Large model question and answer generation system and method, computer equipment and storage medium

Abstract

The invention discloses a large model question and answer generation system and method, computer equipment and a storage medium. According to the scheme, firstly, a life semantic network of a target person is constructed, and life events, associated persons, works, backgrounds, places and other entities and mutual relations of the target person are integrated in a structured mode; when a user question is received, the system parses its semantics and retrieves related knowledge fragments based on the semantic network. Then, inputting the retrieved knowledge fragments and logic rules in the network into a large model together to constrain the reasoning and generating process of the large model so as to obtain a preliminary answer; and finally, performing fact verification on the preliminary answer to ensure that the preliminary answer is consistent with the fact recorded in the semantic network, thereby outputting an accurate and reliable final answer. According to the invention, the accuracy and reliability of character theme question answering are effectively improved.

Description

Technical Field

[0001] This application relates to a large-scale question-answering generation system, method, computer device, and storage medium, belonging to the fields of natural language processing and artificial intelligence technology. Background Technology

[0002] With the deep integration of Large Language Model (LLM) and Retrieval-Augmented Generation (RAG) technologies, question-answering systems for historical figures are increasingly being used in fields such as cultural dissemination, digital humanities, and academic research. In question-answering scenarios related to historical figures, existing solutions mainly rely on general large-scale knowledge bases or retrieval methods based on simple text similarity. The typical process involves retrieving relevant information fragments from the knowledge base through keyword matching, followed by integration by the LLM to generate the final answer.

[0003] However, the lives and deeds of historical figures are usually an organic whole containing complex chronological events, close interpersonal relationships, the background of specific works (or writings, achievements), and historical geography, among other multidimensional information. Existing technological solutions face the following key challenges: Insufficient retrieval accuracy and "relevance drift": Historical figures in general knowledge bases are often fragmented and weakly correlated. Retrieval methods based on keywords or shallow semantic similarity struggle to understand the deeper semantic context behind user questions (such as causal and temporal relationships between events). This leads to "relevance drift" during the RAG retrieval phase, meaning that while the retrieved knowledge fragments contain the query keywords, they are not strongly related to the true intent of the question and the context, failing to accurately support the generation of high-quality answers.

[0004] Large-scale models suffer from "illusions" and a lack of logical reasoning: Due to the lack of structured and logical constraints on historical figures' knowledge, large-scale models are prone to generating "illusion" content that contradicts the facts of historical figures' lives when generating answers. Furthermore, existing solutions struggle to effectively handle complex questions involving deep logic (such as "how a certain important event influenced the person's later thoughts or behaviors") because the system cannot provide the necessary structured information, such as event chronology and causal relationships, to support deep reasoning within the large-scale model.

[0005] To address the above problems, existing technologies have made several attempts at improvement, but still have significant limitations: Some solutions improve performance by expanding the corpus size or optimizing text similarity calculation algorithms, but they fail to fundamentally model the semantic relationships between historical figures' lives in a deep, structured way. Therefore, they cannot solve the inherent fragmentation problem of knowledge, resulting in limited improvement in retrieval accuracy. Other solutions attempt to construct simplified biographical knowledge graphs, but fail to achieve deep collaboration with the RAG retrieval process. Their retrieval processes often fail to fully utilize the rich semantic relationship paths within the knowledge graph for deep reasoning and generally lack targeted fact-checking and logical constraint mechanisms for content generated by large models.

[0006] In addition, the knowledge bases of existing solutions are mostly statically constructed, which cannot be dynamically optimized and supplemented with long-tail knowledge and details related to historical figures based on continuous user question and answer feedback. This results in poor adaptability of the system to users' personalized and in-depth query needs, making it difficult to meet the needs for accurate and in-depth exploration of details such as the life of historical figures. Summary of the Invention

[0007] To address the problems of insufficient retrieval accuracy, factual bias, and lack of biographical logical reasoning in existing question-and-answer systems for biographical subjects, this application provides a RAG large-scale model question-and-answer generation system, method, computer equipment, and storage medium that combines a semantic network of biographical subjects to achieve accurate retrieval and reliable generation of questions and answers related to target figures, especially Su Dongpo.

[0008] The first aspect of this application discloses a large-scale model question-answering generation system. The system includes: a biographical semantic network construction unit, used to construct a biographical semantic network of a target person, wherein the target person is the core entity, semantic relationships are defined between the core entities, and entity attributes are labeled to form a structured knowledge system with temporal logic and causal relationships. The core entities include at least one of biographical events, related figures, creative works, historical background, and geographical location; a semantic enhancement retrieval unit, used to receive user question-answering requests, parse the core entities and semantic requirements in the request, mine related semantic paths based on the target person's biographical semantic network, recall matching knowledge fragments according to path weight priority, and generate a retrieval result set; a large-scale model generation unit, used to take the retrieval result set and the logical rules of the target person's biographical semantic network as input, load a pre-trained large-scale model, and generate preliminary answer results; and a fact verification unit, used to extract the core entities and their semantic relationships from the preliminary answer results, perform consistency verification with factual information in the target person's biographical semantic network, correct content inconsistent with biographical facts, and output the final question-answering result.

[0009] Furthermore, the target figures are historical and cultural figures from ancient times. The biographical semantic network construction unit is specifically used for: extracting core entities related to the life of the historical and cultural figures from authoritative documents; labeling entity attributes, which include at least time, type, and description; defining semantic relationships between the core entities, which include at least occurrence, association with, creation of, and cause of; and storing the core entities, entity attributes, and semantic relationships in a graph database to form the structured knowledge system.

[0010] Furthermore, the semantic enhancement retrieval unit includes a path weight dynamic adjustment module, which is specifically used to: set benchmark weights for different semantic paths, wherein: a first benchmark weight W1 is assigned to the first semantic path related to turning points in the target person's life or their creative works; a second benchmark weight W2 is assigned to the second semantic path related to the target person's associates; and a third benchmark weight W3 is assigned to the third semantic path related to the target person's life experiences; and the relationship between the benchmark weights is: W3 <W2<W1。

[0011] Furthermore, the path weight dynamic adjustment module is specifically used to: maintain the weight when the core entity is identified, including a turning point event or a classic work, with the priority order being: first semantic path > second semantic path > third semantic path; adjust the weight when the intention of the associated person is identified, with the priority order being: second semantic path > first semantic path > third semantic path; and adjust the weight when the intention of the experiential event is identified, with the priority order being: third semantic path > first semantic path > second semantic path.

[0012] Furthermore, the step of recalling matching knowledge fragments based on path weight priority includes: assigning higher dynamic weights to semantic paths including turning points and classic works compared to other semantic paths, and prioritizing the recall of corresponding knowledge fragments.

[0013] Furthermore, the fact verification unit also includes a logical consistency verification module, which is specifically used to: verify whether the relationships involved in the preliminary answer are consistent with the logical rules defined in the biographical semantic network, wherein the relationships include temporal logic and / or causal association; and when a logical conflict is detected, adjust the content according to the logical rules in the biographical semantic network.

[0014] Furthermore, the target person is Su Dongpo.

[0015] The second aspect of this application discloses a large-scale model question-answering generation method. The method includes: constructing a target person's biographical semantic network, wherein the target person's biographical semantic network takes the target person as the core entity, defines the semantic relationships between the core entities, and labels the entity attributes to form a structured knowledge system with temporal logic and causal relationships. The core entities include at least one of biographical events, related people, creative works, historical background, and geographical location; receiving a user's question-answering request, parsing the core entities and semantic requirements in the request, mining related semantic paths based on the target person's biographical semantic network, recalling matching knowledge fragments according to path weight priority, and generating a retrieval result set; using the retrieval result set and the logical rules of the target person's biographical semantic network as input, loading a pre-trained large-scale model to generate preliminary answer results; extracting the core entities and their semantic relationships of the preliminary answer results, performing consistency verification with the factual information in the target person's biographical semantic network, correcting content that does not conform to biographical facts, and outputting the final question-answering result.

[0016] A third aspect of this application discloses a computer-readable storage medium comprising a stored program, wherein the program, when running, controls the execution of the large model question-answering generation method of the above embodiments in the processor of the device.

[0017] A fourth aspect of this application discloses a computer device, the computer device including a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed by the large model question-answering generation method of the above embodiments.

[0018] Compared with the prior art, the embodiments of this application have the following beneficial effects: 1) Improved retrieval accuracy: By mining the association paths of the biographical semantic network, the problem of traditional keyword-based retrieval accuracy is solved. The fragmentation problem of keyword retrieval enables RAG retrieval to accurately retrieve biographical knowledge that is strongly related to the context of the question, improving retrieval relevance by more than 30%. 2) Significant hallucination suppression effect: Based on rule constraints and fact verification of a biographical semantic network, large-scale... The accuracy of the generated answers has been improved by more than 40%, effectively avoiding erroneous content that does not conform to Su Dongpo's life story; 3) High adaptability: The structured biographical semantic network supports sequential, causal and other logical question answering, which can meet the deep question answering needs in academic research, cultural dissemination and other scenarios, and can adapt to diverse query scenarios without reconstructing the model. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0020] Figure 1 This is a structural diagram of a large-scale question-answering generation system provided in an embodiment of this application.

[0021] Figure 2 This is a schematic diagram of the core entity association of a semantic network of Su Dongpo's life, provided as an embodiment of this application.

[0022] Figure 3 A flowchart of a large model question-answering generation method provided in an embodiment of this application. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] The following are explanations of key terms used in this embodiment: 1. Biographical Semantic Network: A structured knowledge network centered on the target person, integrating entities and semantic relationships such as their life events, related people, works, background, and locations, possessing temporal logic and causal relationships.

[0026] 2. RAG (Retrieval-Augmented Generation): A retrieval-enhanced generation technique that supplements model input with retrieved external knowledge to improve the accuracy of generated content.

[0027] 3. Semantic Path Mining: Based on the biographical semantic network, starting from the core entities of the user's question, we can mine the links between entities (such as "Su Dongpo → Wutai Poetry Case → Demoted to Huangzhou → Composed 'Ding Feng Bo'", or "Li Bai → Served as a Hanlin Academician → Offended the powerful → Granted gold and released → Composed 'Dreaming of Tianmu Mountain'") to achieve accurate retrieval.

[0028] Example 1: Figure 1 This is a structural diagram of a large-model question-answering generation system provided in an embodiment of this application. Figure 1 As shown, the system includes: The biographical semantic network construction unit 100 is used to construct a biographical semantic network of a target person. The target person's biographical semantic network takes the target person as the core entity, defines the semantic relationships between the core entities, and marks the entity attributes to form a structured knowledge system with temporal logic and causal relationship. The core entities include at least one of the following: biographical events, related people, creative works, historical background, and geographical location.

[0029] In this embodiment, the target person is a famous historical and cultural figure from ancient times.

[0030] Taking Su Dongpo as an example, the biographical semantic network construction unit is specifically used for: Step 1. Extract the core entities related to the life of the ancient historical and cultural figures from authoritative documents.

[0031] For example, core entities (such as "Wutai Poetry Case", "Su Zhe", "Huangzhou", "Ode to the Red Cliff" etc.) can be extracted by crawling authoritative documents such as "Chronological Biography of Su Shi" and "Biography of Su Shi in the History of Song".

[0032] Step 2. Label entity attributes, which include at least time, type, and description. In this step, the entity attributes are event time and work genre.

[0033] For example, the entity attribute of the event time is such as "Wutai Poetry Case - Time of Occurrence: 1079".

[0034] Step 3. Define the semantic relationships between the core entities, wherein the semantic relationships include at least those that occur at, are associated with, are created at, and are caused by. For example, semantic relations are defined (such as "Su Shi - exiled to - Huangzhou" and "Su Shi - wrote - Ode to the Red Cliff").

[0035] Step 4. Store the core entities, entity attributes, and semantic relationships in a graph database to form the structured knowledge system.

[0036] The semantic enhancement retrieval unit 200 is used to receive user question-and-answer requests, parse the core entities and semantic requirements in the requests, mine related semantic paths based on the semantic network of the target person's life, recall matching knowledge fragments according to the path weight priority, and generate a retrieval result set.

[0037] For example, when a user asks a question (such as "What poems did Su Dongpo create during his exile in Huangzhou?"), the core entities "Su Dongpo", "Huangzhou", and "exile" are parsed. The association path "Su Dongpo - exiled to - Huangzhou - created - poems" is mined from the semantic network of his life. Based on this path, relevant knowledge fragments such as "Ding Feng Bo: Mo Ting Chuan Lin Da Ye Sheng" and "Nian Nu Jiao: Chibi Huai Gu" are recalled.

[0038] Furthermore, the step of recalling matched knowledge fragments based on path weight priority includes: assigning higher dynamic weights to semantic paths, including turning points and classic works, compared to other semantic paths, and prioritizing the recall of corresponding knowledge fragments.

[0039] For example, a path weight dynamic adjustment module is added to assign higher weights to semantic paths related to core biographical events (such as the Wutai Poetry Case) and classic works (such as "Prelude to Water Melody: When Will the Moon Be Bright?") and prioritize the retrieval of key information.

[0040] Furthermore, the semantic enhancement retrieval unit includes a path weight dynamic adjustment module, which is specifically used to: set benchmark weights for different semantic paths, wherein: a first benchmark weight W1 is assigned to the first semantic path related to turning points in the target person's life or their creative works; a second benchmark weight W2 is assigned to the second semantic path related to the target person's associates; and a third benchmark weight W3 is assigned to the third semantic path related to the target person's life experiences; and the relationship between the benchmark weights is: W3 <W2<W1。

[0041] The first semantic path focuses on pivotal events and classic works, such as "Su Dongpo - Encounter - Wutai Poetry Case - Demotion to Huangzhou - Creation - Ci Poetry - 'Nian Nu Jiao: Reminiscences of Chibi'", which has the highest weight in the baseline state. This baseline design ensures that the most influential events and works are presented first for open-ended questions (such as "Introducing Su Dongpo").

[0042] Furthermore, the path weight dynamic adjustment module is specifically used to: maintain the weight when the core entity is identified, including a turning point event or a classic work, with the priority order being: first semantic path > second semantic path > third semantic path; adjust the weight when the intention of the associated person is identified, with the priority order being: second semantic path > first semantic path > third semantic path; and adjust the weight when the intention of the experiential event is identified, with the priority order being: third semantic path > first semantic path > second semantic path.

[0043] When the request identifies a core entity containing a "turning point event" or a "classic work", the system will focus on the semantic path of the turning point event and the classic work (referred to as the first type of path) with the highest weight, followed by the path involving related figures (the second type), and the path involving general experiential events (the third type) with the lowest weight.

[0044] Priority: First type of path > Second type of path > Third type of path.

[0045] Application scenario: For example, when querying "the impact of the Wutai Poetry Case on Su Dongpo's creative works", the system will prioritize recalling highly relevant causal chains such as "Wutai Poetry Case - leading to - exile - creation during the period - 'Nian Nu Jiao'".

[0046] When identifying the intentions of associated persons, the system gives the highest weight to the second type of path (person / location association), followed by the first type of path, and the lowest weight to the third type of path.

[0047] Priority: Second type of path > First type of path > Third type of path.

[0048] Application scenario: For example, when querying "the relationship between Su Zhe and Su Dongpo", the system will prioritize recalling paths that emphasize the relationship between the two people, such as "Su Shi - his younger brother is - Su Zhe" and "shared experience - Wutai Poetry Case", rather than prioritizing the display of Su Shi's personal works.

[0049] When identifying the intent of an experiential event, the system gives the highest weight to the third type of path (general experience), followed by the first type of path, and the lowest weight to the second type of path.

[0050] Priority: Third type of path > First type of path > Second type of path.

[0051] Application scenarios: For example, when querying "Su Dongpo's political achievements", the system will prioritize recalling specific experience event paths such as "Su Shi - served in - Hangzhou - presided over the construction of - Su Causeway".

[0052] The core of this dynamic adjustment mechanism is that the system does not fixate on a certain type of path as always optimal. Instead, it adaptively determines the semantic relationship type that should be most focused on based on the user's query intent, thereby optimizing the retrieval strategy and significantly improving the relevance and accuracy of the answer.

[0053] The large model generation unit 300 is used to take the retrieval result set and the logical rules of the semantic network of the target person's life as input, load the pre-trained large model, and generate preliminary answer results.

[0054] In this embodiment, the knowledge fragments output by the semantic enhancement retrieval unit are received and used as contextual constraints input into the large model. At the same time, the core logical rules in the biographical semantic network (such as event sequence rules and interpersonal relationship rules) are loaded to guide the large model to perform logical reasoning and answer generation.

[0055] For example, the recalled knowledge fragments and the temporal rules in the semantic network ("during his exile to Huangzhou: 1080-1084") are used as inputs to load a pre-trained large model and generate an initial answer.

[0056] The fact verification unit 400 is used to extract the core entities and their semantic relationships of the preliminary answer results, perform consistency verification with the factual information in the semantic network of the target person's life, correct the content that is inconsistent with the life facts, and output the final question and answer results.

[0057] For example, the "work-creation time" association in the preliminary answer is extracted, compared with the factual information stored in the semantic network, and errors such as possible time misalignment are corrected to output the final answer.

[0058] Furthermore, the fact verification unit also includes a logical consistency verification module, which is specifically used to: verify whether the relationships involved in the preliminary answer are consistent with the logical rules defined in the biographical semantic network, wherein the relationships include temporal logic and / or causal association; and when a logical conflict is detected, adjust the content according to the logical rules in the biographical semantic network.

[0059] For example, the fact verification unit has added a "logical consistency verification" function to handle questions involving causal relationships (such as "Why did Su Dongpo write 'Jiang Cheng Zi: A Dream Recorded on the Night of the Twentieth Day of the First Month of the Year Yi Mao'?"). The causal relationship in the verification answer is consistent with the semantic network relationship of "deceased wife Wang Fu passed away - creation of the word". Whether they are consistent or not, ensure the accuracy of the reasoning.

[0060] Example 2: Figure 3 This is a flowchart illustrating a large-model question-answering generation method provided in an embodiment of this application. Figure 3 As shown, the method includes: S301. Construct a semantic network of the life of the target person, wherein the semantic network of the life of the target person is the core entity, the semantic relationship between the core entities is defined, and the entity attributes are labeled to form a structured knowledge system with temporal logic and causal relationship. The core entities include at least one of the following: life events, related people, creative works, historical background and geographical location.

[0061] S302. Receive user question and answer requests, parse the core entities and semantic requirements in the requests, mine related semantic paths based on the semantic network of the target person's life, recall matching knowledge fragments according to the path weight and priority, and generate a search result set.

[0062] S303. Using the retrieval result set and the logical rules of the semantic network of the target person's life as input, load the pre-trained large model to generate preliminary answer results.

[0063] S304. Extract the core entities and semantic relationships of the preliminary answer results, perform consistency verification with the factual information in the semantic network of the target person's life, correct the content that does not conform to the facts of the life, and output the final question and answer results.

[0064] Example 3: Embodiments of this application also provide a computer device, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of the present invention during runtime.

[0065] The aforementioned memory can refer to devices inside a computer used to store data and programs, including RAM, hard disks, etc. RAM can be used to temporarily store running programs and data, while hard disks can be used to store programs and data long-term. Memory enables the computer to read and write data and execute programs. The aforementioned processor is responsible for executing instructions in computer programs and performing data processing. It can also be responsible for controlling and executing various operations, including arithmetic operations, logical operations, and data transmission.

[0066] Example 4: Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of the present invention.

[0067] The aforementioned computer storage media can refer to the media used in computer memory to store certain discontinuous physical quantities. Computer storage media mainly include semiconductors, magnetic cores, magnetic drums, magnetic tapes, laser discs, etc. Computer-readable storage media include stored programs, which can be a set of instructions that a computer can recognize and execute, running on an electronic computer to meet certain information needs.

[0068] Example 5: Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of the present invention.

[0069] The aforementioned computer program products can refer to software programs that have been written, tested, and released, and can run on computers or other devices. Computer program products can include application programs, operating systems, utility software, etc., used to achieve specific functions or solve specific problems.

[0070] Example 6: Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of the present invention.

[0071] The aforementioned non-volatile computer-readable storage medium can refer to a medium for storing data. Non-volatile computer-readable storage media can retain data without loss when power is off and can be used to store long-term data, such as operating systems, applications, and user files. Non-volatile storage media can include hard disk drives, solid-state drives, optical disks, and flash memory storage devices, etc.

[0072] Example 7: Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of the present invention.

[0073] The aforementioned computer program can refer to a set of instructions used to tell the computer to perform specific tasks or operations. Computer programs can be written by programmers using specific programming languages ​​and can include algorithms, data structures, logic, and control flow. Computer programs can be used for a variety of purposes, including application software, operating systems, etc.

[0074] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0076] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0077] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0078] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0079] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A large-scale question-answering generation system, characterized in that, include: The biographical semantic network construction unit is used to construct a biographical semantic network of a target person. The target person's biographical semantic network takes the target person as the core entity, defines the semantic relationships between the core entities, and marks the entity attributes to form a structured knowledge system with temporal logic and causal relationship. The core entities include at least one of the following: biographical events, related people, creative works, historical background, and geographical location. The semantic enhancement retrieval unit is used to receive user question-and-answer requests, parse the core entities and semantic requirements in the requests, mine related semantic paths based on the semantic network of the target person's life, recall matching knowledge fragments according to the path weight and priority, and generate a retrieval result set. The large model generation unit is used to take the retrieval result set and the logical rules of the semantic network of the target person's life as input, load the pre-trained large model, and generate preliminary answer results; The fact verification unit is used to extract the core entities and their semantic relationships from the preliminary answer results, perform consistency verification with the factual information in the semantic network of the target person's life, correct content that is inconsistent with the life facts, and output the final question and answer results.

2. The large-scale question-answering generation system according to claim 1, characterized in that, The target person is a famous historical and cultural figure from ancient times, and the biographical semantic network construction unit is specifically used for: Extract the core entities related to the life stories of the aforementioned historical and cultural figures from authoritative documents; Label the entity attributes, which include at least time, type, and description; The semantic relationships defined between the core entities include at least the relationships of occurrence, association with, creation of, and cause of. The core entities, entity attributes, and semantic relationships are stored in a graph database to form the structured knowledge system.

3. The large-scale question-answering generation system according to claim 1, characterized in that, The semantically enhanced retrieval unit includes a path weight dynamic adjustment module, which is specifically used for: Assign baseline weights to different semantic paths, where: Assign a first baseline weight W1 to the first semantic path related to the turning points in the target person's life and the creation of their works; For the second semantic path related to the associated persons of the target person, assign a second baseline weight W2; Assign a third baseline weight W3 to the third semantic path related to the life experiences of the target person; The relative weights of each benchmark are: W3 < W2 < W1.

4. The large-scale question-answering generation system according to claim 3, characterized in that, The path weight dynamic adjustment module is also specifically used for: When core entities are identified, including pivotal events or classic works, their weights are maintained, with the priority order being: first semantic path > second semantic path > third semantic path; When identifying the intent of related individuals, the weights are adjusted, with the priority order being: second semantic path > first semantic path > third semantic path; When identifying the intent of an experiential event, the weights are adjusted, with the priority order being: third semantic path > first semantic path > second semantic path.

5. The large-scale question-answering generation system according to claim 1, characterized in that, The process of recalling matched knowledge fragments based on path weight priority includes: Compared to other semantic paths, semantic paths including pivotal events and classic works are given higher dynamic weights, prioritizing the recall of corresponding knowledge fragments.

6. The large-scale question-answering generation system according to claim 1, characterized in that, The fact verification unit further includes a logical consistency verification module, which is specifically used for: Verify whether the relationships involved in the preliminary response are consistent with the logical rules defined in the biographical semantic network, including temporal logic and / or causal associations; When a logical conflict is detected, the content is adjusted according to the logical rules in the biographical semantic network.

7. The large-scale question-answering generation system according to any one of claims 1 to 6, characterized in that, The target person is Su Dongpo.

8. A method for generating large-scale question-answering models, characterized in that, include: Construct a semantic network of the life of a target person, wherein the target person is the core entity, the semantic relationship between the core entities is defined, and the entity attributes are labeled to form a structured knowledge system with temporal logic and causal relationship. The core entities include at least one of the following: life events, related people, creative works, historical background and geographical location. Receive user question and answer requests, parse the core entities and semantic requirements in the requests, mine related semantic paths based on the semantic network of the target person's life, recall matching knowledge fragments according to the path weight and priority, and generate a search result set; The search result set and the logical rules of the semantic network of the target person's life are used as inputs to load a pre-trained large model and generate preliminary answer results. Extract the core entities and their semantic relationships from the preliminary answer results, perform consistency verification with the factual information in the semantic network of the target person's life, correct any content that does not conform to the facts of the life, and output the final question and answer results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the large model question-answering generation method as described in claim 8.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the program to implement the large model question-answering generation method as described in claim 8.

Images