A digital cultural heritage activation and dynamic deduction system and method based on neural symbols and multi-agent collaboration

Abstract

This invention discloses a system and method for the activation and dynamic deduction of digital cultural heritage based on neural symbols and multi-agent collaboration. First, the system transforms unstructured ancient Chinese documents into machine-readable neural symbol vectors through a two-layer semantic mapping mechanism, combined with fine-tuning of a large language model and semantic dependency analysis. Second, it instantiates intelligent agents representing cultural relics or historical figures, dynamically calculating their health status and updating their memory bank based on environmental parameters. Third, it drives a multi-agent network composed of researchers, simulated actors, and critics to collaborate and engage in game theory under the constraints of a geographic information system and spatiotemporal rules, generating historically accurate and logically consistent historical narratives. Finally, it outputs multimodal results and updates the knowledge base, realizing the transformation of cultural heritage from static archives to dynamic deduction, improving the accuracy of historical semantic understanding and the depth of narrative logic.

Description

Technical Field

[0001] This invention relates to the intersection of digital cultural heritage protection and artificial intelligence technology, and more specifically, to a system and method for the activation and dynamic deduction of digital cultural heritage based on neural symbols and multi-agent collaboration. Background Technology

[0002] Currently, the protection and utilization of digital cultural heritage mainly focuses on digital archiving and static display, such as establishing 3D model databases of cultural relics or digital museums. These methods have limitations, emphasizing recording over activation, making it difficult to effectively present the historical context and dynamic evolution of cultural heritage. In recent years, attempts have begun to emerge to use generative artificial intelligence (such as large language models) for content generation, but when applied to the field of cultural heritage, the following key technological bottlenecks have been exposed:

[0003] Insufficient accuracy in historical semantic understanding: When dealing with ancient Chinese documents, specific historical contexts, or unstructured descriptions, large language models lack a precise semantic foundation, which can easily generate illusory content that does not conform to historical facts, making it difficult to guarantee the logical rigor and factual accuracy of the generated narrative.

[0004] Lack of dynamic evolution and risk perception capabilities: Existing digital displays are mostly static slices, which cannot simulate the dynamic changes that cultural relics undergo in the long course of history or in a virtual environment, such as the cumulative effects of risk factors such as environmental erosion and human damage. The system does not have the ability to remember and extrapolate the state of cultural heritage throughout its "full life cycle".

[0005] The narrative logic is simplistic and lacks collaborative verification: Existing generation methods are mostly one-way, linear content outputs, lacking multi-role, multi-perspective interaction and game mechanisms, making it difficult to reproduce the complex causal chains of historical events. At the same time, the generated content lacks effective spatiotemporal logic and historical rule verification, which can easily lead to spatiotemporal inconsistencies or errors that violate ritual and regulations.

[0006] This invention aims to overcome the aforementioned deficiencies of the prior art and provide a digital cultural heritage revitalization system that can accurately understand the semantics of classical Chinese, possess historical risk memory, and conduct rigorous spatiotemporal and logical deductions through multi-agent collaboration. Summary of the Invention

[0007] This invention aims to solve the above problems and provide a digital cultural heritage revitalization system that can accurately understand the semantics of classical Chinese, possess historical risk memory, and conduct rigorous spatiotemporal and logical deductions through multi-agent collaboration.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A system and method for the activation and dynamic deduction of digital cultural heritage based on neural symbols and multi-agent collaboration is characterized by comprising: a two-layer semantic mapping module, a risk memory-enhanced agent construction module, a multi-agent spatiotemporal collaborative deduction module, and a dynamic visualization and feedback interaction module.

[0010] The two-layer semantic mapping module is used to receive unstructured cultural heritage raw data. By combining a fine-tuned large language model with semantic dependency analysis, it extracts event elements from the raw data and maps the event elements into machine-readable neural symbol vectors. Event elements include trigger words, participants, time and space elements.

[0011] The Risk Memory Enhanced Intelligent Agent Construction Module is used to instantiate cultural relic intelligent agents or historical figure intelligent agents based on the autonomous intelligent agent architecture empowered by a large language model. The cultural relic intelligent agent or historical figure intelligent agent has a built-in long-term memory unit and a risk assessment unit. The risk assessment unit integrates the ABC risk assessment model to dynamically calculate the health status of the cultural relic intelligent agent or historical figure intelligent agent based on environmental parameters and update its memory bank.

[0012] The multi-agent spatiotemporal collaborative inference module includes at least three types of functionally heterogeneous role agents: researcher agent, simulated actor agent, and critic agent. These role agents collaborate and play games around cultural relic agent or historical figure agent based on neural symbolic reasoning, and generate dynamic narrative logic under the constraints of geographic information system and historical spatiotemporal rules.

[0013] The dynamic visualization and feedback interaction module is used to transform dynamic narrative logic into multimodal output and receive feedback signals to update the knowledge base of the cultural relic agent or historical figure agent.

[0014] Furthermore, the two-layer semantic mapping module is used to receive unstructured cultural heritage raw data. By combining a fine-tuned large language model with semantic dependency analysis, it extracts event elements from the raw data and maps the event elements into machine-readable neural symbol vectors. Event elements include trigger words, participants, time and space elements.

[0015] The language mapping unit receives multimodal raw data, uses a fine-tuned large language model to convert classical Chinese or heterogeneous historical texts into modern standard Chinese, and uses semantic dependency analysis to extract the semantic dependency graph of events, identifying the subject, patient, time and space elements of the events.

[0016] The logical mapping unit adopts a neural-vector-symbolic architecture, which binds the extracted semantic elements into neural symbol vectors through the superposition operation of tensor products, forming a structured knowledge base that includes ontology files, protection processes, and management units.

[0017] Furthermore, the risk memory-enhanced intelligent agent construction module is used to instantiate cultural relic intelligent agents or historical figure intelligent agents based on the autonomous intelligent agent architecture empowered by a large language model. The cultural relic intelligent agent or historical figure intelligent agent has a built-in long-term memory unit and a risk assessment unit. The risk assessment unit integrates the ABC risk assessment model to dynamically calculate the health status of the cultural relic intelligent agent or historical figure intelligent agent based on environmental parameters and update its memory bank.

[0018] The architecture of autonomous intelligent agents (LAAs) empowered by large language models instantiates an intelligent agent for each cultural relic or historical figure.

[0019] The health status parameter H(t) of a cultural relic agent or a historical figure agent is expressed as:

[0020]

[0021] Among them, H initial α represents the initial health state value of the agent. k R represents the value weight and sensitivity coefficient of the cultural relic to the k-th type of risk. k (τ) is the risk function, representing the intensity of the k-th risk factor at time τ;

[0022] The ABC risk assessment model defines the risk function as R(t), expressed as:

[0023] R(t)=Σ(A i ·B i ·C i );

[0024] Among them, A i This represents the frequency of the i-th type of risk occurring;

[0025] B i This represents the extent of functional loss caused by the risk;

[0026] C i Represents the proportion of the affected cultural relic's physical area;

[0027] During the simulation process, the intelligent agents for cultural relics or historical figures update their health status parameters based on the calculation results of the risk function, and trigger preset interactive behaviors when the health status parameters are lower than the preset threshold.

[0028] Furthermore, the multi-agent spatiotemporal collaborative inference module includes at least three types of functionally heterogeneous role agents: researcher agent, simulated actor agent, and critic agent. These role agents collaborate and play games around cultural relic agent or historical figure agent based on neural symbolic reasoning, generating dynamic narrative logic under the constraints of geographic information system and historical spatiotemporal rules.

[0029] The researcher's intelligent agent is equipped with a retrieval-enhanced generation interface, which is responsible for retrieving historical data from external historical knowledge bases or archaeological databases to provide factual basis for inferences;

[0030] The simulated actor intelligent agent generates dialogue and behavioral decisions from a first-person perspective based on the personality data and ideological rules of specific historical figures.

[0031] The critic agent is configured with a geographic information system and a historical spatiotemporal rule constraint algorithm, and performs logical review of the behavior generated by the simulated actor agent based on neural symbolic reasoning.

[0032] The system drives three functional roles: researcher agent, simulated actor agent, and critic agent. These agents collaborate and play against each other around the cultural relic agent or historical figure agent. During the interaction, the system calls the risk assessment unit to dynamically calculate the health status of the cultural relic agent or historical figure agent based on the ABC risk assessment model and current environmental parameters. The system also dynamically updates the memory bank based on the calculation results.

[0033] The critic agent's logical review includes:

[0034] The task of historical causal deduction is decomposed into a series of logical propositions or executable pseudocode steps, and the logic is verified by the solver to prevent the large language model from generating factual illusions.

[0035] Furthermore, the dynamic visualization and feedback interaction module includes a logic transformation unit and a human feedback reinforcement learning unit:

[0036] The logic transformation unit includes:

[0037] Transform the decisions generated by the simulated actor agent into a first-person narrative that matches its personality traits.

[0038] By calling upon data from a geographic information system, the paths of historical events can be dynamically marked on a 3D digital map;

[0039] Based on the health status parameter H(t), the evolution trajectory of cultural relics is displayed in real time;

[0040] Human feedback reinforcement learning units include:

[0041] When there is controversy over the multimodal output results, human experts are brought in for review.

[0042] The expert's corrections are encoded as reward signals to fine-tune the agent's policy network. The corrected knowledge is then written back to the underlying knowledge graph using knowledge fusion technology, enabling the dynamic evolution of the knowledge base for cultural relic agents or historical figure agents.

[0043] This invention also provides a method for the activation and dynamic deduction of digital cultural heritage based on neural symbols and multi-agent collaboration, comprising the following steps:

[0044] S1: Receive unstructured cultural heritage raw data, combine fine-tuned large language model and semantic dependency analysis, extract event elements from the raw data and map them into machine-readable neural symbol vectors;

[0045] S2: Based on the autonomous intelligent agent architecture empowered by the large language model, an instantiation of cultural relic intelligent agent or historical figure intelligent agent is performed. It includes a long-term memory unit and a risk assessment unit. The risk assessment unit integrates the ABC risk assessment model and dynamically calculates the health status of the cultural relic intelligent agent or historical figure intelligent agent according to environmental parameters and updates its memory bank.

[0046] S3: Drives researcher agents, simulated actor agents, and critic agents to collaborate and play games around cultural relic agents or historical figure agents based on neural symbolic reasoning, generating dynamic narrative logic under the constraints of geographic information systems and historical spatiotemporal rules.

[0047] S4: Transforms dynamic narrative logic into multimodal output and receives feedback signals to update the knowledge base of the cultural relic agent or historical figure agent.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. Improved accuracy of historical semantic understanding: Through a two-layer mapping mechanism of language layer-logic layer, this invention effectively overcomes the illusion problem when relying solely on large models for ancient Chinese text processing, and significantly improves the accuracy of extracting and understanding ancient Chinese events by utilizing semantic dependency analysis.

[0050] 2. Enabled the living evolution of cultural heritage: For the first time, the ABC risk model in the field of cultural relic protection was combined with intelligent agent memory. This makes digital cultural relics no longer static 3D models, but intelligent entities that can exhibit dynamic processes of aging, damage, or restoration according to environmental changes and historical processes, greatly enhancing the realism and educational value of digital heritage.

[0051] 3. Enhanced logical coherence and depth of historical narratives: By leveraging the collaboration and checks and balances mechanism of multi-agent systems and combining it with spatiotemporal data constraints from geographic information systems, the system can generate logically rigorous, historically accurate, and thought-provoking historical narratives. This avoids the spatiotemporal inconsistencies that are common in traditional generative AI, achieving a leap from simple text generation to in-depth historical deduction.

[0052] 4. Promotes the integration and innovation of interdisciplinary knowledge: This system integrates linguistics, history, geographic information science, and computer science to build a scalable digital humanities research platform. It can not only be used to generate popular science content but also assist professional scholars in conducting quantitative analysis of historical events, thus driving the transformation of the digital humanities research paradigm. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the logical architecture provided by the present invention, which uses a two-layer semantic mapping mechanism to structure unstructured historical documents and constructs an intelligent agent embedded in the ABC risk assessment model and a multi-agent collaborative inference.

[0054] Figure 2 This invention provides a flowchart illustrating the process of generating narrative logic through multi-agent collaborative deduction and transforming it into multimodal output, while combining expert feedback and dynamic ontology updates to achieve knowledge base evolution. Detailed Implementation

[0055] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] This invention proposes a human-machine-object integrated digital cultural heritage revitalization architecture, the core of which lies in constructing an intelligent agent network with dual-layer semantic mapping capabilities and risk-based full-lifecycle memory. The system deconstructs historical documents and cultural relics archives, transforming them into computable and deductive dynamic narrative scenarios.

[0057] Example 1:

[0058] Reference Figure 1 This embodiment details how to address semantic loss and machine comprehension bias during the transformation of unstructured historical documents into machine-readable structured knowledge. The system employs a two-layer mapping mechanism between the language layer and the logic layer.

[0059] Language mapping unit, cross-temporal language alignment and dependency analysis, the system receives multimodal raw data D raw This includes scanned copies of ancient books, oral history recordings, and 3D point cloud data of cultural relics.

[0060] Text translation and cleaning: using a fine-tuned large language model to translate and clean ancient Chinese text T. ancient Translated into Modern Standard Chinese Text T modern This process references the HistoLens framework and performs specialized entity alignment for phonetic loan characters and official titles from specific historical periods.

[0061] Semantic dependency analysis for modern standard Chinese text T modernPerform semantic dependency analysis to extract core event elements. Define the parse function as F. SDP The output is a dependency graph G. dep :

[0062] G dep =F SDP (T modern )={(w i ,w j ,r k )}|w∈Words,r∈Relations

[0063] Among them, w i Representative core word; w j It is a modifier associated with it; r k It includes labels such as agent, patient, time, and space to identify trigger words and participants in the text.

[0064] The logical mapping unit uses a neural-vector-symbolic architecture to map the semantic dependency graph G. dep The elements in the vector are mapped to neural symbol vectors.

[0065] Event vectorization maps the extracted entities and relationships to a high-dimensional vector space, defining an event vector V. event The superposition of the tensor products of each element:

[0066]

[0067] in, It represents semantic roles such as "craftsman" and "skill". Representing specific entities such as "Li Jie" and "Yingzao Fashi" This is for tensor product operations.

[0068] The archival dimensions are constructed by referencing the four-dimensional framework of digital cultural heritage archives. Vectorized data is classified and stored as ontological archival data, protection process data, and management unit data, forming a machine-readable knowledge graph foundation.

[0069] Example 2:

[0070] Reference Figure 1 This embodiment constructs an intelligent agent with historical memory and evolutionary perception capabilities, referring to... Figure 1 Based on the Large Language Model-enabled Autonomous Agent (LAAs) architecture, an agent is instantiated for each cultural relic or historical figure.

[0071] The core of this intelligent agent includes a long-term memory unit and a risk assessment unit:

[0072] Long-term memory units, based on vector databases, store the entire lifecycle archive of the cultural relic from its creation and transfer to restoration;

[0073] The risk assessment unit integrates the ABC risk assessment model.

[0074] The health status parameter H(t) of a cultural relic agent or a historical figure agent is expressed as:

[0075]

[0076] Among them, H initial α represents the initial health state value of the agent. k R represents the value weight and sensitivity coefficient of the cultural relic to the k-th type of risk. k (τ) is the risk function, representing the intensity of the k-th risk factor at time τ;

[0077] The ABC risk assessment model defines the risk function as R(t), expressed as:

[0078] R(t)=Σ(A i ·B i ·C i );

[0079] Among them, A i This represents the frequency of the i-th type of risk occurring;

[0080] B i This represents the extent of functional loss caused by the risk;

[0081] C i Represents the proportion of the affected cultural relic's physical area;

[0082] During the simulation process, the intelligent agents of cultural relics or historical figures update their health status parameters based on the calculation results of the risk function. When the health status parameter H(t) is lower than the preset threshold, the intelligent agents of cultural relics or historical figures will actively trigger a distress signal or present a damaged state in the virtual display, thereby realizing a paradigm shift from passive recording to proactive risk warning.

[0083] Example 3:

[0084] Reference Figure 2 This embodiment utilizes a multi-agent network to simulate the evolution of historical events, addressing the problem of poor logical consistency in content generated by a single model and the tendency to exhibit historical nihilism.

[0085] The role-based division of labor and intelligent agent networking system automatically generates a heterogeneous intelligent agent network based on historical topics input by the user, such as "Maritime Silk Road trade in the Song Dynasty".

[0086] Researcher AI Agent: Connects to external knowledge bases such as the "History of the Song Dynasty" and the "Records of the Maritime Trade Office," and is responsible for retrieving historical facts to ensure the accuracy of data sources;

[0087] Simulated actor intelligent agent: Based on the personality data of specific historical figures, such as "Xin Qiji's boldness and concern for the country", it generates dialogue in the first person and is also constrained by the ideological rules of Confucianism's view of righteousness and profit or Legalism's pro-commerce policy, generating dialogue and decisions that conform to its position.

[0088] The critic agent is responsible for verifying logic and historical facts. This agent has built-in geographic information system data to verify whether the itinerary generated by the simulated actor agent conforms to physical logic.

[0089] In the game theory deduction under spatiotemporal constraints, a spatiotemporal constraint function is introduced when generating the narrative. If the agent claims to "arrive in Hangzhou from Kaifeng within one day", the critic agent will execute the following verification logic:

[0090]

[0091] Obtain the movement requests generated by the simulated actor agent and extract the starting location L. start (Kaifeng), Destination L end (Hangzhou) and claimed to have taken T duration (1 day);

[0092] Query the geographic information system database to obtain the historical geographic distance Dist(L) between the two locations. start L end );

[0093] Query the Song Dynasty transportation database to obtain the theoretical maximum speed V of commonly used transportation vehicles. max_transport(Era) ;

[0094] Calculate the claimed speed

[0095] When claiming speed Greater than the theoretical maximum speed V max_transport(Era) If the behavior is deemed logically flawed, the critic agent will reject the proposed action and require the actor agent to be regenerated, such as by changing the arrival time to three days later.

[0096] Example 4:

[0097] Reference Figure 2 The narrative logic generated by the critic agent and through game theory is transformed into a multimodal result that is perceptible to humans. Specific examples include:

[0098] The decisions generated by the simulated actor agent are transformed into first-person narratives that conform to its personality traits; for example, when the scenario of "the Maritime Silk Road trade being blocked" is explored, war mobilization or trade decision texts containing patriotic sentiments are automatically generated.

[0099] By calling on data from a geographic information system, the paths of historical events can be dynamically marked on a 3D digital map; for example, a real-time rendering of the sailing route of a Song Dynasty merchant ship departing from Quanzhou Port, passing through the South China Sea to the Arabian Peninsula, can be displayed, along with the duration of stay at each stop.

[0100] Based on the health status parameter H(t), the evolution trajectory of cultural relics is displayed in real time; for example, it can show a bronze artifact over a virtual simulation period of up to 100 years, under the influence of the environmental humidity risk term R. k The cumulative effect of corrosion is reflected in the curve showing a decrease in surface corrosion coverage and corresponding health status parameter H(t).

[0101] Human Feedback Reinforcement Learning (RLHF) introduces a human expert feedback mechanism to correct model biases in specific cultural contexts, such as misinterpretations of the cultural metaphor of dragons. The human expert's corrective opinions are encoded as a reward signal r, which is used to fine-tune the policy network π.

[0102]

[0103] Where J(π) represents the objective function, which is the total score that the system wants to maximize; Let (x, y) represent the mathematical expectation, where (x, y) represents the distribution from the current policy D. π The combination of "historical context x" and "generated action / narrative y" sampled from the dataset, r(x,y) represents the output value of the reward model, and π(y|x) represents the policy network currently being trained. ref (y|x) represents the reference model, and β is the KL divergence coefficient, which is used to balance the reward signal and the original model distribution.

[0104] Through this mechanism, the system can learn generation strategies that conform to specific cultural values;

[0105] As the dynamic ontology updates progress, the newly generated reasonable historical logic will be written back into the underlying knowledge graph through knowledge fusion technology, resolving entity co-reference and ambiguity issues, and realizing the self-growth of the digital cultural heritage knowledge base.

[0106] Technical Effects: This invention systematically solves the problems pointed out in the background art through the technical chain described in Examples 1 to 4. From the accurate semantic parsing of unstructured data (Example 1), to endowing cultural heritage with dynamic life characteristics (Example 2), to generating logically self-consistent narratives through multi-agent game theory and strong constraints (Example 3), and finally achieving human-machine collaborative evolution (Example 4), it completes a full closed loop from static data to dynamic deduction, significantly improving the depth, accuracy, and interactivity of digital cultural heritage revitalization.

[0107] This invention also provides a method for the activation and dynamic deduction of digital cultural heritage based on neural symbols and multi-agent collaboration, comprising the following steps:

[0108] S1: Receive unstructured cultural heritage raw data, combine fine-tuned large language model and semantic dependency analysis, extract event elements from the raw data and map them into machine-readable neural symbol vectors;

[0109] S2: Based on the autonomous intelligent agent architecture empowered by the large language model, an instantiation of cultural relic intelligent agent or historical figure intelligent agent is performed. It includes a long-term memory unit and a risk assessment unit. The risk assessment unit integrates the ABC risk assessment model and dynamically calculates the health status of the cultural relic intelligent agent or historical figure intelligent agent according to environmental parameters and updates its memory bank.

[0110] S3: Drives researcher agents, simulated actor agents, and critic agents to collaborate and play games around cultural relic agents or historical figure agents based on neural symbolic reasoning, generating dynamic narrative logic under the constraints of geographic information systems and historical spatiotemporal rules.

[0111] S4: Transforms dynamic narrative logic into multimodal output and receives feedback signals to update the knowledge base of the cultural relic agent or historical figure agent.

[0112] The method for activating and dynamically extrapolating digital cultural heritage based on neural symbols and multi-agent collaboration of the present invention corresponds to and has the same operation and effect as the aforementioned system for activating and dynamically extrapolating digital cultural heritage based on neural symbols and multi-agent collaboration. Therefore, the method for activating and dynamically extrapolating digital cultural heritage based on neural symbols and multi-agent collaboration will not be described again here.

[0113] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of a necessary general-purpose hardware platform, or by a combination of hardware and software. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a computer product. The present invention can be implemented in the form of a computer program product on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Other embodiments may also be used. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A system for the activation and dynamic deduction of digital cultural heritage based on neural symbols and multi-agent collaboration, characterized in that, Includes the following modules: The two-layer semantic mapping module is used to receive unstructured cultural heritage raw data, extract event elements from the raw data by combining a fine-tuned large language model with semantic dependency analysis, and map the event elements into machine-readable neural symbol vectors. The event elements include trigger words, participants, time and space elements. The risk memory-enhanced intelligent agent construction module is used to instantiate cultural relic intelligent agents or historical figure intelligent agents based on the autonomous intelligent agent architecture empowered by a large language model. The cultural relic intelligent agent or historical figure intelligent agent has a built-in long-term memory unit and a risk assessment unit. The risk assessment unit integrates the ABC risk assessment model to dynamically calculate the health status of the cultural relic intelligent agent or historical figure intelligent agent according to environmental parameters and update its memory bank. The multi-agent spatiotemporal collaborative inference module includes at least three types of functionally heterogeneous role agents: researcher agent, simulated actor agent, and critic agent. These role agents cooperate and play games around the cultural relic agent or historical figure agent based on neural symbolic reasoning, and generate dynamic narrative logic under the constraints of geographic information system and historical spatiotemporal rules. The dynamic visualization and feedback interaction module is used to transform the dynamic narrative logic into multimodal output and receive feedback signals to update the knowledge base of the cultural relic agent or historical figure agent.

2. The system for revitalizing and dynamically reconstructing digital cultural heritage based on neural symbols and multi-agent collaboration as described in claim 1, characterized in that, The two-layer semantic mapping module includes: The language mapping unit receives multimodal raw data, converts ancient Chinese or heterogeneous historical texts into modern standard Chinese through a fine-tuned large language model, and uses semantic dependency analysis to extract the semantic dependency graph of events, identifying the subject, patient, time and space elements of the events. The logical mapping unit adopts a neural-vector-symbolic architecture, which binds the extracted semantic elements into neural symbol vectors through the superposition operation of tensor products, forming a structured knowledge base that includes ontology files, protection processes, and management units.

3. The system for revitalizing and dynamically reconstructing digital cultural heritage based on neural symbols and multi-agent collaboration as described in claim 1, characterized in that, In the risk memory-enhanced intelligent agent construction module: The health status parameter H(t) of the cultural relic agent or historical figure agent is expressed as: Among them, H initial α represents the initial health state value of the agent. k R represents the value weight and sensitivity coefficient of the cultural relic to the k-th type of risk. k (τ) is the risk function, representing the intensity of the k-th risk factor at time τ; The ABC risk assessment model defines the risk function as R(t), expressed as: R(t)=Σ(A i ·B i ·C i ); Among them, A i This represents the frequency of the i-th type of risk occurring; B i This represents the extent of functional loss caused by the risk; C i Represents the proportion of the affected cultural relic's physical area; During the deduction process, the intelligent agent representing cultural relics or historical figures updates its health status parameters based on the calculation results of the risk function, and triggers a preset interactive behavior when the health status parameters are lower than a preset threshold.

4. The system for revitalizing and dynamically reconstructing digital cultural heritage based on neural symbols and multi-agent collaboration as described in claim 1, characterized in that, The multi-agent spatiotemporal collaborative inference module includes: The researcher's intelligent agent is equipped with a retrieval-enhanced generation interface, which is responsible for retrieving historical data from external historical knowledge bases or archaeological databases to provide factual basis for inferences; The simulated actor agent generates dialogue and behavioral decisions from a first-person perspective based on the personality data and ideological rules of specific historical figures. The critic agent is configured with a geographic information system and a historical spatiotemporal rule constraint algorithm to logically review the behavior generated by the simulated actor agent based on neural symbolic reasoning. The three functional roles of the intelligent agent—the researcher agent, the simulated actor agent, and the critic agent—are driven to collaborate and play games around the intelligent agent of the cultural relic or the intelligent agent of the historical figure. During the interaction, the risk assessment unit is invoked to dynamically calculate the health status of the intelligent agent of the cultural relic or the intelligent agent of the historical figure based on the ABC risk assessment model and combined with the current environmental parameters, and the memory bank is dynamically updated according to the calculation results.

5. The system for revitalizing and dynamically reconstructing digital cultural heritage based on neural symbols and multi-agent collaboration as described in claim 4, characterized in that, The logical review performed by the critic agent includes: The task of historical causal deduction is decomposed into a series of logical propositions or executable pseudocode steps, and the logic is verified by the solver to prevent the large language model from generating factual illusions.

6. The system for revitalizing and dynamically reconstructing digital cultural heritage based on neural symbols and multi-agent collaboration as described in claim 1, characterized in that, The dynamic visualization and feedback interaction module includes a logic transformation unit and a human feedback reinforcement learning unit: The logic transformation unit includes: The decisions generated by the simulated actor agent are transformed into first-person narratives that conform to its personality traits. By calling upon data from a geographic information system, the paths of historical events can be dynamically marked on a 3D digital map; Based on the health status parameter H(t), the evolution trajectory of the cultural relic is displayed in real time; The human feedback reinforcement learning unit includes: When there is controversy over the multimodal output results, human experts are brought in for review. The expert's corrections are encoded as reward signals to fine-tune the agent's policy network. The corrected knowledge is then written back to the underlying knowledge graph using knowledge fusion technology, enabling the dynamic evolution of the knowledge base of the cultural relic agent or historical figure agent.

7. A method for the revitalization and dynamic deduction of digital cultural heritage based on neural symbols and multi-agent collaboration, characterized in that, Includes the following steps: S1: Receive unstructured cultural heritage raw data, combine a fine-tuned large language model with semantic dependency analysis, extract event elements from the raw data and map them into machine-readable neural symbol vectors; S2: An intelligent agent architecture based on a large language model is used to instantiate an intelligent agent for cultural relics or historical figures, which includes a long-term memory unit and a risk assessment unit. The risk assessment unit integrates the ABC risk assessment model and dynamically calculates the health status of the intelligent agent for cultural relics or historical figures based on environmental parameters and updates its memory bank. S3: Drive the researcher agent, simulated actor agent, and critic agent to cooperate and play games around the cultural relic agent or historical figure agent based on neural symbolic reasoning, and generate dynamic narrative logic under the constraints of geographic information system and historical spatiotemporal rules. S4: Transform the dynamic narrative logic into a multimodal output and receive feedback signals to update the knowledge base of the cultural relic agent or historical figure agent.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in claim 7.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in claim 7.

Images