Content information query method and system, and file database construction method

Abstract

Provided in the present application is a content information query method. The method comprises: acquiring a query request, which comprises query information related to a question; on the basis of the query request, acquiring from a file database query content related to the query request, wherein the file database comprises a file set composed of structured parsing files, each structured parsing file comprises a text file having text content and element markers, and the element markers are used for identifying element content in a non-textual form; inputting the query content into a large language model to acquire initial answer information corresponding to the question, wherein the initial answer information comprises textual answer content in a textual form and element markers generated as marker-type answer content; and on the basis of the textual answer content and the element markers generated as the marker-type answer content, generating final answer information for the query request to serve as content information.

Description

Methods and systems for querying content information, and methods for constructing file databases. Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to methods and systems for querying content information, methods for constructing file databases, computing devices, and computer storage media. Background Technology

[0002] With the rapid development of computer and internet technologies, artificial intelligence and large-scale modeling techniques are widely used in daily life and work. However, this technology is less commonly used in some industrial sectors (e.g., display and chip manufacturing). Besides the inherent complexity of problems in industrial fields, another major reason is the often highly proprietary nature of industrial data and the rapid pace of technological and product updates, making direct training on large models unsuitable due to high costs and time consumption. For example, during the initial training, the world's thinnest chip might be 3 nanometers. After training, the data might be updated to 2 nanometers. Currently, there is no technology to make a model completely forget what it has already learned through retraining, which leads to low recall rates and inaccurate recall in large models. Summary of the Invention

[0003] In view of this, this disclosure provides methods and systems for querying content information, methods for constructing file databases, computing devices, and computer storage media, in order to overcome some or all of the defects mentioned above, as well as other possible defects.

[0004] According to a first aspect of this disclosure, a method for querying content information is provided, comprising: obtaining a query request, the query request including query information related to a question; obtaining query content related to the query request from a file database according to the query request, wherein the file database includes a file collection composed of structured parsed files, the structured parsed files including text files having text content and element markers, the element markers being used to identify non-textual element content; inputting the query content into a large language model to obtain initial answer information corresponding to the question, wherein the initial answer information includes text answer content in text form and element markers generated as identifiers of the answer content; and generating final answer information for the query request based on the text answer content and the element markers generated as identifiers of the answer content, as the content information.

[0005] In some embodiments, generating final answer information for the query request based on the text answer content and an element marker generated as an identifier answer content includes: obtaining element content corresponding to the element marker generated as an identifier answer content; and generating final answer information for the query request based on the text answer content and the obtained element content.

[0006] In some embodiments, obtaining the element content corresponding to the element marker generated as the marker answer content includes: obtaining the address of the element content corresponding to the element marker generated as the marker answer content from an element mapping file based on the element marker generated as the marker answer content, wherein the element mapping file is used to store the mapping relationship between element markers and element content addresses; and obtaining the element content corresponding to the element marker generated as the marker answer content based on the obtained element content address.

[0007] In some embodiments, the method further includes: presenting the final answer information in the form of an interface, such that the interface includes the text answer content and the acquired element content.

[0008] In some embodiments, generating final answer information for the query request based on the text answer content and an element marker generated as an identifier of the answer content includes: generating a corresponding element content marker based on the element marker generated as an identifier of the answer content; and generating the final answer information based on the text answer content and the element content marker.

[0009] In some embodiments, the method further includes: presenting the final answer information in the form of an interface, the interface including the text answer content and the element content marker; and, in response to the selection of the element content marker, presenting element content corresponding to the element marker generated as the marker answer content in the interface.

[0010] In some embodiments, the element content is obtained through the following steps, which include: obtaining the element content address corresponding to the element flag generated as the flag answer content from an element mapping file, the element mapping file being used to store the mapping relationship between element flags and element content addresses; and obtaining the element content corresponding to the element flag generated as the flag answer content based on the obtained element content address.

[0011] In some embodiments, the size of the element content in the interface can be determined by at least one of the following methods: determining the size of the element content based on its original size; automatically determining the size of the element content based on the size of the interface and the sizes of other content in the interface besides the element content; or adjusting the size of the element content according to a received user instruction.

[0012] In some embodiments, the element content includes at least one of an image, a mathematical formula, and a hyperlink.

[0013] In some embodiments, obtaining query content related to the query request from a file database according to the query request includes: vectorizing the query request to obtain a query request vector; obtaining vector data related to the query request from a file vector library based on the query request vector, wherein the file vector library is obtained after vectorizing the structured parsed file; and obtaining query content related to the query request from the file database based on the vector data.

[0014] In some embodiments, the query information includes text-type query information. Retrieving query content related to the query request from a file database based on the query request includes: retrieving query content related to the query request from text files with text content and element identifiers in the file database based on the query information.

[0015] In some embodiments, the query information includes image-type query information, and the file database further includes element files with non-text element content, wherein there is a mapping relationship between the text content and element markers in the text files and the element content in the element files. Retrieving query content related to the query request from the file database according to the query request includes: retrieving image information related to the query request from the element files with non-text element content based on the query information; and retrieving query content related to the query request from text files with text content and element markers in the file database based on the image information and utilizing the mapping relationship.

[0016] In some embodiments, the query information includes text-type query information and image-type query information, and the file database also includes element files with non-text element content, wherein there is a mapping relationship between the text content and element tags in the text files and the element content in the element files. According to the query request, obtaining query content related to the query request from the file database includes: determining a first weight for text-type query information and a second weight for image-type query information, wherein the corresponding weight in the first and second weights represents the acquisition condition when obtaining the corresponding query content during a corresponding type of query; obtaining first query content related to the text-type query request from text files with text content and element tags in the file database according to the text-type query information and the first weight; obtaining image information related to the query request from element files with non-text element content according to the image-type query information and the second weight; and obtaining second query content related to the query request from text files with text content and element tags in the file database based on the image information and using the mapping relationship. The first and second query contents are determined as query content related to the query request.

[0017] According to a second aspect of this disclosure, a method for constructing a file database is provided. The method includes: acquiring an original file; performing content recognition on the original file to determine text content and non-text element content in the original file; generating corresponding element tags for the element content, and replacing the element content in the original file with the corresponding element tags to obtain a text file having text content and element tags, wherein the element tags are used to identify the element content; obtaining a structured parsing file based on the text file having text content and element tags, and generating the file database based on the structured parsing file.

[0018] According to a third aspect of this disclosure, a content information query system is provided, comprising: a front-end processor configured to acquire a query request, the query request including query information related to a question, and, based on the query request, acquire query content related to the query request from a file database and send the query content, wherein the file database includes a file collection composed of structured parsing files, the structured parsing files including text files having text content and element markers, the element markers being used to identify non-textual element content; a back-end processor configured to receive the query content, input the query content into a large language model to acquire initial answer information corresponding to the question, and send the initial answer information, wherein the initial answer information includes text answer content in text form and element markers generated as identifiers of the answer content; the front-end processor is further configured to receive the answer information and, based on the text answer content and the element markers generated as identifiers of the answer content, generate final answer information for the query request as the content information.

[0019] According to a fourth aspect of this disclosure, a computing device is provided, including a processor; and a memory configured to store computer-executable instructions thereon, which, when executed by the processor, perform any of the methods described above.

[0020] According to a fifth aspect of this disclosure, a computer-readable storage medium is provided that stores computer-executable instructions that, when executed, perform any of the methods described above.

[0021] According to a sixth aspect of this disclosure, a computer program product is provided, the computer program product including computer-executable instructions that, when executed, implement any of the methods described above.

[0022] In the method and system for querying content information claimed in this disclosure, firstly, query content related to the query request is obtained from a file database based on the query request; then, the query content is input into a large language model to obtain initial answer information corresponding to the question, wherein the initial answer information includes text answer content in text form and element flags generated as markers of the answer content; finally, based on the text answer content and the element flags generated as markers of the answer content, final answer information for the query request is generated as the content information. In the embodiments of this disclosure, the query request is not directly input into the large model to obtain content information. Instead, the query content related to the query request is first obtained from a file database using the query request, and then the query content is input into the large model to obtain initial answer information corresponding to the question. The initial answer information includes two parts: text answer content in text form and element flags generated as markers of the answer content, and final answer information is generated based on the initial answer information. This ensures that the large model can accurately understand and parse the query or question, improving the accuracy and efficiency of the large model in recalling content information.

[0023] These and other advantages of this disclosure will become clear from the embodiments described below, and will be illustrated with reference to the embodiments described below. Attached Figure Description

[0024] Embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings, in which:

[0025] Figure 1 illustrates exemplary application scenarios in which the technical solutions according to embodiments of the present disclosure can be implemented;

[0026] Figure 2 illustrates a schematic flowchart of a content information query method according to an embodiment of the present disclosure;

[0027] Figure 3 illustrates a schematic diagram of the implementation of a content information query method according to an embodiment of the present disclosure;

[0028] Figure 4 illustrates a schematic flowchart of a single-modal type query according to an embodiment of the present disclosure;

[0029] Figure 5 illustrates a schematic flowchart of a multimodal type query according to an embodiment of the present disclosure;

[0030] Figure 6 illustrates an exemplary flowchart of a method for constructing a file database as described in Figure 2 according to an embodiment of the present disclosure;

[0031] Figure 7 illustrates an exemplary flowchart of a method for obtaining a structured parsed file based on a text file having text content and element tags according to an embodiment of the present disclosure;

[0032] Figure 8 illustrates a situation where image elements in the original file cannot be correctly identified;

[0033] Figures 9A-9B illustrate an example of integrating paragraphs in a text file with segmented content by performing title recognition according to an embodiment of the present disclosure.

[0034] Figure 10 shows the document parsing results after using a text file splitting tool;

[0035] Figure 11 shows the parsing results after optical character recognition;

[0036] Figure 12 illustrates an exemplary schematic diagram of constructing a file vector library according to an embodiment of the present disclosure;

[0037] Figure 13 illustrates an exemplary implementation scheme for querying content information according to an embodiment of the present disclosure;

[0038] Figure 14 illustrates a schematic diagram of an interface for receiving query requests according to an embodiment of the present disclosure;

[0039] Figure 15 illustrates an exemplary structural block diagram of a content information query system according to an embodiment of the present disclosure;

[0040] Figure 16 illustrates an example system that includes an example computing device representing one or more systems and / or devices that can implement the various technologies described herein. Detailed Implementation

[0041] The following description provides specific details of various embodiments of this disclosure to enable those skilled in the art to fully understand and implement the various embodiments of this disclosure. It should be understood that the technical solutions of this disclosure can be implemented without some of these details. In some cases, this disclosure does not show or describe in detail some well-known structures or functions to avoid such unnecessary descriptions that would obscure the description of the embodiments of this disclosure. The terminology used in this disclosure should be understood in its broadest and most reasonable manner, even when used in conjunction with specific embodiments of this disclosure.

[0042] As described earlier, artificial intelligence and large-scale modeling techniques are less commonly used in daily life and work, particularly in industrial sectors (e.g., display and chip manufacturing). This is because industrial problems are often complex, data is frequently highly proprietary, and technologies and products in the industrial sector iterate very rapidly, making them unsuitable for direct training on large models. Furthermore, specific technical problems exist in industrial sectors that place higher demands on artificial intelligence and large-scale modeling techniques. For example, the industrial sector employs numerous industry terms and jargon. While "DO" in general often translates as the verb "to do," in the display field it means "Data Open," which can be difficult for non-experts to understand. As another example, defects in display products can involve multiple fields such as optics and materials science, making the description of these defects complex and requiring large models to possess interdisciplinary knowledge and analytical capabilities.

[0043] Figure 1 illustrates an exemplary application scenario 100 in which the technical solution according to the embodiments of the present disclosure can be implemented. As shown in Figure 1, the application scenario includes a terminal 110 and a server 120, wherein the terminal 110 is communicatively coupled to the server 120 via a network 130.

[0044] As an example, terminal 110 may send a query request to server 120, the query request including query information related to a question. Upon receiving the query request, server 120 retrieves query content related to the query request from a file database, wherein the file database comprises a collection of structured parsed files, including text files with text content and element markers, the element markers being used to identify non-textual element content; then, the query content is input into a large language model to obtain initial answer information corresponding to the question, wherein the initial answer information includes text answer content in text form and generated element markers as identifiers of the answer content; finally, based on the text answer content and the generated element markers as identifiers of the answer content, final answer information for the query request is generated as the content information. The server can return the generated content information to terminal 110.

[0045] The scenario described above is merely one example in which the embodiments of this disclosure can be implemented, and is not restrictive. For example, in some exemplary scenarios, the technical solutions described above may also be implemented independently on terminal 110 or server 120. Alternatively, part of the technical solution may be implemented on terminal 110 and part on server 120.

[0046] It should be noted that terminal 110 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Server 120 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. Terminals and servers can be directly or indirectly connected via wired or wireless communication, and this application does not impose any limitations. The network 130 can be, for example, a wide area network (WAN), a local area network (LAN), a wireless network, a public telephone network, an intranet, and any other type of network well known to those skilled in the art.

[0047] Figure 2 illustrates a schematic flowchart of a content information query method 200 according to an embodiment of the present disclosure. As shown in Figure 2, the method 200 includes the following steps.

[0048] In step 210, a query request is obtained, which includes query information related to the question. The query request is used to request content information related to the question. The question may be, for example, an industrial problem. The industrial field may include, for example, the display field, the chip field, etc. As an example, the question could be, "A defect has occurred in the production of the display unit, with white striped patterns running across the entire glass. What could be the cause?"

[0049] In step 220, based on the query request, query content related to the query request is retrieved from a file database. The file database comprises a collection of structured parsing files, each containing text files with text content and element tags. The element tags are used to identify non-textual element content. The structured parsing files typically include the query content. The query content may be, for example, content similar to the query information or the context of the query information within the structured parsing files; this is not a limitation. The element content includes at least one of images, mathematical formulas, and hyperlinks. Correspondingly, the element tags may be at least one of image tags, mathematical formula tags, and hyperlink tags. The query content may or may not include element tags; this is not limited.

[0050] In some embodiments, the structured parsing file in the file database is obtained by parsing the original file. The original file may be a file or document containing known content or knowledge. For example, the original file may store knowledge documents, experience data, and user-uploaded knowledge related to the analysis of poor display performance. Specifically, the original file may first undergo content recognition to determine the text content and non-textual element content. Then, corresponding element tags are generated for the element content, and the element content in the original file is replaced with the corresponding element tags to obtain a text file with text content and element tags, wherein the element tags are used to identify the element content. Finally, the structured parsing file is obtained based on the text file with text content and element tags.

[0051] In some embodiments, when retrieving query content related to the query request from a file database based on the query request, the query request can first be vectorized to obtain a query request vector; then, vector data related to the query request can be retrieved from a file vector library based on the query request vector, wherein the file vector library is obtained after vectorizing the structured parsed file; finally, query content related to the query request can be retrieved from the file database based on the vector data.

[0052] As an example, vectorizing the query request can specifically include vectorizing query information related to the question. For example, various embedding models can be used to vectorize the query information related to the question, such as Word2Vec and BGE-M3 models. The file vector library is obtained by vectorizing the structured parsed file; in other words, the vector data in the file vector library originates from the corresponding structured parsed file. The vector library can be, for example, MilnVS, which is compatible with multiple modalities of retrieval and querying. The vector library can be stored in the cloud to ensure high availability and disaster recovery capabilities. In some embodiments, when retrieving query content related to the query request from the file database based on the vector data, the query content related to the query request can be retrieved from the corresponding structured parsed file in the file database based on the vector data, wherein the corresponding structured parsed file includes the structured parsed file from which the vector data originates. This results in more accurate query content. Of course, query content can also be retrieved from other structured parsed files in the file database using the vector data; this is not limiting. The structured parsing file can be obtained, for example, by performing structured parsing on files or documents that include known content or knowledge in the industrial field. This enables the accurate understanding and parsing of specific terms or technical issues in the industrial field.

[0053] In some embodiments, the method for vectorizing the structured parsing file is the same as the method for vectorizing the query request; that is, the same embedding model can be used to vectorize both the structured parsing file and the query request.

[0054] In some embodiments, the query information includes text-type query information. Retrieving query content related to the query request from a file database based on the query request includes: retrieving query content related to the query request from a text file containing text content and element identifiers in the file database based on the query information. In this embodiment, a single-modal query is implemented. In this document, the text-type query information may be, for example, text-type query information. The text file may be, for example, a Doc, Docx, Excel, or Json text file. Optionally, in a specific implementation, the query request can be vectorized to obtain a query request vector; then, vector data related to the query request can be retrieved from a file vector library based on the query request vector, where the file vector library is obtained by vectorizing the structured parsed file (e.g., a text file containing text content and element identifiers); finally, query content related to the query request can be retrieved from the file database based on the vector data.

[0055] In some embodiments, the query information includes image-type query information. The file database also includes element files with non-text element content, where the text content and element markers in the text files have a mapping relationship with the element content in the element files. Retrieving query content related to the query request from the file database according to the query request includes: retrieving image information related to the query request from the element files with non-text element content based on the query information; and retrieving query content related to the query request from text files with text content and element markers in the file database based on the image information and utilizing the mapping relationship. In this embodiment, a single-modal query is implemented. In this document, the image-type query information may be, for example, query information in the form of an image. The image type may be, for example, JPG, PNG, or other image types. Optionally, in this embodiment, the query request can be vectorized to obtain a query request vector. Then, image vector data related to the query request is obtained from a file vector library based on the query request vector. The image vector data in the file vector library is obtained by vectorizing image files with non-text image content. Next, based on the image vector data, text vector data related to the image vector data is obtained from the file database using a mapping relationship. Finally, based on the text vector data, query content related to the query request is obtained from text files with text content and element tags in the file database. First, image vector data related to the query request is obtained based on the query request. Then, the image vector data is mapped back to text vector data, and the text vector data is used to obtain query content related to the query request. This avoids the inaccuracy of directly using the image for querying and improves the recall accuracy of the large model.

[0056] In some embodiments, the query information includes text-type query information and image-type query information, and the file database also includes element files with non-text element content, wherein there is a mapping relationship between the text content and element markers in the text files and the element content in the element files. According to the query request, obtaining query content related to the query request from the file database includes: determining a first weight for text-type query information and a second weight for image-type query information, wherein the corresponding weight in the first and second weights represents the acquisition condition when obtaining the corresponding query content during a query of the corresponding type; obtaining first query content related to the text-type query request from text files with text content and element markers in the file database according to the text-type query information and the first weight; obtaining image information related to the query request from element files with non-text element content according to the image-type query information and the second weight; and obtaining second query content related to the query request from text files with text content and element markers in the file database based on the image information and using the mapping relationship. The first and second query contents are determined as the query content related to the query request. Alternatively, in the specific implementation, the vectorization processing method for single-modal type queries can be referred to above, which will not be repeated here.

[0057] This approach enables bimodal queries, suitable for situations where the text lacks explicit information and requires secondary searching using images. For example, inputting a poorly displayed image and then querying with the text "What other defects are similar to the defects in this image?" The first and second weights can be determined as needed or semantically. For instance, if the question is "What is in the image?", the second weight should be greater than the first. If the image resolution is low and lacks useful information, the first weight should be greater than the second. Higher weights simplify the retrieval criteria. These criteria can represent the probability of retrieving the corresponding type of content during a query or be related to a similarity threshold. As an example, when the image contains limited information, the first weight could be 0.8 and the second weight 0.2. In this scenario, when retrieving the first query content related to the text-type query request, it can be retrieved if the similarity between the content in the text file and the text-type query information is greater than 0.2 (1-0.8=0.2); similarly, when retrieving image information related to the query request, it can be retrieved only if the similarity between the element content in the element file and the image-type query information is greater than 0.8 (1-0.2=0.8). This indicates that the query process places greater emphasis on text-type query information.

[0058] In step 230, the query content is input into a large language model to obtain initial answer information corresponding to the question. This initial answer information includes textual answer content and generated element markers that indicate the answer content. Optionally, the query content and question-related query information can be input together into the large model to obtain initial answer information corresponding to the question. The large model can be, for example, various suitable generative large language models, such as common large language models based on GPT (Generative Pre-Training) technology. It should be noted that when the query content includes element markers, the generated element markers that indicate the answer content may be element markers included in the query content, or they may be different from the element markers included in the query content.

[0059] In step 240, based on the text answer content and the element markers generated as identifiers of the answer content, final answer information for the query request is generated as the content information. In other words, the final answer information here consists of two parts: one part is the text answer content, and the other part is related to the element markers generated as identifiers of the answer content. This improves the accuracy of the generated content information.

[0060] In some embodiments, when generating the final answer information for the query request, the element content corresponding to the element marker generated as the identifier answer content can be obtained first; then, the final answer information for the query request can be generated based on the text answer content and the obtained element content. As an example, if the element marker generated as the identifier answer content is an image marker, the image content corresponding to the image marker can be obtained first, and then the text answer content and the image content can be used together as the final answer information for the query request. Further, as an example, the final answer information can be presented in the form of an interface, such that the interface includes the text answer content and the obtained element content. The interface is, for example, a user interface displayed on a display device.

[0061] In some embodiments, when retrieving the element content corresponding to the element marker generated as the marker answer content, the address of the element content corresponding to the element marker generated as the marker answer content can first be obtained from an element mapping file based on the element marker generated as the marker answer content. The element mapping file stores the mapping relationship between element markers and element content addresses. Then, based on the obtained element content address, the element content corresponding to the element marker generated as the marker answer content can be retrieved. The element mapping file can be pre-generated or configured to store the mapping relationship between element markers and element content addresses. Based on the element mapping address, the element content corresponding to the element marker stored at that address can be retrieved. Therefore, after obtaining the element content address, the element content corresponding to the element marker generated as the marker answer content can be retrieved.

[0062] In some embodiments, when generating the final answer information for the query request, a corresponding element content marker can first be generated based on the element marker generated as an identifier of the answer content; then, the final answer information can be generated based on the text answer content and the element content marker. The element content marker here can be a specific representation of the element marker, such as an icon, diagram, thumbnail, or hyperlink. The element content marker can be displayed in a user interface. As an example, the method can further include: presenting the final answer information in the form of an interface, the interface including the text answer content and the element content marker; and, in response to the selection of the element content marker, presenting the element content corresponding to the element marker generated as an identifier of the answer content in the interface. As an example, when the element content marker is clicked in the interface, the element content corresponding to the element marker can be presented in the interface. In this way, it is not necessary to display all the queried information at once, improving processing efficiency and saving processing resources.

[0063] Similarly, the element content can be obtained through the following steps. These steps include: first, obtaining the address of the element content corresponding to the element flag generated as the flag answer content from an element mapping file, where the element mapping file stores the mapping relationship between element flags and element content addresses; then, obtaining the element content corresponding to the element flag generated as the flag answer content based on the obtained element content address. The element mapping file can be pre-generated or configured to store the mapping relationship between element flags and element content addresses. Based on the element mapping address, the element content corresponding to the element flag stored at that address can be obtained. Therefore, after obtaining the element content address, the element content corresponding to the element flag generated as the flag answer content can be obtained.

[0064] In some embodiments, the size of the element content in the interface (i.e., the display size of the element content) is adjustable. For example, the size of the element content can be determined based on its original size. For instance, the size of an image can be made the same as or proportional to its original size. Alternatively, the size of the element content can be automatically determined based on the size of the interface and the sizes of other content in the interface, so that the size of the element content is aligned with or matches the sizes of other content. Furthermore, the size of the element content can be adjusted according to received user instructions. User instructions can directly specify the size of the element content.

[0065] The content information query method of this disclosure first retrieves query content related to the query request from a file database based on the query request; then, the query content is input into a large language model to obtain initial answer information corresponding to the question, wherein the initial answer information includes text answer content in text form and element flags generated as marker answer content; finally, based on the text answer content and the element flags generated as marker answer content, final answer information for the query request is generated as the content information. In this disclosure, instead of directly inputting the query request into the large model to obtain content information, the method first retrieves query content related to the query request from a file database using the query request, and then inputs the query content into the large model to obtain initial answer information corresponding to the question. The initial answer information includes two parts: text answer content in text form and element flags generated as marker answer content, and the final answer information is generated based on the initial answer information. This ensures that the large model can accurately understand and parse the query or question, improving the accuracy and efficiency of the large model in recalling content information.

[0066] Figure 3 illustrates a schematic diagram of the implementation of a content information query method 200 according to an embodiment of this disclosure. As shown in Figure 3, in block 310, a query request is received. In block 320, the query request is vectorized using a suitable embedding model to obtain a query request vector as shown in block 330. Then, vector data related to the query request is obtained from a file vector library 340 based on the query request vector, and query content related to the query request is obtained from a file database 350 based on the vector data, as shown in block 360. The file database includes a collection of at least one structured parsed file, which includes text files with text content and element tags, the element tags being used to identify non-textual element content. The query content is included in the text files with text content and element tags. The file vector library is obtained by vectorizing the at least one structured parsed file. The file vector library and the file database can be implemented as a single database, such as a content database as shown in block 370. Then, the query content is input into the large model as shown in box 380 to obtain initial answer information 385 corresponding to the question, wherein the initial answer information includes text answer content in text form and element flags generated as marker answer content. The text answer content and element flags generated as marker answer content can be used to generate final answer information for the query request, as the content information 390.

[0067] As an example, Figure 4 illustrates a schematic flowchart of a single-modal query according to an embodiment of the present disclosure. As shown in Figure 4, in step 410, a query request is obtained, the query request including query information related to a question. In step 420, the type of query information is determined, such as text-type query information or image-type query information. When the type of query information is determined to be image-type, in step 430, image information related to the query request is obtained from an element file with non-textual element content based on the query information, and in step 440, based on the image information, query content related to the query request is obtained from a text file with text content and element tags in a file database using the mapping relationship. When the type of query information is determined to be text-type, in step 450, query content related to the query request is obtained from a text file with text content and element tags in a file database based on the query information. In step 460, the query content is input into a large model to obtain initial answer information corresponding to the question. The initial answer information includes text answer content in text form and generated element tags as markers of the answer content. In step 470, based on the text answer content and the element flag generated as the flag answer content, the final answer information for the query request is generated as the content information.

[0068] As an example, Figure 5 illustrates a schematic flowchart of a multimodal query according to an embodiment of the present disclosure. As shown in Figure 5, in step 510, a query request is obtained, the query request including both text-type query information and image-type query information. In step 520, a first weight L0 for the text-type query information and a second weight L1 for the image-type query information are determined, where L1 = 1 - L0. The corresponding weights in the first and second weights represent the acquisition conditions when obtaining the corresponding query content during a query of the corresponding type. In step 530, based on the text-type query information and the first weight, first query content related to the text-type query request is obtained from text files with text content and element tags in a file database. In step 540, based on the image-type query information and the second weight, image information related to the query request is obtained from element files with non-text element content. In step 550, based on the image information, second query content related to the query request is obtained from text files with text content and element tags in a file database using the mapping relationship. In step 560, the first query content and the second query content are determined as query content related to the query request. Then, in step 570, the query content is input into the large model to obtain initial answer information corresponding to the question. The initial answer information includes text answer content in text form and element flags generated as markers of the answer content. In step 580, based on the text answer content and the element flags generated as markers of the answer content, final answer information for the query request is generated as the content information.

[0069] Figure 6 illustrates an exemplary flowchart of a method 600 for constructing a file database as described in Figure 2 according to an embodiment of the present disclosure. Method 600 may be implemented on the same device as method 200 or on a different device. As shown in Figure 6, the method includes steps 610-640.

[0070] In step 610, the original file is obtained. The original file may be a file or document containing known content or knowledge in the industrial field. For example, the original file may store knowledge documents, experiential data, and user-uploaded knowledge related to the analysis of poor display performance.

[0071] In step 620, content recognition is performed on the original file to determine the text content and non-text element content in the original file.

[0072] In step 630, a corresponding element flag is generated for the element content, and the element content in the original file is replaced with the corresponding element flag to obtain a text file with text content and element flags, wherein the element flags are used to identify the element content.

[0073] In step 640, a structured parsing file is obtained based on the text file containing text content and element identifiers, and the file database is generated based on the structured parsing file.

[0074] In some embodiments, the text file with text content and element markers is typically a text file with segmented content. The method shown in FIG7 can be used to obtain the structured parsing file based on the text file with text content and element markers in step 640. FIG7 illustrates an exemplary flowchart of a method 700 for obtaining a structured parsing file based on the text file with text content and element markers according to an embodiment of the present disclosure. As shown in FIG7, in step 710, paragraphs in the text file can be integrated by performing title recognition on the text file with text content and element markers to obtain a preliminary parsing file. Then, in step 720, the structured parsing file is obtained based on the text file and the preliminary parsing file. The structured parsing file is beneficial to the efficiency and accuracy of subsequent searches. As an example, obtaining the structured parsing file based on the text file and the preliminary parsing file may include establishing a mapping relationship between text and elements in the preliminary parsing file and the element file; then, the element file with the mapping relationship and the preliminary parsing file are determined as the structured parsing file. Of course, the element file and the preliminary parsing file can also be directly determined as the structured parsing file.

[0075] When performing content recognition on the original file, some content can be directly obtained (e.g., content whose basic elements are characters, such as TXT), while some content cannot be directly obtained (e.g., image types such as JPG and PNG), or the obtained content may have various formatting issues that affect the accuracy of the content. Figure 8 illustrates the situation where image elements in the original file cannot be correctly recognized. As shown in Figure 8, image element 801 is often not correctly recognized because it is not in the vocabulary, thus affecting the integrity of the content and therefore cannot be directly obtained. In addition, mathematical formulas and hyperlinks in the original file, if directly obtained, usually also have various formatting issues that affect the accuracy of the content, because common formula characters such as "α" and "β" and special symbols such as "→" and "↑" in formulas and hyperlinks are often not correctly recognized.

[0076] Therefore, if the original file contains at least one element selected from a group of elements consisting of images, mathematical formulas, and hyperlinks, the content of the corresponding element in the original file is replaced with an element marker corresponding to the corresponding element in the at least one element to obtain a text file with text content and element markers. Then, the content of the corresponding element in the at least one element is stored in a structured manner to obtain at least one element file as a structured parsing file, and optionally, an element mapping file can be generated based on the mapping relationship between the element markers and the element content. The element mapping file is usually also stored in a content database.

[0077] As an example, if the original file contains images, the images in the original file are replaced with image tags corresponding to those images to obtain a text file with text content and image tags. For example, the image tag for image one could be represented as "\n<&imageimage1&>". Then, the images are stored in a structured manner to obtain an image file as a structured parsing file. In this step, it is ensured that all images are preserved and that image tags are left at the correct text positions in the original file. For example, OCR (Optical Character Recognition) or text file segmentation tools such as Text Splitter can be used to perform content recognition on the original file.

[0078] As an example, if the elements in the original file are mathematical formulas, the mathematical formulas in the original file are replaced with mathematical formula markers corresponding to the mathematical formulas to obtain a text file with text content and mathematical formula markers. For example, the original file records that when debugging compounds A and B, the following ratios need to be referenced: import math; a = 3; b = 4; c = math.sqrt(a**2 + b**2); print(c). The content after replacing the mathematical formulas can be represented as follows: Calculate a using <&formulaPythagorean_Theorem&>, then use <&formulaQuadratic_Equation&> to perform another operation with b and c... Then, the mathematical formulas are stored in a structured manner to obtain a mathematical formula file as a structured parsing file. For example, the mathematical formula Theorem can be stored in a JSON file as follows:

[0079] {

[0080] "Pythagorean Theorem":{

[0081] "python":"import math;a=3;b=4;c=math.sqrt(a**2+b**2);print(c)",

[0082] "latex":"c=\\sqrt{a^2+b^2}"

[0083] }

[0084] As an example, if the original file contains hyperlinks, the hyperlinks in the original file can be replaced with the hyperlink flags corresponding to the hyperlinks to obtain a text file with text content and hyperlink flags.

[0085] When performing content recognition on the original file, segmenting the text solely based on general methods such as formatting or maximum text length limits may not yield ideal results. Therefore, in one embodiment, paragraphs in a text file with segmented content can be integrated by performing title recognition on the text file containing text content and element markers. Generally, titles often have common characteristics, such as being on a single line, having a short text length, lacking punctuation at the end, and having a font size inconsistent with the surrounding text (possibly bold). The title and the text under that title can be integrated into a single paragraph. Paragraph recognition can often be performed using line breaks or bullet points ("·", "①", "1.1", etc.). It is important to note that modifying the judgment of colons before line break recognition (i.e., colon + line break will not result in segmentation) can prevent forced segmentation when displaying images, causing semantic incoherence.

[0086] As an example, Figures 9A-9B illustrate an example of integrating paragraphs in a text file with segmented content by performing title recognition according to an embodiment of the present disclosure. The content of the text file with segmented content is shown in Figure 9A. The text after integration using title recognition is shown in Figure 9B, and the content shown in Figure 9B can be saved as a JSON file.

[0087] Furthermore, for text files, after obtaining the results using a text file segmentation tool (e.g., Text Splitter), they can be recognized again using OCR. This allows a large model to generate and integrate the two recognition results, reducing manual intervention. For example, Figure 10 shows the document parsing result after using a text file segmentation tool, and Figure 11 shows the parsing result after using OCR. It can be seen that the document parsing accuracy is high, but there are issues such as paragraph misalignment. The OCR error rate is higher, but paragraph order is rarely incorrect. Therefore, a large model can generate and integrate the two recognition results. When reviewing images, for static images with key annotations such as circles or arrows added via "graphics," the rectangular area of ​​the image can be identified, and then a "screenshot" operation can be performed on that area (or a manual screenshot can be taken directly). Image reconstruction can then be completed using image super-resolution reconstruction technology or image fusion technology (a backup solution is to save both the "screenshot" and the original image and display them to the user simultaneously). In addition, when reviewing images, company logos, encryption tags, and image names can be clearly and effectively labeled. Furthermore, after reviewing the content, the JSON file can be converted into an Excel file to further refine the mapping relationship. This can be done by placing more important keywords in the structured document on the left side of the column according to the reading order, and deleting invalid keywords, etc.

[0088] When vectorizing a structured parsed file to obtain vector data, text symbols that do not conform to the markup language format can be removed to obtain a cleaned parsed file. Then, the cleaned parsed file is vectorized to obtain vector data. As an example, front-end developers often use Markdown format to display reference content for better presentation. Therefore, Markdown format cleaning can be performed before vectorization to avoid data containing Markdown keywords. For example, a product with the ID "##123456" might have "##" interpreted as a title in Markdown. Since Markdown syntax primarily uses symbols and spaces, often only spaces between symbols and text need to be removed. For example, "The device with ID ##1234567 has malfunctioned" can be cleaned to "The device with ID 1234567 has malfunctioned."

[0089] The vector data can be stored in the file vector library. For example, vector data obtained by vectorizing a text file can be stored in a text vector library, and vector data obtained by vectorizing an image file can be stored in an image vector library. Of course, the text vector library and the image vector library can also be implemented as a single vector library.

[0090] By utilizing the method 600 described in the embodiments of this disclosure, a file database can be constructed efficiently and quickly. A file vector library can be obtained by vectorizing structured parsed files in the file database. The content database may include a file vector library and a file database, and optionally may include an element mapping file, which represents the mapping relationship between element identifiers and element content. As an example, FIG12 illustrates an exemplary schematic diagram of constructing a file vector library according to an embodiment of this disclosure. As shown in FIG12, in block 1201, an original file is input. In block 1202, the original file is parsed (including content recognition and paragraph integration, etc.) to obtain a set of structured text files 1203 and structured image files 1204. Then, a mapping relationship between text and images is established between the structured text files 1203 and the structured image files 1204, as shown in block 1205, and the structured text files 1203 and structured image files 1204 with the mapping relationship are stored in the content database 1206. Next, the structured text file 1203 and the structured image file 1204 are vectorized to obtain vector data, and the vector data is stored in the content database 1206.

[0091] Figure 13 illustrates an exemplary implementation scheme for querying content information according to an embodiment of the present disclosure. As shown in Figure 13, in step 1301, a query request is received. In step 1302, it is determined whether the query request should be queued. This can be achieved by checking whether a queue parameter is returned within a time window. If the query request does not enter the query queue, the query request process ends in step 1303. If the query request enters the query queue, the question queried by the query request is reviewed in step 1304. Reviewing the question queried by the query request may include reviewing various blacklisted terms, etc. If the review is successful, the query content related to the query request is obtained from the file database in step 1305. If the review fails, the process returns directly in step 1306. If the query content related to the query request is obtained in step 1305, the query content may optionally be rearranged and / or filtered in step 1307; otherwise, the process returns directly in step 1306. Then, in step 1308, the query content is input into a large model to obtain initial answer information corresponding to the question, and content information related to the question is generated based on the initial answer information. In step 1309, it can be further determined whether the query request is a streaming request. If so, in step 1310, the query is executed according to each request until all queries are completed, and then all content information is returned. If it is not a streaming request, the content information is returned directly in step 1311.

[0092] In practice, upon receiving a query request, a queue parameter can be sent back to the frontend, informing it that the request has entered the query queue. At this point, the frontend can display "Retrieving relevant content..." to the user. After obtaining the query content related to the query request, a corresponding parameter can be returned, and the frontend can display "Reordering the query content..." to the user. After reordering and filtering are complete, a reference content parameter can be returned, which the frontend should directly display, along with "Generating content for you..." to the user. At this stage, if the user deems the query content of poor quality, the process can be terminated early to conserve resources.

[0093] Figure 14 illustrates a schematic diagram of an interface for receiving query requests according to an embodiment of this disclosure. Taking the analysis of defects in a display manufacturing production line as an example, the query request might ask, for example, "A defect has occurred in the production of the display unit; white striped textures run across the entire glass. What could be the cause?" The content database stores structured parsed files constructed based on the original files, as well as vector data obtained by vectorizing the structured parsed files. The original files store knowledge documents, experience data, and user-uploaded knowledge related to defect analysis of the display. The knowledge documents related to defect analysis include descriptions and data of the defects, as well as corresponding causes and improvement measures.

[0094] Upon receiving a query request, it can be vectorized to obtain a query request vector. Then, the query request vector is used to retrieve relevant query content from the content database. This query content includes similar undesirable phenomena, their corresponding causes, and improvement measures. For example, the similarity between the query request vector and vector data in the database (i.e., vector data derived from the descriptions and data of undesirable phenomena, and their corresponding causes and improvement measures) can be used to retrieve relevant query content. Optionally, sorting and filtering methods can be used to obtain the top N query content items with the highest similarity ranking, where N is a positive integer. Then, the query content, and optionally the question from the query request, can be input into a large model to obtain content information related to the question.

[0095] In some embodiments, the acquired content information related to the question can be evaluated, i.e., its relevance to the queried question can be assessed. If the relevance is greater than a relevance threshold, the content information and the question are vectorized, and the content information, the question, and the vectorized data are added to the database for subsequent queries. Furthermore, the data involved in the query process can be encrypted to ensure data security during transmission.

[0096] Figure 15 illustrates an exemplary structural block diagram of a content information query system 1500 according to an embodiment of the present disclosure. As shown in Figure 15, the content information query system includes a front-end processor 1510 and a back-end processor 1520. The front-end processor may be a processor from a terminal or server segment as described in Figure 1. The back-end processor may, for example, be a processor from a server segment as described in Figure 1 or a processor from a large model provider.

[0097] The front-end processor 1510 is configured to acquire a query request, the query request including query information related to a question, and, based on the query request, acquire query content related to the query request from a file database and send the query content, wherein the file database includes a file collection consisting of structured parsed files, the structured parsed files including text files with text content and element tags, the element tags being used to identify non-text element content;

[0098] The backend processor 1520 is configured to receive the query content, input the query content into a large language model to obtain initial answer information corresponding to the question, and send the initial answer information, wherein the initial answer information includes text answer content in text form and generated element flags as markers of the answer content;

[0099] The front-end processor 1510 is further configured to receive the answer information and generate final answer information for the query request based on the text answer content and the element flag generated as the identifier of the answer content, as the content information.

[0100] Figure 16 illustrates an example system 1600, which includes an example computing device 1610 representing one or more systems and / or devices that can implement the various technologies described herein. The computing device 1610 may be, for example, a server of a service provider, a device associated with a server, a system-on-a-chip, and / or any other suitable computing device or computing system. At least a portion of the content information query system 1500 described above with reference to Figure 15 may take the form of computing device 1610.

[0101] The example computing device 1610 shown includes a processing system 1611, one or more computer-readable media 1612, and one or more I / O interfaces 1613 that are communicatively coupled to each other. Although not shown, the computing device 1610 may also include a system bus or other data and command transfer system that couples the various components to each other. The system bus may include any or a combination of different bus architectures, such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and / or a processor or local bus utilizing any of the various bus architectures. Various other examples, such as control and data lines, are also conceived.

[0102] Processing system 1611 represents the functionality of performing one or more operations using hardware. Therefore, processing system 1611 is illustrated as including hardware elements 1614 that can be configured as processors, function blocks, etc. This may include other logic devices implemented in hardware as application-specific integrated circuits (ASICs) or formed using one or more semiconductors. Hardware element 1614 is not limited by the materials in which it is formed or the processing mechanism employed therein. For example, a processor may consist of semiconductors and / or transistors (e.g., integrated circuits (ICs)). In such a context, processor-executable instructions may be electronically executable instructions.

[0103] Computer-readable medium 1612 is illustrated as including memory / storage device 1615. Memory / storage device 1615 represents a memory / storage capacity associated with one or more computer-readable media. Memory / storage device 1615 may include volatile media (such as random access memory (RAM)) and / or non-volatile media (such as read-only memory (ROM), flash memory, optical disk, magnetic disk, etc.). Memory / storage device 1615 may include fixed media (e.g., RAM, ROM, fixed hard disk drive, etc.) and removable media (e.g., flash memory, removable hard disk drive, optical disk, etc.). Computer-readable medium 1612 may be configured in various other ways as further described below.

[0104] One or more I / O interfaces 1613 represent the functionality to allow users to input commands and information to the computing device 1610 using various input devices and optionally also to present information to the user and / or other components or devices using various output devices. Examples of input devices include keyboards, cursor control devices (e.g., mice), microphones (e.g., for voice input), scanners, touch functionality (e.g., capacitive or other sensors configured to detect physical touch), cameras (e.g., capable of detecting non-touch-related motion as gestures using visible or invisible wavelengths (such as infrared frequencies), etc. Examples of output devices include display devices (e.g., monitors or projectors), speakers, printers, network interface cards, haptic-responsive devices, etc. Therefore, the computing device 1610 can be configured to support user interaction in various ways as further described below.

[0105] The computing device 1610 also includes an application 1616. The application 1616 may be, for example, a software instance of a method for querying content information, and may implement the techniques described herein in combination with other elements in the computing device 1610.

[0106] This document describes various technologies within the general context of software and hardware components or program modules. Generally, these modules include routines, programs, objects, elements, components, data structures, etc., that perform specific tasks or implement specific abstract data types. As used herein, the terms "module," "function," and "component" generally refer to software, firmware, hardware, or a combination thereof. The technologies described herein are characterized as platform-independent, meaning that these technologies can be implemented on a variety of computing platforms with various processors.

[0107] Implementations of the described modules and technologies may be stored on or transmitted across some form of computer-readable medium. Computer-readable medium may include a variety of media accessible by the computing device 1610. By way of example and not limitation, computer-readable medium may include "computer-readable storage medium" and "computer-readable signal medium".

[0108] In contrast to simple signal transmission, carrier waves, or signals themselves, a "computer-readable storage medium" refers to a medium and / or device capable of persistently storing information, and / or a tangible storage device. Therefore, a computer-readable storage medium refers to a non-signal-bearing medium. Computer-readable storage media include hardware such as volatile and non-volatile, removable and non-removable media and / or storage devices implemented using methods or techniques suitable for storing information (such as computer-readable instructions, data structures, program modules, logic elements / circuits, or other data). Examples of computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, DVD or other optical storage devices, hard disks, magnetic tape cassettes, magnetic tapes, disk storage devices or other magnetic storage devices, or other storage devices, tangible media, or articles of art suitable for storing desired information and accessible by a computer.

[0109] "Computer-readable signal medium" refers to a signal-bearing medium configured to transmit instructions, such as via a network, to computing device 1610. A signal medium typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave, data signal, or other transmission mechanism. Signal media also include any information transmission medium. The term "modulated data signal" refers to a signal in which one or more of its characteristics are set or altered to encode information. By way of example and not limitation, communication media include wired media such as wired networks or direct connections, and wireless media such as acoustic, RF, infrared, and other wireless media.

[0110] As previously described, hardware element 1614 and computer-readable medium 1612 represent instructions, modules, programmable device logic, and / or fixed device logic implemented in hardware, which in some embodiments can be used to implement at least some aspects of the techniques described herein. Hardware elements may include components of integrated circuits or systems-on-a-chip, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs), and other implementations or other hardware devices in silicon. In this context, hardware elements can serve as processing devices for executing program tasks defined by instructions, modules, and / or logic embodied by the hardware element, and as hardware devices for storing instructions for execution, such as the previously described computer-readable storage medium.

[0111] The foregoing combinations can also be used to implement the various techniques and modules described herein. Therefore, software, hardware, or program modules and other program modules can be implemented as one or more instructions and / or logic embodied on some form of computer-readable storage medium and / or by one or more hardware elements 1614. The computing device 1610 can be configured to implement specific instructions and / or functions corresponding to the software and / or hardware modules. Thus, for example, by using the computer-readable storage medium and / or hardware element 1614 of a processing system, modules can be implemented at least partially in hardware as modules executable as software by the computing device 1610. Instructions and / or functions can be executable / operable by one or more articles of art (e.g., one or more computing devices 1610 and / or processing systems 1611) to implement the techniques, modules, and examples described herein.

[0112] In various embodiments, the computing device 1610 can be configured in various ways. For example, the computing device 1610 can be implemented as a computer-type device, including personal computers, desktop computers, multi-screen computers, laptop computers, netbooks, etc. The computing device 1610 can also be implemented as a mobile device, including mobile devices such as mobile phones, portable music players, portable gaming devices, tablet computers, multi-screen computers, etc. The computing device 1610 can also be implemented as a television-type device, including devices with or connected to a generally large screen in a leisure viewing environment. These devices include televisions, set-top boxes, game consoles, etc.

[0113] The techniques described herein can be supported by these various configurations of computing device 1610, and are not limited to specific examples of the techniques described herein. Functionality can also be implemented, wholly or partially, on the “cloud” 1620 using distributed systems, such as through platform 1622 as described below.

[0114] Cloud 1620 includes and / or represents platform 1622 for resource 1624. Platform 1622 abstracts the underlying functionality of the hardware (e.g., server) and software resources of cloud 1620. Resource 1624 may include applications and / or data that can be used when performing computer processing on a server located remotely from computing device 1610. Resource 1624 may also include services provided via the Internet and / or via subscriber networks such as cellular or Wi-Fi networks.

[0115] Platform 1622 can abstract resources and functions to connect computing device 1610 to other computing devices. Platform 1622 can also be used to abstract resource hierarchy to provide a corresponding level of hierarchy for any encountered needs for resource 1624 implemented via platform 1622. Therefore, in interconnect device embodiments, the implementation of the functions described herein can be distributed throughout system 1600. For example, functions can be implemented partly on computing device 1610 and partly through platform 1622, which abstracts the functions of cloud 1620.

[0116] This disclosure provides a computer-readable storage medium having computer-readable instructions stored thereon, which, when executed, implement any of the methods described above.

[0117] This disclosure provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computing device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computing device to perform any of the methods provided in the various alternative implementations described above.

[0118] It should be understood that, for clarity, embodiments of this disclosure have been described with reference to different functional units. However, it will be apparent that, without departing from this disclosure, the functionality of each functional unit may be implemented in a single unit, in multiple units, or as part of other functional units. For example, functionality described as being performed by a single unit may be performed by multiple different units. Therefore, references to a particular functional unit are considered merely as references to the appropriate unit used to provide the described functionality, and not as indicating a strict logical or physical structure or organization. Thus, this disclosure may be implemented in a single unit, or may be physically and functionally distributed among different units and circuits.

[0119] It will be understood that although the terms first, second, third, etc., may be used herein to describe various devices, elements, components, or parts, these devices, elements, components, or parts should not be limited by these terms. These terms are used only to distinguish one device, element, component, or part from another device, element, component, or part.

[0120] Although this disclosure has been described in conjunction with some embodiments, it is not intended to be limited to the specific forms set forth herein. Rather, the scope of this disclosure is limited only by the appended claims. Additionally, although individual features may be included in different claims, these may be advantageously combined, and inclusion in different claims does not imply that such a combination of features is not feasible and / or advantageous. The order of features in the claims does not imply that the features must be in any particular order of their operation. Furthermore, in the claims, the word "comprising" does not exclude other elements, and the terms "a" or "an" do not exclude a plurality. Reference numerals in the claims are provided only by way of explicit example and should not be construed as limiting the scope of the claims in any way.

Claims

1. A method for querying content information, comprising: Obtain a query request, which includes query information related to the question; According to the query request, query content related to the query request is obtained from the file database, wherein the file database includes a file collection composed of structured parsed files, the structured parsed files include text files with text content and element tags, and the element tags are used to identify non-text element content; The query content is input into the large language model to obtain the initial answer information corresponding to the question, wherein the initial answer information includes text answer content in text form and generated element flags as markers of the answer content; Based on the text answer content and the element flags generated as identifiers of the answer content, the final answer information for the query request is generated as the content information.

2. The method according to claim 1, wherein, Based on the text answer content and the element flags generated as identifiers of the answer content, the final answer information for the query request is generated, including: Retrieve the element content corresponding to the element flag generated as the flag answer content; Based on the text answer content and the obtained element content, the final answer information for the query request is generated.

3. The method according to claim 2, wherein, Retrieving the element content corresponding to the element flag generated as the flag answer content includes: Based on the element flag generated as the flag answer content, the element content address corresponding to the element flag generated as the flag answer content is obtained from the element mapping file, and the element mapping file is used to store the mapping relationship between element flags and element content addresses; Based on the obtained element content address, retrieve the element content corresponding to the element flag generated as the flag answer content.

4. The method according to claim 2, further comprising: The final answer information is presented in the form of an interface, such that the interface includes the text answer content and the obtained element content.

5. The method according to claim 1, wherein final answer information for the query request is generated based on the text answer content and the element marker generated as an identifier of the answer content, comprising: Generate corresponding element content tags based on the element tags generated as markers of the answer content; The final answer information is generated based on the text answer content and the element content tags.

6. The method according to claim 5, further comprising: The final answer information is presented in the form of an interface, which includes the text answer content and the element content markers; In response to the selection of the element content marker, the element content corresponding to the element marker generated as the marker answer content is presented in the interface.

7. The method according to claim 6, wherein the element content is obtained through the following steps, the steps comprising: Based on the element flag generated as the flag answer content, the element content address corresponding to the element flag generated as the flag answer content is obtained from the element mapping file, and the element mapping file is used to store the mapping relationship between element flags and element content addresses; Based on the obtained element content address, retrieve the element content corresponding to the element flag generated as the flag answer content.

8. The method according to claim 6, wherein, The size of the element content in the interface can be determined by at least one of the following methods: Determine the size of the element content based on its original size; The size of the element content is automatically determined based on the size of the interface and the size of other content in the interface besides the element content. Adjust the size of the element content according to the received user instructions.

9. The method according to claim 1, wherein, The element content includes at least one of images, mathematical formulas, and hyperlinks.

10. The method according to claim 1, wherein obtaining query content related to the query request from the file database according to the query request includes: The query request is vectorized to obtain a query request vector; Based on the query request vector, vector data related to the query request is obtained from the file vector library, which is obtained after vectorizing the structured parsed file; Based on the vector data, retrieve the query content related to the query request from the file database.

11. The method according to claim 1, wherein, The query information includes text-type query information, and wherein, based on the query request, query content related to the query request is retrieved from the file database, including: Based on the query information, retrieve the query content related to the query request from the text files containing text content and element tags in the file database.

12. The method according to claim 1, wherein, The query information includes image-type query information. The file database also includes element files with non-text element content. There is a mapping relationship between the text content and element markers in the text files and the element content in the element files. Furthermore, according to the query request, query content related to the query request is retrieved from the file database, including: Based on the query information, obtain image information related to the query request from an element file containing non-textual element content; Based on the image information, the query content related to the query request is obtained from a text file with text content and element tags in the file database using the mapping relationship.

13. The method according to claim 1, wherein, The query information includes text-type query information and image-type query information, and the file database also includes element files with non-text element content. There is a mapping relationship between the text content and element markers in the text files and the element content in the element files. Furthermore, according to the query request, retrieving query content related to the query request from the file database includes: Determine a first weight for text-type query information and a second weight for image-type query information, wherein the corresponding weights in the first and second weights represent the acquisition conditions when obtaining the corresponding query content when performing a query of the corresponding type. Based on the query information of the text type and the first weight, obtain the first query content related to the query request of the text type from the text file with text content and element tags in the file database; Based on the query information of the image type and the second weight, obtain image information related to the query request from element files with non-text element content; Based on the image information, the second query content related to the query request is obtained from a text file with text content and element tags in the file database using the mapping relationship. The first query content and the second query content are determined as query content related to the query request.

14. A method for constructing a file database, the method comprising: Obtain the original file; Perform content recognition on the original file to determine the text content and non-text element content in the original file; A corresponding element flag is generated for the element content, and the element content in the original file is replaced with the corresponding element flag to obtain a text file with text content and element flags, wherein the element flag is used to identify the element content; Based on the text file containing text content and element identifiers, a structured parsing file is obtained, and the file database is generated according to the structured parsing file.

15. A content information query system, comprising: A front-end processor is configured to acquire a query request, the query request including query information related to a question, and, based on the query request, acquire query content related to the query request from a file database and send the query content, wherein the file database includes a collection of files composed of structured parsed files, the structured parsed files including text files with text content and element tags, the element tags being used to identify non-textual element content; A backend processor is configured to receive the query content, input the query content into a large language model to obtain initial answer information corresponding to the question, and send the initial answer information, wherein the initial answer information includes text answer content in text form and generated element flags as identifiers of the answer content; The front-end processor is further configured to receive the answer information and, based on the text answer content and the element flag generated as a marker of the answer content, generate final answer information for the query request, as the content information.

16. A computing device, comprising Memory, which is configured to store computer-executable instructions; A processor configured to perform the method as described in any one of claims 1-14 when the computer-executable instructions are executed by the processor.

17. A computer-readable storage medium storing computer-executable instructions that, when executed, perform the method as described in any one of claims 1-14.

18. A computer program product, characterized in that, The computer program product includes computer-executable instructions that, when executed, implement the method according to any one of claims 1 to 14.

Images