Multi-scale multi-channel information enhancement method, system and electronic device based on large model

Abstract

The present application discloses a multi-scale multi-channel information enhancement method, system and electronic device based on a large model. The method adopts the following steps: obtaining a target input object, the target input object at least contains some text information; obtaining the organizational structure and corresponding specific content of the target input object according to the text information; generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content; storing the additional information and the specific content in a structured manner according to the corresponding organizational structure to form a corresponding knowledge base; and in response to the first large language model receiving an interaction request, searching the knowledge base for some information related to the interaction request as request return information and passing it to the first large language model to generate feedback information. The present application solves the technical problem that the traditional knowledge base question-and-answer system based on a large language model cannot extract scattered information and integrate it to obtain a more comprehensive answer.

Description

Technical Field

[0001] The present application relates to the application field of large language models, and in particular to a multi-scale multi-path information enhancement method, system and electronic device based on a large model. Background Art

[0002] With the development of language model technology, the application of large language models has become very extensive. For example, the number of parameters of the GPT-3 series models launched by OpenAI has increased to hundreds of billions or even trillions of levels, greatly promoting their widespread application in the field of NLP (Natural Language Processing).

[0003] Natural language-based chatbots are developing rapidly. Users can use natural language to interact with chatbots and complete certain specific tasks. Based on the interface provided by the large language model, using various interfaces to call / use the knowledge base has become a common and practical processing method during the execution of this task. Therefore, a type of knowledge base question-and-answer system with the help of a large language model has been derived. Users can use natural language to call / use the knowledge base through chatbots, thereby making full use of the content of the customized knowledge base. These needs are widely seen in scenarios such as robot customer service and professional knowledge training.

[0004] However, when traditional knowledge bases are used to enter documents, they often segment the text first and then directly enter the segmented documents. In addition, the large language model itself has the characteristic of forgetting long texts. Therefore, this type of knowledge base question-answering system based on the large language model often cannot extract and integrate satisfactory answers from the knowledge base when the answers to the user's questions are relatively scattered, which leads to limitations in the use of information in the knowledge base. Summary of the invention

[0005] The embodiments of the present application provide a large-model-based multi-scale and multi-channel information enhancement method, system, and electronic device for solving the technical problem that the existing knowledge base question-answering system based on a large language model cannot extract scattered information and integrate it to obtain a more comprehensive answer.

[0006] According to one aspect of the present application, a multi-scale multi-channel information enhancement method based on a large model is provided, comprising:

[0007] Obtaining a target input object, wherein the target input object at least includes some text information;

[0008] Obtaining the organizational structure of the target input object and specific contents corresponding to different levels of the organizational structure according to the text information;

[0009] generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content;

[0010] The additional information and the specific content are respectively structured and stored according to corresponding organizational structures to form an additional information database and a specific content database, and the additional information database and the specific content database form a knowledge base corresponding to the target input object; wherein, during the structured storage, the specific content and any type of additional information are metadata to each other according to the hierarchical correspondence of the organizational structure;

[0011] In response to a data acquisition request generated by the first large language model according to the interaction request it receives, a number of information related to the interaction request are searched from the additional information database and the specific content database respectively and passed as request return information to the first large language model, so that the first large language model generates feedback information of the interaction request according to the request return information.

[0012] Furthermore, the additional information includes:

[0013] At least one of summary, tag, question-answer pair.

[0014] Furthermore, generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content includes:

[0015] The second largest language model is called and generated using prompt word engineering.

[0016] Furthermore, generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content includes:

[0017] The additional information is a summary, and the summary is generated in order from the bottom layer to the top layer of the organizational structure.

[0018] Furthermore, generating the summary in order from the bottom layer to the top layer of the organizational structure includes:

[0019] Generate a corresponding first summary for each natural paragraph one by one;

[0020] A second summary is generated for a paragraph combination in the organizational structure in order from bottom to top of the organizational structure level, wherein the paragraph combination is composed of a set of paragraphs under child nodes belonging to the same parent node in the organizational structure.

[0021] Furthermore, generating a second summary by combining the paragraphs in the organizational structure includes:

[0022] directly generating a second summary for the paragraph combination;

[0023] or,

[0024] All generated digests under child nodes belonging to the same parent node are obtained, and the second digest is generated by using all the digests.

[0025] Furthermore, searching for a number of information related to the interaction request from the additional information database and the specific content database and passing them as request return information to the first language model respectively includes:

[0026] The additional information includes a first type of additional information and a second type of additional information, the first type of additional information is a label, and the second type of additional information is a summary or a question-answer pair;

[0027] Calculating a first correlation between the first-category additional information and the interaction request, and obtaining the first-category additional information whose first correlation exceeds a first threshold as target first-category additional information;

[0028] According to the corresponding relationship between the metadata, the specific content and the second type of additional information corresponding to the target first type of additional information are obtained as the target specific content and the target second type of additional information;

[0029] Calculating a second correlation between the target specific content and the interaction request, and a third correlation between the target second category additional information and the interaction request;

[0030] According to the order of the first relevance, the second relevance and the third relevance, the corresponding target first category additional information, target specific content and target second category additional information are sent to the first large language model in order from high to low, so that the first large language model dynamically generates feedback information of the interaction request according to the received content.

[0031] A second aspect of the present application provides an information feedback system based on a large language model, comprising:

[0032] A target input module, used to obtain a target input object, wherein the target input object at least includes some text information;

[0033] A first generating module, configured to obtain the organizational structure of the target input object and specific contents corresponding to different levels of the organizational structure according to the text information;

[0034] A second generating module, configured to generate at least one type of additional information corresponding to the level of each organizational structure according to the specific content;

[0035] A storage module, used for structurally storing the additional information and the specific content according to corresponding organizational structures to form an additional information database and a specific content database, wherein the additional information database and the specific content database form a knowledge base corresponding to the target input object; wherein, during structured storage, the specific content and any type of additional information are metadata to each other according to the hierarchical correspondence of the organizational structure;

[0036] The return module is used to respond to the data acquisition request generated by the first large language model according to the interaction request it receives, and respectively search for a number of information related to the interaction request from the additional information database and the specific content database as request return information to pass to the first large language model, so that the first large language model generates feedback information of the interaction request according to the request return information.

[0037] The third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the electronic device implements the method described in the first aspect of the present application.

[0038] The fourth aspect of the present application provides a computer-readable storage medium, which is used to store a computer program. When the computer program is run on a computer, the computer executes the method described in the first aspect of the present application.

[0039] In an embodiment of the present application, a target input object is obtained, the target input object at least including some text information; the organizational structure of the target input object and the specific contents corresponding to different levels of the organizational structure are obtained according to the text information; at least one type of additional information corresponding to the level of each organizational structure is generated according to the specific contents; the additional information and the specific contents are structuredly stored according to the corresponding organizational structures to form an additional information database and a specific content database, and the additional information database and the specific content database form a knowledge base corresponding to the target input object; wherein, during the structured storage, the specific contents and any type of additional information are metadata to each other according to the hierarchical correspondence of the organizational structure; in response to a data acquisition request generated by the first large language model according to the interaction request received by it, several information related to the interaction request is searched from the additional information database and the specific content database respectively as request return information and passed to the first large language model, so that the first large language model generates feedback information of the interaction request according to the request return information. This application solves the technical problem that the traditional knowledge base question-answering system based on a large language model cannot extract scattered information and integrate it to obtain a more comprehensive answer, thereby being able to provide multi-dimensional information with more granularity, which is conducive to integrating the large language model to obtain a more satisfactory answer. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0041] Figure 1 It is a flowchart of a multi-scale and multi-channel information enhancement method based on a large model according to an embodiment of the present application.

[0042] Figure 2 It is a schematic diagram of a multi-scale and multi-path information enhancement system based on a large language model according to an embodiment of the present application.

[0043] Figure 3 is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0044] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0045] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0046] The knowledge base question-answering system based on the big language model is a type of question-answering system that combines the knowledge base and big language model technology. With the rapid popularization of big language models, this type of system, based on the information in the knowledge base and the powerful processing ability of natural semantics through the big language model, has enabled fields such as intelligent question-answering to develop rapidly.

[0047] When writing many documents, due to the various levels of chapters and paragraphs, the relevant information about a certain descriptive object is often scattered in different locations of the document. For example, a concept introduced in the first chapter of a document will not be reintroduced in the third chapter, but some of the corpus will be used. This writing style does not hinder human reading, but for large language models, they can only answer based on the retrieved text of the third chapter, and cannot obtain the definition of the concept in the first chapter. The answers generated in this way are often missing information.

[0048] The reason for the above reasons is that when the document is entered into the existing knowledge base question-answering system, it is often segmented first, and then the document after segmentation is directly entered. This results in the loss of some chapter titles, paragraph information, etc. scattered in the document itself, which can provide a lot of contextual information. When the knowledge base system generates answers to questions, it often directly uses the retrieved text and gives it to the large language model to generate answers. In this way, only limited use can be made of the document information, that is, the answer can only be generated by referring to the short-distance context of the retrieved text. That is, there is a characteristic of information missing when extracting long paragraph information, resulting in the fact that although the large language model can theoretically recognize natural semantics and organize natural language as answers, it ultimately cannot really be as good as human summary ability.

[0049] Therefore, the present embodiment conceives to extract the document into a tree structure based on the chapter titles, paragraphs and other information of the document before entering the knowledge base, and store the knowledge base based on this, that is, retaining the structured context information during storage, which helps to inherit some text attributes. Furthermore, when querying, since there is a hierarchical association between the tree structures, it is very convenient to extract the context of the associated information along the hierarchical association. After that, the large language model can generate a generative answer based on the richer content after extraction, which makes it easier to obtain a more comprehensive answer.

[0050] like Figure 1 FIG. 1 is a flowchart of a multi-scale multi-channel information enhancement method based on a large model according to an embodiment of the present application. Figure 1As shown, the multi-scale multi-channel information enhancement method based on a large model includes the following steps:

[0051] Step S101: Obtain a target input object, wherein the target input object at least includes some text information.

[0052] The target input object can be in the form of DOC, PDF, TXT, XLS, PPT, pictures, web pages, etc. The text information in the target input object is extracted as the information source stored in the knowledge base. According to the different forms of text information, corresponding extraction tools can be selected for extraction. During the extraction, non-text parts and parts without clear semantics can be removed as needed.

[0053] Step S102: Obtain the organizational structure of the target input object and specific contents corresponding to different levels of the organizational structure according to the text information.

[0054] The organizational structure of a text refers to the overall organizational form and logical arrangement of the text. In terms of expression, the organizational structure of a text often includes a series of chapters and paragraphs, where paragraphs can be divided into large paragraphs and small paragraphs, and each paragraph consists of at least one sentence. According to the levels of chapters and paragraphs, the tree structure of the text can be extracted from the text information as its organizational structure, and the specific content corresponding to each level in the organizational structure is also clear.

[0055] Step S103: Generate at least one type of additional information corresponding to the level of each organizational structure according to the specific content.

[0056] For small paragraphs with less specific content, the additional information generated based on such small paragraphs is usually more similar to the small paragraph itself, and the granularity of such additional information is smaller. For small paragraphs with more specific content, or large paragraphs with multiple small paragraphs, the information they contain is more than that of small paragraphs with less specific content. Therefore, the additional information generated based on such paragraphs can usually reflect the information related to the entire paragraph, and the granularity of such additional information is relatively large. By the same token, for higher-level organizational structures such as chapters, the corresponding additional information has a larger span and granularity. Therefore, obtaining additional information at different levels means having both information with smaller granularity and information with larger granularity but taking into account the context.

[0057] Step S104: The additional information and the specific content are respectively structuredly stored according to corresponding organizational structures to form an additional information database and a specific content database, and the additional information database and the specific content database form a knowledge base corresponding to the target input object; wherein, during structured storage, the specific content and any type of additional information are metadata to each other according to the hierarchical correspondence of the organizational structure.

[0058] The acquisition and storage of this additional information in the database is an important supplement to the original text itself. It is an important reference for providing to the large language model for answer generation later. It can make up for the problem of missing information when entering the knowledge base and generating answers.

[0059] Storing additional information and specific content in different databases respectively enables multiple recalls to be performed simultaneously when responding to knowledge base searches. Furthermore, since additional information and specific content have a metadata relationship with each other, no matter when additional information or specific content is hit, another type of data with a metadata relationship with each other will be automatically retrieved and hit, which helps to improve retrieval efficiency.

[0060] Step S105: In response to a data acquisition request generated by the first large language model according to the received interaction request, a number of information related to the interaction request is searched from the additional information database and the specific content database respectively as request return information and passed to the first large language model, so that the first large language model generates feedback information of the interaction request according to the request return information.

[0061] When searching for data based on interactive requests, the knowledge base searches for corresponding corpus information based on semantic distance and other methods as the source for the subsequent large language model to generate answers. Since the corpus information has more sources and is richer in granularity, it is beneficial for the first large language model to generate more accurate answers, thus overcoming the problem that traditional knowledge bases cannot provide contextual information scattered in multiple places.

[0062] In some preferred embodiments, the additional information includes:

[0063] At least one of summary, tag, question-answer pair.

[0064] Among them, the abstract is a summary of the corresponding specific content. The specific content can be abbreviated, but the expression form is still one or several sentences.

[0065] Tags are labels for specific content. They are usually expressed in words. Compared with summaries, tags are more concise and summarize the dimensions of specific content. The dimensions of each specific content can be set according to the document content. The specific settings can be made in the prompt word project.

[0066] Question and answer pairs are content in the form of questions and answers constructed based on the corresponding specific content.

[0067] In some preferred embodiments, generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content includes:

[0068] The second largest language model is called and generated using prompt word engineering.

[0069] The different additional information may use the same or different second language models, and the prompt words used for the different additional information may be adjusted and set according to the experience of those skilled in the art.

[0070] In some preferred embodiments, generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content includes:

[0071] The additional information is a summary, and the summary is generated in order from the bottom layer to the top layer of the organizational structure.

[0072] That is, after performing structured extraction of the document based on its chapter titles, paragraphs, and other information, we extract summaries for each paragraph, each small chapter, and each large chapter. The summary of each chapter fully refers to its text content and the summary of the sub-chapter. When querying, the summary information related to the text is screened as necessary and submitted to the large language model together with the text to generate answers, so that the long-distance context information in the document can be fully utilized.

[0073] In some preferred embodiments, generating summaries in order from the bottom layer to the top layer of the organizational structure includes:

[0074] Generate a corresponding first summary for each natural paragraph one by one;

[0075] A second summary is generated for a paragraph combination in the organizational structure in order from bottom to top of the organizational structure level, wherein the paragraph combination is composed of a set of paragraphs under child nodes belonging to the same parent node in the organizational structure.

[0076] In some preferred embodiments, the step of generating a second summary for the paragraph combination in the organizational structure includes:

[0077] directly generating a second summary for the paragraph combination;

[0078] or,

[0079] All generated digests under child nodes belonging to the same parent node are obtained, and the second digest is generated by using all the digests.

[0080] When the size of the second summary generated directly by paragraph combination is controllable, the direct generation by paragraph combination is preferred; when the size of the second summary generated directly by paragraph combination exceeds a set threshold, it is selected to pre-compress the information volume by obtaining all the generated summaries under the child nodes belonging to the same parent node, and generate the second summary by using the all the summaries, so that the size of the second summary is within a controllable range.

[0081] In some optional embodiments, if the above method still cannot obtain a summary with controllable size, the summary is further generated with itself as the content until the size of the finally generated summary is controllable.

[0082] In some preferred embodiments, searching for a number of information related to the interaction request from the additional information database and the specific content database respectively as request return information and passing it to the first language model includes:

[0083] The additional information includes a first type of additional information and a second type of additional information, the first type of additional information is a label, and the second type of additional information is a summary or a question-answer pair;

[0084] Calculating a first correlation between the first-category additional information and the interaction request, and obtaining the first-category additional information whose first correlation exceeds a first threshold as target first-category additional information;

[0085] According to the corresponding relationship between the metadata, the specific content and the second type of additional information corresponding to the target first type of additional information are obtained as the target specific content and the target second type of additional information;

[0086] Calculating a second correlation between the target specific content and the interaction request, and a third correlation between the target second category additional information and the interaction request;

[0087] According to the order of the first relevance, the second relevance and the third relevance, the corresponding target first category additional information, target specific content and target second category additional information are sent to the first large language model in order from high to low, so that the first large language model dynamically generates feedback information of the interaction request according to the received content.

[0088] Since the label reflects the dimensional information, the label can be used as the initial screening condition during the search to eliminate the obviously inappropriate results. However, since the amount of information in the label is too small, the second relevance and the third relevance are respectively obtained based on the initial screening results using the target specific content and the target second category additional information with greater information density. Therefore, no matter the first relevance, the second relevance or the third relevance, there is a corresponding relationship between the corresponding target information (including the target first category additional information, the target specific content and the target second category additional information). Finally, the corresponding target information is sent to the large language model in order according to the level of relevance. Therefore, the target information with higher relevance is sent first. At this time, the large language model can quickly generate semantics through these contents, and the answer generation speed is fast, but some information particles will be sacrificed. As more target information is received and processed by the large language model, the time required for answer generation becomes relatively longer, but the accuracy of the answer and the comprehensiveness of the information will be better. During implementation, different response speeds can be set as needed to adapt to the needs of different users.

[0089] Another embodiment of the present invention is as follows Figure 2 As shown, a multi-scale multi-path information enhancement system based on a large language model is provided, comprising:

[0090] The target input module is used to obtain a target input object, wherein the target input object at least includes some text information.

[0091] The first generating module is used to obtain the organizational structure of the target input object and the specific contents corresponding to different levels of the organizational structure according to the text information.

[0092] The second generating module is used to generate at least one type of additional information corresponding to the level of each organizational structure according to the specific content.

[0093] A storage module is used to structure the additional information and the specific content according to corresponding organizational structures to form an additional information database and a specific content database, and the additional information database and the specific content database form a knowledge base corresponding to the target input object; wherein, during structured storage, the specific content and any type of additional information are metadata to each other according to the hierarchical correspondence of the organizational structure.

[0094] The return module is used to respond to the data acquisition request generated by the first large language model according to the interaction request it receives, and respectively search for a number of information related to the interaction request from the additional information database and the specific content database as request return information to pass to the first large language model, so that the first large language model generates feedback information of the interaction request according to the request return information.

[0095] In some preferred embodiments, the additional information includes:

[0096] At least one of summary, tag, question-answer pair.

[0097] In some preferred embodiments, the first generating module comprises:

[0098] The prompt word module is used to call the second largest language model and generate it using the prompt word project.

[0099] In some preferred embodiments, the second generating module includes:

[0100] The summary module is used for generating the summary in the order from the bottom layer to the top layer of the organizational structure when the additional information is a summary.

[0101] In some preferred embodiments, the summary module includes:

[0102] A first summary module, used to generate a corresponding first summary for each natural paragraph one by one;

[0103] The second summary module is used to generate a second summary for the paragraph combination in the organizational structure in the order of the organizational structure level from bottom to top, wherein the paragraph combination is composed of a paragraph set under the child nodes belonging to the same parent node in the organizational structure.

[0104] In some preferred embodiments, the second summary module includes:

[0105] A direct generation module, used for directly generating a second summary for the paragraph combination;

[0106] or,

[0107] The indirect generation module is used to obtain all the generated digests under the child nodes belonging to the same parent node, and generate the second digest through all the digests.

[0108] In some preferred embodiments, the additional information includes a first type of additional information and a second type of additional information, the first type of additional information is a label, and the second type of additional information is a summary or a question-answer pair; the return module includes:

[0109] A first correlation calculation module, used for calculating a first correlation between the first-category additional information and the interaction request, and obtaining the first-category additional information whose first correlation exceeds a first threshold as target first-category additional information;

[0110] An additional information calculation module, used for obtaining the specific content and the second type of additional information corresponding to the target first type of additional information as the target specific content and the target second type of additional information according to the corresponding relationship between the metadata;

[0111] A second correlation calculation module, used to calculate a second correlation between the target specific content and the interaction request, and a third correlation between the target second category additional information and the interaction request;

[0112] A sending module is used to send the corresponding target first-category additional information, target specific content and target second-category additional information to the first large language model in descending order according to the first relevance, the second relevance and the third relevance, so that the first large language model dynamically generates feedback information of the interaction request according to the received content.

[0113] The embodiments of the present application solve the technical problem that the traditional knowledge base question-answering system based on a large language model cannot extract scattered information and integrate it to obtain a more comprehensive answer, thereby being able to provide multi-dimensional information with more granularity, which is conducive to integrating the large language model to obtain a more satisfactory answer.

[0114] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0115] Another embodiment of the present invention is an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the electronic device implements a method as described in any method embodiment of the present invention.

[0116] In one embodiment, the electronic device may be a server. In this embodiment, the structure of the electronic device may be as follows: Figure 3 As shown, it includes a memory 2001 , a communication module 2003 and one or more processors 2002 .

[0117] The memory 2001 is used to store computer programs executed by the processor 2002. The memory 2001 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and programs required for running the instant messaging function, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0118] The memory 2001 may be a volatile memory, such as a random-access memory (RAM); the memory 2001 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or the memory 2001 may be any other medium that can be used to carry or store a desired computer program in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2001 may be a combination of the above memories.

[0119] The processor 2002 may include one or more central processing units (CPU) or a digital processing unit, etc. The processor 2002 is configured to implement the above-mentioned audio data processing method when calling the computer program stored in the memory 2001 .

[0120] The communication module 2003 is used to communicate with terminal devices and other servers.

[0121] The specific connection medium between the memory 2001, the communication module 2003 and the processor 2002 is not limited in the embodiment of the present application. Figure 3 In the embodiment, the memory 2001 and the processor 2002 are connected via a bus 2004. The bus 2004 is Figure 3 The thick arrows in the figure are used to describe the connection between other components, which is only for illustrative purposes and is not intended to be limiting. The bus 2004 can be divided into an address bus, a data bus, a control bus, etc. For ease of description, Figure 3 Only one thick arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0122] Based on the same inventive concept as the above method embodiment, an embodiment of the present invention further provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, when the computer program is run on a computer, the electronic device implements the method in any of the above method embodiments. The computer-readable storage medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0123] Based on the same inventive concept as the above-mentioned method embodiment, an embodiment of the present invention further provides a computer program product, the computer program product including a computer program. When the program product is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of the control method according to various exemplary embodiments of the present application described above in this specification. The program product may adopt any combination of one or more readable media. These computer program commands may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the commands executed by the processor of the computer or other programmable data processing device generate commands for implementing the steps in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0124] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

Claims

1. A multi-scale multi-channel information enhancement method based on a large model, characterized in that: Obtaining a target input object, wherein the target input object at least includes some text information; Obtaining the organizational structure of the target input object and specific contents corresponding to different levels of the organizational structure according to the text information; generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content; The additional information and the specific content are respectively structured and stored according to corresponding organizational structures to form an additional information database and a specific content database, and the additional information database and the specific content database form a knowledge base corresponding to the target input object; wherein, during the structured storage, the specific content and any type of additional information are metadata to each other according to the hierarchical correspondence of the organizational structure; In response to a data acquisition request generated by the first large language model according to the received interaction request, searching for a number of information related to the interaction request from the additional information database and the specific content database respectively and transmitting the information as request return information to the first large language model, so that the first large language model generates feedback information of the interaction request according to the request return information; The searching for a number of information related to the interaction request from the additional information database and the specific content database as request return information and transmitting it to the first language model includes: The additional information includes a first type of additional information and a second type of additional information, the first type of additional information is a label, and the second type of additional information is a summary or a question-answer pair; Calculating a first correlation between the first-category additional information and the interaction request, and obtaining the first-category additional information whose first correlation exceeds a first threshold as target first-category additional information; According to the corresponding relationship between the metadata, the specific content and the second type of additional information corresponding to the target first type of additional information are obtained as the target specific content and the target second type of additional information; Calculating a second correlation between the target specific content and the interaction request, and a third correlation between the target second category additional information and the interaction request; According to the order of the first relevance, the second relevance and the third relevance, the corresponding target first category additional information, target specific content and target second category additional information are sent to the first large language model in order from high to low, so that the first large language model dynamically generates feedback information of the interaction request according to the received content.

2. The method according to claim 1, characterized in that The additional information includes: at least one of a summary, a tag, and a question-answer pair.

3. The method according to claim 2, characterized in that The generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content includes: calling the second largest language model and generating it by using prompt word engineering.

4. The method according to claim 2, characterized in that: The generating at least one type of additional information corresponding to the level of each organizational structure according to the specific content includes: the additional information is a summary, and the summary is generated in order from the bottom layer to the top layer of the organizational structure.

5. The method according to claim 4, characterized in that Generating the summary in order from the bottom layer to the top layer of the organizational structure includes: generating a corresponding first summary for each natural paragraph one by one; A second summary is generated for a paragraph combination in the organizational structure in order from bottom to top of the organizational structure level, wherein the paragraph combination is composed of a set of paragraphs under child nodes belonging to the same parent node in the organizational structure.

6. The method according to claim 5, characterized in that Generating a second summary for the paragraph combination in the organizational structure includes: directly generating a second summary for the paragraph combination; or, All generated digests under child nodes belonging to the same parent node are obtained, and the second digest is generated using all the digests.

7. The information feedback system based on the large language model is characterized by: include: A target input module, used to obtain a target input object, wherein the target input object at least includes some text information; A first generating module, configured to obtain the organizational structure of the target input object and specific contents corresponding to different levels of the organizational structure according to the text information; A second generating module, configured to generate at least one type of additional information corresponding to the level of each organizational structure according to the specific content; A storage module, used for structurally storing the additional information and the specific content according to corresponding organizational structures to form an additional information database and a specific content database, wherein the additional information database and the specific content database form a knowledge base corresponding to the target input object; wherein, during structured storage, the specific content and any type of additional information are metadata to each other according to the hierarchical correspondence of the organizational structure; a return module, configured to respond to a data acquisition request generated by the first large language model according to the received interaction request, and to search for a number of information related to the interaction request from the additional information database and the specific content database respectively as request return information and pass it to the first large language model, so that the first large language model generates feedback information of the interaction request according to the request return information; The searching for a number of information related to the interaction request from the additional information database and the specific content database as request return information and transmitting it to the first language model includes: The additional information includes a first type of additional information and a second type of additional information, the first type of additional information is a label, and the second type of additional information is a summary or a question-answer pair; Calculating a first correlation between the first-category additional information and the interaction request, and obtaining the first-category additional information whose first correlation exceeds a first threshold as target first-category additional information; According to the corresponding relationship between the metadata, the specific content and the second type of additional information corresponding to the target first type of additional information are obtained as the target specific content and the target second type of additional information; Calculating a second correlation between the target specific content and the interaction request, and a third correlation between the target second category additional information and the interaction request; According to the order of the first relevance, the second relevance and the third relevance, the corresponding target first category additional information, target specific content and target second category additional information are sent to the first large language model in order from high to low, so that the first large language model dynamically generates feedback information of the interaction request according to the received content.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the electronic device implements the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and when the computer program is executed on a computer, the computer is caused to execute the method according to any one of claims 1 to 6.

Images