text block is selected from multiple text blocks.

[0097] Optionally, text block features are a set of numerical values ​​obtained after analyzing and extracting features from text blocks. These values ​​reflect the key characteristics and structure of the text blocks. Feature extraction refers to the process of selecting or calculating numerical values ​​or attributes that can represent the essential characteristics of the data from the original data. When performing feature extraction on document query information, feature extraction refers to identifying and quantifying key elements that help understand and locate the user's query intent. Document query features refer to a set of specific attributes or indicators obtained through the feature extraction process. Document query features can effectively characterize the core elements of document query information. Document query features can be word distribution, semantic vector representation, etc., in the text content, or structured information extracted from visual elements (such as table layout, image description, etc.). The matching result is used to describe the degree of feature matching between the document query features and any text block features. The matching result can be a specific matching value, such as 90%, or a matching level, such as very good match, fairly good match, no match, etc., which is selected according to the actual situation, and this disclosure does not impose any limitations on it.

[0098] In practical applications, there are various ways to obtain features from multiple text blocks, and the specific method chosen depends on the actual situation. This disclosure does not impose any limitations on this approach. One possible implementation of this specification involves acquiring multiple text blocks stored in an information database and extracting features from each text block in real time to obtain multiple text block features. Another possible implementation of this specification involves reading multiple text block features from other databases or data acquisition devices.

[0099] There are various ways to extract features from document query information and obtain document query features. The specific method should be selected according to the actual situation, and this disclosure does not impose any limitations on it. In one possible implementation of this specification, the text query features of the document query information can be generated based on the term frequency-inverse document frequency (TF-IDF) method. In another possible implementation of this specification, the document query features of the document query information can be generated based on semantics using a pre-trained language model.

[0100] There are multiple ways to match document query features with multiple text block features to obtain multiple matching results. The specific method should be selected according to the actual situation, and this disclosure does not impose any limitations on this. In one possible implementation of this specification, the document query features and any text block features can be input into a vector retrieval model to obtain the matching results between the text block features and the document query features output by the vector retrieval model. In another possible implementation of this specification, the matching results between the text block features and each document query feature can be calculated using similarity calculation methods, including but not limited to cosine similarity calculation methods and Euclidean distance calculation methods.

[0101] Furthermore, there are multiple ways to filter target text blocks from multiple text blocks based on multiple matching results. The specific method should be selected according to the actual situation, and this disclosure does not impose any limitations on this. In one possible implementation of this specification, multiple text blocks can be sorted in descending order of matching degree based on the matching results, and the top N text blocks in the sorted order can be determined as target text blocks, where N is a positive integer. In another possible implementation of this specification, text blocks whose matching results meet preset matching conditions can be determined as target text blocks, where preset matching conditions include matching value greater than a preset matching threshold, matching level higher than, and relatively good match.

[0102] By applying the scheme disclosed herein, target text blocks are selected from multiple text blocks through vector recall based on the text query features of document query information and the text block features of each text block, thereby improving the selection efficiency and accuracy of target text blocks.

[0103] In one optional embodiment of this specification, the process of filtering the target text block from multiple text blocks based on multiple matching results may include the following steps:

[0104] The vector query engine is invoked to sort multiple text blocks based on multiple matching results, and the target text block is selected from the sorted text blocks.

[0105] Alternatively, a vector query engine refers to a tool or system that performs searches based on matching results. It works by filtering target text blocks from multiple text blocks based on the matching results, selecting those blocks that highly match the document query features.

[0106] In practical applications, there are various ways to sort multiple text blocks based on multiple matching results. The specific method should be chosen according to the actual situation, and this disclosure does not impose any limitations on it. In one possible implementation of this specification, multiple text blocks can be sorted in descending order of matching degree based on the matching results. In another possible implementation of this specification, multiple text blocks can be sorted in ascending order of matching degree based on the matching results.

[0107] Furthermore, when selecting the target text block from the sorted text blocks, if the sorting is based on the degree of matching from high to low, the top N text blocks in the sorted order can be identified as the target text block. If the sorting is based on the degree of matching from low to high, the bottom N text blocks in the sorted order can be identified as the target text block.

[0108] By applying the scheme disclosed herein, a vector query engine is used to filter out several target text blocks that have a high similarity to document query features, thereby enabling efficient and accurate acquisition of target text blocks.

[0109] In one optional embodiment of this specification, before filtering the target text block from the multiple text blocks included in the target document based on document query information, the multiple text blocks included in the target document can be obtained. In practical applications, there are various ways to obtain the multiple text blocks included in the target document. In one possible implementation of this specification, the multiple text blocks included in the target document can be read from other databases or data acquisition devices.

[0110] In another possible implementation of this specification, the target document can be parsed and segmented to obtain multiple text blocks. That is, before filtering the target text block from the multiple text blocks included in the target document based on the document query information, the following steps may also be included:

[0111] Retrieve the target document corresponding to the document query information;

[0112] Parse the target document to obtain its text.

[0113] The target document text is divided into text blocks to obtain the multiple text blocks contained in the target document.

[0114] Optionally, text block segmentation can be understood as text splitting. The target document, or target document, and the process of obtaining its text can be called document parsing. The target document text refers to the processable plain text corresponding to the target document; it can be in text format or markup language format.

[0115] In practical applications, when retrieving the target document corresponding to document query information, a search engine or database query tool can be used to find the target document matching the document query information in a specified document library. The search process can be carried out in various ways, such as full-text indexing and metadata matching. The process of parsing the target document and obtaining the target document text can be carried out in various ways, such as visual processing model parsing, DOM parsing, document parsing libraries (such as PyPDF2 and python-docx in Python), and OCR parsing.

[0116] Furthermore, there are various ways to divide the target document text into text blocks to obtain the multiple text blocks contained in the target document. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations on this approach. In one possible implementation of this specification, the target document text can be divided into multiple text blocks based on natural delimiters (such as paragraphs, sentences, chapters, etc.) in the target document text. In another possible implementation of this specification, a deep learning model can be used to understand the semantic structure of the target document text and divide the target document text into multiple semantically complete text blocks.

[0117] By applying the scheme disclosed herein, multiple text blocks are obtained by parsing the target document and segmenting it into text blocks, thereby dividing the target document into multiple independent query units. This enables more efficient searching and matching of text fragments related to the input query during the query phase.

[0118] In one optional embodiment of this specification, the above-described method of dividing the target document text into text blocks to obtain multiple text blocks comprising the target document may include the following steps:

[0119] Based on a preset sliding window, the target document text is divided into text blocks to obtain multiple text blocks included in the target document. Among these multiple text blocks, there is content overlap between adjacent text blocks.

[0120] Optionally, a preset sliding window refers to a fixed-size "window" used to view or process a continuous data stream segment by segment. The preset sliding window can slide across the data in steps, thus enabling traversal of the entire dataset. The size of the preset sliding window is smaller than the size of the text block, meaning that each new window contains part of the content from the previous window along with some new content, resulting in overlapping text blocks.

[0121] For example, assuming that 512 tokens in the target document text are taken as a text block, and the preset sliding window is 448, then each text block obtained by dividing the target document text into text blocks according to the preset sliding window will contain the last 64 tokens of the previous text block.

[0122] By applying the solution disclosed herein, the target document text is divided into text blocks using a preset sliding window, thereby enabling content overlap between adjacent text blocks and ensuring the contextual continuity of the text block content.

[0123] Step 306: Use text processing models and visual processing models to process the document query information and target text blocks to obtain the document query results for the target query task.

[0124] Optionally, a text processing model refers to a computational model used to understand and analyze target text blocks. Text processing models can perform tasks such as part-of-speech tagging, named entity recognition, and semantic analysis on the text content of the target text block, helping to understand and extract information from the text. Text processing models can be large-scale language models, such as the BERT model and the T5 model. A visual processing model refers to a computational model used to parse and understand the visual data corresponding to the target text block. Visual processing models can be visual language models, such as the CLIP model and the ViLT model.

[0125] Visual data includes visual elements such as images and charts, as well as the layout and structure between visual and text elements. Document query results refer to the final output obtained after processing the document query information and the target text block. Document query results are typically partial document content or information summaries related to the document query information, directly responding to the user's query needs.

[0126] In practical applications, there are multiple ways to process document query information and target text blocks using text processing models and visual processing models to obtain document query results for the target query task. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations on this approach. In one possible implementation of this specification, the text processing model can determine whether it can generate document query results, and based on the determination result, choose to have either the text processing model or the visual processing model generate the document query results. In another possible implementation of this specification, the text processing model and the visual processing model can generate query results in parallel, and then the document query result is determined based on the query results generated by the two models respectively.

[0127] By applying the solution disclosed herein, text block parsing of the target document is performed to locate the target text block corresponding to the document query information, thereby improving the accuracy of query task processing. Furthermore, while utilizing a text processing model to understand the text content within the target text block, a visual processing model is also used to parse the complex document structure or non-text elements such as images and tables contained within the target text block, providing a more comprehensive query task processing service and further improving the comprehensiveness and accuracy of document query results. Experiments have shown that the query task processing method proposed in this disclosure improves the document query performance from 0.72 to 0.79, achieving a 7 percentage point improvement.

[0128] In one optional embodiment of this specification, a thought chain can guide the text processing model to reflect on the selected target text block. If the text processing model determines that it cannot process the text query information based on the provided information, then routing enables the visual processing model to generate document query results based on the visual data associated with the target text block. Conversely, if the text processing model determines that it can process the text query information based on the provided information, then the text processing model generates document query results based on the target text block. That is, the above-mentioned use of the text processing model and the visual processing model to process the document query information and the target text block to obtain the document query results for the target query task may include the following steps:

[0129] Input the model validation prompts, document query information, and target text blocks into the text processing model to obtain the model validation results;

[0130] When the model validation results indicate that the visual processing model is used, the visual processing model is used to process the document query information and the target text block to obtain the document query results for the target query task.

[0131] Optionally, model validation prompts are used to guide the text processing model to verify whether it can find relevant content in the target text block that supports answering the document query information. For example, the model validation prompt could be, "In the following reference document {target text block}, if you do not find relevant information that supports answering the question {document query information}, please reply: 'Unable to answer'." The specific model validation prompts are selected according to the actual situation, and this disclosure does not impose any limitations on them. The model validation result refers to the result obtained by the text processing model in verifying whether it can process text query information based on the target text block. The model validation result can be either "unable to answer" or "able to answer." "Unable to answer" indicates that the text processing model believes it cannot process the text query information based on the target text block; in this case, the model validation result indicates that the query task will be processed by the visual processing model. "Able to answer" indicates that the text processing model believes it can process the text query information based on the target text block; in this case, the model validation result indicates that the query task will be processed by the text processing model.

[0132] In practical applications, there are various ways to process document query information and target text blocks using a visual processing model to obtain document query results for the target query task. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations. In one possible implementation, visual data corresponding to the target text block can be obtained; the document query information and visual data are then input into the visual processing model to obtain the document query results for the target query task. In another possible implementation, the target document can be converted into a target document image, and the target document image can be stored as multimodal image features using a multimodal encoder. Subsequently, a multimodal retrieval system recalls the target image features associated with the target text block, and the document query information and target image features are input into the visual processing model to obtain the document query results for the target query task.

[0133] By applying the solution disclosed herein, the text processing model is guided to think further through a thought chain approach, rather than directly generating document query results. This ensures that the visual processing model is only invoked when needed, keeping the query task processing time within a controllable range. At the same time, it combines the advantages of both text processing and visual processing models, achieving a significant performance improvement in the end-to-end RAG chain.

[0134] In one optional embodiment of this specification, the above-described method of using a visual processing model to process document query information and target text blocks to obtain document query results for the target query task may include the following steps:

[0135] Obtain the visual data corresponding to the target text block;

[0136] By inputting document query information and visual data into the visual processing model, the document query results for the target query task are obtained.

[0137] Optionally, the visual data corresponding to the target text block refers to non-text elements such as images, icons, tables, and symbols directly associated with the target text block. Visual data can be visual data located near or embedded in the target text block in the target document, or it can be visual data that includes the target text block, such as the document page where the target text block is located.

[0138] In practical applications, there are various ways to obtain the visual data corresponding to the target text block, and the specific method should be selected according to the actual situation. This disclosure does not impose any limitations on this method. In one possible implementation of this specification, the metadata of the target text block can be obtained, and the corresponding visual data can be linked through the metadata. Metadata refers to data that is directly related to the text content of the target text block but is independent of its main content. Metadata can provide various additional information about the target text block, including but not limited to the filename, document type, and page number of the text block. In another possible implementation of this specification, the text description information of the candidate visual data can be obtained, and the text description information of the target text block and each candidate visual data can be matched to filter out the visual data corresponding to the target text block from the candidate visual data based on the matching results.

[0139] By applying the solution disclosed herein, visual data corresponding to the target text block is introduced during the query task processing. The visual processing model is then used to jointly process the text and visual data, thereby capturing the complex relationship between the two and improving the accuracy of document query results.

[0140] In one optional embodiment of this specification, after inputting the model validation prompts, document query information, and target text blocks into the text processing model to obtain the model validation results, the following steps may be further included:

[0141] If the model validation results indicate that the text processing model is used, the text processing model is used to process the document query information and the target text block to obtain the document query results for the target query task.

[0142] In practical applications, there are various ways to process document query information and target text blocks using a text processing model to obtain the document query results for the target query task. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations. In one possible implementation of this specification, the document query information and target text block can be directly input into the text processing model to obtain the document query results for the target query task. In another possible implementation of this specification, since the text content in the target text block may be long, the document query information and the target key information in the target text block can be input into the text processing model to obtain the text query results for the target query task.

[0143] By applying the solution disclosed herein, when the text processing model can process document query information based on the target text block, the text processing model can be used directly to generate text query results, thereby improving the efficiency of the query task processing process.

[0144] In one optional embodiment of this specification, the above-described text processing model for processing document query information and target text blocks to obtain document query results for the target query task may include the following steps:

[0145] Extract key information from the target text block to obtain key target information;

[0146] Input the document query information and target key information into the text processing model to obtain the document query results for the target query task.

[0147] Optionally, key information extraction refers to the process of identifying and extracting information fragments from a target text block that represent its core content or important attributes. Target key information refers to the specific information fragments obtained after key information extraction. Target key information can reflect the core content of the target text block. Target key information includes, but is not limited to, entities (such as names of people and places), relationships (such as cause and effect of events), and values ​​(such as dates and amounts) in the target text block.

[0148] In practical applications, there are various ways to extract key information from target text blocks, and the specific method chosen depends on the actual situation. This disclosure does not impose any limitations on this approach. One possible implementation of this specification involves using regular expressions to extract key information from the target text block. These regular expressions are used to identify information conforming to a specific format, such as phone numbers and email addresses. Another possible implementation utilizes neural network models, such as recurrent neural networks, long short-term memory networks, and bidirectional encoders, to gain a deeper understanding of the contextual information of the target text block and extract more accurate key information.

[0149] By applying the scheme disclosed herein, document query information and target key information are input into a text processing model to obtain document query results for the target query task. While ensuring the quality of the target text block, the text input of the text processing model is made more concise, thereby improving the processing efficiency of the text processing model.

[0150] Referring to Figure 4, which shows a flowchart of a query task processing method according to an embodiment of this specification, the query task processing method proposed in this disclosure includes five stages: document parsing, text slicing and vectorization, vector retrieval, text processing model judgment, and model processing. These five stages will now be described in detail.

[0151] Document parsing: Retrieves the target document corresponding to the document query information; parses the target document to obtain the target document text.

[0152] Text slicing and vectorization: The target document text is divided into text blocks to obtain multiple text blocks. Feature extraction is performed on each text block to obtain the text block features of each text block. The multiple text blocks, the text block features of each text block, and metadata are stored in a vector database.

[0153] Vector retrieval: Extract features from document query information to obtain document query features; match the document query features with features of multiple text blocks to obtain multiple matching results, where each matching result corresponds one-to-one with a text block; and select the target text block from the multiple text blocks based on the multiple matching results.

[0154] Text processing model judgment: The model validation prompts, document query information, and target text block are input into the text processing model. The text processing model then uses its own understanding capabilities to determine whether it can answer the document query based on the target text block. Typically, this stage involves an either-or choice: either the text processing model or the visual processing model generates the document query results.

[0155] Model Processing: If yes, extract key information from the target text block to obtain the target key information; input the document query information and the target key information into the text processing model for processing to obtain the document query results for the target query task. If no, associate the target metadata of the target text block with the corresponding visual data; input the document query information and visual data into the visual processing model for processing to obtain the document query results for the target query task.

[0156] By applying the solution disclosed herein, and combining the respective advantages of text processing models and multimodal visual processing models in handling different document types, the text processing model can be specifically designed for documents containing rich text content, while the visual processing model can be specifically designed for documents with complex document structures or complex charts, thus improving the comprehensiveness of query task processing. Furthermore, the visual processing model is only invoked when needed, ensuring that the processing time remains within a controllable range.

[0157] In one optional embodiment of this specification, the above-described processing of document query information and target text blocks using text processing models and visual processing models to obtain document query results for the target query task may include the following steps:

[0158] By using a text processing model, the document query information and target text block are processed to obtain the first query result;

[0159] By using a visual processing model, the document query information and target text block are processed to obtain the second query result;

[0160] Based on the first and second query results, generate the document query results for the target query task.

[0161] Optionally, the first query result refers to the result obtained by the text processing model after processing the document query information and the target text block based on its own text understanding capabilities. The second query result refers to the result obtained by the visual processing model after processing the corresponding visual data of the document query information and the target text block based on its own multimodal data processing capabilities.

[0162] In practical applications, the implementation method of "using a text processing model to process document query information and target text blocks to obtain the first query result" can refer to the above implementation method of "using a text processing model to process document query information and target text blocks to obtain the document query result of the target query task"; the implementation method of "using a visual processing model to process document query information and target text blocks to obtain the second query result" can refer to the above implementation method of "using a visual processing model to process document query information and target text blocks to obtain the document query result of the target query task", and will not be elaborated further in this disclosure.

[0163] By applying the solution disclosed herein, document query results are generated based on the first and second query results from different models, thereby fully utilizing the advantages of both models to provide more comprehensive and accurate query task processing services, and improving the accuracy and reliability of document query results.

[0164] In practical applications, there are multiple ways to generate document query results for the target query task based on the first query result and the second query result. The specific method to be selected depends on the actual situation, and this disclosure does not impose any restrictions on it.

[0165] In one possible implementation of this specification, document query results can be filtered from the first query results and the second query results. That is, generating document query results for the target query task based on the first query results and the second query results may include the following steps:

[0166] The confidence level of the first query result is evaluated to obtain the first confidence level index, and the confidence level of the second query result is evaluated to obtain the second confidence level index.

[0167] Based on the first confidence index and the second confidence index, the document query results for the target query task are filtered from the first query results and the second query results.

[0168] Optionally, confidence assessment refers to the process of evaluating the reliability of the model's output. Confidence assessment measures the system's confidence in the model's predictions or answers. The first confidence index is the result obtained by assessing the confidence of the first query result. The first confidence index is used to quantify the reliability of the first query result. The second confidence index is the result obtained by assessing the confidence of the second query result. The second confidence index is used to quantify the reliability of the second query result. Both the first and second confidence indices are typically expressed numerically (e.g., probability values ​​between 0 and 1), with higher values ​​indicating higher confidence, i.e., higher reliability.

[0169] In practical applications, there are multiple ways to assess the confidence level of the first query result and obtain the first confidence index. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations on this approach. In one possible implementation, the first query result and document query information can be input into a text processing model to obtain the first confidence index output by the text processing model. In another possible implementation, preset keywords can be obtained, and the first confidence index can be determined based on the number of preset keywords included in the first query result; the more preset keywords, the higher the first confidence index. The preset keywords are set based on domain knowledge and are words with high reliability in the current domain. The implementation method for "assessing the confidence level of the second query result and obtaining the second confidence index" can refer to the implementation method for "assessing the confidence level of the first query result and obtaining the first confidence index," and will not be elaborated upon here.

[0170] Furthermore, there are multiple ways to filter the document query results for the target query task from the first and second query results based on the first and second confidence indices. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations on this. In one possible implementation of this specification, the first and second confidence indices can be compared, and the query result with the larger confidence index can be determined as the document query result. In another possible implementation of this specification, a confidence index threshold can be obtained, and the first and second confidence indices can be compared, with the query result having a confidence index greater than the threshold determined as the document query result. If the confidence indices of both the first and second query results are greater than the confidence index threshold, then any query result is randomly selected as the document query result. If the confidence indices of both the first and second query results are less than the confidence index threshold, then the step of filtering the target text block from multiple text blocks included in the target document based on the document query information is returned, and the query task is processed again. The confidence index threshold is set according to the actual situation, and this disclosure does not impose any restrictions on it.

[0171] By applying the scheme disclosed herein, the document query results for the target query task are filtered from the first query results and the second query results based on the first confidence index of the first query results and the second confidence index of the second query results, thus ensuring the accuracy of the document query results.

[0172] In another possible implementation of this specification, the document query result can be obtained by merging the first query result and the second query result. That is, generating the document query result for the target query task based on the first query result and the second query result may include the following steps:

[0173] The first and second query results are input into a text processing model for result fusion to obtain the document query results for the target query task.

[0174] Optionally, result fusion refers to the process of combining the results of the first and second queries to form a comprehensive and optimized document query result. Result fusion aims to fully utilize the advantages of each result to improve the accuracy and comprehensiveness of the final output.

[0175] For example, suppose the target document is a legal document containing text and diagrams (such as a payment flowchart). The document query information is "What are the payment terms stipulated in this contract?" The first query result generated by the text processing model is "Party A shall pay Party B's service fee within the first five working days of each month," and the second query result generated by the visual processing model is "Payment cycle: once a month; first payment date: December 1, 2024." The first and second query results are input into the text processing model and fused to obtain the document query result: "Party A shall pay Party B's service fee within the first five working days of each month, with a payment cycle of once a month and a first payment date of December 1, 2024."

[0176] By applying the solution disclosed herein, the first query result and the second query result are input into a text processing model for result fusion to obtain the document query result for the target query task. This achieves the fusion of processing results from different models, making the document query result a comprehensive and optimized result, providing a more comprehensive and accurate query task processing service, and improving the accuracy and reliability of the document query result.

[0177] Referring to Figure 5, which shows a flowchart of another query task processing method provided in one embodiment of this specification, the query task processing method proposed in this disclosure includes four stages: document parsing, text slicing and vectorization, vector retrieval, and model processing. These four stages will now be described in detail.

[0178] Document parsing: Retrieves the target document corresponding to the document query information; parses the target document to obtain the target document text.

[0179] Text slicing and vectorization: The target document text is divided into text blocks to obtain multiple text blocks. Feature extraction is performed on each text block to obtain the text block features of each text block. The multiple text blocks, the text block features of each text block, and metadata are stored in a vector database.

[0180] Vector retrieval: Extract features from document query information to obtain document query features; match the document query features with features of multiple text blocks to obtain multiple matching results, where each matching result corresponds one-to-one with a text block; and select the target text block from the multiple text blocks based on the multiple matching results.

[0181] Model Processing: Key information is extracted from the target text block to obtain key target information. The document query information and the key target information are input into a text processing model for processing to obtain the first query result. The target metadata of the target text block is linked to the corresponding visual data. The document query information and visual data are input into a visual processing model for processing to obtain the second query result. Based on the first and second query results, the document query result for the target query task is generated.

[0182] By applying the scheme disclosed herein, the outputs from different models (text processing model for text processing and visual processing model for visual understanding) are combined in an ensemble manner to generate comprehensive and optimized document query results. This fully leverages the advantages of both models, providing more comprehensive and accurate query task processing services and improving the accuracy and reliability of document query results.

[0183] The following description, in conjunction with Figure 6, uses the application of the query task processing method provided in this specification in a document question-and-answer scenario as an example to further illustrate the query task processing method. Figure 6 shows a flowchart of a document question-and-answer method provided in one embodiment of this specification, specifically including the following steps:

[0184] Step 602: Receive the document question sent by the terminal device for the target document.

[0185] Step 604: Based on the document question, filter out the target text block from the multiple text blocks included in the target document.

[0186] Step 606: Using text processing and visual processing models, process the document question and target text block to obtain the answer to the document question, and send the answer to the terminal device.

[0187] Optionally, a document question refers to a question or query request raised by a user on a terminal device regarding a target document. A document question can be a detailed question, a summary question, an explanation question, etc., concerning the document content. The answer to the question is the final output obtained after processing the document question and the target text block. The implementation methods of steps 602 to 606 can refer to the implementation methods of steps 302 to 304 above, and will not be repeated here.

[0188] By applying the solution disclosed herein, text block parsing of the target document is performed to locate the target text block corresponding to the document question, thereby improving the accuracy of document question answering. Furthermore, while utilizing a text processing model to understand the text content within the target text block, a visual processing model is also used to parse the complex document structure or non-text elements such as images and tables contained within the target text block, providing a more comprehensive document question answering service and further improving the comprehensiveness and accuracy of the question answers.

[0189] Referring to Figure 7, which shows a flowchart of an information processing method based on a processing model according to an embodiment of this specification, the information processing method based on the processing model is applied to a task platform and specifically includes the following steps:

[0190] Step 702: Receive the model request sent by the terminal device.

[0191] Step 704: Based on the model request, determine the text processing model and the visual processing model from multiple processing models. The text processing model and the visual processing model are used to query the execution process of the task processing method.

[0192] Optionally, the text processing model and the visual processing model are processing models suitable for the target scene. The model request includes at least one of the following: the scene identifier of the target scene, the scene input data of the target scene, and model specification parameters. There are multiple ways to determine the text processing model and the visual processing model from multiple processing models based on the model request; the specific method is selected according to the actual situation, and this disclosure does not impose any limitations on this. In one possible implementation of this specification, the corresponding text processing model and visual processing model can be searched from at least one processing model included in the model library based on the model request; in another possible implementation of this specification, the text processing model and the visual processing model can be trained and obtained based on the model request; in yet another optional implementation of this specification, the text processing model and the visual processing model can be constructed based on the model request.

[0193] For example, based on the scene identifier of the target scene, at least one pre-trained processing model can be searched from the model library. Then, based on the model specification parameters, an initial processing model can be selected from the at least one processing model. Finally, based on the scene input data of the target scene, the selected initial processing model can be trained to obtain a text processing model and a visual processing model suitable for user needs. The training process of the processing model can be obtained according to the training method of the processing model shown in Figure 2, which will not be described in detail in this disclosure.

[0194] By applying the solution disclosed herein, text processing models and visual processing models can be obtained according to user needs, realizing personalized model services and providing users with an efficient, flexible and easy-to-use model service method, thereby improving the user experience.

[0195] In one optional embodiment of this specification, the model request includes a scene identifier of the target scene; the process of determining the text processing model and the visual processing model from multiple processing models based on the model request may include the following steps:

[0196] Based on the scene identifier of the target scene, the text processing model and visual processing model suitable for the target scene are searched from the model library. The model library stores multiple processing models suitable for different query processing scenarios.

[0197] Optionally, a scenario identifier refers to a unique or specific label used to distinguish different query task scenarios. The model library is a database for storing and managing various pre-trained deep learning models. Multiple processing models adapted to different query task scenarios cover different query application scenarios and needs. The model library allows users to select the appropriate model according to their needs, or directly call the model for query task processing through the application programming interface.

[0198] Multiple processing models adapted to different query task scenarios are stored in the model library. Each model is optimized for a specific application environment. For example, based on the scenario identifier "document question answering" of the target scenario, the model library can be used to find text processing models and visual processing models suitable for the document question answering scenario.

[0199] By applying the solution disclosed herein, based on scenario requirements, the appropriate text processing model and visual processing model can be accurately found through scenario identification, making query task processing more accurate and more scenario-appropriate, thereby improving user experience and query task processing quality.

[0200] In one optional embodiment of this specification, the model request includes scene input data of the target scene; the determination of the text processing model and the visual processing model from multiple processing models based on the model request may include the following steps:

[0201] From multiple processing models, an initial text processing model and an initial visual processing model suitable for the target scenario are determined.

[0202] Based on the scene input data of the target scene, the initial text processing model and the initial visual processing model are trained to obtain the text processing model and the visual processing model.

[0203] Optionally, the scenarios to which each processing model is adapted may differ. For example, processing model one is suitable for scenarios one and two, while processing model two is suitable for scenarios two and three. The initial processing model refers to the model among multiple processing models that is suitable for the target scenario. If the target scenario is scenario one, then the initial processing model is processing model one adapted for scenario one. The initial processing model may not only be applicable to the target scenario but also to other scenarios, making it a general processing model applicable to different scenarios. The initial processing model can be used for query task processing, but the results may not be very good. In this case, the initial processing model can be optimized based on the scenario input data of the target scenario. For example, optimizing the initial processing model based on the scenario input data of a document question-answering scenario can yield text processing and visual processing models suitable for document question-answering scenarios. The scenario input data of the target scenario can be understood as the model training data for sample query tasks in the target scenario.

[0204] By applying the solution disclosed herein, based on scenario requirements, a general initial processing model is further trained using scenario input data to obtain a text processing model and a visual processing model adapted to the scenario. This makes the text processing model and the visual processing model more closely aligned with the scenario, thereby improving the user experience and the processing quality of the target query task.

[0205] In one optional embodiment of this specification, the model request includes model specification parameters; the process of determining the text processing model and the visual processing model from multiple processing models based on the model request may include the following steps:

[0206] Based on the model specification parameters, the corresponding text processing model and visual processing model are searched from the model library. The model library stores multiple processing models with different model specification parameters.

[0207] Optionally, model specifications refer to various parameters that define the model's structure and behavior. These parameters can be broadly categorized into two types: model parameters (learnable parameters) and hyperparameters. Model parameters are those automatically adjusted during model training via backpropagation, including but not limited to weight matrices and biases. For example, in a simple fully connected layer, the weight matrix is ​​a two-dimensional tensor connecting neurons in the input and output layers; the biases are one-dimensional vectors providing additional offset values ​​for each output neuron. Hyperparameters are parameters set before model training begins, controlling the model's learning process and architecture. Hyperparameters include, but are not limited to, the learning rate and the number of neurons per layer, selected based on specific circumstances.

[0208] By applying the solution disclosed herein, based on the model specification parameters, the corresponding text processing model and visual processing model can be accurately found, ensuring the efficient and stable operation of the text processing model and visual processing model and improving the user experience.

[0209] In one optional embodiment of this specification, after determining the text processing model and the visual processing model from multiple processing models based on model requests, the following steps may be further included:

[0210] Deploy text processing and visual processing models, and build query task processing interfaces based on the text processing and visual processing models so that terminal devices can schedule the text processing and visual processing models to execute target query tasks.

[0211] Optionally, the query task processing interface is an interactive programming interface for the terminal device to schedule text processing models and visual processing models to perform target query task processing, typically provided in the form of an application programming interface (API). Through the query task processing interface, users can input task data for the target query task, such as document questions sent for a target document, to perform document question-and-answer task processing.

[0212] In practical applications, there are various ways to deploy text processing models and visual processing models, and the specific method chosen depends on the actual situation. This disclosure does not impose any limitations on this approach. One possible implementation of this specification involves using infrastructure provided by a cloud service provider to deploy the text processing models and visual processing models on cloud-side devices. Another possible implementation involves using a lightweight framework to deploy the text processing models and visual processing models on edge devices. For example, the text processing models and visual processing models can be deployed on a distributed system, and a query task processing interface can be built based on these models and provided to terminal devices, enabling the terminal devices to schedule the text processing models and visual processing models to execute target query tasks.

[0213] By applying the solution provided in this disclosure, deploying text processing models and visual processing models, and building query task processing interfaces based on the text processing models and visual processing models, terminal devices can efficiently call the text processing models and visual processing models, thereby improving the processing quality and response speed of target query tasks.

[0214] Referring to Figure 8, Figure 8 shows a schematic diagram of the structure of a task platform provided in one embodiment of this specification. The task platform 800 includes a request interface 802 and a response component 804.

[0215] Request interface 802 is configured to receive model requests sent by terminal devices, wherein the model request includes at least one of the following: scene identifier of the target scene, scene input data of the target scene, and model specification parameters.

[0216] Response component 804 is configured to determine the text processing model and the visual processing model from multiple processing models based on the model request. The text processing model and the visual processing model are used to query the execution process of the task processing method.

[0217] In one optional embodiment of this specification, the task platform further includes a query task processing interface, which is constructed based on a text processing model and a visual processing model.

[0218] The query task processing interface is configured to allow terminal devices to schedule and execute target query tasks.

[0219] By applying the solution disclosed herein, the task platform can acquire text processing models and visual processing models according to user needs, realize personalized model services, provide users with an efficient, flexible and easy-to-use model service platform, and improve user experience.

[0220] The above is an illustrative scheme of a task platform according to this embodiment. Optionally, the technical solution of this task platform and the technical solution of the information processing method based on the processing model described above belong to the same concept. For details not described in detail in the technical solution of the task platform, please refer to the description of the technical solution of the information processing method based on the processing model described above.

[0221] Corresponding to the above-described query task processing method embodiments, this specification also provides a query task processing device embodiment. Figure 9 shows a schematic diagram of the structure of a query task processing device provided in one embodiment of this specification. As shown in Figure 9, the device includes:

[0222] The first acquisition component 902 is configured to acquire document query information for the target query task;

[0223] The first filtering component 904 is configured to filter out target text blocks from multiple text blocks included in the target document based on document query information.

[0224] The first processing component 906 is configured to use a text processing model and a visual processing model to process document query information and target text blocks to obtain document query results for the target query task.

[0225] Optionally, the first processing component 906 is further configured to input the model verification prompt information, document query information, and target text block into the text processing model to obtain the model verification result; if the model verification result indicates that the visual processing model is used for processing, the visual processing model is used to process the document query information and target text block to obtain the document query result of the target query task.

[0226] Optionally, the first processing component 906 is further configured to acquire visual data corresponding to the target text block; input the document query information and visual data into the visual processing model to obtain the document query result of the target query task.

[0227] Optionally, the device further includes a third processing component configured to, when the model validation result indicates that the text processing model has been used for processing, utilize the text processing model to process the document query information and the target text block to obtain the document query result for the target query task.

[0228] Optionally, the first processing component 906 is further configured to extract key information from the target text block to obtain target key information; input the document query information and the target key information into the text processing model to obtain the document query results of the target query task.

[0229] Optionally, the first processing component 906 is further configured to use a text processing model to process the document query information and the target text block to obtain a first query result; use a visual processing model to process the document query information and the target text block to obtain a second query result; and generate a document query result for the target query task based on the first query result and the second query result.

[0230] Optionally, the first processing component 906 is further configured to evaluate the confidence level of the first query result to obtain a first confidence index, and evaluate the confidence level of the second query result to obtain a second confidence index; and to filter the document query results of the target query task from the first query result and the second query result based on the first confidence index and the second confidence index.

[0231] Optionally, the first processing component 906 is further configured to input the first query result and the second query result into a text processing model for result fusion to obtain the document query result of the target query task.

[0232] Optionally, the first filtering component 904 is further configured to: acquire multiple text block features, wherein each text block feature corresponds one-to-one with a text block, and the text block features are obtained based on feature extraction of the text blocks; extract features from the document query information to obtain document query features; match the document query features with the multiple text block features to obtain multiple matching results, wherein each matching result corresponds one-to-one with a text block; and filter out the target text block from the multiple text blocks based on the multiple matching results.

[0233] Optionally, the first filtering component 904 is further configured to invoke a vector query engine, sort multiple text blocks based on multiple matching results, and filter out the target text block from the sorted multiple text blocks.

[0234] Optionally, the device further includes: a second acquisition component configured to acquire a target document corresponding to the document query information; parse the target document to obtain the target document text; and divide the target document text into text blocks to obtain multiple text blocks included in the target document.

[0235] Optionally, the second acquisition component is further configured to divide the target document text into text blocks according to a preset sliding window, and obtain multiple text blocks included in the target document, wherein there is content overlap between adjacent text blocks in the multiple text blocks.

[0236] By applying the solution disclosed herein, text block parsing of the target document is performed to locate the target text block corresponding to the document query information, thereby improving the accuracy of query task processing. Furthermore, while utilizing a text processing model to understand the text content within the target text block, a visual processing model is also used to parse the complex document structure or non-text elements such as images and tables contained within the target text block, providing a more comprehensive query task processing service and further improving the comprehensiveness and accuracy of document query results.

[0237] The above is an illustrative scheme of a query task processing device according to this embodiment. Optionally, the technical solution of this query task processing device and the technical solution of the query task processing method described above belong to the same concept. For details not described in detail in the technical solution of the query task processing device, please refer to the description of the technical solution of the query task processing method described above.

[0238] Corresponding to the above-described document question-and-answer method embodiments, this specification also provides document question-and-answer device embodiments. Figure 10 shows a schematic diagram of the structure of a document question-and-answer device provided in one embodiment of this specification. As shown in Figure 10, the device includes:

[0239] The first receiving component 1002 is configured to receive a document question sent by the terminal device for the target document;

[0240] The second filtering component 1004 is configured to filter out target text blocks from multiple text blocks included in the target document based on the document question.

[0241] The second processing component 1006 is configured to use a text processing model and a visual processing model to process the document question and the target text block, obtain the answer to the document question, and send the answer to the terminal device.

[0242] By applying the solution disclosed herein, text block parsing of the target document is performed to locate the target text block corresponding to the document question, thereby improving the accuracy of document question answering. Furthermore, while utilizing a text processing model to understand the text content within the target text block, a visual processing model is also used to parse the complex document structure or non-text elements such as images and tables contained within the target text block, providing a more comprehensive document question answering service and further improving the comprehensiveness and accuracy of the question answers.

[0243] The above is an illustrative scheme of a document question-and-answer device according to this embodiment. Optionally, the technical solution of this document question-and-answer device belongs to the same concept as the technical solution of the document question-and-answer method described above. For details not described in detail in the technical solution of the document question-and-answer device, please refer to the description of the technical solution of the document question-and-answer method described above.

[0244] Corresponding to the above-described embodiments of information processing methods based on processing models, this specification also provides embodiments of information processing apparatus based on processing models. Figure 11 shows a schematic diagram of the structure of an information processing apparatus based on a processing model provided in one embodiment of this specification. As shown in Figure 11, the apparatus is applied to a task platform and includes:

[0245] The second receiving component 1102 is configured to receive model requests sent by the terminal device;

[0246] Component 1104 is configured to determine a text processing model and a visual processing model from multiple processing models based on a model request. The text processing model and the visual processing model are used to query the execution process of the task processing method.

[0247] Optionally, the model request includes a scene identifier of the target scene; the determining component 1104 is further configured to search for text processing models and visual processing models adapted to the target scene from the model library based on the scene identifier of the target scene, wherein the model library stores multiple processing models adapted to different query processing scenarios.

[0248] Optionally, the model request includes scene input data of the target scene; the determining component 1104 is further configured to determine an initial text processing model and an initial visual processing model adapted to the target scene from multiple processing models; and to train the initial text processing model and the initial visual processing model based on the scene input data of the target scene to obtain the text processing model and the visual processing model.

[0249] Optionally, the model request includes model specification parameters; the determining component 1104 is further configured to search for corresponding text processing models and visual processing models from the model library based on the model specification parameters, wherein the model library stores multiple processing models with different model specification parameters.

[0250] Optionally, the device further includes: a deployment component configured to deploy a text processing model and a visual processing model, and based on the text processing model and the visual processing model, to build a query task processing interface so that the terminal device can schedule the text processing model and the visual processing model to execute the target query task.

[0251] By applying the solution disclosed herein, text processing models and visual processing models can be obtained according to user needs, realizing personalized model services and providing users with an efficient, flexible and easy-to-use model service method, thereby improving the user experience.

[0252] The above is an illustrative scheme of an information processing device based on a processing model according to this embodiment. Optionally, the technical solution of this information processing device based on a processing model belongs to the same concept as the technical solution of the information processing method based on a processing model described above. For details not described in detail in the technical solution of the information processing device based on a processing model, please refer to the description of the technical solution of the information processing method based on a processing model described above.

[0253] Figure 12 shows a structural block diagram of a computing device 1200 provided in one embodiment of this specification.

[0254] The computing device 1200 includes: a memory 1210 and a processor 1220;

[0255] The memory 1210 is used to store computer programs / instructions, and the processor 1220 is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor 1220, they implement the steps of the above-mentioned query task processing method, document question and answer method, or information processing method based on processing model.

[0256] In one or more embodiments of this specification, the computing device can be understood as an integrated smart terminal, including but not limited to a server, desktop computer, personal computer (PC), all-in-one model machine, mobile phone, tablet computer or other portable smart terminal, etc., and the computing device may have the model described in the above embodiments of this disclosure pre-installed.

[0257] Specifically, this computing device can pre-install various types of models, including but not limited to models in natural language processing, visual processing, speech processing, code processing, and multimodal task processing, thus providing diverse model selection. In different product forms, this computing device can support one or more model usage methods, including but not limited to model training, model invocation, model fine-tuning, model deployment, model inference, and application. In some product forms, this computing device also supports model management, including but not limited to multi-type model management (supporting the management of discriminative, generative, and other types of models), model version control (supporting the control of different model versions), and model evaluation (evaluating model performance and effectiveness based on model evaluation tools). In other product forms, this computing device can also create applications based on models, providing Application Programming Interface (API) invocation capabilities. Models can be invoked into created applications through the API interface, and application management tools are provided for application management and monitoring.

[0258] Furthermore, the computing device may also include data management (supporting the creation and management of model tuning datasets), a training center (providing abundant training resources to help users learn and master artificial intelligence technology), and basic control capabilities (providing enterprise-level basic control capabilities to ensure the security and efficient operation of the system). Through the above functions, it provides a comprehensive and integrated device for artificial intelligence development, training, deployment, and application.

[0259] Figure 13 shows a structural block diagram of an electronic device 1300 provided according to an embodiment of this specification.

[0260] The memory 1310 and the processor 1320 are connected via a bus 1330;

[0261] The memory 1310 is used to store computer programs / instructions, and the processor 1320 is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor 1320, they implement the steps of the above-mentioned query task processing method, document question-and-answer method, or information processing method based on the processing model.

[0262] Specifically, the components of the electronic device 1300 include, but are not limited to, a memory 1310 and a processor 1320. The processor 1320 and the memory 1310 can be connected via a bus 1330.

[0263] Electronic device 1300 may also include access device 1340, which enables electronic device 1300 to communicate with database 1350 storing data via one or more networks 1360. Examples of such networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. Access device 1340 may include any type of wired or wireless network interface (e.g., one or more Network Interface Cards (NICs), such as IEEE 802.11 Wireless Local Area Networks (WLAN) wireless interface, World Interoperability for Microwave Access (Wi-MAX) interface, Ethernet interface, Universal Serial Bus (USB) interface, cellular network interface, Bluetooth interface, Near Field Communication (NFC) interface, etc.

[0264] In one embodiment of this specification, the aforementioned components of the electronic device 1300, as well as other components not shown in FIG. 13, may also be connected to each other, for example, via a bus. It should be understood that the block diagram of the electronic device shown in FIG. 13 is merely for illustrative purposes and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0265] Electronic device 1300 can be any type of stationary or mobile electronic device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable electronic devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary electronic devices such as desktop computers or PCs. Electronic device 1300 can also be a mobile or stationary electronic device.

[0266] The above is an illustrative scheme of an electronic device according to this embodiment. Optionally, the technical solution of this electronic device belongs to the same concept as the above-described query task processing method, document question answering method, and information processing method based on processing model. For details not described in detail in the technical solution of the electronic device, please refer to the description of the technical solution of the above-described query task processing method, document question answering method, or information processing method based on processing model.

[0267] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the above-described query task processing method, document question-and-answer method, or information processing method based on a processing model.

[0268] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. Optionally, the technical solution of this storage medium belongs to the same concept as the technical solutions of the query task processing method, document question answering method, and information processing method based on processing model described above. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solutions of the query task processing method, document question answering method, or information processing method based on processing model described above.

[0269] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described query task processing method, document question-and-answer method, or information processing method based on a processing model.

[0270] The above is an illustrative scheme of a computer program product according to this embodiment. Optionally, the technical solution of this computer program product belongs to the same concept as the technical solutions of the query task processing method, document question answering method, and information processing method based on processing model described above. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solutions of the query task processing method, document question answering method, or information processing method based on processing model described above.

[0271] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0272] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. Optionally, the content included in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0273] Optionally, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and components involved are not necessarily essential to this disclosure.

[0274] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0275] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this disclosure. These embodiments have been selected and specifically described in this specification to better explain the principles and practical applications of this disclosure, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents. Industrial applicability

[0276] The solution provided in this disclosure can be applied to the process of task querying. By parsing text blocks in the target document, the target text block corresponding to the document query information can be located from the target document, thereby improving the accuracy of query task processing. While using a text processing model to understand the text content in the target text block, a visual processing model is used to parse the non-text elements contained in the target text block, providing a more comprehensive query task processing service and further improving the comprehensiveness and accuracy of document query results.

Claims

1. A query task processing method, comprising: Retrieve document query information for the target query task; Based on the document query information, the target text block is selected from multiple text blocks included in the target document; By using text processing models and visual processing models, the document query information and the target text block are processed to obtain the document query results for the target query task.

2. The method according to claim 1, wherein processing the document query information and the target text block using a text processing model and a visual processing model to obtain the document query result for the target query task includes: The model validation prompt, the document query information, and the target text block are input into the text processing model to obtain the model validation result. In response to the model validation result, the visual processing model is used to process the document query information and the target text block to obtain the document query result of the target query task.

3. The method according to claim 2, wherein processing the document query information and the target text block using the visual processing model to obtain the document query result of the target query task includes: Obtain the visual data corresponding to the target text block; The document query information and the visual data are input into the visual processing model to obtain the document query results for the target query task.

4. The method according to claim 2, after inputting the model validation prompt information, the document query information, and the target text block into the text processing model to obtain the model validation result, further includes: In response to the model validation result, the text processing model is used to process the document query information and the target text block to obtain the document query result of the target query task.

5. The method according to claim 4, wherein processing the document query information and the target text block using the text processing model to obtain the document query result of the target query task includes: Key information is extracted from the target text block to obtain key target information; The document query information and the target key information are input into the text processing model to obtain the document query results for the target query task.

6. The method according to claim 1, wherein processing the document query information and the target text block using a text processing model and a visual processing model to obtain the document query result for the target query task includes: Using the text processing model, the document query information and the target text block are processed to obtain a first query result; Using the visual processing model, the document query information and the target text block are processed to obtain a second query result; Based on the first query result and the second query result, generate the document query result for the target query task.

7. The method according to claim 6, wherein generating the document query result for the target query task based on the first query result and the second query result includes: The confidence level of the first query result is evaluated to obtain a first confidence index, and the confidence level of the second query result is evaluated to obtain a second confidence index. Based on the first confidence index and the second confidence index, the document query results for the target query task are filtered from the first query results and the second query results.

8. The method according to claim 6, wherein generating the document query result for the target query task based on the first query result and the second query result includes: The first query result and the second query result are input into the text processing model for result fusion to obtain the document query result of the target query task.

9. The method according to claim 1, wherein filtering the target text block from multiple text blocks included in the target document based on the document query information comprises: Multiple text block features are obtained, wherein each text block feature corresponds one-to-one with a text block, and the text block features are obtained based on feature extraction of the text blocks; Feature extraction is performed on the document query information to obtain document query features; The document query features are matched with the multiple text block features to obtain multiple matching results, wherein each matching result corresponds one-to-one with a text block; Based on the multiple matching results, the target text block is selected from the multiple text blocks.

10. The method according to claim 9, wherein filtering the target text block from the plurality of text blocks based on the plurality of matching results comprises: The vector query engine is invoked to sort the multiple text blocks based on the multiple matching results, and the target text block is selected from the sorted multiple text blocks.

11. The method according to any one of claims 1 to 10, wherein before filtering the target text block from the plurality of text blocks included in the target document based on the document query information, the method further comprises: Retrieve the target document corresponding to the document query information; Parse the target document to obtain the target document text; The target document text is divided into text blocks to obtain the multiple text blocks included in the target document.

12. The method according to claim 11, wherein dividing the target document text into text blocks to obtain the plurality of text blocks included in the target document comprises: According to a preset sliding window, the target document text is divided into text blocks to obtain the multiple text blocks included in the target document, wherein there is content overlap between adjacent text blocks in the multiple text blocks.

13. A document question-and-answer method, comprising: The receiving terminal device sends a document related to the target document; Based on the document question, select the target text block from the multiple text blocks included in the target document; Using text processing and visual processing models, the document question and the target text block are processed to obtain the answer to the document question, and the answer is sent to the terminal device.

14. An information processing method based on a processing model, applied to a task platform, comprising: Receive model requests sent by terminal devices; Based on the model request, a text processing model and a visual processing model are determined from multiple processing models, wherein the text processing model and the visual processing model are used in the execution of the method as described in any one of claims 1 to 12.

15. The method of claim 14, wherein the model request includes a scene identifier of the target scene; The step of determining the text processing model and the visual processing model from multiple processing models based on the model request includes: Based on the scene identifier of the target scene, a text processing model and a visual processing model suitable for the target scene are searched from the model library. The model library stores multiple processing models suitable for different query processing scenarios.

16. The method of claim 14, wherein the model request includes scene input data of the target scene; The step of determining the text processing model and the visual processing model from multiple processing models based on the model request includes: From the multiple processing models, an initial text processing model and an initial visual processing model suitable for the target scenario are determined; Based on the scene input data of the target scene, the initial text processing model and the initial visual processing model are trained to obtain the text processing model and the visual processing model.

17. The method of claim 14, wherein the model request includes model specification parameters; The step of determining the text processing model and the visual processing model from multiple processing models based on the model request includes: Based on the model specification parameters, the corresponding text processing model and visual processing model are searched from the model library, wherein the model library stores multiple processing models with different model specification parameters.

18. The method according to any one of claims 14 to 17, wherein after determining the text processing model and the visual processing model from a plurality of processing models based on the model request, the method further comprises: The text processing model and the visual processing model are deployed, and a query task processing interface is constructed based on the text processing model and the visual processing model, so that the terminal device can schedule the text processing model and the visual processing model to execute the target query task.

19. A task platform, comprising a request interface and a response component; The request interface is configured to receive model requests sent by the terminal device, wherein... The model request includes at least one of the following: the scene identifier of the target scene, the scene input data of the target scene, and the model specification parameters. The response component is configured to determine a text processing model and a visual processing model from a plurality of processing models based on the model request, wherein the text processing model and the visual processing model are used in the execution of the method as described in any one of claims 1 to 12.

20. The task platform according to claim 19, further comprising a query task processing interface, the query task processing interface being constructed based on the text processing model and the visual processing model; The query task processing interface is configured to allow the terminal device to schedule and execute the target query task.

21. A computing device, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 18.

22. An electronic device, comprising: A memory and a processor, the memory and the processor being connected via a bus; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 18.

23. A computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 18.

24. [Amended according to Rule 26, 30.12.2025] A computer program product comprising a computer program / instructions which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 18.

Images