Knowledge retrieval method and device based on large model middleware and medium

Abstract

The embodiment of the invention discloses a knowledge retrieval method and device based on large model middleware and a medium, relates to the technical field of large model processing, and is used for solving the problem that existing knowledge retrieval questions and answers are poor in flexibility and optimizability. The method comprises the following steps: adjusting a processing strategy node of preset large model middleware based on a document structure of an original document, and analyzing the original document according to the adjusted processing strategy node to obtain a knowledge fragment corresponding to the original document; performing vectorization processing on the current user question so as to determine a knowledge fragment associated with the current user question according to the similarity between the processed current user question and the knowledge fragment; analyzing the associated knowledge fragments and the current user question based on a preset large model middleware so as to output an answer result corresponding to the current user question; wherein each answer statement in the answer result is marked with a knowledge fragment referenced by the answer statement.

Description

Technical Field

[0001] The present specification relates to the field of large model processing technology, and in particular to a knowledge retrieval method, device and medium based on large model middleware. Background Art

[0002] With the rapid development of big model technology, the threshold of natural language processing technology has been greatly reduced, and various big model applications with advanced natural language processing capabilities have developed explosively. However, in actual applications, big model application development pays more attention to intelligent horizontal alignment in terms of effects than conventional application development, which is not fully supported in current technical implementations. Especially in the knowledge base question and answer scenario, there are problems such as poor question and answer effects, no reliable references, and heavy functions but light effects, which greatly hinder the widespread application of big model capabilities.

[0003] Although the existing language models provide a complete set of standard encapsulated large model capabilities, including large model calls, document parsing, etc., to help developers quickly build large model knowledge question and answer applications, there are problems such as high usage barriers and inability to optimize the effects. Developers need to be very familiar with the algorithm models related to document parsing in order to build a complete processing flow and integrate it into the original business process. At the same time, the framework is mainly aimed at reuse and efficiency improvement in the development stage, and lacks the ability to debug and optimize the intermediate states of each stage of the whole process, resulting in the inability to optimize the overall end-to-end question and answer effect. In addition, for the existing low-code SaaS platform solutions, although the rapid construction and hosting of large model question and answer applications are achieved through drag-and-drop configuration, which improves ease of use, the flexibility is insufficient and usually only supports limited or proprietary large model calls, and it is difficult to adjust based on demand, which makes the flexibility and optimizability of question and answer poor. Summary of the invention

[0004] In order to solve the above problems, one or more embodiments of this specification provide a knowledge retrieval method based on large model middleware.

[0005] One or more embodiments of this specification adopt the following technical solutions:

[0006] One or more embodiments of this specification provide a knowledge retrieval method based on a large model middleware, the method comprising:

[0007] Adjusting the processing strategy node of the preset large model middleware based on the document structure of the original document, so as to parse the original document according to the adjusted processing strategy node to obtain the knowledge fragment corresponding to the original document;

[0008] Performing vectorization processing on the current user question to determine the knowledge fragment associated with the current user question according to the similarity between the processed current user question and the knowledge fragment;

[0009] Based on the preset large model middleware, the associated knowledge fragments and the current user question are analyzed to output an answer result corresponding to the current user question; wherein each answer statement in the answer result is identified with the knowledge fragment it references.

[0010] Optionally, in one or more embodiments of the present specification, the adjusting the processing strategy node of the preset large model middleware based on the document structure of the original document specifically includes:

[0011] Acquire the document structure of the original document, so as to call the default configuration data corresponding to the original document in the preset database based on the document structure;

[0012] If it is determined that the preset large model middleware is in an online state, the default configuration data and the configuration parameters of each processing strategy node in the preset database are loaded into a configuration page;

[0013] The configuration information uploaded by the configuration page is received, and the default configuration data is updated with the configuration information to obtain updated configuration data, so as to configure and set the processing strategy node according to the updated configuration data to obtain an adjusted processing strategy node.

[0014] Optionally, in one or more embodiments of the present specification, parsing the original document according to the adjusted processing strategy node to obtain the knowledge fragment corresponding to the original document specifically includes:

[0015] Creating a corresponding identifier based on a user request corresponding to the current user question, and parsing the original document based on a corresponding processing strategy node to obtain a knowledge fragment corresponding to the original document output after parsing by the corresponding processing strategy node;

[0016] The knowledge fragment is identified and recorded based on the identification, so as to store the identification information corresponding to the knowledge fragment and the original document in a preset database.

[0017] Optionally, in one or more embodiments of the present specification, after parsing the original document according to the adjusted processing strategy node, the method further includes:

[0018] Parsing the original document based on the corresponding processing strategy nodes to obtain processing results output after parsing of each processing strategy node;

[0019] Based on the identification, the processing result corresponding to each processing strategy node is identified and recorded, and the identification record and the processing result corresponding to each processing strategy node are stored in a preset database.

[0020] Optionally, in one or more embodiments of the present specification, parsing the original document based on the corresponding processing strategy node to obtain the knowledge fragment corresponding to the original document output after the corresponding processing strategy node is parsed specifically includes:

[0021] Performing geometric layout analysis and logical layout analysis on the original document respectively through corresponding processing strategy nodes to obtain geometric layout analysis results and logical layout analysis results;

[0022] Extracting structural information of the original document according to the geometric layout analysis result and the logical layout analysis result;

[0023] The original document is segmented according to the structural information to obtain knowledge fragments output after the corresponding processing strategy nodes are parsed.

[0024] Optionally, in one or more embodiments of the present specification, after segmenting the original document according to the structural information, the method further includes:

[0025] Segmentation is performed based on the structural information to obtain initial knowledge fragments, so as to determine the title information of each initial knowledge fragment and extract key information of each initial knowledge fragment; wherein the key information includes: author information and publication time information;

[0026] The initial knowledge fragment is supplemented according to the title information and the key information to obtain the knowledge fragment.

[0027] Optionally, in one or more embodiments of the present specification, the vectorization processing of the current user question to determine the knowledge fragment associated with the current user question according to the similarity between the processed current user question and the knowledge fragment specifically includes:

[0028] Performing vectorization processing on the current user question to obtain a vector representation of the current user question, and determining the similarity between the vector representation and the knowledge fragment based on a preset similarity calculation method;

[0029] The knowledge fragments are sorted based on the similarity to filter and obtain the knowledge fragments associated with the current user question.

[0030] Optionally, in one or more embodiments of the present specification, analyzing the associated knowledge fragments and the current user question based on the preset large model middleware to output an answer result corresponding to the current user question specifically includes:

[0031] Analyze the associated knowledge fragments and the current user question based on the preset large model middleware to obtain an answer statement corresponding to the current user question;

[0032] The associated knowledge fragments and the answer statement are processed based on a preset reference method to identify the associated knowledge fragments at corresponding positions of the answer statement, and obtain an answer result corresponding to the current user question.

[0033] One or more embodiments of this specification provide a knowledge retrieval device based on a large model middleware, the device comprising:

[0034] at least one processor; and,

[0035] a memory communicatively connected to the at least one processor; wherein,

[0036] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute any of the above-mentioned knowledge retrieval methods based on large model middleware.

[0037] One or more embodiments of the present specification provide a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to be able to: execute any of the above-mentioned knowledge retrieval methods based on large model middleware.

[0038] At least one of the above technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects:

[0039] According to the document structure of the original document, the processing strategy nodes of the big model middleware are adjusted. Different processing strategies can be flexibly applied to different types of documents, which improves the adaptability and flexibility of the system and enables debuggability and optimization in question-and-answer application scenarios, thereby improving the reliability of question-and-answer results. Based on the preset big model middleware, the associated knowledge fragments and user questions are analyzed, so that the preset big model refers to multiple knowledge fragments and questions to answer, improving the accuracy and relevance of the answers. In addition, by identifying the referenced knowledge fragments, the answers are made more transparent, and users can understand what specific information the answers are based on, realizing the reference traceability function, and greatly improving the professionalism and authority of the questions and answers. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art description. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0041] Figure 1 A schematic diagram of a knowledge retrieval method based on a large model middleware provided in an embodiment of this specification;

[0042] Figure 2 A schematic diagram of a knowledge retrieval strategy flow based on a large model middleware provided in an embodiment of this specification;

[0043] Figure 3 A schematic diagram of the structure of a knowledge retrieval device based on a large model middleware provided in an embodiment of this specification;

[0044] Figure 4 A schematic diagram of the structure of a non-volatile storage medium provided in an embodiment of this specification. DETAILED DESCRIPTION

[0045] The embodiments of this specification provide a knowledge retrieval method, device and medium based on large model middleware.

[0046] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0047] like Figure 1 As shown, in the embodiment of this specification, a schematic diagram of a knowledge retrieval method based on a large model middleware is provided. Figure 1 It can be seen that in one or more embodiments of this specification, a knowledge retrieval method based on a large model middleware includes the following steps:

[0048] S101: adjusting the processing strategy nodes of the preset large model middleware based on the document structure of the original document, so as to parse the original document according to the adjusted processing strategy nodes and obtain the knowledge fragments corresponding to the original document.

[0049] The contents of different document structures vary greatly, and it is difficult to solve the parsing and search database construction of all types of documents through a general solution. Therefore, it is crucial to confirm and optimize the effects of documents of different structures and types. In order to be able to debug and optimize based on the implementation in the question and answer application scenario, and then improve the reliability of the question and answer results. In the embodiments of this specification, the processing strategy node of the large model middleware is first adjusted according to the document structure of the original document, and the effect of question and answer parsing is optimized through targeted debugging, so that the original document can be parsed according to the adjusted processing strategy node to obtain the knowledge fragment corresponding to the original document. Among them, such as Figure 2 As shown, it should be noted that each middleware contains a complete Retrieval Augmented Generation (RAG) policy process. The configuration parameters of the RAG policy are recorded in the database and can be modified through the configuration page. After the configuration is modified, the configuration will be loaded by the policy service in the "online process" to change the default policy process. Possible configuration items include: document format content parsing configuration such as document type settings, table structure settings in the document, chapter structure settings, etc., document content segmentation configuration such as segmentation length settings, segmentation chapter level settings, etc., document segmentation index building strategy embedding strategy model selection, index length configuration, etc.

[0050] Specifically, in one or more embodiments of the present specification, adjusting the processing strategy node of the preset large model middleware based on the document structure of the original document specifically includes the following process:

[0051] First, the document structure of the original document is obtained, so that the default configuration data corresponding to the original document in the preset database is called according to the document structure. If it is determined that the preset large model middleware is in an online state, the default configuration data and the configuration parameters of each processing strategy node in the preset database are loaded into the configuration page. Then, the configuration information uploaded by the configuration page is received, and the default configuration data is updated with the configuration information to obtain the updated configuration data, so that the processing strategy node is configured and set according to the updated configuration data to obtain the adjusted processing strategy node. This process allows the default configuration data in the preset database to be dynamically called according to the document structure of the original document. This means that different processing strategies can be flexibly applied for different types of documents, which improves the adaptability and flexibility of the system. By loading the default configuration data and the configuration parameters of each processing strategy node in the preset database into the configuration page, and receiving the configuration information uploaded by the user for updating, the consistency and accuracy of the configuration data are ensured. This helps to avoid system abnormalities or performance degradation caused by configuration errors. In addition, by ensuring that the middleware is in an online state and performing configuration updates, it helps to maintain the overall stability and reliability of the system.

[0052] Furthermore, in one or more embodiments of the present specification, parsing the original document according to the adjusted processing strategy node to obtain the knowledge fragment corresponding to the original document specifically includes the following process:

[0053] First, in order to accurately locate the effectiveness issues of different strategy stages, a corresponding identifier will be created based on the user request corresponding to the current user question, that is, a unique identifier will be established for each user request. Then, the original document is parsed according to the corresponding processing strategy node to obtain the knowledge fragment corresponding to the original document output after the corresponding processing strategy node is parsed, and the knowledge fragment is marked and recorded according to the identifier to store the knowledge fragment and the identification information corresponding to the original document in the preset database.

[0054] In this process, by creating a unique identifier for each user request, we can track the processing and results of each user request, including the extraction and parsing effects of knowledge fragments. This helps to evaluate the performance and effects of different processing strategy nodes and provide data support for optimization and adjustment. In addition, the identification information of knowledge fragments and original documents is stored in the preset database, which realizes the centralized management and efficient storage of data. This helps to reduce data redundancy and duplication and improve the efficiency and accuracy of data management.

[0055] Furthermore, in one or more embodiments of the present specification, after parsing the original document according to the adjusted processing strategy node, the method further includes:

[0056] The original document is parsed based on the corresponding processing strategy node to obtain the processing results outputted after parsing of each processing strategy node. Then, the processing results corresponding to each processing strategy node are identified and recorded according to the identifier, and the identification record and the processing results corresponding to each processing strategy node are stored in a preset database. Based on this method, the output of each processing strategy node can be recorded in the database according to the unique identifier, and all records with the same unique identifier can be displayed together in the order of calling through a visual page. At the same time, when the API obtains an answer, all policy result records of the answer can be synchronously returned for user reference.

[0057] Further, in one or more embodiments of the present specification, parsing the original document based on the corresponding processing strategy node to obtain the knowledge fragment corresponding to the original document output after the corresponding processing strategy node is parsed specifically includes:

[0058] Through the corresponding processing strategy nodes, the original document is subjected to geometric layout analysis and logical layout analysis respectively, and the geometric layout analysis results and logical layout analysis results are obtained. It should be noted that the geometric layout analysis focuses on the physical layout of the document, such as the position, size and arrangement of elements such as text, pictures, and tables on the page. Through geometric layout analysis, that is, based on image recognition processes such as OCR, deep learning models, etc., the system can identify different areas in the document such as the title area, the body area, the header and footer, etc., and how the elements in these areas are related to each other. Logical layout analysis focuses on the content structure and semantic relationship of the document. For example, through natural language processing technology, the chapter title, paragraph topic, list item, etc. in the document can be identified, and the hierarchical relationship between them can be determined. The geometric layout analysis results and logical layout analysis results obtained through the above process can extract the structural information of the original document, that is, the position, size, type, hierarchical relationship, semantic relationship, etc. of each element in the document. Then, the original document is segmented according to the structural information to obtain the knowledge fragments output after the corresponding processing strategy node is parsed. The document knowledge that needs to be referenced when answering questions with a large model usually contains a large amount of text content, which far exceeds the input limit of common large models. Therefore, by segmenting the original document, the complete document is segmented into a large number of document knowledge slices. In the embodiments of this specification, the layout of the file is analyzed to extract the structural information of the document, such as titles and tables at all levels, and then the document is segmented. This avoids the common problem of segmenting the document based on a fixed length so that the complete information is segmented into different fragments, and maintains the integrity of the information.

[0059] Furthermore, in one or more embodiments of the present specification, after segmenting the original document according to the structural information, the method further includes:

[0060] Based on the structural information obtained in the above process, the document is segmented to obtain initial knowledge fragments, so as to determine the title information of each initial knowledge fragment and extract the key information of each initial knowledge fragment; wherein the key information includes: author information, publication time information, etc. Then the initial knowledge fragments are supplemented according to the title information and key information to obtain knowledge fragments. In this process, the document is segmented based on the structure of the document, and the multi-level titles of the fragment, as well as the author, time point and other information are supplemented to the document fragment. In this way, not only does the document fragment have very good information integrity, but the supplementary information can also solve various reference problems in the document fragment, achieving the purpose of reference resolution. By segmenting the document according to the document structure and supplementing key information, the retrieval effect of the large model question-answering process is improved.

[0061] S102: performing vectorization processing on the current user question to determine the knowledge fragment associated with the current user question according to the similarity between the processed current user question and the knowledge fragment.

[0062] After the original document is processed based on the above step S101 to obtain the knowledge fragments, in order to improve the accuracy and availability of the answer of the large model, the answer can be obtained based on the corresponding knowledge fragments. In the embodiment of this specification, the current user question is vectorized to obtain a vector representation of the current user question, so that the user question and the knowledge fragment can be efficiently calculated in a high-dimensional space. The similarity between the processed current user question and the knowledge fragment is calculated by metrics such as cosine similarity and Euclidean distance. By sorting the knowledge fragments based on the similarity, the knowledge fragments associated with the current user question can be screened and determined. By accurately matching user questions and knowledge fragments, it is helpful to find the knowledge fragments most relevant to the user question in a short time, which facilitates the provision of knowledge fragments to the large model, and helps the large model to answer with reference to multiple knowledge fragments and questions.

[0063] S103: Analyze the associated knowledge fragments and the current user question based on the preset large model middleware to output an answer result corresponding to the current user question; wherein each answer statement in the answer result is marked with the knowledge fragment it references.

[0064] In order to improve the accuracy of answers and enhance the interpretability of answers, in the embodiments of this specification, the associated knowledge fragments and the current user questions are analyzed by presetting the large model middleware, so as to output the answer results corresponding to the current user questions. It should be noted that each answer statement in the answer result is marked with the knowledge fragment referenced by the answer statement. By identifying the referenced knowledge fragments, the answers are made more transparent, and users can understand what specific information the answers are based on, realizing the reference traceability function, and greatly improving the professionalism and authority of the questions and answers.

[0065] Specifically, in one or more embodiments of the present specification, the associated knowledge fragments and the current user question are analyzed based on the preset large model middleware to output an answer result corresponding to the current user question, which specifically includes the following process:

[0066] First, based on the preset large model middleware, the associated knowledge fragments and the current user question are analyzed to obtain the answer sentence corresponding to the current user question. Then, the associated knowledge fragments and the answer sentence are processed by the preset reference method, so that the associated knowledge fragments are marked at the corresponding position of the answer sentence, and the answer result corresponding to the current user question is obtained.

[0067] like Figure 3 As shown, the embodiment of this specification provides a schematic diagram of the structure of a knowledge retrieval device based on a large model middleware. Figure 3It can be seen that in one or more embodiments of this specification, a knowledge retrieval device based on a large model middleware includes:

[0068] at least one processor; and,

[0069] a memory communicatively connected to the at least one processor; wherein,

[0070] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute any of the above-mentioned methods.

[0071] like Figure 4 As shown, the present specification provides a schematic diagram of the structure of a non-volatile storage medium. Figure 4 It can be seen that in one or more embodiments of the present specification, a non-volatile storage medium stores computer executable instructions 401, and the computer executable instructions 401 can: execute any of the above-mentioned methods.

[0072] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0073] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0074] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A knowledge retrieval method based on large model middleware, characterized in that: The method comprises: Adjusting the processing strategy node of the preset large model middleware based on the document structure of the original document, so as to parse the original document according to the adjusted processing strategy node to obtain the knowledge fragment corresponding to the original document; Performing vectorization processing on the current user question to determine the knowledge fragment associated with the current user question according to the similarity between the processed current user question and the knowledge fragment; Based on the preset large model middleware, the associated knowledge fragments and the current user question are analyzed to output an answer result corresponding to the current user question; wherein each answer statement in the answer result is identified with the knowledge fragment it references.

2. A knowledge retrieval method based on large model middleware according to claim 1, characterized in that: The processing strategy node of the preset large model middleware is adjusted based on the document structure of the original document, specifically including: Acquire the document structure of the original document, so as to call the default configuration data corresponding to the original document in the preset database based on the document structure; If it is determined that the preset large model middleware is in an online state, the default configuration data and the configuration parameters of each processing strategy node in the preset database are loaded into a configuration page; The configuration information uploaded by the configuration page is received, and the default configuration data is updated with the configuration information to obtain updated configuration data, so as to configure and set the processing strategy node according to the updated configuration data to obtain an adjusted processing strategy node.

3. The knowledge retrieval method based on large model middleware according to claim 1 is characterized in that: The parsing of the original document according to the adjusted processing strategy node to obtain the knowledge fragment corresponding to the original document specifically includes: Creating a corresponding identifier based on a user request corresponding to the current user question, and parsing the original document based on a corresponding processing strategy node to obtain a knowledge fragment corresponding to the original document output after parsing by the corresponding processing strategy node; The knowledge fragment is identified and recorded based on the identification, so as to store the identification information corresponding to the knowledge fragment and the original document in a preset database.

4. According to the knowledge retrieval method based on large model middleware in claim 3, after parsing the original document according to the adjusted processing strategy node, it also includes: Parsing the original document based on the corresponding processing strategy nodes to obtain processing results output after parsing of each processing strategy node; Based on the identification, the processing result corresponding to each processing strategy node is identified and recorded, and the identification record and the processing result corresponding to each processing strategy node are stored in a preset database.

5. The knowledge retrieval method based on large model middleware according to claim 3 is characterized in that: The parsing of the original document based on the corresponding processing strategy node to obtain the knowledge fragment corresponding to the original document output after the parsing of the corresponding processing strategy node specifically includes: Performing geometric layout analysis and logical layout analysis on the original document respectively through corresponding processing strategy nodes to obtain geometric layout analysis results and logical layout analysis results; Extracting structural information of the original document according to the geometric layout analysis result and the logical layout analysis result; The original document is segmented according to the structural information to obtain knowledge fragments output after the corresponding processing strategy nodes are parsed.

6. A knowledge retrieval method based on large model middleware according to claim 5, characterized in that: After segmenting the original document according to the structural information, the method further includes: Segmentation is performed based on the structural information to obtain initial knowledge fragments, so as to determine the title information of each initial knowledge fragment and extract key information of each initial knowledge fragment; wherein the key information includes: author information and publication time information; The initial knowledge fragment is supplemented according to the title information and the key information to obtain the knowledge fragment.

7. The knowledge retrieval method based on large model middleware according to claim 1 is characterized in that: The vectorization processing of the current user question to determine the knowledge fragment associated with the current user question according to the similarity between the processed current user question and the knowledge fragment specifically includes: Performing vectorization processing on the current user question to obtain a vector representation of the current user question, and determining the similarity between the vector representation and the knowledge fragment based on a preset similarity calculation method; The knowledge fragments are sorted based on the similarity to filter and obtain the knowledge fragments associated with the current user question.

8. The knowledge retrieval method based on large model middleware according to claim 1 is characterized in that: Analyzing the associated knowledge fragments and the current user question based on the preset large model middleware to output an answer result corresponding to the current user question, specifically including: Analyze the associated knowledge fragments and the current user question based on the preset large model middleware to obtain an answer statement corresponding to the current user question; The associated knowledge fragments and the answer statement are processed based on a preset reference method to identify the associated knowledge fragments at corresponding positions of the answer statement, and obtain an answer result corresponding to the current user question.

9. A knowledge retrieval device based on large model middleware, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: execute the knowledge retrieval method based on large model middleware as described in any one of claims 1-7 above.

10. A non-volatile storage medium storing computer executable instructions, characterized in that: The computer executable instructions are capable of: executing any of the knowledge retrieval methods based on large model middleware described in claims 1-7.

Images