Method and System for Implementing Intelligent Question Answering Based on Fine-Tuning of Large Language Models

Abstract

The present invention relates to the field of interactive question-answering technology, and discloses a method for realizing intelligent question-answering based on fine-tuning of a large language model, including: determining the subject association position of the industry knowledge subject according to the number of associated entities and the number of associated attributes, connecting the industry knowledge subjects related by using relationship attributes, and obtaining an association vector. Establishing a vector association relationship between the association vector and the relationship attribute through an initial large language model, obtaining the current industry knowledge subject, extracting the current relationship attribute, indexing the target industry knowledge subject according to the current subject association position and the current association vector, and feeding it back to the user. The present invention also proposes a system, an electronic device, and a computer-readable storage medium for realizing intelligent question-answering based on fine-tuning of a large language model. The present invention can effectively solve the problems of low answer accuracy and slow retrieval speed existing in the existing interactive response system, and can be applied to intelligent interactive question-answering of various industry knowledge bases to provide users with efficient and accurate information services.

Description

Technical Field

[0001] The present invention relates to the technical field of interactive question - answering, and particularly to a method, system, electronic device and computer - readable storage medium for realizing intelligent question - answering based on fine - tuning of a large - language model. Background Art

[0002] Interactive Voice Respond (IVR) means that a user asks a series of questions related to a theme, and the system answers the user's multi - aspect questions about the same entity or event one by one in the form of a dialogue. An industry knowledge base refers to a knowledge base established based on the knowledge data of a certain industry. Currently, with the development of artificial intelligence, the combination of the industry knowledge base and interactive response technology is becoming closer and closer, and using the industry knowledge base to automatically answer the user's interactive question statements has become the current research trend.

[0003] Currently, the main method for automatic reply of interactive response is to perform syntactic and grammatical optimizations such as anaphora resolution on the user's interactive question statements, then extract keywords from the optimized interactive question statements, and use the keywords to match answers in a pre - constructed database. This way of answer feedback for interactive response has problems such as low answer accuracy and slow retrieval speed. Summary of the Invention

[0004] The present invention provides a method, system and computer - readable storage medium for realizing intelligent question - answering based on fine - tuning of a large - language model, and its main purpose is to solve the problems of low answer accuracy and slow retrieval speed in the answer feedback method of interactive response.

[0005] To achieve the above object, a method for realizing intelligent question - answering based on fine - tuning of a large - language model provided by the present invention includes:

[0006] Obtain industry knowledge data, sequentially extract industry knowledge subjects from the industry knowledge data, and identify the number of associated entities and the number of associated attributes of the industry knowledge subjects, where the industry knowledge subjects can be industry knowledge entities or industry knowledge attributes;

[0007] Determine the subject - associated position of the industry knowledge subject in a pre - constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes, and obtain an initial associated coordinate knowledge graph;

[0008] Obtain the relationship attributes between industry knowledge subjects in the industry knowledge data, and use the relationship attributes to connect the industry knowledge subjects in the initial associated coordinate knowledge graph, and obtain a target associated coordinate knowledge graph;

[0009] Obtain the association vector between two industry knowledge entities in the target associated coordinate knowledge graph, and fine-tune the pre-constructed initial large language model according to the association vector, industry knowledge entity and relationship attribute to obtain the target large language model;

[0010] Obtain the interactive question statement of the user, and determine whether there is an industry knowledge entity in the interactive question statement;

[0011] If there is no industry knowledge entity in the interactive question statement, obtain the set of pre-interactive statements containing the interactive question statement, and extract the current industry knowledge entity from the set of pre-interactive statements;

[0012] If there is an industry knowledge entity in the interactive question statement, extract the current industry knowledge entity from the interactive question statement;

[0013] Extract the current relationship attribute of the current industry knowledge entity in the interactive question statement, input the current industry knowledge entity and the current relationship attribute into the target large language model to obtain the current association vector;

[0014] Obtain the current entity association position of the current industry knowledge entity, index the target industry knowledge entity in the target associated coordinate knowledge graph according to the current entity association position and the current association vector, and feedback the target industry knowledge entity to the user.

[0015] Optionally, the determining the entity association position of the industry knowledge entity in the pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of association attributes to obtain the initial associated coordinate knowledge graph includes:

[0016] Use the number of associated entities as the horizontal coordinate point in the industry knowledge association coordinate system;

[0017] Use the number of association attributes as the vertical coordinate point in the industry knowledge association coordinate system;

[0018] Determine the entity association position of the industry knowledge entity in the industry knowledge association coordinate system according to the horizontal coordinate point and the vertical coordinate point;

[0019] Calibrate the entity association positions of all industry knowledge entities in the industry knowledge association coordinate system to obtain the initial associated coordinate knowledge graph.

[0020] Optionally, the obtaining the relationship attributes between industry knowledge entities in the industry knowledge data and using the relationship attributes to connect the industry knowledge entities in the initial associated coordinate knowledge graph to obtain the target associated coordinate knowledge graph includes:

[0021] Obtain a set of associated knowledge entities associated with the industry knowledge entity;

[0022] Successively extract associated knowledge entities from the set of associated knowledge entities, and identify the relationship attributes between the industry knowledge entity and the associated knowledge entities;

[0023] Use the relationship attributes to connect the industry knowledge entity and the associated knowledge entities, and obtain the target associated coordinate knowledge graph.

[0024] Optionally, obtaining the association vector between two industry knowledge entities in the target associated coordinate knowledge graph includes:

[0025] Obtain the main body association positions of two industry knowledge entities that are connected in the target associated coordinate knowledge graph;

[0026] Use the main body association positions to calculate the association vector between the two industry knowledge entities according to a pre-constructed association vector calculation formula.

[0027] Optionally, the association vector calculation formula is as follows:

[0028]

[0029] Among them, represents the association vector from industry knowledge entity a to industry knowledge entity b, (x a , y a ) represents the main body association position of industry theme a, (x b , y b ) represents the main body association position of industry theme b, represents the horizontal unit vector, represents the vertical unit vector.

[0030] Optionally, determining whether there is an industry knowledge entity in the interactive question statement includes:

[0031] Remove the preset stop words and interrogative words from the interactive question statement to obtain an initial content word set;

[0032] Determine whether there is an industry knowledge entity in the initial content word set;

[0033] If there is an industry knowledge entity in the initial content word set, it is determined that there is an industry knowledge entity in the interactive question statement;

[0034] If there is no industry knowledge entity in the initial content word set, it is determined that there is no industry knowledge entity in the interactive question statement.

[0035] Optionally, obtaining the set of pre - interaction statements containing the interactive question statement and extracting the current industry knowledge subject from the set of pre - interaction statements includes:

[0036] Extracting the first pre - interaction statement of the interactive question statement from the set of pre - interaction statements;

[0037] Determining whether there is an industry knowledge subject in the first pre - interaction statement;

[0038] If there is no industry knowledge subject in the first pre - interaction statement, extract pre - interaction statements in reverse order from the set of pre - interaction statements until there is an industry knowledge subject in the pre - interaction statement, and use the industry knowledge subject as the current industry knowledge subject;

[0039] If there is an industry knowledge subject in the first pre - interaction statement, extract the current industry knowledge subject from the first pre - interaction statement.

[0040] Optionally, indexing the target industry knowledge subject in the target association coordinate knowledge graph according to the current subject association position and the current association vector includes:

[0041] Using the current subject association position as the vector starting point of the current association vector to obtain an indexed association vector;

[0042] Identifying the vector end point of the indexed association vector and identifying the industry knowledge subject with the same subject association position as the vector end point;

[0043] Using the industry knowledge subject with the same subject association position as the vector end point as the target industry knowledge subject.

[0044] Optionally, extracting the current relationship attribute of the current industry knowledge subject in the interactive question statement, the method further includes:

[0045] Determining whether there is a current relationship attribute matching the current industry knowledge subject in the interactive question statement;

[0046] If there is no current relationship attribute matching the current industry knowledge subject in the interactive question statement, continue to extract the current industry knowledge subject from the set of pre - interaction statements until a current industry knowledge subject matching the current relationship attribute is obtained, and extract the current relationship attribute of the current industry knowledge subject in the interactive question statement;

[0047] If there is a current relationship attribute matching the current industry knowledge subject in the interactive question statement, extract the current relationship attribute of the current industry knowledge subject in the interactive question statement.

[0048] To solve the above problems, the present invention also provides a system for realizing intelligent question answering based on fine-tuning of a large language model. The system includes:

[0049] An initial associated coordinate knowledge graph construction module, configured to obtain industry knowledge data, sequentially extract industry knowledge entities in the industry knowledge data, identify the number of associated entities and the number of associated attributes of the industry knowledge entities, where the industry knowledge entities may be industry knowledge entities or industry knowledge attributes; determine the main body associated position of the industry knowledge entities in a pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes to obtain an initial associated coordinate knowledge graph;

[0050] A target associated coordinate knowledge graph construction module, configured to obtain the relationship attributes between industry knowledge entities in the industry knowledge data, and use the relationship attributes to connect the industry knowledge entities in the initial associated coordinate knowledge graph to obtain a target associated coordinate knowledge graph;

[0051] A vector association relationship establishment module, configured to obtain the association vector between two industry knowledge entities in the target associated coordinate knowledge graph, and fine-tune a pre-constructed initial large language model according to the association vector, industry knowledge entities and relationship attributes to obtain a target large language model;

[0052] A current industry knowledge entity extraction module, configured to obtain an interactive question statement of a user, and determine whether there is an industry knowledge entity in the interactive question statement; if there is no industry knowledge entity in the interactive question statement, obtain a set of pre-interaction statements including the interactive question statement, and extract the current industry knowledge entity from the set of pre-interaction statements; if there is an industry knowledge entity in the interactive question statement, extract the current industry knowledge entity from the interactive question statement;

[0053] A target industry knowledge entity feedback module, configured to extract the current relationship attribute of the current industry knowledge entity in the interactive question statement, input the current industry knowledge entity and the current relationship attribute into the target large language model to obtain a current association vector; obtain the current main body associated position of the current industry knowledge entity, and index the target industry knowledge entity in the target associated coordinate knowledge graph according to the current main body associated position and the current association vector, and feedback the target industry knowledge entity to the user.

[0054] To solve the above problems, the present invention also provides an electronic device, and the electronic device includes:

[0055] At least one processor; and,

[0056] A memory communicatively connected to the at least one processor; wherein,

[0057] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to implement the method for realizing intelligent question answering based on fine-tuning of a large language model as described above.

[0058] To solve the above problems, the present invention further provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is executed by a processor in an electronic device to implement the method for realizing intelligent question answering based on fine-tuning of a large language model as described above.

[0059] Compared with the background art, in the answer feedback method of interactive response, there are problems of low answer accuracy and slow retrieval speed. In the process of extracting the target industry knowledge subject in the embodiment of the present invention, it is necessary to first construct a target associated coordinate knowledge graph, so as to achieve the purpose of indexing the target industry knowledge subject in the target associated coordinate knowledge graph according to the current subject association position and the current association vector. The current subject association position refers to the industry knowledge subject in the user's interactive question statement. When constructing the target associated coordinate knowledge graph, first identify the number of associated entities and the number of associated attributes of the industry knowledge subject, and then use the number of associated entities and the number of associated attributes to determine the subject association position of the industry knowledge subject in the target associated coordinate knowledge graph. According to the subject association position relationship between two industry knowledge subjects in the target associated coordinate knowledge graph, calculate the association vector. Fine-tune the pre-constructed initial large language model according to the association vector, the industry knowledge subject and the relationship attribute to obtain a target large language model. By establishing the association vector, the indexing efficiency of the target industry knowledge subject is greatly simplified. At this time, the current industry knowledge subject and the corresponding current relationship attribute can be extracted from the interactive question statement or the preposed interactive statement set. Determine the current subject association position and the current association vector according to the current industry knowledge subject and the corresponding current relationship attribute, and finally index the target industry knowledge subject in the target associated coordinate knowledge graph according to the current subject association position and the current association vector and feedback it to the user. Therefore, the method, system, electronic device and computer-readable storage medium for realizing intelligent question answering based on fine-tuning of a large language model proposed by the present invention can solve the problems of low answer accuracy and slow retrieval speed in the answer feedback method of interactive response. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic flowchart of a method for realizing intelligent question answering based on fine-tuning of a large language model provided by an embodiment of the present invention;

[0061] Figure 2 It is a functional module diagram of a system for realizing intelligent question answering based on fine-tuning of a large language model provided by an embodiment of the present invention;

[0062] Figure 3 The structural schematic diagram of the electronic device for implementing the method of intelligent question answering based on fine-tuning of large language models provided by an embodiment of the present invention.

[0063] The implementation of the object, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0064] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0065] An embodiment of the present application provides a method for implementing intelligent question answering based on fine-tuning of large language models. The execution subject of the method for implementing intelligent question answering based on fine-tuning of large language models includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for implementing intelligent question answering based on fine-tuning of large language models can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0066] Embodiment 1:

[0067] Refer to Figure 1 As shown, it is a flowchart of a method for implementing intelligent question answering based on fine-tuning of large language models provided by an embodiment of the present invention. In this embodiment, the method for implementing intelligent question answering based on fine-tuning of large language models includes:

[0068] S1. Obtain industry knowledge data, sequentially extract industry knowledge entities in the industry knowledge data, and identify the number of associated entities and the number of associated attributes of the industry knowledge entities, where the industry knowledge entities can be industry knowledge entities or industry knowledge attributes.

[0069] In the embodiment of the present invention, the industry knowledge data refers to professional knowledge data in a certain industry field, such as: knowledge data in industry fields such as economic and financial industries, IT and communication industries, healthcare industries, legal and intellectual property industries, engineering and construction industries, education and training industries, culture and entertainment industries, energy and chemical industries, and agriculture, forestry, animal husbandry and fishery industries. The specific granularity of division can be changed according to actual situations. For example: the agriculture, forestry, animal husbandry and fishery industry can be divided into agriculture, forestry, animal husbandry and fishery; the engineering and construction industry can be further divided into fields such as building codes and standards, construction management, and building maintenance.

[0070] Specifically, the industry knowledge data may include relevant knowledge data such as industry definitions, distribution characteristics, policies and regulations, and industry standards within the industry. For example, when the industry knowledge data is in the field of building codes and standards, the industry knowledge data may be: "The welding methods of steel structures mainly include manual arc welding, gas shielded welding, submerged arc welding, plasma welding, electron beam welding, bolt welding, electroslag welding, and submerged arc welding"; "The steel structure painting project is usually divided into anti-corrosion coating painting and fireproof coating painting"; "The waterproof grade of underground projects can be divided into four levels"; "The specifications of thin and small-sized plates are a thickness of less than 10 mm and a side length of less than 400 mm."

[0071] It can be understood that the industry knowledge subject includes two categories: industry knowledge entities and industry knowledge attributes. The industry knowledge subject refers to the entities or attributes obtained by splitting the industry knowledge data in the form of a knowledge graph. The industry knowledge entity is the entity of the industry knowledge data, and the industry knowledge attribute is the attribute of the industry knowledge data. For example, when the industry knowledge data is: "The welding methods of steel structures mainly include manual arc welding, gas shielded welding, submerged arc welding, plasma welding, electron beam welding, bolt welding, electroslag welding, and submerged arc welding", the industry knowledge entities are "welding methods of steel structures", "manual arc welding", "gas shielded welding", "submerged arc welding", "plasma welding", "electron beam welding", "bolt welding", "electroslag welding", and "submerged arc welding". When the industry knowledge data is: "The waterproof grade of underground projects can be divided into four levels", the industry knowledge attribute is "four levels", and the industry knowledge entity is "the waterproof grade of underground projects".

[0072] It can be explained that the associated entity number refers to the number of industry knowledge entities associated with the industry knowledge subject, and the associated attribute number refers to the number of industry knowledge attributes associated with the industry knowledge subject. For example, when the industry knowledge data is: "The welding methods of steel structures mainly include manual arc welding, gas shielded welding, submerged arc welding, plasma welding, electron beam welding, bolt welding, electroslag welding, and submerged arc welding", the associated entity number of the industry knowledge subject "welding methods of steel structures" is 8; when the industry knowledge data is: "The waterproof grade of underground projects can be divided into four levels", the associated attribute number of the industry knowledge subject "the waterproof grade of underground projects" is 1; when the industry knowledge data is: "The welding methods of steel structures mainly include 8 types", the associated attribute number of the industry knowledge subject "welding methods of steel structures" is 1, and the corresponding 1 industry knowledge attribute is "8 types".

[0073] S2. Determine the subject association position of the industry knowledge subject in the pre-constructed industry knowledge association coordinate system according to the associated entity number and the associated attribute number, and obtain the initial associated coordinate knowledge graph.

[0074] It is understandable that the industry knowledge association coordinate system refers to a two-dimensional coordinate system constructed with the number of associated entities as the abscissa unit and the number of associated attributes as the ordinate unit. The main body association position refers to the coordinate position of the industry knowledge main body in the industry knowledge association coordinate system. For example, when the number of associated entities of the industry knowledge main body "welding method of steel structure" is 8 and the number of associated attributes is 1, the main body association position of the industry knowledge main body "welding method of steel structure" is (8, 1).

[0075] In an embodiment of the present invention, the determining the main body association position of the industry knowledge main body in the pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes, and obtaining the initial associated coordinate knowledge graph includes:

[0076] Taking the number of associated entities as the horizontal coordinate point in the industry knowledge association coordinate system;

[0077] Taking the number of associated attributes as the vertical coordinate point in the industry knowledge association coordinate system;

[0078] Determining the main body association position of the industry knowledge main body in the industry knowledge association coordinate system according to the horizontal coordinate point and the vertical coordinate point;

[0079] Calibrating the main body association positions of all industry knowledge main bodies in the industry knowledge association coordinate system to obtain the initial associated coordinate knowledge graph.

[0080] S3. Obtaining the relationship attributes between industry knowledge main bodies in the industry knowledge data, and using the relationship attributes to connect the industry knowledge main bodies in the initial associated coordinate knowledge graph to obtain a target associated coordinate knowledge graph.

[0081] It is interpretable that the relationship attributes include relationship association and attribute association. The relationship association refers to the association between an industry knowledge main body and an industry knowledge entity, and the attribute association refers to the association between an industry knowledge main body and an industry knowledge attribute. For example, when the industry knowledge data is: "The welding methods of steel structures mainly include manual arc welding, gas shielded welding, sub-shielded arc welding, plasma welding, electron beam welding, bolt welding, electroslag welding, submerged arc welding", the relationship association is "mainly include", and when the industry knowledge data is: "The waterproof grade of underground engineering can be divided into four levels", the attribute association is "can be divided into".

[0082] In an embodiment of the present invention, the obtaining the relationship attributes between industry knowledge main bodies in the industry knowledge data, and using the relationship attributes to connect the industry knowledge main bodies in the initial associated coordinate knowledge graph to obtain a target associated coordinate knowledge graph includes:

[0083] Obtain a set of associated knowledge entities associated with the industry knowledge entity;

[0084] Successively extract associated knowledge entities from the set of associated knowledge entities, and identify the relationship attributes between the industry knowledge entity and the associated knowledge entities;

[0085] Use the relationship attributes to connect the industry knowledge entity and the associated knowledge entities, and obtain the target associated coordinate knowledge graph.

[0086] S4. Obtain the association vector between two industry knowledge entities in the target associated coordinate knowledge graph, and fine-tune the pre-constructed initial large language model according to the association vector, industry knowledge entity and relationship attribute to obtain the target large language model.

[0087] It can be understood that the association vector refers to a vector constructed based on the entity association positions of two industry knowledge entities with relationship attributes. Each association vector corresponds to a relationship attribute. For example: when the industry knowledge entities are "welding method of steel structure" and "plasma welding", the relationship attribute corresponding to the association vector pointing from "welding method of steel structure" to "plasma welding" is "mainly includes", and the vector direction is from the entity association position of "welding method of steel structure" to the entity association position of "plasma welding"; the relationship attribute corresponding to the association vector pointing from "plasma welding" to "welding method of steel structure" is "belongs to", and the vector direction is from the entity association position of "plasma welding" to the entity association position of "welding method of steel structure".

[0088] Furthermore, the initial large language model refers to an artificial intelligence model that can be trained using a large corpus. It integrates various language knowledge and language rules and has the ability to understand, generate, and process natural language. Such models can understand human natural language input and generate semantically relevant outputs based on the input content. The working principle of the large language model mainly includes pre-training, fine-tuning, and learning based on prompt words. In the pre-training stage, the large language model learns through a large amount of text materials to establish its own knowledge base, which serves as the basis for processing various natural languages. In the fine-tuning stage, the model can be adjusted according to text data in a specific task or field (for example: text data in the field of building codes and standards) to improve the accuracy and integrity of the output. At the same time, the quality of the prompt words is also crucial for the output of the large language model.

[0089] In the embodiments of the present invention, the obtaining of the association vector between two industry knowledge entities in the target associated coordinate knowledge graph includes:

[0090] Obtain the entity association positions of two industry knowledge entities that are connected in the target associated coordinate knowledge graph;

[0091] Using the subject association position, calculate the association vector between two industry knowledge subjects according to a pre-constructed association vector calculation formula.

[0092] Specifically, the association vector calculation formula is as follows:

[0093]

[0094] Wherein, represents the association vector from industry knowledge subject a to industry knowledge subject b, (x a , y a ) represents the subject association position of industry theme a, (x b , y b ) represents the subject association position of industry theme b, represents the horizontal unit vector, represents the vertical unit vector.

[0095] It can be understood that through the correspondence between the association vector and the industry knowledge subject and the relationship attribute, fine-tuning training is performed on the initial large language model, so as to establish a one-to-one correspondence between the association vector and the industry knowledge subject and the relationship attribute in the initial large language model, and achieve the purpose of retrieving the association vector according to the industry knowledge subject and the relationship attribute. S5. Obtain the interactive question statement of the user, and judge whether there is an industry knowledge subject in the interactive question statement.

[0096] It can be understood that the interactive question statement refers to the question statement in the interactive Q&A. The interactive Q&A allows the user to ask a series of questions related to the theme and answer the user's questions about multiple aspects of the same entity or event one by one in the form of a dialogue.

[0097] In the embodiment of the present invention, judging whether there is an industry knowledge subject in the interactive question statement includes:

[0098] Remove the preset stop words and interrogative words in the interactive question statement to obtain an initial content word set;

[0099] Judge whether there is an industry knowledge subject in the initial content word set;

[0100] If there is an industry knowledge subject in the initial content word set, it is determined that there is an industry knowledge subject in the interactive question statement;

[0101] If there is no industry knowledge subject in the initial content word set, it is determined that there is no industry knowledge subject in the interactive question statement.

[0102] It is understandable that the stop words refer to certain words or terms that are automatically filtered out before or after processing natural language data (or text) in information retrieval to save storage space and improve search efficiency. They can be "de", "jishi", "shi", etc. The interrogative words can be "shenme", "zenme", "ruhe", etc. The stop words and interrogative words can be summarized using corresponding word lists.

[0103] Further, when determining whether there is an industry knowledge entity in the initial content word set, the content words in the initial content word set can be compared one by one with the preset industry knowledge entities. If they are the same, then there is an industry knowledge entity; if not, then there is no industry knowledge entity.

[0104] If there is no industry knowledge entity in the interactive question statement, then execute S6, obtain the set of previous interactive statements containing the interactive question statement, and extract the current industry knowledge entity from the set of previous interactive statements.

[0105] It is explainable that the set of previous interactive statements refers to a set of statements with a predetermined number before the interactive question statement, which can be the first 10 interactive question statements of the interactive question statement.

[0106] In the embodiment of the present invention, the obtaining the set of previous interactive statements containing the interactive question statement and extracting the current industry knowledge entity from the set of previous interactive statements includes:

[0107] Extract the first previous interactive statement of the interactive question statement from the set of previous interactive statements;

[0108] Determine whether there is an industry knowledge entity in the first previous interactive statement;

[0109] If there is no industry knowledge entity in the first previous interactive statement, then extract the previous interactive statements in reverse order from the set of previous interactive statements until there is an industry knowledge entity in the previous interactive statement, and use the industry knowledge entity as the current industry knowledge entity;

[0110] If there is an industry knowledge entity in the first previous interactive statement, then extract the current industry knowledge entity from the first previous interactive statement.

[0111] It is understandable that the first previous interactive statement refers to the previous interactive question statement of the interactive question statement. The reverse extraction of the previous interactive statements refers to extracting the interactive question statements from the set of previous interactive statements in the order from near to far from the interactive question statement.

[0112] If there is an industry knowledge entity in the interactive question statement, then execute S7, and extract the current industry knowledge entity from the interactive question statement.

[0113] S8. Extract the current relationship attribute of the current industry knowledge subject in the interactive question statement, input the current industry knowledge subject and the current relationship attribute into the target large language model to obtain a current association vector.

[0114] Further, when the interactive question statement is: "How many levels can the waterproof grade of underground engineering be divided into?", the current industry knowledge subject is "the waterproof grade of underground engineering", and the current relationship attribute is: "can be divided into".

[0115] In the embodiment of the present invention, for the method of extracting the current relationship attribute of the current industry knowledge subject in the interactive question statement, the method further includes:

[0116] Judge whether there is a current relationship attribute in the interactive question statement that matches the current industry knowledge subject;

[0117] If there is no current relationship attribute in the interactive question statement that matches the current industry knowledge subject, continue to extract the current industry knowledge subject in the pre - interaction statement set until a current industry knowledge subject that matches the current relationship attribute is obtained, and extract the current relationship attribute of the current industry knowledge subject in the interactive question statement;

[0118] If there is a current relationship attribute in the interactive question statement that matches the current industry knowledge subject, extract the current relationship attribute of the current industry knowledge subject in the interactive question statement.

[0119] Further, the current industry knowledge subject extracted in the interactive question statement may not be collocatable with the current relationship attribute. In this case, it is necessary to re - extract the current industry knowledge subject in the pre - interaction statement set. The collocation relationship between the industry knowledge subject and the relationship attribute can be judged by establishing a collocation relationship table of the industry knowledge subject and the relationship attribute.

[0120] S9. Obtain the current subject association position of the current industry knowledge subject, index the target industry knowledge subject in the target association coordinate knowledge graph according to the current subject association position and the current association vector, and feedback the target industry knowledge subject to the user.

[0121] In the embodiment of the present invention, for indexing the target industry knowledge subject in the target association coordinate knowledge graph according to the current subject association position and the current association vector, it includes:

[0122] Use the current subject association position as the vector starting point of the current association vector to obtain an index association vector;

[0123] Identify the vector end point of the index association vector, and identify the industry knowledge entity whose entity association position is the same as the vector end point;

[0124] Use the industry knowledge entity whose entity association position is the same as the vector end point as the target industry knowledge entity.

[0125] It is understandable that the feedback method of the target industry knowledge entity can be text or voice.

[0126] Compared with the background technology, there are problems such as low answer accuracy and slow retrieval speed in the answer feedback method of interactive response. In the process of extracting the target industry knowledge entity in the embodiment of the present invention, it is necessary to first construct a target association coordinate knowledge graph, so as to achieve the purpose of indexing the target industry knowledge entity in the target association coordinate knowledge graph according to the current entity association position and the current association vector. The current entity association position refers to the industry knowledge entity in the user's interactive question sentence. When constructing the target association coordinate knowledge graph, first identify the number of associated entities and the number of associated attributes of the industry knowledge entity, and then use the number of associated entities and the number of associated attributes to determine the entity association position of the industry knowledge entity in the target association coordinate knowledge graph. Calculate the association vector through the entity association position relationship between two industry knowledge entities in the target association coordinate knowledge graph, and fine-tune the pre-constructed initial large language model according to the association vector, industry knowledge entity and relationship attributes to obtain the target large language model. By establishing the association vector, the indexing efficiency of the target industry knowledge entity is greatly simplified. At this time, the current industry knowledge entity and the corresponding current relationship attribute can be extracted from the interactive question sentence or the pre-interactive sentence set. Determine the current entity association position and the current association vector according to the current industry knowledge entity and the corresponding current relationship attribute, and finally index the target industry knowledge entity in the target association coordinate knowledge graph according to the current entity association position and the current association vector and feedback it to the user. Therefore, the method, system, electronic device and computer-readable storage medium for realizing intelligent question answering based on fine-tuning of a large language model proposed by the present invention can solve the problems of low answer accuracy and slow retrieval speed in the answer feedback method of interactive response.

[0127] Embodiment 2:

[0128] As Figure 2 shown, it is a functional module diagram of a system for realizing intelligent question answering based on fine-tuning of a large language model provided by an embodiment of the present invention.

[0129] The system 100 for realizing intelligent question answering based on fine-tuning of a large language model according to the present invention can be installed in an electronic device. According to the functions realized, the system 100 for realizing intelligent question answering based on fine-tuning of a large language model can include an initial associated coordinate knowledge graph construction module 101, a target associated coordinate knowledge graph construction module 102, a vector association relationship establishment module 103, a current industry knowledge subject extraction module 104, and a target industry knowledge subject feedback module 105. The modules according to the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0130] The initial associated coordinate knowledge graph construction module 101 is configured to obtain industry knowledge data, sequentially extract industry knowledge subjects from the industry knowledge data, identify the number of associated entities and the number of associated attributes of the industry knowledge subjects, where the industry knowledge subjects can be industry knowledge entities or industry knowledge attributes; determine the subject association position of the industry knowledge subjects in a pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes, and obtain an initial associated coordinate knowledge graph;

[0131] The target associated coordinate knowledge graph construction module 102 is configured to obtain the relationship attributes between industry knowledge subjects in the industry knowledge data, and use the relationship attributes to connect the industry knowledge subjects in the initial associated coordinate knowledge graph, and obtain a target associated coordinate knowledge graph;

[0132] The vector association relationship establishment module 103 is configured to obtain the association vector between two industry knowledge subjects in the target associated coordinate knowledge graph, and fine-tune a pre-constructed initial large language model according to the association vector, industry knowledge subjects, and relationship attributes, and obtain a target large language model;

[0133] The current industry knowledge subject extraction module 104 is configured to obtain an interactive question statement of a user, and determine whether there is an industry knowledge subject in the interactive question statement; if there is no industry knowledge subject in the interactive question statement, obtain a pre-interactive statement set including the interactive question statement, and extract the current industry knowledge subject from the pre-interactive statement set; if there is an industry knowledge subject in the interactive question statement, extract the current industry knowledge subject from the interactive question statement;

[0134] The target industry knowledge entity feedback module 105 is used to extract the current relationship attributes of the current industry knowledge entity in the interactive question statement, input the current industry knowledge entity and the current relationship attributes into the target large language model to obtain a current association vector; obtain the current entity association position of the current industry knowledge entity, and index the target industry knowledge entity in the target association coordinate knowledge graph according to the current entity association position and the current association vector, and feedback the target industry knowledge entity to the user.

[0135] Specifically, each module in the system 100 for implementing intelligent question answering based on fine-tuning of a large language model in the embodiments of the present invention adopts the same technical means as the Figure 1 method for implementing intelligent question answering based on fine-tuning of a large language model described above, and can produce the same technical effects, which will not be elaborated here.

[0136] Embodiment 3:

[0137] As Figure 3 shown, it is a schematic structural diagram of an electronic device for implementing a method for intelligent question answering based on fine-tuning of a large language model provided by an embodiment of the present invention.

[0138] The electronic device 1 may include a processor 10, a memory 11, a bus 12, and a communication interface 13, and may further include a computer program stored in the memory 11 and executable on the processor 10, such as a program for intelligent question answering based on fine-tuning of a large language model.

[0139] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc. The memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as the mobile hard disk of the electronic device 1. The memory 11 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 1. Further, the memory 11 may also include both the internal storage unit and the external storage device of the electronic device 1. The memory 11 can be used not only to store application software installed on the electronic device 1 and various types of data, such as the code of a program for intelligent question answering based on fine-tuning of a large language model, but also to temporarily store data that has been output or will be output.

[0140] In some embodiments, the processor 10 may be composed of an integrated circuit. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple packaged integrated circuits with the same or different functions, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting various components of the entire electronic device through various interfaces and circuits, and executing various functions of the electronic device 1 and processing data by running or executing programs or modules stored in the memory 11 (such as programs for realizing intelligent question answering based on fine-tuning of large language models) and calling data stored in the memory 11.

[0141] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is set to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0142] Figure 3 Only the electronic device with components is shown. Those skilled in the art can understand that Figure 3 the shown structure does not constitute a limitation on the electronic device 1, and it may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0143] For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for powering each component. Preferably, the power source can be logically connected to the at least one processor 10 through a power management system, so as to realize functions such as charge management, discharge management, and power consumption management through the power management system. The power source may also include any components such as one or more DC or AC power sources, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator. The electronic device 1 may also include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0144] Furthermore, the electronic device 1 may further include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used to establish a communication connection between the electronic device 1 and other electronic devices.

[0145] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display the information processed in the electronic device 1 and to display a visual user interface.

[0146] It should be understood that the above embodiments are only for illustration purposes and are not limited by this structure in the scope of the patent application.

[0147] The program for realizing intelligent question answering by fine-tuning based on a large language model stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can:

[0148] Obtain industry knowledge data, sequentially extract industry knowledge entities in the industry knowledge data, and identify the number of associated entities and the number of associated attributes of the industry knowledge entities, where the industry knowledge entities may be industry knowledge entities or industry knowledge attributes;

[0149] Determine the main body association position of the industry knowledge entity in a pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes, and obtain an initial associated coordinate knowledge graph;

[0150] Obtain the relationship attributes between industry knowledge entities in the industry knowledge data, and use the relationship attributes to connect the industry knowledge entities in the initial associated coordinate knowledge graph, and obtain a target associated coordinate knowledge graph;

[0151] Obtain the association vector between two industry knowledge entities in the target associated coordinate knowledge graph, and fine-tune a pre-constructed initial large language model according to the association vector, industry knowledge entities and relationship attributes to obtain a target large language model;

[0152] Obtain an interactive question statement of a user, and determine whether there is an industry knowledge entity in the interactive question statement;

[0153] If there is no industry knowledge entity in the interactive question statement, obtain a set of pre-interactive statement containing the interactive question statement, and extract the current industry knowledge entity in the set of pre-interactive statement;

[0154] If there is an industry knowledge entity in the interactive question statement, extract the current industry knowledge entity in the interactive question statement;

[0155] Extract the current relationship attributes of the current industry knowledge subject in the interactive question statement, input the current industry knowledge subject and the current relationship attributes into the target large language model to obtain a current association vector;

[0156] Obtain the current subject association position of the current industry knowledge subject, index the target industry knowledge subject in the target association coordinate knowledge graph according to the current subject association position and the current association vector, and feedback the target industry knowledge subject to the user.

[0157] Specifically, for the specific implementation method of the above instructions by the processor 10, reference can be made to Figures 1 to 2 the description of the relevant steps in the corresponding embodiment, which will not be elaborated here.

[0158] Furthermore, if the modules / units integrated in the electronic device 1 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or system capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory).

[0159] The present invention also provides a computer-readable storage medium, and the readable storage medium stores a computer program, which when executed by a processor of an electronic device, can implement:

[0160] Obtain industry knowledge data, sequentially extract industry knowledge subjects in the industry knowledge data, and identify the number of associated entities and the number of associated attributes of the industry knowledge subjects, where the industry knowledge subjects can be industry knowledge entities or industry knowledge attributes;

[0161] Determine the subject association position of the industry knowledge subject in a pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes to obtain an initial association coordinate knowledge graph;

[0162] Obtain the relationship attributes between industry knowledge subjects in the industry knowledge data, and use the relationship attributes to connect the industry knowledge subjects in the initial association coordinate knowledge graph to obtain a target association coordinate knowledge graph;

[0163] Obtain the association vector between two industry knowledge subjects in the target association coordinate knowledge graph, and fine-tune a pre-constructed initial large language model according to the association vector, industry knowledge subjects and relationship attributes to obtain a target large language model;

[0164] Obtain the interactive question statement of the user, and determine whether there is an industry knowledge subject in the interactive question statement;

[0165] If there is no industry knowledge subject in the interactive question statement, obtain the set of previous interactive statements containing the interactive question statement, and extract the current industry knowledge subject from the set of previous interactive statements;

[0166] If there is an industry knowledge subject in the interactive question statement, extract the current industry knowledge subject from the interactive question statement;

[0167] Extract the current relationship attribute of the current industry knowledge subject in the interactive question statement, input the current industry knowledge subject and the current relationship attribute into the target large language model to obtain the current association vector;

[0168] Obtain the current subject association position of the current industry knowledge subject, index the target industry knowledge subject in the target association coordinate knowledge graph according to the current subject association position and the current association vector, and feedback the target industry knowledge subject to the user.

[0169] In several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0170] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0171] In addition, each functional module in various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware, or in the form of hardware plus software functional modules.

[0172] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for implementing intelligent question answering based on fine-tuning of a large language model, characterized in that: The method comprises: Acquire industry knowledge data, extract industry knowledge subjects in the industry knowledge data in sequence, and identify the number of associated entities and the number of associated attributes of the industry knowledge subjects, wherein the industry knowledge subjects are industry knowledge entities or industry knowledge attributes; Determine the subject association position of the industry knowledge subject in the pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes, and obtain an initial association coordinate knowledge graph; Acquire the relationship attributes between industry knowledge subjects in the industry knowledge data, and use the relationship attributes to correlate the industry knowledge subjects in the initial association coordinate knowledge graph to obtain a target association coordinate knowledge graph; The relationship attributes include relationship association and attribute association. The relationship association refers to the association between the industry knowledge subject and the industry knowledge entity, and the attribute association refers to the association between the industry knowledge subject and the industry knowledge attribute. Obtaining an association vector between two industry knowledge subjects in the target association coordinate knowledge graph, and fine-tuning a pre-built initial large language model according to the association vector, the industry knowledge subject, and the relationship attribute to obtain a target large language model; Obtaining an interactive question statement from a user, and determining whether there is an industry knowledge subject in the interactive question statement; If there is no industry knowledge subject in the interactive question statement, obtaining a preceding interactive statement set containing the interactive question statement, and extracting the current industry knowledge subject from the preceding interactive statement set; If there is an industry knowledge subject in the interactive question statement, extracting the current industry knowledge subject from the interactive question statement; Extracting the current relationship attributes of the current industry knowledge subject from the interactive question sentence, inputting the current industry knowledge subject and the current relationship attributes into the target large language model, and obtaining a current association vector; The current subject association position of the current industry knowledge subject is obtained, the target industry knowledge subject is indexed in the target association coordinate knowledge graph according to the current subject association position and the current association vector, and the target industry knowledge subject is fed back to the user.

2. The method for implementing intelligent question answering based on fine-tuning of a large language model as claimed in claim 1, characterized in that: The determining the subject association position of the industry knowledge subject in the pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes to obtain an initial association coordinate knowledge graph includes: Taking the number of associated entities as horizontal coordinate points in the industry knowledge associated coordinate system; Taking the associated attribute number as the vertical coordinate point in the industry knowledge associated coordinate system; Determine the subject association position of the industry knowledge subject in the industry knowledge association coordinate system according to the horizontal coordinate point and the vertical coordinate point; The subject association positions of all industry knowledge subjects are calibrated in the industry knowledge association coordinate system to obtain the initial association coordinate knowledge graph.

3. The method for implementing intelligent question answering based on fine-tuning of a large language model as claimed in claim 2, characterized in that: The step of obtaining the relationship attributes between the industry knowledge subjects in the industry knowledge data, and using the relationship attributes to connect the industry knowledge subjects in the initial association coordinate knowledge graph to obtain the target association coordinate knowledge graph includes: Acquire a set of associated knowledge subjects associated with the industry knowledge subject; Extracting related knowledge subjects in the related knowledge subject set in turn, and identifying the relationship attributes between the industry knowledge subject and the related knowledge subject; The industry knowledge subject and the associated knowledge subject are connected with each other using the relationship attribute to obtain the target associated coordinate knowledge graph.

4. The method for implementing intelligent question answering based on fine-tuning of a large language model as claimed in claim 3, characterized in that: The step of obtaining the association vector between two industry knowledge subjects in the target association coordinate knowledge graph includes: Obtaining the subject association positions of two industry knowledge subjects that are relatedly connected in the target association coordinate knowledge graph; The subject association position is utilized to calculate the association vector between two industry knowledge subjects according to a pre-constructed association vector calculation formula.

5. The method for implementing intelligent question answering based on fine-tuning of a large language model as claimed in claim 4, characterized in that: The calculation formula of the correlation vector is as follows: in, represents the association vector from industry knowledge subject a to industry knowledge subject b, (x a ,y a ) represents the subject-related position of industry subject a, (x b ,y b ) represents the subject-related position of industry subject b, represents the transverse unit vector, represents the longitudinal unit vector.

6. The method for implementing intelligent question answering based on fine-tuning of a large language model as claimed in claim 1, characterized in that: The determining whether there is an industry knowledge subject in the interactive question statement includes: Removing preset stop words and question words from the interactive question sentence to obtain an initial content word set; Determining whether there is an industry knowledge subject in the initial content word set; If the initial content word set contains an industry knowledge subject, determining that the interactive question statement contains an industry knowledge subject; If there is no industry knowledge subject in the initial content word set, it is determined that there is no industry knowledge subject in the interactive question sentence.

7. The method for implementing intelligent question answering based on fine-tuning of a large language model as claimed in claim 6, characterized in that: The obtaining of a pre-interaction statement set including the interactive question statement, and extracting the current industry knowledge subject from the pre-interaction statement set includes: Extracting a first preceding interactive sentence of the interactive question sentence from the preceding interactive sentence set; Determining whether there is an industry knowledge subject in the first preceding interactive statement; If there is no industry knowledge subject in the first preceding interaction statement, extracting preceding interaction statements in reverse order from the preceding interaction statement set until an industry knowledge subject exists in the preceding interaction statement, and taking the industry knowledge subject as the current industry knowledge subject; If the first preceding interaction statement contains an industry knowledge subject, the current industry knowledge subject is extracted from the first preceding interaction statement.

8. The method for implementing intelligent question answering based on fine-tuning of a large language model as claimed in claim 3, characterized in that: The step of indexing the target industry knowledge subject in the target association coordinate knowledge graph according to the current subject association position and the current association vector includes: Taking the current subject association position as the vector starting point of the current association vector, to obtain an index association vector; Identify the vector endpoint of the index association vector, and identify the industry knowledge subject whose subject association position is the same as the vector endpoint; The industry knowledge subject whose subject association position is the same as the vector end point is used as the target industry knowledge subject.

9. The method for implementing intelligent question answering based on fine-tuning of a large language model as claimed in claim 7, characterized in that: The method further comprises: extracting the current relational attributes of the current industry knowledge subject from the interactive question statement; Determining whether there is a current relationship attribute matching the current industry knowledge subject in the interactive question statement; If there is no current relation attribute matching the current industry knowledge subject in the interactive question statement, continue to extract the current industry knowledge subject in the preceding interactive statement set until a current industry knowledge subject matching the current relation attribute is obtained, and extract the current relation attribute of the current industry knowledge subject in the interactive question statement; If there is a current relationship attribute matching the current industry knowledge subject in the interactive question sentence, the current relationship attribute of the current industry knowledge subject is extracted from the interactive question sentence.

10. A system for implementing intelligent question answering based on fine-tuning of a large language model, characterized in that: The system comprises: An initial association coordinate knowledge graph construction module is used to obtain industry knowledge data, extract industry knowledge subjects in the industry knowledge data in sequence, identify the number of associated entities and the number of associated attributes of the industry knowledge subjects, wherein the industry knowledge subjects are industry knowledge entities or industry knowledge attributes; determine the subject association position of the industry knowledge subject in the pre-constructed industry knowledge association coordinate system according to the number of associated entities and the number of associated attributes, and obtain an initial association coordinate knowledge graph; A target-associated coordinate knowledge graph construction module is used to obtain the relationship attributes between industry knowledge subjects in the industry knowledge data, and use the relationship attributes to connect the industry knowledge subjects in the initial associated coordinate knowledge graph to obtain a target-associated coordinate knowledge graph; The relationship attributes include relationship association and attribute association. The relationship association refers to the association between the industry knowledge subject and the industry knowledge entity, and the attribute association refers to the association between the industry knowledge subject and the industry knowledge attribute. A vector association relationship establishment module is used to obtain the association vector between two industry knowledge subjects in the target association coordinate knowledge graph, and fine-tune the pre-built initial large language model according to the association vector, industry knowledge subjects and relationship attributes to obtain a target large language model; The current industry knowledge subject extraction module is used to obtain the user's interactive question statement and determine whether there is an industry knowledge subject in the interactive question statement; if there is no industry knowledge subject in the interactive question statement, obtain a preceding interactive statement set containing the interactive question statement and extract the current industry knowledge subject from the preceding interactive statement set; if there is an industry knowledge subject in the interactive question statement, extract the current industry knowledge subject from the interactive question statement; The target industry knowledge subject feedback module is used to extract the current relationship attributes of the current industry knowledge subject in the interactive question statement, input the current industry knowledge subject and the current relationship attributes into the target large language model to obtain the current association vector; obtain the current subject association position of the current industry knowledge subject, index the target industry knowledge subject in the target association coordinate knowledge graph according to the current subject association position and the current association vector, and feed back the target industry knowledge subject to the user.

Images