A fusion reasoning system of large language model and knowledge graph

Abstract

The present invention discloses a fusion reasoning system of a large language model and a knowledge graph, belonging to the field of computer system technology, comprising a data acquisition system for users to submit user questions and collect popular science data, a knowledge graph system, a graph editing and indexing display system, a large model reasoning and judgment module, and a storage and identification control system; the data acquisition system is communicatively connected to the storage and identification control system, the storage and identification control system is communicatively connected to the large model reasoning and judgment module and the graph editing and indexing display system, and the knowledge graph system is communicatively connected to the large model reasoning and judgment module, the graph editing and indexing display system, and the storage and identification control system; the storage and identification control system is used to store user questions and popular science data, extract their semantic entities N, and identify the number of semantic entities N. Through the above-mentioned method, the technology of the present invention realizes the fusion reasoning of user questions or popular science data through a large model and a knowledge graph.

Description

Technical Field

[0001] The present invention relates to the field of computer system technology, and in particular to a fusion reasoning system of a large language model and a knowledge graph. Background Art

[0002] When users currently search for problems through search engines, they retrieve a large amount of information, the authenticity of which is difficult for them to distinguish, and they cannot intuitively reason about their problems.

[0003] Currently, websites are flooded with a large amount of false information, especially social networks are flooded with a large amount of pseudo-popular science. Some pseudo-popular science copywriting is widely circulated and can easily mislead the public. It is also difficult for users to complain about pseudo-popular science copywriting.

[0004] Among many existing technologies, Chinese patent application CN109635171B discloses a fusion reasoning system and method for intelligent tags of news programs, which includes an intelligent recognition executor, a historical tag library, an internal knowledge base, an internal case library and an analysis reasoner. The intelligent recognition executor performs the recognition task of various news program materials and extracts basic tags from video images, voice and text information; the historical tag library stores materials, metadata and tags; the internal knowledge base is used to supplement the intelligent recognition results and provide more information for subsequent analysis and reasoning; the internal case library is a case collection established based on the historical tag library; the analysis reasoner is used for the fusion reasoning of intelligent tags, including a rule-based reasoner and a deep learning-based reasoner.

[0005] However, this patent application cannot determine whether there is pseudo-popular science content in, for example, popular science news or popular science copywriting.

[0006] Based on this, the present invention designs a fusion reasoning system of a large language model and a knowledge graph to solve the above problems. Summary of the Invention

[0007] In response to the above-mentioned shortcomings of the existing technology, the present invention provides a fusion reasoning system of a large language model and a knowledge graph.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] A fusion reasoning system of a large language model and a knowledge graph, including a data collection system for users to submit user questions and collect popular science data, a knowledge graph system, a graph editing and indexing display system, a large model reasoning and judgment module, and a storage and recognition control system;

[0010] The data acquisition system is in communication with the storage and identification control system, the storage and identification control system is in communication with the large model reasoning and judgment module and the atlas editing and indexing display system, and the knowledge graph system is in communication with the large model reasoning and judgment module, the atlas editing and indexing display system, and the storage and identification control system;

[0011] The storage recognition control system is used to store user questions and popular science data, extract their semantic entities N, and identify the number of semantic entities N;

[0012] The storage recognition control system is further used to select whether to perform contextual semantic recognition on user questions or popular science data according to the number of semantic entities N, and to determine whether there is a relationship L between two semantic entities N;

[0013] The large model reasoning and judgment module is used to perform reasoning analysis on whether the relationship L matches the relationship S between the two corresponding node entities F in the knowledge graph system, and to add attributes to the two node entities F and their relationship S based on the large model if they do not match;

[0014] The large model reasoning and judgment module is also used to provide support for entity extraction and semantic analysis of the storage recognition control system;

[0015] The graph editing index display system also includes a system for creating an index for the knowledge graph system node entity F and the corresponding semantic entity N, and for retrieving and editing the graph of the knowledge graph system according to the inference result of the large model inference judgment module;

[0016] The graph editing index display system edits the graph of the knowledge graph system using the added attributes;

[0017] The graph editing index display system is also used to display the graph of the knowledge graph system or the edited knowledge graph system graph to the user.

[0018] Furthermore, the data acquisition system includes a data acquisition module and a user search module, and the data acquisition module and the user search module are both communicatively connected to the storage identification control system;

[0019] The data collection module is used for users to submit popular science data;

[0020] The user search module is used for users to submit user questions.

[0021] Furthermore, the knowledge graph system includes a knowledge graph and a graphics library for storing the knowledge graph, and the graphics library is communicatively connected with the large model reasoning and judgment module, the graph editing and indexing display system, and the storage and identification control system;

[0022] The storage and identification control system includes a data cloud storage module, an identification module, a central control module and a language large model. The data cloud storage module is communicatively connected to the data acquisition module, the user search module, the semantic extraction module, the identification module and the central control module. The semantic extraction module and the identification module are communicatively connected to the large model reasoning and judgment module, and the large model reasoning and judgment module provides support for the semantic extraction module, the identification module and the central control module. The identification module is communicatively connected to the central control module, and the central control module is communicatively connected to the graphics library, the atlas editing index display system and the large model reasoning and judgment module.

[0023] Furthermore, the large model reasoning and judgment module includes a database, a language large model and a reasoning and judgment module. The database is communicatively connected to the language large model, the language large model is communicatively connected to the semantic extraction module, the recognition module, and the reasoning and judgment module, and the reasoning and judgment module is communicatively connected to the graphics library, the central control module, and the atlas editing index display system.

[0024] Furthermore, the atlas editing index display system includes a display module, an index module and an atlas editing module; the display module, index module and atlas editing module are all connected to the graphics library in communication, and the display module, index module and atlas editing module are all connected to the central control module in communication.

[0025] Furthermore, it also includes a copy publishing platform for publishing popular science data, and the copy publishing platform is communicatively connected with the data acquisition module and the atlas editing module.

[0026] Furthermore, the data cloud storage module is used for cloud storage of hot science popularization data and user questions;

[0027] The semantic extraction module is used to extract semantic entities N from the text content of user questions and popular science data by calling the language model;

[0028] The recognition module is used to identify the number of semantic entities N;

[0029] When the number of semantic entities N is 0, the recognition module sends a task termination instruction to the central control module;

[0030] When the number of semantic entities N is 1, the recognition module sends a quick search instruction and the semantic entity N to the central control module;

[0031] When the number of semantic entities N is greater than 1, the recognition module calls the language model to perform contextual semantic recognition on the text of user questions and popular science data, and determines whether there is a relationship L between two semantic entities N:

[0032] If the relationship L exists, the recognition module sends the entity relationship determination instruction, semantic entity N and relationship L to the central control module;

[0033] If the relationship L does not exist, the recognition module sends a quick search instruction and the semantic entity N to the central control module.

[0034] Furthermore, the central control module is used to call the text vector encoder model of the language large model to encode the semantic entity N and the node entity F when receiving a fast search instruction or a relationship determination instruction, to obtain the encoded semantic entity n and the encoded node entity f, the central control module calculates the cosine similarity of the encoded semantic entity n and the encoded node entity f, and performs semantic matching on the encoded semantic entity n and the encoded node entity f according to the cosine similarity, and when the semantic match is found, the central control module sends an indexing instruction to the indexing module;

[0035] When the central control module receives the quick search instruction, it determines whether the semantic entity N belongs to the user question:

[0036] When the answer is yes, the central control module sends a display instruction 1 to the display module;

[0037] If the answer is no, the central control module terminates the task, sends the correct instruction to the display module, and the display module displays the correct words;

[0038] When the central control module receives the entity relationship determination instruction, the central control module sends the determination instruction, the semantic entity N and the relationship L to the reasoning determination module;

[0039] The index module is used to create an index path between the matched calculated and encoded semantic entity n and the calculated and encoded node entity f, and is also used to send the index path to the reasoning and judgment module, the display module and the graph editing module;

[0040] The index path is used by the graphics library to retrieve the graphics at the location of the node entity F in the knowledge graph;

[0041] The display module is used to retrieve the node entity F of the knowledge graph from the graphics library according to the index path, and when receiving the display instruction sent by the central control module, the display module displays the position of the node entity F in the knowledge graph to the user.

[0042] Furthermore, the reasoning and judgment module is used to, when receiving the judgment instruction, semantic entity N and relationship L sent by the central control module, retrieve two node entities F of the knowledge graph from the graph library through the index path, and call the text vector encoder model of the language large model to encode the relationship S between the two node entities F and the relationship L between the two semantic entities N to obtain the encoded relationship s and the encoded relationship l. The reasoning and judgment module calculates the cosine similarity of the encoded relationship s and the relationship l, and determines whether the two have the same semantics based on the cosine similarity;

[0043] If the two have the same semantics, the reasoning and judgment module determines that the relationship L is correct. At this time, the reasoning and judgment module determines whether the semantic entity N belongs to the user question:

[0044] If so, the reasoning and judgment module sends a display instruction to the central control module, and the central control module sends a display instruction three to the display module, and the display module displays the positions of the two node entities F in the knowledge graph to the user.

[0045] If not, the reasoning and judgment module sends a task termination instruction to the central control module, the central control module sends a correct instruction to the display module, and the display module displays the correct words to the user;

[0046] If the semantics of the two are different, the reasoning and judgment module determines that the relationship L is wrong. At this time, the reasoning and judgment module calls the language model to read the detailed attribute X of the relationship S corresponding to the wrong relationship L and the detailed attributes B1 and B2 of the two node entities F corresponding to the relationship S from the database, and sends the detailed attributes B1, B2, X and graph editing instructions to the central control module, and then the central control module sends them to the graph editing module.

[0047] Furthermore, the graph editing module is used to retrieve the node entity F from the graph library according to the index path and take a screenshot of the position of the node entity F in the knowledge graph when receiving the graph editing instruction. The graph editing module locates the node entity F and the relationship S in the screenshot through a text recognition algorithm, and then uses a graphic text editing tool to paste the detailed attribute B1, detailed attribute B2 and detailed attribute X into the corresponding positions of the two node entities F and the relationship S in the screenshot, and marks the relationship S with the correct relationship word, thereby obtaining an edited screenshot;

[0048] At this point, the graph editing module determines whether the semantic entity F belongs to the user question:

[0049] When the answer is yes, the atlas editing module sends the edited screenshot and the second display instruction to the display module, and the display module displays the edited screenshot to the user;

[0050] When the answer is no, the graph editing module sends the edited screenshot to the removal application email address of the copywriting publishing platform.

[0051] Compared with the existing technology, the present invention has the following advantages: through the graph editing index display system, it is possible to create an index between the semantic entity N and the corresponding node entity of the knowledge graph system, thereby facilitating the retrieval of the node entity of the knowledge graph system;

[0052] The graph editing, indexing and display system also realizes the calling or editing of the image content of the knowledge graph system according to the judgment result of the relationship L by the large model reasoning and judgment module; and the editing function integrates the knowledge graph technology and the large model technology to better reflect the entities and the relationship between entities, and realizes the integrated reasoning of user questions or popular science data through the large model and the knowledge graph. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0054] Figure 1 A block diagram of a large language model and knowledge graph fusion reasoning system of the present invention;

[0055] Figure 2 It is a partial schematic diagram of the knowledge graph of the present invention;

[0056] Figure 3 This is an edited screenshot of the present invention.

[0057] The numbers in the figure represent:

[0058] 1. Database 2. Data collection module 3. Document publishing platform 4. Semantic extraction module 5. Data cloud storage module 6. User search module 7. Recognition module 8. Central control module 9. Language model 10. Graphics library 11. Reasoning and judgment module 12. Knowledge graph 13. Display module 14. Index module 15. Graph editing module DETAILED DESCRIPTION

[0059] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0060] Example 1: In some embodiments, please refer to the accompanying drawings of the specification. Figure 1-Figure 3 , a fusion reasoning system of a large language model and a knowledge graph, including a data collection system for users to submit user questions and collect popular science data, a knowledge graph system, a graph editing and indexing display system, a large model reasoning and judgment module, and a storage and recognition control system;

[0061] The data acquisition system is in communication with the storage and identification control system, the storage and identification control system is in communication with the large model reasoning and judgment module and the atlas editing and indexing display system, and the knowledge graph system is in communication with the large model reasoning and judgment module, the atlas editing and indexing display system, and the storage and identification control system;

[0062] The storage recognition control system is used to store user questions and popular science data, extract their semantic entities N, and identify the number of semantic entities N;

[0063] The storage recognition control system is further used to select whether to perform contextual semantic recognition on user questions or popular science data according to the number of semantic entities N, and to determine whether there is a relationship L between two semantic entities N;

[0064] The large model reasoning and judgment module is used to perform reasoning analysis on whether the relationship L matches the relationship S between the two corresponding node entities F in the knowledge graph system, and to add attributes to the two node entities F and their relationship S based on the large model if they do not match;

[0065] The large model reasoning and judgment module is also used to provide support for entity extraction and semantic analysis of the storage recognition control system;

[0066] The graph editing index display system also includes a system for creating an index for the knowledge graph system node entity F and the corresponding semantic entity N, and for retrieving and editing the graph of the knowledge graph system according to the inference result of the large model inference judgment module;

[0067] The graph editing index display system edits the graph of the knowledge graph system using the added attributes;

[0068] The graph editing index display system is also used to display the graph of the knowledge graph system or the edited knowledge graph system graph to the user.

[0069] When the present invention is used, the semantic entity N of the user question or popular science data is extracted through the storage recognition control system, and based on the semantic entity N, it can be determined whether the user question or popular science data needs to be judged on the relationship L;

[0070] The large model reasoning and judgment module is used to infer whether the relationship L matches the node entity relationship corresponding to the knowledge graph system, so as to judge whether the relationship L is correct. The correctness of the relationship L can reflect whether the user question or the popular science data is wrong. It combines the large model technology to judge the node entity F of the knowledge graph system and

[0071] Through the graph editing index display system, it is possible to create an index between the semantic entity N and the corresponding node entity of the knowledge graph system, thereby facilitating the retrieval of the node entity of the knowledge graph system;

[0072] The graph editing index display system also realizes the calling or editing of the image content of the knowledge graph system based on the relationship L judgment results of the large model inference judgment module; and the editing function integrates the knowledge graph technology and the large model technology to better reflect the entities and the relationships between entities.

[0073] Embodiment 2: In some embodiments, as Figure 1-Figure 3 As shown, as a preferred embodiment of the present invention, the data acquisition system includes a data acquisition module 2 and a user search module 6, and both the data acquisition module 2 and the user search module 6 are communicatively connected to the storage identification control system;

[0074] Data collection module 2 is used for users to submit popular science data;

[0075] The user search module 6 is used for users to submit user questions.

[0076] The knowledge graph system includes a knowledge graph 12 and a graphic library 10 for storing the knowledge graph 12. The graphic library 10 is communicatively connected with the large model reasoning and judgment module, the graph editing and indexing display system, and the storage and recognition control system.

[0077] The storage and identification control system includes a data cloud storage module 5, an identification module 7, a central control module 8 and a language large model 9. The data cloud storage module 5 is communicatively connected to the data acquisition module 2, the user search module 6, the semantic extraction module 4, the identification module 7 and the central control module 8. The semantic extraction module 4 and the identification module 7 are communicatively connected to the large model reasoning and judgment module, and the large model reasoning and judgment module provides support for the semantic extraction module 4, the identification module 7 and the central control module 8. The identification module 7 is communicatively connected to the central control module 8. The central control module 8 is communicatively connected to the graphics library 10, the atlas editing index display system and the large model reasoning and judgment module.

[0078] The large model reasoning and judgment module includes a database 1, a large language model 9 and a reasoning and judgment module 11. The database 1 is communicated with the large language model 9, the large language model 9 is communicated with the semantic extraction module 4, the recognition module 7, and the reasoning and judgment module 11, and the reasoning and judgment module 11 is communicated with the graphics library 10, the central control module 8, and the atlas editing index display system.

[0079] The atlas editing index display system includes a display module 13, an index module 14 and a atlas editing module 15; the display module 13, the index module 14 and the atlas editing module 15 are all connected to the graphics library 10 in communication, and the display module 13, the index module 14 and the atlas editing module 15 are all connected to the central control module 8 in communication.

[0080] It also includes a copy publishing platform 3 for publishing popular science data, and the copy publishing platform 3 is communicatively connected with the data acquisition module 2 and the atlas editing module 15.

[0081] Embodiment 3: In some embodiments, as Figure 1-Figure 3 As shown, as a preferred embodiment of the present invention, the data cloud storage module 5 is used for cloud storage of hot science popularization data and user questions;

[0082] The semantic extraction module 4 is used to extract semantic entities N from the text content of user questions and popular science data by calling the language model 9;

[0083] The recognition module 7 is used to identify the number of semantic entities N;

[0084] When the number of semantic entities N is 0, the recognition module 7 sends a task termination instruction to the central control module 8;

[0085] When the number of semantic entities N is 1, the recognition module 7 sends a quick search instruction and the semantic entity N to the central control module 8;

[0086] When the number of semantic entities N is greater than 1, the recognition module 7 calls the language model 9 to perform contextual semantic recognition on the text of user questions and popular science data, and determines whether there is a relationship L between two semantic entities N:

[0087] If the relationship L exists, the recognition module 7 sends the entity relationship determination instruction, the semantic entity N and the relationship L to the central control module 8;

[0088] If the relationship L does not exist, the recognition module 7 sends a quick search instruction and the semantic entity N to the central control module 8 .

[0089] The central control module 8 is configured to, upon receiving a quick search instruction or a relationship determination instruction, call the text vector encoder model of the language large model 9 to encode the semantic entity N and the node entity F, thereby obtaining the encoded semantic entity n and the encoded node entity f. The central control module 8 calculates the cosine similarity between the encoded semantic entity n and the encoded node entity f, and performs semantic matching on the encoded semantic entity n and the encoded node entity f based on the cosine similarity. When a semantic match is found, the central control module 8 sends an indexing instruction to the indexing module 14.

[0090] When the central control module 8 receives the quick search instruction, the central control module 8 determines whether the semantic entity N belongs to the user question:

[0091] When the answer is yes, the central control module 8 sends a display instruction 1 to the display module 13;

[0092] If the answer is no, the central control module 8 terminates the task, and the central control module 8 sends a correct instruction to the display module 13, and the display module 13 displays the correct words;

[0093] When the central control module 8 receives the entity relationship determination instruction, the central control module 8 sends the determination instruction, the semantic entity N and the relationship L to the reasoning determination module 11;

[0094] The indexing module 14 is used to create an index path between the matched calculated and coded semantic entity n and the calculated and coded node entity f, and is also used to send the index path to the reasoning and judgment module 11, the display module 13 and the graph editing module 15;

[0095] The index path is used by the graphic library 10 to retrieve the graphic at the location of the node entity F in the knowledge graph 12;

[0096] The display module 13 is used to retrieve the node entity F of the knowledge graph 12 from the graphic library 10 according to the index path, and when receiving the display instruction 1 sent by the central control module 8, the display module 13 displays the position of the node entity F in the knowledge graph 12 to the user.

[0097] The reasoning and judgment module 11 is used to, when receiving the judgment instruction, semantic entity N and relationship L sent by the central control module 8, retrieve the two node entities F of the knowledge graph 12 from the graphic library 10 through the index path, and call the text vector encoder model of the language model 9 to encode the relationship S between the two node entities F and the relationship L between the two semantic entities N to obtain the encoded relationship s and the encoded relationship l. The reasoning and judgment module 11 calculates the cosine similarity of the encoded relationship s and the relationship l, and determines whether the two have the same semantics based on the cosine similarity;

[0098] If the two have the same semantics, the reasoning and determination module 11 determines that the relationship L is correct. At this time, the reasoning and determination module 11 determines whether the semantic entity N belongs to the user question:

[0099] If so, the reasoning and judgment module 11 sends a display instruction to the central control module 8, and the central control module 8 sends a display instruction three to the display module 13, and the display module 13 displays the positions of the two node entities F in the knowledge graph 12 to the user.

[0100] If not, the reasoning and judgment module 11 sends a task termination instruction to the central control module 8, and the central control module 8 sends a correct instruction to the display module 13, and the display module 13 displays the correct words to the user;

[0101] If the semantics of the two are different, the reasoning and judgment module 11 determines that the relationship L is wrong. At this time, the reasoning and judgment module 11 calls the language model 9 to read the detailed attribute X of the relationship S corresponding to the wrong relationship L and the detailed attribute B1 and detailed attribute B2 of the two node entities F corresponding to the relationship S from the database 1, and sends the detailed attribute B1, detailed attribute B2, detailed attribute X and graph editing instructions to the central control module 8, and then the central control module 8 sends them to the graph editing module 15.

[0102] The graph editing module 15 is used to retrieve the node entity F from the graph library 10 according to the index path and take a screenshot of the position of the node entity F in the knowledge graph 12 when receiving the graph editing instruction. The graph editing module 15 locates the node entity F and the relationship S in the screenshot through the text recognition algorithm, and then uses the graphic text editing tool to paste the detailed attribute B1, detailed attribute B2 and detailed attribute X into the corresponding positions of the two node entities F and the relationship S in the screenshot, and marks the relationship S with the correct relationship word, thereby obtaining the edited screenshot;

[0103] At this time, the graph editing module 15 determines whether the semantic entity F belongs to the user question:

[0104] When the answer is yes, the atlas editing module 15 sends the edited screenshot and the second display instruction to the display module 13, and the display module 13 displays the edited screenshot to the user;

[0105] When the answer is no, the graph editing module 15 sends the edited screenshot to the delisting application mailbox of the copywriting publishing platform 3.

[0106] When the invention is used, the graph of the knowledge graph 12 is screenshotted and edited through the graph editing module 15. The graph editing module 15 also integrates the detailed attributes read by the language model 9 with the screenshot content, so that the edited screenshot can not only intuitively reflect the correct relationship between entities, but also expand the content of entities and their relationships through the added attributes generated by the language model 9's ability to collect and summarize data;

[0107] When the present invention is used, when processing popular science data, the user only needs to send the popular science data link published by the copy publishing platform 3 to the data acquisition module 2. At the same time, when the atlas editing module 15 finally generates the edited screenshot, the user can choose to send the edited screenshot to the delisting application mailbox of the copy publishing platform 3, thereby realizing that when reasoning about popular science data with content relationship errors, the edited screenshot can be used as strong evidence to prove that the popular science data has errors.

[0108] The present invention realizes the judgment of whether the two forms of data, user questions and popular science data, are correct or not. When the user question is correct, the present invention displays it to the customer by calling the image of the knowledge graph 12; when it is wrong, the edited screenshot is displayed to the user through the display module 13 so that the user can infer the error relationship.

[0109] When the present invention is used, for example, the relationship between microscope and Leeuwenhoek in user questions or popular science data is described as "improvement";

[0110] Then this reasoning system performs cosine similarity judgment on the relationship "improvement" and the correct relationship "invention", and judges that the relationship "improvement" is an incorrect relationship;

[0111] like Figure 3 , edit the screenshot of the knowledge graph 12 to obtain an edited screenshot;

[0112] The edited screenshot can intuitively reflect the correct relationship between microscope and Leeuwenhoek, and expand the attributes of "microscope", "Leeuwenhoek" and the relationship "invention".

[0113] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A fusion reasoning system for a large language model and a knowledge graph, characterized by: It includes a data collection system for users to submit user questions and collect popular science data, a knowledge graph system, a graph editing and indexing display system, a large model reasoning and judgment module, and a storage and recognition control system; The data acquisition system is connected to the storage and identification control system in communication. The storage and identification control system is connected to the large model reasoning and judgment module and the atlas editing and indexing display system. The knowledge graph system is connected to the large model reasoning and judgment module, the atlas editing and indexing display system, and the storage and identification control system. The storage recognition control system is used to store user questions and popular science data, extract their semantic entities N, and identify the number of semantic entities N. The storage recognition control system is also used to select whether to perform contextual semantic recognition on user questions or popular science data based on the number of semantic entities N, and to determine whether there is a relationship L between two semantic entities N. The large model reasoning and judgment module is used to perform reasoning analysis on whether the relationship L matches the relationship S between the two corresponding node entities F in the knowledge graph system. If there is no match, it adds attributes to the two node entities F and their relationship S based on the large model. The large model reasoning and judgment module is also used to provide support for entity extraction and semantic analysis of the storage recognition control system. The graph editing, indexing and displaying system also includes a function for creating an index for the knowledge graph system node entity F and the corresponding semantic entity N, and for retrieving and editing the graph of the knowledge graph system according to the inference result of the large model inference judgment module; the graph editing, indexing and displaying system edits the graph of the knowledge graph system by adding attributes; The graph editing index display system is also used to display the graph of the knowledge graph system or the edited knowledge graph system graph to the user; The storage recognition control system comprises a semantic extraction module (4), a data cloud storage module (5), a recognition module (7), a central control module (8) and a language large model (9); the semantic extraction module (4) is used to extract semantic entities N from the text content of user questions and popular science data by calling the language large model (9); the recognition module (7) is used to identify the number of semantic entities N; when the number of semantic entities N is 0, the recognition module (7) sends a task termination instruction to the central control module (8); when the number of semantic entities N is 1, the recognition module (7) sends a quick search instruction and the semantic entity N to the central control module (8); when the number of semantic entities N is greater than 1, the recognition module (7) Calling the language model (9) to perform contextual semantic recognition on the text of user questions and popular science data, and judging whether there is a relationship L between two semantic entities N: if there is a relationship L, the recognition module (7) sends an entity relationship judgment instruction, the semantic entity N and the relationship L to the central control module (8); if there is no relationship L, the recognition module (7) sends a quick search instruction and the semantic entity N to the central control module (8); the index module (14) is used to create an index path between the matched calculated encoded semantic entity n and the calculated encoded node entity f, and is also used to send the index path to the reasoning judgment module (11), the display module (13) and the map editing module (15); The graph editing index display system includes a display module (13), an index module (14) and a graph editing module (15); when the graph editing module (15) receives a graph editing instruction, the graph editing module (15) retrieves the node entity F from the graph library (10) according to the index path and takes a screenshot of the position of the node entity F in the knowledge graph (12). The graph editing module (15) locates the node entity F and the relationship S in the screenshot through a text recognition algorithm, and then uses a graphic text editing tool to paste the detailed attribute B1, the detailed attribute B2 and the detailed attribute X to the corresponding positions of the two node entities F and the relationship S in the screenshot, and marks the relationship S with the correct relationship word, thereby obtaining an edited screenshot.

2. The large language model and knowledge graph fusion reasoning system according to claim 1 is characterized in that: The data acquisition system comprises a data acquisition module (2) and a user search module (6), and both the data acquisition module (2) and the user search module (6) are communicatively connected to the storage identification control system; The data collection module (2) is used for users to submit popular science data; The user search module (6) is used for users to submit user questions.

3. The fusion reasoning system of a large language model and a knowledge graph according to claim 2 is characterized in that: The knowledge graph system includes a knowledge graph (12) and a graphic library (10) for storing the knowledge graph (12), wherein the graphic library (10) is communicatively connected with a large model reasoning and judgment module, a graph editing and indexing display system, and a storage and identification control system; The data cloud storage module (5) is in communication connection with the data acquisition module (2), the user search module (6), the semantic extraction module (4), the recognition module (7), and the central control module (8); the semantic extraction module (4) and the recognition module (7) are in communication connection with the large model reasoning and judgment module; and the large model reasoning and judgment module provides support for the semantic extraction module (4), the recognition module (7), and the central control module (8); the recognition module (7) is in communication connection with the central control module (8); and the central control module (8) is in communication connection with the graphics library (10), the atlas editing index display system, and the large model reasoning and judgment module.

4. The large language model and knowledge graph fusion reasoning system according to claim 3 is characterized in that: The large model reasoning and judgment module includes a database (1), a language large model (9) and a reasoning and judgment module (11). The database (1) is connected to the language large model (9) in communication. The language large model (9) is connected to the semantic extraction module (4), the recognition module (7) and the reasoning and judgment module (11). The reasoning and judgment module (11) is connected to the graphics library (10), the central control module (8) and the atlas editing index display system.

5. The fusion reasoning system of a large language model and a knowledge graph according to claim 4 is characterized in that: The display module (13), index module (14), and atlas editing module (15) are all connected to the graphics library (10) for communication, and the display module (13), index module (14), and atlas editing module (15) are all connected to the central control module (8) for communication.

6. The large language model and knowledge graph fusion reasoning system according to claim 5 is characterized in that: It also includes a copywriting publishing platform (3) for publishing popular science data, and the copywriting publishing platform (3) is communicatively connected with the data acquisition module (2) and the atlas editing module (15).

7. The large language model and knowledge graph fusion reasoning system according to claim 6 is characterized in that: The data cloud storage module (5) is used for cloud storage of hot science popularization data and user questions.

8. The large language model and knowledge graph fusion reasoning system according to claim 7 is characterized in that: The central control module (8) is used for, when receiving a fast search instruction or a relationship determination instruction, calling the text vector encoder model of the language large model (9) to encode the semantic entity N and the node entity F, obtaining the encoded semantic entity n and the encoded node entity f, calculating the cosine similarity between the encoded semantic entity n and the encoded node entity f, and performing semantic matching on the encoded semantic entity n and the encoded node entity f according to the cosine similarity, and when the semantic matching is found, sending an indexing instruction to the indexing module (14); When the central control module (8) receives the quick search instruction, the central control module (8) determines whether the semantic entity N belongs to the user question: When the answer is yes, the central control module (8) sends a display instruction 1 to the display module (13); When the answer is no, the central control module (8) terminates the task, the central control module (8) sends a correct instruction to the display module (13), and the display module (13) displays the correct words; When the central control module (8) receives the entity relationship determination instruction, the central control module (8) sends the determination instruction, the semantic entity N and the relationship L to the reasoning determination module (11); The index path is used by the graphic library (10) to retrieve the graphic at the location of the node entity F of the knowledge graph (12); The display module (13) is used to retrieve the node entity F of the knowledge graph (12) from the graphic library (10) according to the index path, and when receiving the display instruction sent by the central control module (8), the display module (13) displays the position of the node entity F in the knowledge graph (12) to the user.

9. The large language model and knowledge graph fusion reasoning system according to claim 8 is characterized in that: The reasoning and judging module (11) is used for, when receiving the judgment instruction, semantic entity N and relationship L sent by the central control module (8), the reasoning and judging module (11) calls the two node entities F of the knowledge graph (12) from the graphic library (10) through the index path, the reasoning and judging module (11) calls the text vector encoder model of the language large model (9) to encode the relationship S between the two node entities F and the relationship L between the two semantic entities N, and obtains the encoded relationship s and the encoded relationship l, the reasoning and judging module (11) calculates the cosine similarity of the encoded relationship s and the relationship l, and judges whether the semantics of the two are the same according to the cosine similarity; If the two have the same semantics, the reasoning and judgment module (11) determines that the relationship L is correct. At this time, the reasoning and judgment module (11) determines whether the semantic entity N belongs to the user question: If so, the reasoning and judgment module (11) sends a display instruction to the central control module (8), and the central control module (8) sends a display instruction three to the display module (13), and the display module (13) displays the positions of the two node entities F in the knowledge graph (12) to the user; If not, the reasoning and judgment module (11) sends a task termination instruction to the central control module (8), the central control module (8) sends a correct instruction to the display module (13), and the display module (13) displays the correct words to the user; If the semantics of the two are different, the reasoning and judgment module (11) determines that the relationship L is wrong. At this time, the reasoning and judgment module (11) calls the language large model (9) to read the detailed attribute X of the relationship S corresponding to the wrong relationship L and the detailed attribute B1 and detailed attribute B2 of the two node entities F corresponding to the relationship S from the database (1), and sends the detailed attribute B1, detailed attribute B2, detailed attribute X and the graph editing instruction to the central control module (8), and then the central control module (8) sends it to the graph editing module (15).

10. The fusion reasoning system of a large language model and a knowledge graph according to claim 9 is characterized in that: The graph editing module (15) determines whether the semantic entity F belongs to the user question: When the answer is yes, the atlas editing module (15) sends the edited screenshot and the second display instruction to the display module (13), and the display module (13) displays the edited screenshot to the user; When the answer is no, the atlas editing module (15) sends the edited screenshot to the delisting application mailbox of the copywriting publishing platform (3).

Images