Distributed digital human question and answer method and system based on 5g communication and electronic equipment

Abstract

This application provides a distributed digital human question-answering method, system, and electronic device based on 5G communication. By constructing a central cache set and actively distributing it, the questions faced by a digital human can quickly benefit the entire group, greatly reducing the load on the central server and the average response latency. The central server utilizes the powerful understanding and generation capabilities of a large language model to clean, summarize, and improve the global question-answering data, fundamentally ensuring the high quality and accuracy of the cached content distributed to each node. Based on the attribute information of the digital human nodes, a push strategy is formulated to pre-deploy hot knowledge to front-line digital human nodes, improving the hit rate and saving air interface resources.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a distributed digital human question-answering method, system, and electronic device based on 5G communication. Background Technology

[0002] In large-scale smart exhibition halls, shopping malls, and other similar scenarios, multiple digital humans are typically deployed in a distributed manner to achieve full-domain service coverage. Existing technical solutions mostly focus on improving the intelligence level of individual digital humans, but the question-and-answer knowledge accumulated by each digital human node forms "information silos," making it impossible to summarize, deduplicat, and optimize common issues from a global perspective. This results in inconsistent quality of cached content and a low hit rate. Summary of the Invention

[0003] The purpose of this application is to provide a distributed digital human question-answering method, system, and electronic device based on 5G communication to alleviate the above-mentioned technical problems.

[0004] In a first aspect, the present invention provides a distributed digital human question-answering method based on 5G communication. The method is applied to a system including a central server and multiple distributed digital human nodes, wherein each of the digital human nodes integrates a 5G communication module; each of the digital human nodes accesses the 5G network through its own 5G communication module and establishes a communication connection with the central server.

[0005] The method includes:

[0006] The central server maintains a global central cache set for the historical question-and-answer data of each digital human node; and determines the push strategy according to the attribute information of each digital human node, periodically selects the target cache in the central cache set according to the push strategy, and distributes it to the corresponding digital human node through the 5G network as the initial or updated content of its local cache.

[0007] The digital human node receives the current question and matches it with the local cache; if the match is successful, the answer is returned directly based on the local cache; if the match is unsuccessful, the current question is uploaded to the central server via the 5G network.

[0008] The central server invokes a large language model to process the current question, generates an answer, and transmits it downstream to the digital human node that initiated the request.

[0009] In an optional implementation, the central server maintains a global central cache set for the historical question-and-answer data of each digital human node, including:

[0010] Historical question-and-answer data from different digital human nodes are periodically formatted to eliminate discrepancies in expression;

[0011] Using a large language model, questions with similar semantics but different expressions are automatically classified into the same question cluster, and each question cluster is assigned a globally unique semantic identifier;

[0012] For each question cluster, based on a large language model, all relevant historical answers are summarized, extracted, and refined to generate one or more standardized authoritative answer templates.

[0013] In an optional implementation, the authoritative answer template is structured data that includes fixed information fields and variable parameter placeholders.

[0014] In an optional implementation, the method further includes generating multiple answer versions for the same question cluster, each answer version being suitable for a different audience or a different level of detail.

[0015] In an optional implementation, the attribute information includes at least the exhibition area theme category corresponding to the geographical location where it is deployed.

[0016] In an optional implementation, determining the push strategy based on the attribute information of each digital human node includes:

[0017] Based on preset business rules, dynamic weight coefficients are assigned to different dimensions;

[0018] Determine the static attribute vectors of the digital human node in multiple dimensions;

[0019] The push strategy is determined based on the weight coefficients and the static attribute vector.

[0020] In an optional implementation, the weighting coefficient is dynamically adjusted based on the actual hit rate of the previous push content and user satisfaction feedback.

[0021] In an optional implementation, matching in the local cache includes:

[0022] The judgment threshold can be determined based on the overall load status of the current network and the utilization rate of the central server's computing resources;

[0023] Calculate the semantic similarity between the current problem and the problems in the local cache;

[0024] Determine if there are caches with semantic similarity exceeding the judgment threshold;

[0025] If the match does not exist, the match is considered unsuccessful; if the match exists, the match is considered successful.

[0026] Secondly, the present invention provides a distributed digital human question-and-answer system based on 5G communication. The system includes a central server and multiple distributed digital human nodes, wherein each digital human node integrates a 5G communication module; each digital human node accesses the 5G network through its own 5G communication module and establishes a communication connection with the central server.

[0027] The central server is used to maintain a global central cache set for the historical question and answer data of each digital human node; and to determine the push strategy according to the attribute information of each digital human node, and periodically select target cache in the central cache set according to the push strategy, and distribute it to the corresponding digital human node through the 5G network as the initial content or updated content of its local cache.

[0028] The digital human node is used to receive the current question and match it in the local cache; if the match is successful, the answer is returned directly based on the local cache; if the match is unsuccessful, the current question is uploaded to the central server via the 5G network.

[0029] The central server is also used to call a large language model to process the current problem, generate an answer, and transmit it downstream to the digital human node that initiated the request.

[0030] Thirdly, the present invention provides an electronic device including a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method described in any of the foregoing embodiments.

[0031] This application constructs a central cache set and proactively distributes it, enabling the problems faced by a digital human to quickly benefit the entire group, significantly reducing the load on the central server and the average response latency. The central server leverages the powerful understanding and generation capabilities of a large language model to clean, summarize, and enhance global question-and-answer data, fundamentally ensuring the high quality and accuracy of the cached content distributed to each node. Based on the attribute information of the digital human nodes, a push strategy is formulated to pre-deploy hot topics to frontline digital human nodes, improving the hit rate and saving air interface resources. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1A schematic flowchart of a distributed digital human question-answering method based on 5G communication provided in an embodiment of this application;

[0034] Figure 2 A schematic diagram of a distributed digital human question-answering system based on 5G communication is provided for an embodiment of this application;

[0035] Figure 3 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0036] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0037] Figure 1 This is a schematic flowchart illustrating a distributed digital human question-answering method based on 5G communication, provided as an embodiment of this application. Figure 1 As shown, this method is applied to, for example Figure 2 The system shown includes a central server and multiple distributed digital human nodes. Each digital human node integrates a 5G communication module; each digital human node accesses the 5G network through its own 5G communication module and establishes a communication connection with the central server.

[0038] like Figure 1 As shown, the method includes the following steps:

[0039] S110, the central server maintains a global central cache set for the historical question and answer data of each digital human node; and determines the push strategy according to the attribute information of each digital human node, periodically selects the target cache in the central cache set according to the push strategy, and distributes it to the corresponding digital human node through the 5G network as the initial content or updated content of its local cache.

[0040] The central server can periodically format historical question-and-answer data from different digital human nodes to eliminate differences in expression; it can use a large language model to automatically classify semantically similar but differently expressed questions into the same question cluster and assign a globally unique semantic identifier to each question cluster; for each question cluster, it can summarize, extract and refine all relevant historical answers based on the large language model to generate one or more standardized authoritative answer templates.

[0041] The authoritative answer template is structured data, containing fixed information fields and variable parameter placeholders. For example, "Current temperature is {temperature}℃, {weather_condition}, recommendation {recommendation}", where each placeholder is dynamically filled by the digital human node in combination with local sensor data or real-time API during the actual response, ensuring that the answer is both standardized and highly timely and adaptable to the scenario.

[0042] In some embodiments, the attribute information includes at least the exhibition area theme category corresponding to its geographical location of deployment. Based on this, dynamic weight coefficients can be assigned to different dimensions according to preset business rules; static attribute vectors of multiple dimensions of the digital human node can be determined; and a push strategy can be determined based on the weight coefficients and static attribute vectors.

[0043] The weighting coefficients can also be dynamically adjusted based on the actual hit rate of the previous push content and user satisfaction feedback.

[0044] The central cache set is a global question-and-answer data warehouse maintained by the central server. It contains standardized historical question-and-answer pairs, question clusters, semantic identifiers, and authoritative answer templates, used to distribute cached content to each digital human node. In some embodiments, the specific processing flow is as follows:

[0045] Historical Q&A data formatting: Historical Q&A data (such as user questions and digital human replies) from different digital human nodes are preprocessed in a unified manner, including: text cleaning, such as removing special symbols, redundant spaces, etc.; terminology standardization, which can unify industry terms, such as unifying "exhibition area" and "exhibition hall" as "exhibition area"; and length normalization, which can process the text to a preset length by truncation or padding.

[0046] A large language model can be used to semantically encode the formatted questions. This large language model can include various types, such as a fine-tuned BERT model. Clustering algorithms can be used to group questions with similar semantics but different expressions into question clusters. Each question cluster is assigned a semantic identifier, for example, in the format "Cluster-XXXXX" (XXXXX being a globally unique numeric code), serving as a globally unique identifier for the question cluster.

[0047] For each question cluster, a large language model can be invoked to summarize and refine historical answers, generating an authoritative answer template. For example, this template might include the following:

[0048] Fixed information fields: general descriptions (such as "opening hours" and "exhibition area introduction");

[0049] Variable parameter placeholders: Content that needs to be dynamically replaced based on the attributes of the digital human node (such as "{exhibition area name}" or "{age}").

[0050] Example: The authoritative template for the question cluster "Exhibition Area Opening Hours" is "{Exhibition Area Name} opens daily from {Opening Time} to {Closing Time}, with holidays subject to on-site announcements", where "{Exhibition Area Name}" and "{Opening Time}" are variable parameters.

[0051] In some embodiments, answer versions adapted to different audiences or levels of detail can be generated for the same question family. For example, a children's version with simplified language and emojis; a professional version with detailed data and terminology; a concise version with core information; and a full version with additional notes. Version tags (such as "v_child" and "v_pro") can be added during storage.

[0052] In some embodiments, the push strategy is used to determine which cached content, namely question clusters and corresponding authoritative answer templates, the central server distributes to each digital human node. The core is to optimize the distribution logic based on the attribute information and dynamic feedback of the digital human node.

[0053] Among them, the static attribute vector V is a quantized vector describing the inherent characteristics of the digital human node, and its dimensions can include the following:

[0054] Exhibition area theme categories T: such as culture (1), science and technology (2), entertainment (3);

[0055] Target age group A: Children (1), Adults (2), Seniors (3);

[0056] Historical question-and-answer frequency F: Normalized number of question interactions (0-1, e.g., 100 questions and answers per day corresponds to F=0.8).

[0057] Vector form: ,in .

[0058] Dynamic weight coefficient W: A dynamically adjusted parameter representing the importance of each attribute dimension. It is initially assigned based on preset business rules (e.g., the theme of the exhibition area has the highest weight), and is subsequently optimized based on feedback.

[0059] The formula for updating the weight coefficients can be as follows: ;

[0060] Based on this formula, the weights can be dynamically adjusted by combining the hit rate H of the previous push content and user satisfaction feedback S, thus avoiding over-reliance on a single indicator.

[0061] in, The updated weight coefficients for the i-th dimension;

[0062] The updated weight coefficients for the i-th dimension;

[0063] H represents the local cache hit rate of the pushed content, which can be determined by dividing the number of successful matches by the total number of pushed issues.

[0064] S represents user satisfaction feedback, such as "Satisfied" = 1, "Neutral" = 0.5, "Dissatisfied" = 0;

[0065] Here, (is the adjustment coefficient; where, ( , To avoid excessive weight reduction due to H / S being 0.

[0066] In some embodiments, the matching degree M between the problem cluster and the digital human node can be calculated based on the static attribute vector V and the dynamic weight coefficient W, and the problem cluster with the highest M can be selected as the target cache so that the target cache can be pushed to the corresponding digital human node: ;

[0067] in, Let i be the i-th dimension attribute feature of the problem cluster. for With node attributes Similarity, for example, if the exhibition areas have the same theme... Different ).

[0068] S120: The digital human node receives the current question and matches it in its local cache. If the match is successful, the answer is returned directly based on the local cache. If the match is unsuccessful, the current question is uploaded to the central server via the 5G network.

[0069] The judgment threshold can be determined based on the overall load status of the current network and the computing resource utilization of the central server; the semantic similarity between the current question and the question in the local cache can be calculated; it can be determined whether there is a cached question with a semantic similarity higher than the judgment threshold; if there is no such question, the match is determined to be unsuccessful; if there is such a question, the match is determined to be successful.

[0070] In some embodiments, the judgment threshold is a semantic similarity threshold for determining whether the current problem matches a locally cached problem. It needs to be dynamically adjusted according to network load and central server resources to avoid excessive or insufficient local matching.

[0071] The decision threshold can be determined using the following formula: ;

[0072] Based on this, when network load is high or server resources are scarce, the threshold can be lowered to allow local matching with slightly lower similarity, thus reducing upload requests; conversely, the threshold can be raised to ensure the accuracy of the answer as much as possible.

[0073] in, The base threshold can be preset to, for example, 0.7;

[0074] L represents the overall network load rate, which can be obtained through the 5G network management module, such as the current number of active connections / maximum number of connections.

[0075] U represents the utilization of central server computing resources, such as CPU utilization.

[0076] Here, is the adjustment coefficient; where, To avoid the answer quality from being too low, θ should be avoided.

[0077] In some embodiments, semantic similarity is used to measure the semantic closeness between the current question Q and the locally cached question Q', and semantic features and domain relevance can be combined to improve scenario adaptability.

[0078] The semantic similarity formula is as follows: ;

[0079] Based on this formula, similarity can be determined using cosine distance and neighborhood differences.

[0080] in, Cosine similarity of sentence vectors based on a pre-trained language model;

[0081] Domain relevance similarity can be obtained through domain classification models. For example, if Q and Q' both belong to the "technology domain", then... Cross-domain );

[0082] For weights ( Prioritize semantic matching while also considering domain adaptation.

[0083] Processing the matching results: If the match is successful, that is, if there exists Q' such that... You can directly call the authoritative answer template in the local cache, replace the variable parameters, and return the answer;

[0084] If the match fails, meaning no Q' satisfies the condition... The current problem is uploaded to the central server via the 5G network.

[0085] S130, the central server calls the large language model to process the current question, generates the answer, and transmits it down to the digital human node that initiated the request.

[0086] It can call customized large language models (such as a digital human-specific model fine-tuned based on LLaMA, which optimizes short text question answering and multi-turn dialogue capabilities), input questions and context (such as node attributes and historical dialogues), and generate answers;

[0087] In some embodiments, when the central server processes the problem based on a large language model, the central cache set can also be used as a local knowledge base for enhanced retrieval. In this case, the processing flow of the central server can be as follows:

[0088] Before calling the large language model to process the problem, the central server can start a retrieval and sorting module.

[0089] First, a deep semantic analysis of the question is performed to identify core keywords and intents, such as "reusable rocket", "technical difficulties", and "engine thrust adjustment".

[0090] The question is transformed into a high-dimensional vector, which serves as the query key. A similarity search is performed in the vector database of the central cache set to retrieve the K most semantically relevant question-answer pairs. These pairs, accumulated from the history of all digital human nodes, constitute a rich localized knowledge base. Assuming highly relevant cached entries are retrieved, the results are sorted according to a contextual relevance scoring formula, and the top N are selected as enhanced context. For example, these N cached entries can serve as background knowledge.

[0091] The central server inputs the constructed enhanced prompts into the large language model responsible for core reasoning. The enhanced prompts include the original question, historical dialogue context, and the top N question-answer pairs retrieved from the central cache, effectively improving the accuracy and domain adaptability of the answers. After generating the answer, the large language model returns it to the requesting digital human node, while simultaneously storing the question-answer pair in encrypted form in its local cache for use in matching similar questions later. The entire process balances response efficiency with knowledge accumulation, forming a closed-loop optimization mechanism.

[0092] Among them, the 5G communication module can be a hardware module integrated into the digital human node. It can support the 5G NR protocol, provide high-speed, low-latency data transmission, and is the core of communication between distributed nodes and the central server.

[0093] Distributed digital human nodes are digital human terminals deployed in different geographical locations (such as different exhibition areas). They have local computing (CPU / GPU) and storage (local cache) capabilities, and can independently complete simple question and answer tasks, while complex questions are uploaded to the central server.

[0094] The central cache collection is a global question-and-answer cache database maintained by the central server. It contains question clusters, semantic identifiers, and authoritative answer templates, and serves as the "source" for local caching on each node.

[0095] Static attribute vectors are quantified vectors used to describe the inherent characteristics of digital human nodes (such as exhibition area themes and age groups of service targets), and are used for attribute matching in push strategies.

[0096] The dynamic weighting coefficient can dynamically adjust the attribute importance weights based on business feedback (hit rate, satisfaction) to optimize target cache selection.

[0097] Semantic similarity is an indicator used to measure how close two questions are at the semantic level. It combines semantic features and domain relevance to improve matching accuracy.

[0098] This application constructs a central cache set and proactively distributes it, enabling the problems faced by a digital human to quickly benefit the entire group, significantly reducing the load on the central server and the average response latency. The central server leverages the powerful understanding and generation capabilities of a large language model to clean, summarize, and enhance global question-and-answer data, fundamentally ensuring the high quality and accuracy of the cached content distributed to each node. Based on the attribute information of the digital human nodes, a push strategy is formulated to pre-deploy hot topics to frontline digital human nodes, improving the hit rate and saving air interface resources.

[0099] Based on the same inventive concept, this application also provides a distributed digital human question-answering system based on 5G communication, such as... Figure 2 As shown, the system includes a central server 210 and multiple distributed digital human nodes 220, each of which is equipped with a 5G communication module. Each digital human node 220 accesses the 5G network through its own 5G communication module and establishes a communication connection with the central server 210.

[0100] The central server 210 is used to maintain a global central cache set for the historical question and answer data of each digital human node; and to determine the push strategy according to the attribute information of each digital human node, and periodically select the target cache in the central cache set according to the push strategy, and distribute it to the corresponding digital human node through the 5G network as the initial content or updated content of its local cache.

[0101] The digital human node 220 is used to receive the current question and match it in the local cache; if the match is successful, the answer is returned directly based on the local cache; if the match is unsuccessful, the current question is uploaded to the central server via the 5G network.

[0102] The central server 210 is also used to call the large language model to process the question, generate an answer, and transmit it downstream to the digital human node that initiated the request; the digital human node that initiated the request updates its local cache with the question and the answer as a new question-answer pair.

[0103] In some embodiments, the central server 210 is also used to: periodically format historical question-and-answer data from different digital human nodes to eliminate differences in expression; automatically classify semantically similar but differently expressed questions into the same question cluster using a large language model, and assign a globally unique semantic identifier to each question cluster; for each question cluster, summarize, extract and refine all relevant historical answers based on the large language model to generate one or more standardized authoritative answer templates.

[0104] The authoritative answer template is structured data, which contains fixed information fields and variable parameter placeholders.

[0105] In some embodiments, the central server 210 is also used to generate multiple answer versions for the same question cluster, each answer version being suitable for a different audience or a different level of detail.

[0106] The attribute information includes at least the exhibition area theme category corresponding to the geographical location where it is deployed.

[0107] In some embodiments, the central server 210 is further configured to: assign dynamic weight coefficients to different dimensions according to preset business rules; determine static attribute vectors for multiple dimensions of the digital human node; and determine a push strategy based on the weight coefficients and static attribute vectors.

[0108] In some embodiments, the central server 210 is also configured to: dynamically adjust the weight coefficient based on the actual hit rate of the previously pushed content and user satisfaction feedback.

[0109] In some embodiments, the digital human node 220 is further configured to: determine a judgment threshold based on the overall load status of the current network and the computing resource utilization of the central server; calculate the semantic similarity between the current question and the question in the local cache; determine whether there is a cached question with a semantic similarity higher than the judgment threshold; if there is no such question, determine that the match is unsuccessful; if there is such a question, determine that the match is successful.

[0110] This invention provides an electronic device including a processor and a storage device; the storage device stores a computer program, which, when executed by the processor, performs the method described in any of the above embodiments.

[0111] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 100 includes: a processor 30, a memory 31, a bus 32 and a communication interface 33. The processor 30, the communication interface 33 and the memory 31 are connected through the bus 32. The processor 30 is used to execute executable modules, such as computer programs, stored in the memory 31.

[0112] The memory 31 may include high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 33 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0113] Bus 32 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0114] The memory 31 is used to store programs. After receiving an execution instruction, the processor 30 executes the program. The method executed by the device for defining the flow process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 30 or implemented by the processor 30.

[0115] Processor 30 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 30 or by instructions in software form. Processor 30 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 31. The processor 30 reads the information in memory 31 and, in conjunction with its hardware, completes the steps of the above method.

[0116] The computer program product of the readable storage medium provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For specific implementation, please refer to the foregoing method embodiments, which will not be repeated here.

[0117] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion 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 this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0118] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, 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, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A distributed digital human question-answering method based on 5G communication, characterized in that, The method is applied to a system including a central server and multiple distributed digital human nodes, wherein each digital human node integrates a 5G communication module; each digital human node accesses the 5G network through its own 5G communication module and establishes a communication connection with the central server. The method includes: The central server maintains a global central cache set for the historical question-and-answer data of each digital human node; and determines the push strategy according to the attribute information of each digital human node, periodically selects the target cache in the central cache set according to the push strategy, and distributes it to the corresponding digital human node through the 5G network as the initial or updated content of its local cache. The digital human node receives the current question and matches it with the local cache; if the match is successful, the answer is returned directly based on the local cache; if the match is unsuccessful, the current question is uploaded to the central server via the 5G network. The central server invokes a large language model to process the current question, generates an answer, and transmits it downstream to the digital human node that initiated the request.

2. The method according to claim 1, characterized in that, The central server maintains a global central cache set for the historical question-and-answer data of each digital human node, including: Historical question-and-answer data from different digital human nodes are periodically formatted to eliminate discrepancies in expression; Using a large language model, questions with similar semantics but different expressions are automatically classified into the same question cluster, and each question cluster is assigned a globally unique semantic identifier; For each question cluster, based on a large language model, all relevant historical answers are summarized, extracted, and refined to generate one or more standardized authoritative answer templates.

3. The method according to claim 2, characterized in that, The authoritative answer template is structured data, which contains fixed information fields and variable parameter placeholders.

4. The method according to claim 3, characterized in that, Also includes: Generate multiple answer versions for the same question family, each suitable for a different audience or a different level of detail.

5. The method according to claim 1, characterized in that, The attribute information includes at least the exhibition area theme category corresponding to its geographical location.

6. The method according to claim 5, characterized in that, The push strategy is determined based on the attribute information of each digital human node, including: Based on preset business rules, dynamic weight coefficients are assigned to different dimensions; Determine the static attribute vectors of the digital human node in multiple dimensions; The push strategy is determined based on the weight coefficients and the static attribute vector.

7. The method according to claim 6, characterized in that, The weighting coefficients are dynamically adjusted based on the actual hit rate of the previous push content and user satisfaction feedback.

8. The method according to claim 1, characterized in that, Matching in the local cache includes: The judgment threshold is determined based on the overall load status of the current network and the utilization rate of the central server's computing resources. Calculate the semantic similarity between the current problem and the problems in the local cache; Determine if there are caches with semantic similarity exceeding the judgment threshold; If the match does not exist, the match is considered unsuccessful; if the match exists, the match is considered successful.

9. A distributed digital human question-answering system based on 5G communication, characterized in that, The system includes a central server and multiple distributed digital human nodes, each of which integrates a 5G communication module; each digital human node accesses the 5G network through its own 5G communication module and establishes a communication connection with the central server. The central server is used to maintain a global central cache set for the historical question and answer data of each digital human node; and to determine the push strategy according to the attribute information of each digital human node, and periodically select target cache in the central cache set according to the push strategy, and distribute it to the corresponding digital human node through the 5G network as the initial content or updated content of its local cache. The digital human node is used to receive the current question, match it with the local cache, and if the match is successful, return the answer directly based on the local cache. If the match fails, the current problem will be uploaded to the central server via the 5G network; The central server is also used to call a large language model to process the current problem, generate an answer, and transmit it downstream to the digital human node that initiated the request.

10. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method of any one of claims 1 to 8.

Images