An artificial intelligence-based knowledge question and answer data processing method and system

Abstract

The application relates to the technical field of artificial intelligence knowledge question answering, and discloses a knowledge question answering data processing method and system.The method comprises the following steps: collecting user query data and local knowledge base information through a distributed node, obtaining semantic vector representation through semantic coding; and obtaining a preliminary answer result, grouping to obtain similar answer clusters; if the threshold value is exceeded, sharing cluster centers, judging deviation modes, correcting knowledge representation and updating coding; establishing a unified answer framework by fusing global consistent features, and when the matching degree is lower than a preset threshold value, iteratively fusing to obtain a refined answer; distributing the refined answer to obtain a globally consistent output result and extracting a version mark, updating the node to judge whether the deviation is eliminated, if the deviation still exists, integrating additional data for iterative optimization, and finally outputting a target result.The method can improve the accuracy and consistency of knowledge question answering.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and knowledge question answering data processing technology, and in particular to a knowledge question answering data processing method and system based on artificial intelligence. Background Technology

[0002] Currently, knowledge-based question-answering systems, as an important branch of artificial intelligence, are widely used in education, customer service, medical diagnosis, and other scenarios. Their core is to build a semantic network based on a knowledge base and semantic understanding capabilities, enabling rapid and accurate responses to user queries and providing support for improving the accuracy and semantic relevance of question answers. With the surge in data volume and the diversification of user needs, building efficient, accurate knowledge-based question-answering systems that support multi-node collaboration has become an urgent requirement for industry development, especially in real-time response scenarios where the system needs to balance the efficiency of processing massive amounts of data with the consistency of answers.

[0003] In one existing technology, the knowledge question answering system mainly adopts a centralized architecture-based processing mode, which coordinates data processing through a single or a small number of core nodes, and relies on a knowledge base to complete semantic matching and answer generation, thus ensuring basic response efficiency when the query volume is small.

[0004] However, centralized architectures lack multi-node collaboration capabilities and effective deviation correction mechanisms, making them unsuitable for distributed deployment scenarios. When the system expands to multiple nodes, the knowledge base versions of each node are prone to synchronization issues and local data updates lag, leading to semantically biased answers for the same user's query. Furthermore, it's difficult for nodes to share deviation correction experience; even if one node corrects a semantic bias, it cannot be quickly synchronized to other nodes. This is especially problematic when processing queries for newly added knowledge points in the knowledge base, significantly increasing the risk of incorrect answer output and ultimately resulting in poor global answer consistency and a high risk of erroneous output. Summary of the Invention

[0005] This invention provides a knowledge question-answering data processing method and system based on artificial intelligence to solve the problems of poor global answer consistency and high risk of erroneous output caused by existing technologies.

[0006] In a first aspect, the present invention provides a knowledge question-answering data processing method based on artificial intelligence, comprising:

[0007] By collecting user query data and local knowledge base information through distributed nodes, semantic encoding is performed on the user query data to obtain a semantic vector representation;

[0008] Preliminary answer results are obtained based on the semantic vector representation. Similar answer clusters are determined by grouping the preliminary answer results. It is then detected whether the similar answer clusters involve knowledge base version deviation.

[0009] If the number of similar answer clusters exceeds a preset threshold, the cluster center information is shared and the answer inconsistency deviation pattern is determined to obtain the corrected knowledge representation;

[0010] Globally consistent features are extracted and fused from the corrected knowledge representation to determine a unified answer framework;

[0011] If the matching degree between the unified answer framework and the semantic vector representation is lower than the preset matching degree threshold, then the globally consistent features are iteratively fused and the node consistency is verified to obtain a refined answer;

[0012] The refined answer is distributed to each distributed node and the node receives and confirms the receipt, resulting in a preliminary globally consistent output result.

[0013] Extract the answer version marker from the preliminary globally consistent output and update the distributed nodes to obtain the optimized system state;

[0014] If the deviation pattern in the optimized system status display still exists, then additional query data is integrated and iterated from the semantic encoding step to update the unified answer framework, and finally globally consistent output results are obtained.

[0015] In one optional implementation, the step of collecting user query data and local knowledge base information through distributed nodes, and semantically encoding the user query data to obtain a semantic vector representation includes:

[0016] The system collects user query data and local knowledge base information through distributed nodes, performs consistent allocation, and outputs an initial data set.

[0017] Based on the initial data set, semantic encoding is performed on the user query data to generate a semantic vector representation, and semantic encoded data containing the semantic vector representation is output;

[0018] Based on the semantically encoded data containing semantic vector representation, data between nodes is synchronized, deviations caused by rare queries are identified and processed in deviation detection, and usable semantically encoded data containing semantic vector representation is output.

[0019] In one optional implementation, the step of obtaining preliminary answer results from each distributed node based on the semantic vector representation, and grouping them to determine similar answer clusters includes:

[0020] Based on the semantic vector representation, preliminary answer data matching the semantics is obtained from distributed nodes, the distance between semantic vectors is calculated to filter related data, and a preliminary answer set is output.

[0021] Based on the preliminary answer set, the preliminary answer results are grouped according to semantic similarity to form similar answer clusters, and the answer cluster set is output.

[0022] Detect the semantic consistency of similar answer clusters in the answer cluster set, determine whether there is a deviation due to differences in knowledge base versions, and output the deviation judgment result.

[0023] In one optional implementation, the step of sharing cluster center information and determining the answer inconsistency deviation pattern if the number of similar answer clusters exceeds a preset threshold, and obtaining the corrected knowledge representation, includes:

[0024] If the number of similar answer clusters in the answer cluster set exceeds a preset threshold, the cluster center information of each similar answer cluster is shared through inter-node broadcasting to determine the scope of cluster center information sharing;

[0025] Based on the information sharing range of the cluster center, and combined with the comparison of different versions of the knowledge base content, the deviation patterns of inconsistent answers at each node are detected, and a set of deviation patterns is output.

[0026] Deviation features are extracted from the deviation pattern set, the initial knowledge representation is corrected, and the corrected knowledge representation is output.

[0027] The corrected knowledge representation is associated with additional query data, the iterative encoding parameters and process are adjusted, the iterative encoding process is updated, and the updated encoded data is output.

[0028] In one optional implementation, the step of extracting and fusing globally consistent features from the corrected knowledge representation to determine a unified answer framework includes:

[0029] From the corrected knowledge representation, extract globally consistent features that are common across nodes and output a set of globally consistent features;

[0030] Based on the globally consistent feature set, calculate the distance between the semantic vectors corresponding to each feature, fuse the globally consistent features by similarity weighting, and output the fused feature representation;

[0031] Based on the fused feature representation, a structured unified answer framework is constructed and output;

[0032] The unified answer framework is supplemented with additional query data collected synchronously between nodes to supplement semantic information, enhance the accuracy of deviation pattern detection, and output an optimized unified answer framework.

[0033] In one optional implementation, if the matching degree between the unified answer framework and the semantic vector representation is lower than a preset matching degree threshold, then the globally consistent features are iteratively fused and node consistency is verified to obtain a refined answer, including:

[0034] If the matching degree between the unified answer framework and the semantic vector representation is lower than the preset matching degree threshold, the globally consistent features are iteratively fused through the inter-node data exchange protocol to supplement semantic information and output an updated feature set.

[0035] The candidate answers corresponding to the updated feature set are distributed to each distributed node according to a preset distribution rule. The consistency between the data of each node and the updated feature set is verified, the semantically consistent answers are filtered, and a refined set of answers is output.

[0036] The refined answer set is iteratively encoded in accordance with the knowledge base version control requirements to eliminate remaining semantic bias patterns and output the optimized refined answer.

[0037] In one optional implementation, the step of distributing the refined answer to each distributed node and processing the node to receive confirmation, thereby obtaining a preliminary globally consistent output result, includes:

[0038] The refined answer is distributed to all nodes in the distributed system, and the received results of each node are hash-verified to generate a set of node reception confirmation identifiers.

[0039] Based on the set of confirmation identifiers received by the nodes, after confirming that all nodes have received the complete and refined answer, the answer distribution process and results are recorded to obtain a globally consistent output result;

[0040] Extract the answer version tag used to identify the answer version from the globally consistent output results, and output the answer version tag.

[0041] In one optional implementation, the step of extracting the answer version tag from the preliminary globally consistent output result and updating the distributed nodes to obtain the optimized system state includes:

[0042] Based on the answer version tag, the output results of each distributed node are aggregated and merged to generate a globally consistent answer record;

[0043] Based on the globally consistent answer record, the answer version marker is compared with a preset version threshold to detect semantic consistency deviation and output the semantic consistency deviation detection result.

[0044] Based on the semantic consistency deviation detection results, adjust the state parameters of the distributed nodes to update the nodes, determine whether the semantic deviation has been eliminated, and output the optimized system state.

[0045] In the optimized system state, additional query data collected synchronously between nodes is integrated to provide data support for potential iterative coding processes and output iterative coding preparation data.

[0046] In an optional implementation, if the optimized system status display deviation pattern still exists, then additional query data is integrated and iterated from the semantic encoding step to update the unified answer framework, obtaining a final globally consistent output result, including:

[0047] If the optimized system status display still shows a semantic deviation pattern, additional query data related to the deviation pattern is collected synchronously between nodes to determine the deviation pattern dataset;

[0048] Based on the deviation pattern dataset, extract related data from the knowledge base, update the entries in the knowledge base corresponding to the deviation patterns, and output the updated knowledge base dataset.

[0049] Based on the updated knowledge base dataset, the unified answer framework is iteratively encoded and updated, and the semantic vector representation in the framework is adjusted in combination with additional query data to output the updated unified answer framework.

[0050] Based on the updated unified answer framework, the semantic processing logic of each distributed node is optimized to ensure that the semantics of the node output are consistent, thus obtaining the final optimized system.

[0051] Secondly, the present invention provides a knowledge question-answering data processing system based on artificial intelligence, comprising:

[0052] The data acquisition and encoding module collects user query data and local knowledge base information through distributed nodes, performs semantic encoding on the user query data, and obtains a semantic vector representation.

[0053] Cluster generation module: Based on the semantic vector representation, it obtains preliminary answer results from each distributed node, groups them, and determines similar answer clusters;

[0054] Knowledge correction module: If the number of similar answer clusters exceeds a preset threshold, the cluster center information is shared and the answer inconsistency deviation pattern is determined to obtain the corrected knowledge representation;

[0055] Framework construction module: Extracts and fuses globally consistent features from the corrected knowledge representation to determine a unified answer framework;

[0056] Refined Answer Module: If the matching degree between the unified answer framework and the semantic vector representation is lower than the preset matching degree threshold, then the globally consistent features are iteratively fused and the node consistency is verified to obtain the refined answer;

[0057] Global output module: Distributes the refined answer to each distributed node and processes the node's confirmation of receipt to obtain a preliminary globally consistent output result;

[0058] System update module: Extracts the answer version tag from the preliminary globally consistent output result and updates the distributed nodes to obtain the optimized system state;

[0059] Enhanced optimization module: If the deviation mode of the optimized system status display still exists, then integrate additional query data and iterate from the semantic encoding steps to update the unified answer framework and obtain the final globally consistent output result.

[0060] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the artificial intelligence-based knowledge question-answering data processing method described in any one of the above.

[0061] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the knowledge question-answering data processing method based on artificial intelligence described in any one of the above-mentioned methods.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] (1) This invention collects data from distributed nodes and distributes data in a consistent manner. It combines semantic encoding to generate semantic vector representations and synchronizes data processing between nodes to avoid rare query deviations. This breaks through the limitations of the traditional centralized architecture and achieves distributed and accurate perception of user queries, providing a unified semantic foundation for subsequent answer processing.

[0064] (2) This invention forms a closed loop of “data encoding-deviation correction-framework construction” by generating similar answer clusters, judging deviation patterns, and fusing global consistent features. Then, through refined answer generation and node updates, it ensures the global consistency of answers in a multi-node environment and effectively reduces the risk of erroneous output caused by semantic deviation.

[0065] (3) This invention continuously adapts to the dynamic updates of the knowledge base and changes in user query needs through iterative coding updates and final optimization of system construction, thereby improving the system's response efficiency and answer accuracy in high-concurrency scenarios and meeting the high-precision requirements of knowledge question and answer services in fields such as education, customer service, and medical care. Attached Figure Description

[0066] Figure 1 This is a schematic diagram of a knowledge question-answering data processing method based on artificial intelligence provided in the first embodiment of the present invention;

[0067] Figure 2 This is a schematic diagram of the structure of a knowledge question-answering data processing system based on artificial intelligence, provided in the second embodiment of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0069] Reference Figure 1 The first embodiment of the present invention provides a knowledge question-answering data processing method based on artificial intelligence, including the following steps:

[0070] S1: Collect user query data and local knowledge base information through distributed nodes, perform semantic encoding on the user query data, and obtain semantic vector representation;

[0071] S2, based on the semantic vector representation, obtain preliminary answer results from each distributed node, group them, and determine similar answer clusters;

[0072] S3, if the number of similar answer clusters exceeds a preset threshold, share cluster center information and determine the answer inconsistency deviation pattern to obtain the corrected knowledge representation;

[0073] S4. Extract and fuse globally consistent features from the corrected knowledge representation to determine a unified answer framework;

[0074] S5, if the matching degree between the unified answer framework and the semantic vector representation is lower than the preset matching degree threshold, then iteratively fuse the globally consistent features and verify the node consistency to obtain the refined answer;

[0075] S6, Distribute the refined answer to each distributed node and process the node's confirmation of receipt to obtain a preliminary globally consistent output result;

[0076] S7, extract the answer version tag from the preliminary globally consistent output result and update the distributed node to obtain the optimized system state;

[0077] S8. If the deviation mode of the optimized system status display still exists, then integrate the additional query data and iterate from the semantic encoding step to update the unified answer framework and obtain the final globally consistent output result.

[0078] In step S1, user query data and local knowledge base information are collected through distributed nodes. The user query data is semantically encoded to obtain a semantic vector representation. Rare query discrepancies in data processing between synchronization nodes are addressed, including:

[0079] S11 collects user query data and local knowledge base information through distributed nodes, performs consistent allocation, and outputs the initial data set;

[0080] S12, based on the initial data set, semantically encode the user query data to generate a semantic vector representation, and output semantically encoded data containing the semantic vector representation;

[0081] S13, based on the semantically encoded data containing semantic vector representation, synchronize the data between nodes, identify and process the deviations caused by rare queries in the deviation detection, and output usable semantically encoded data containing semantic vector representation.

[0082] In step S11, user query data and local knowledge base information are collected through distributed nodes, and consistent allocation is performed to output the initial data set.

[0083] It should be noted that distributed nodes are deployed in multiple regions. Each node collects user query text in real time, such as "broadband fault troubleshooting methods" and "medical insurance reimbursement process", as well as locally stored knowledge base information. Data is allocated in a consistent manner according to data characteristics to ensure that the data load of each node is balanced. The allocated data is stored in the format of "node number-collection time-user query-knowledge base related entry" to form an initial data set.

[0084] For example, node "ND01" collected the query "How to reset broadband password" and the associated knowledge base entry "Broadband Password Reset Guide V2.1" at 10:00:00 on 2025-09-01, and recorded it as "ND01, 2025-09-01 10:00:00, How to reset broadband password, Broadband Password Reset Guide V2.1"; node "ND02" collected the query "What materials are needed for medical insurance reimbursement in other places" and the associated entry "Medical insurance reimbursement material list V3.0" at the same time, and recorded it as "ND02, 2025-09-01 10:00:00, What materials are needed for medical insurance reimbursement in other places, Medical insurance reimbursement material list V3.0". The integration of the two types of data is the initial data set.

[0085] In step S12, based on the initial data set, the user query data is semantically encoded to generate a semantic vector representation, and the semantically encoded data containing the semantic vector representation is output.

[0086] It should be noted that the user query text is extracted from the initial dataset and converted into a machine-readable numerical form to capture semantic features, generating a high-dimensional semantic vector representation. The vector dimension and numerical value are determined according to the query semantics. The generated data is organized in the format of "node number-query text-semantic vector-collection time" to form semantic encoded data containing semantic vector representation.

[0087] For example, the query "How to reset broadband password" is encoded into a 512-dimensional vector "[0.12,0.35,...,0.28]", recorded as "ND01, How to reset broadband password, [0.12,0.35,...,0.28], 2025-09-01 10:00:00". The values ​​in the vector are normalized example values, and their specific range depends on the semantic encoding model used. The query "What materials are needed for medical insurance reimbursement in other places" is encoded into a vector "[0.41,0.22,...,0.33]", recorded as "ND02, What materials are needed for medical insurance reimbursement in other places, [0.41,0.22,...,0.33], 2025-09-01 10:00:00". The integration of the two types of data is the semantic encoded data.

[0088] In step S13, based on the semantically encoded data containing semantic vector representation, the data between nodes is synchronized, deviations caused by rare queries are identified and processed in the deviation detection, and usable semantically encoded data containing semantic vector representation is output.

[0089] It should be noted that each node shares semantic encoded data through a data synchronization mechanism to ensure data consistency between nodes; rare queries with abnormal semantic vectors, such as queries with special expressions or involving niche knowledge points, are screened and their semantic deviations from regular queries are analyzed; the deviations are corrected by supplementing the associated knowledge base information and adjusting the semantic encoding parameters, and finally, data without deviation or with deviations that meet the requirements are retained to form usable semantic encoded data.

[0090] For example, node "ND03" collected a rare query "Enterprise leased broadband password reset steps". Its semantic vector deviated significantly from the regular "broadband password reset" query. By synchronizing the "Enterprise Leased Line Knowledge Base" information of other nodes and supplementing the encoding dimensions, the deviation was corrected. If the semantic vector of a query deviates from the regular data of all nodes by more than a preset range, it is marked as invalid data and removed. In the final output of usable data, the semantic vector deviation corresponding to "Enterprise leased broadband password reset steps" was reduced to the compliant range and recorded as "ND03, Enterprise leased broadband password reset steps, [0.15, 0.32, ..., 0.26], 2025-09-01 10:05:00".

[0091] In step S2, based on the semantic vector representation, preliminary answer results are obtained from each distributed node and grouped to determine similar answer clusters, including:

[0092] S21, based on the semantic vector representation, obtain preliminary answer data that matches the semantics from the distributed nodes, calculate the distance between semantic vectors to filter related data, and output a preliminary answer set;

[0093] S22, Based on the preliminary answer set, group the preliminary answer results according to semantic similarity to form similar answer clusters, and output the answer cluster set;

[0094] S23, detect the semantic consistency of similar answer clusters in the answer cluster set, determine whether there is a deviation caused by differences in knowledge base versions, and output the deviation judgment result.

[0095] In step S21, based on the semantic vector representation, preliminary answer data that matches the semantics is obtained from distributed nodes, the distance between semantic vectors is calculated to filter related data, and a preliminary answer set is output.

[0096] It should be noted that, based on the semantic vector representation generated in step S1, in the local knowledge base of each distributed node, answer data with semantic similarity to the vector is matched. For example, querying "broadband password reset" matches answers such as "reset via APP" and "contact customer service to reset". The similarity between the semantic vector corresponding to the answer data and the query semantic vector is calculated using cosine similarity. The associated data with similarity higher than the preset threshold is retained and organized in the format of "query semantic vector-node number-preliminary answer-cosine similarity value" to form a preliminary answer set.

[0097] For example, the semantic vector for the query "How to reset broadband password" is "[0.12,0.35,...,0.28]". The answer "Log in to the APP and click to reset password" (vector distance 0.08) is matched from node "ND01", and the answer "Contact customer service to verify identity and then reset" (vector distance 0.12) is matched from node "ND02". Both are below the threshold of 0.2, and are recorded as "[0.12,0.35,...,0.28],ND01, Log in to the APP and click to reset password, 0.08" and "[0.12,0.35,...,0.28],ND02, Contact customer service to verify identity and then reset, 0.12". Integrating these results in the initial set of answers.

[0098] In step S22, based on the preliminary answer set, the preliminary answer results are grouped according to semantic similarity to form similar answer clusters, and the answer cluster set is output.

[0099] It should be noted that all preliminary answers in the preliminary answer set are extracted, and the similarity between the semantic vectors of each answer is calculated. Answers with similarity higher than a preset threshold are grouped together, and each group forms a similar answer cluster. Each cluster is labeled with information such as "cluster number - core answer - number of answers contained". After being organized by cluster, an answer cluster set is formed.

[0100] For example, in the initial answer set, the semantic similarity between "Log in to the APP and click to reset password" and "Reset broadband password online through the APP" is 0.95, and the threshold is 0.8. This threshold is determined by multiple tests based on a large amount of historical knowledge question and answer data, ensuring semantic consistency and effective distinguishability of the answers. It is classified as "Cluster 01", with the core answer being "Log in to the APP and click to reset password", containing 2 answers. The similarity between "Contact customer service to verify identity and then reset" and "Call customer service hotline to reset password" is 0.92, classified as "Cluster 02", with the core answer being "Contact customer service to verify identity and then reset", containing 2 answers. The two clusters are integrated to form the answer cluster set.

[0101] In step S23, the semantic consistency of similar answer clusters in the answer cluster set is detected, it is determined whether there is a deviation due to differences in knowledge base versions, and the deviation judgment result is output.

[0102] It should be noted that the knowledge base version corresponding to the answers in each similar answer cluster is checked. For example, the answer in "Cluster 01" comes from knowledge base V2.0, and the answer in "Cluster 02" comes from V1.0. The standard answers for the same query in different versions of the knowledge base are compared to analyze whether the answers within or between clusters have semantic deviations due to version updates (e.g., V2.0 added "APP reset", while V1.0 only retained "customer service reset", resulting in differences in answers between clusters). If there are version-related deviations, the deviation type is marked, such as "version lag deviation" or "new feature missing deviation". Otherwise, it is marked as "no deviation", forming the deviation judgment result.

[0103] For example, the answer for "Cluster 01" comes from Knowledge Base V2.0 (including APP reset function), and the answer for "Cluster 02" comes from V1.0 (without APP reset). Compared with the standard answer (V2.0 prioritizes "APP reset"), it is determined that "Cluster 02" has a "version lag deviation" and is recorded as "Cluster 02: Version lag deviation exists (Knowledge Base V1.0 has not updated APP reset function)"; the answer for "Cluster 01" is consistent with the standard and is recorded as "Cluster 01: No deviation". The combined result is the deviation judgment result.

[0104] In step S3, if the number of similar answer clusters exceeds a preset threshold, cluster center information is shared and the answer inconsistency deviation pattern is determined to obtain the corrected knowledge representation, including:

[0105] S31, if the number of similar answer clusters in the answer cluster set exceeds a preset threshold, the cluster center information of each similar answer cluster is shared through inter-node broadcasting to determine the sharing range of cluster center information;

[0106] S32, based on the cluster center information sharing range and combined with the comparison of different versions of the knowledge base content, detect the deviation patterns of inconsistent answers at each node and output the deviation pattern set;

[0107] S33, extract deviation features from the deviation pattern set, correct the initial knowledge representation, and output the corrected knowledge representation;

[0108] S34, associate the corrected knowledge representation with the additional query data, adjust the iterative encoding parameters and process, update the iterative encoding process, and output the updated encoded data.

[0109] In step S31, if the number of similar answer clusters in the answer cluster set exceeds a preset threshold, the cluster center information of each similar answer cluster is shared through inter-node broadcasting to determine the scope of cluster center information sharing.

[0110] It should be noted that the total number of similar answer clusters in the statistical answer cluster set is compared with a preset threshold (e.g., 5). This threshold is determined after multiple rounds of experimental verification, taking into account the processing capacity of the comprehensive knowledge question-answering system and the impact of the number of answer clusters on system efficiency and answer consistency in actual business. If the threshold is exceeded, a broadcast mechanism between nodes is triggered. Each similar answer cluster uses the semantic vector of the core answer within the cluster as the cluster center information, which is broadcast to all distributed nodes. After receiving the cluster center information, each node compares it with the center vector of its local answer clusters and filters out clusters with a semantic relevance higher than the preset threshold, such as 0.8. This threshold is set based on the semantic matching effect between the cluster center information in historical data and the local cluster center vector, under the premise of ensuring effective information sharing and avoiding redundant interference, to determine the effective sharing range of cluster center information.

[0111] For example, the answer cluster set contains 6 similar answer clusters (5 exceeding the threshold). The center vector of "cluster 01" is "[0.13,0.34,...,0.27]" (corresponding to "APP password reset"), and the center vector of "cluster 02" is "[0.21,0.42,...,0.31]" (corresponding to "customer service password reset"). These are broadcast to all nodes. The local cluster center vector of node "ND03" has a correlation degree of 0.91 with "cluster 01" and 0.88 with "cluster 02", both exceeding the threshold. These two clusters are included in the sharing scope and recorded as "Sharing scope: cluster 01, cluster 02 (node ​​ND03's correlation degree meets the standard)".

[0112] In step S32, based on the cluster center information sharing range and by comparing the content of different versions of the knowledge base, deviation patterns of inconsistent answers at each node are detected, and a set of deviation patterns is output.

[0113] It should be noted that, within the shared scope, the knowledge base versions corresponding to each similar answer cluster are extracted, such as "cluster 01" corresponding to V2.0 and "cluster 03" corresponding to V1.5. The answer logic and core information of the same query in different versions are compared. Clusters with differences in answer expression, process or priority are selected, and the causes of the differences are analyzed and summarized into patterns such as "version lag deviation", "data missing deviation", and "priority misalignment deviation". Each deviation pattern is organized in the format of "deviation type-involved cluster number-knowledge base version difference-difference content" to form a deviation pattern set.

[0114] For example, within the shared scope, the answer for "Cluster 03" (corresponding to Knowledge Base V1.5) is "You need to bring your ID card to the business hall to reset your password," and the answer for "Cluster 01" (V2.0) is "Reset online via the APP without offline operation." A comparison reveals that "Cluster 03" lacks an online process due to outdated version, and is therefore judged as "Version Lag Deviation." The answer for "Cluster 04" (V2.0) omits "identity verification steps," and is judged as "Data Missing Deviation." The integrated records are "Deviation Pattern 1: Version Lag Deviation (Cluster 03, V1.5 vs V2.0, missing online reset process); Deviation Pattern 2: Data Missing Deviation (Cluster 04, V2.0, missing identity verification steps)," which is the deviation pattern set.

[0115] In step S33, deviation features are extracted from the deviation pattern set, the initial knowledge representation is corrected, and the corrected knowledge representation is output.

[0116] It should be noted that for each type of deviation in the deviation pattern set, key deviation features are extracted. For example, "version lag deviation" corresponds to "missing new version feature items", and "data missing deviation" corresponds to "missing core process steps". Based on the standard answer of the latest version of the knowledge base, the initial knowledge representation, such as the structured items and semantic relationships of the local knowledge base of each node, is adjusted to supplement missing information, correct erroneous logic, and align priority order. After correction, it is organized in the format of "knowledge item - semantic vector - associated version - correction description" to form the corrected knowledge representation.

[0117] For example, for the "version lag bias" of "cluster 03", the feature "missing V2.0 online reset process" is extracted, and the structured entry "APP password reset steps, login-verification-set new password" and the corresponding semantic vector are added to the initial knowledge representation; for the "data missing bias" of "cluster 04", the "identity verification (SMS / face recognition)" step is added to the knowledge entry, and finally recorded as "knowledge entry: broadband password reset (V2.0), semantic vector: [0.13,0.34,...,0.27], correction description: supplement online process and identity verification steps", which is the corrected knowledge representation.

[0118] In step S34, the corrected knowledge representation is associated with additional query data, the iterative encoding parameters and process are adjusted, the iterative encoding process is updated, and the updated encoded data is output.

[0119] It should be noted that additional query data collected synchronously between collection nodes, such as user supplementary questions like "Does resetting the password on the APP require facial recognition?" or "Offline resetting business hours," are used to extract semantic information related to the corrected knowledge representation. The semantic vectors of the additional query data are then matched with the vectors of the corrected knowledge representation, and the parameters of the iterative encoding (such as semantic weights and vector dimensions) are adjusted to ensure that the encoding process can capture the new information. Based on the adjusted parameters, the user query data is re-encoded and organized in the format of "query text - corrected knowledge association items - updated semantic vector - encoding parameters" to form updated encoded data.

[0120] For example, the semantic vector of the additional query data "Does APP password reset require facial recognition?" is "[0.18,0.31,...,0.25]", which is associated with the corrected knowledge representation of "Broadband Password Reset (V2.0)" and the encoding weight is adjusted (the weight of the facial recognition-related dimension is increased from 0.2 to 0.4); the query "How to reset broadband password" is re-encoded, generating the updated semantic vector "[0.13,0.36,...,0.29]", which is recorded as "Query: How to reset broadband password, associated knowledge: Broadband password reset (V2.0, including facial recognition), updated vector: [0.13,0.36,...,0.29]", which, after integration, becomes the updated encoded data.

[0121] In step S4, globally consistent features are extracted from the corrected knowledge representation and fused to determine a unified answer framework, including:

[0122] S41, extract cross-node common global consistent features from the corrected knowledge representation and output a global consistent feature set;

[0123] S42, Based on the globally consistent feature set, calculate the distance between the semantic vectors corresponding to each feature, fuse the globally consistent features by similarity weighting, and output the fused feature representation;

[0124] S43, Based on the fused feature representation, construct a structured unified answer framework and output the unified answer framework;

[0125] S44, for the unified answer framework, incorporate additional query data synchronously collected between nodes to supplement semantic information, enhance the accuracy of deviation pattern detection, and output the optimized unified answer framework.

[0126] In step S41, global consistent features that are common across nodes are extracted from the corrected knowledge representation, and a global consistent feature set is output.

[0127] It should be noted that the corrected knowledge representation, such as the knowledge entry "Broadband Password Reset V2.0", selects core semantic features that are recognized by all distributed nodes, including key operation steps, such as logging into the APP, identity verification, setting a new password, priority rules (APP reset takes precedence over customer service reset), and applicable scenarios (individual users and enterprise users). It is organized according to "feature type - feature description - node consensus" to form a globally consistent feature set.

[0128] For example, extract the "Operation steps: Log in to the APP → Verify identity (SMS / Face recognition) → Set a new password" (100% consensus among nodes) and "Priority: APP reset > Customer service reset > Offline reset" (98% consensus), and record them as "Globally consistent feature 1: Operation steps - Log in to the APP → Verify → Set a new password (100% consensus); Globally consistent feature 2: Priority - APP reset takes precedence (98% consensus)", and integrate them to form the set of globally consistent features.

[0129] In step S42, based on the globally consistent feature set, the distance between the semantic vectors corresponding to each feature is calculated, and the globally consistent features are fused by weighting similarity to output the fused feature representation.

[0130] It should be noted that a semantic vector is generated for each globally consistent feature, such as the vector for "APP reset steps" being "[0.13,0.34,...,0.27]". The vector distance between features is calculated, and related features with similarity exceeding a threshold (such as 0.85) are filtered out. The features are then linearly weighted and fused according to their consensus to generate a fused feature vector containing all core information, and output in the format of "fused feature description - fused vector - weight allocation".

[0131] For example, the vector similarity between "APP reset steps" (consensus degree 100%, weight 0.6) and "priority rules" (consensus degree 98%, weight 0.4) is 0.92. The weighted fused vector is "[0.13×0.6+0.21×0.4,0.34×0.6+0.42×0.4,...,0.27×0.6+0.31×0.4]", which is recorded as "fused feature: APP reset process and priority, fused vector: [0.162,0.372,...,0.286]", which is the fused feature representation.

[0132] In step S43, a structured unified answer framework is constructed based on the fused feature representation, and the unified answer framework is output.

[0133] It should be noted that, based on the semantic logic of the fusion features, a structured framework is built that includes "core solution - alternative solutions - operation steps - precautions"; the core solution corresponds to high-priority features (APP reset), the alternative solutions correspond to secondary-priority features, and the steps and precautions match globally consistent operation details; they are organized in a fixed format to ensure that each node can be directly reused.

[0134] For example, the unified answer framework is as follows: "1. Core solution (APP reset): Log in to the broadband APP → Enter 'Password Management' → Select 'Reset Password' → Complete identity verification (SMS / Face Recognition) → Set a new password; 2. Alternative solutions: Customer service reset (call hotline 400-XXX), offline reset (bring your ID card to the business hall); 3. Notes: The new password must contain letters and numbers, and you need to repeat the operation if the verification times out," which is the unified answer framework.

[0135] In step S44, the unified answer framework is supplemented with additional query data synchronously collected between nodes to supplement semantic information, enhance the accuracy of deviation pattern detection, and output the optimized unified answer framework.

[0136] It should be noted that when importing additional query data, such as "Does APP reset require facial recognition?" and "Offline business hours", supplementary information related to the framework is extracted. This information is then filled into the framework, such as adding "Supports facial recognition / SMS verification" after "Identity verification" and "Business hours 9:00-18:00" after "Offline reset". After supplementation, it is checked whether potential deviations are covered, thus forming an optimized framework.

[0137] For example, the optimized unified answer framework can be achieved by adding "Supports facial recognition or SMS verification code, enterprise users need to submit an additional photo of the official seal" after "identity verification" in the original framework, and "Business hall address can be obtained in the 'Branch Inquiry' of the APP after "offline reset" and "Open on weekends".

[0138] In step S5, if the matching degree between the unified answer framework and the semantic vector representation is lower than a preset matching degree threshold, then the globally consistent features are iteratively fused and node consistency is verified to obtain a refined answer, including:

[0139] S51, if the matching degree between the unified answer framework and the semantic vector representation is lower than the preset matching degree threshold, the globally consistent features are iteratively fused through the inter-node data exchange protocol to supplement semantic information, and the updated feature set is output.

[0140] S52, distribute the answer candidates corresponding to the updated feature set to each distributed node according to the preset distribution rules, verify the consistency between the data of each node and the updated feature set, and output the refined answer set;

[0141] S53, the refined answer set is iteratively encoded in accordance with the knowledge base version control requirements to eliminate the remaining semantic deviation patterns and output the optimized refined answer.

[0142] In step S51, if the matching degree between the unified answer framework and the semantic vector representation is lower than the preset matching degree threshold, the globally consistent features are iteratively fused through the inter-node data exchange protocol to supplement semantic information, and an updated feature set is output.

[0143] It should be noted that the matching degree between the semantic vector corresponding to the unified answer framework and the query semantic vector generated in step S1 is calculated. If it is lower than the threshold, such as 0.8, this threshold is determined by analyzing the impact of the matching degree between the framework semantic vector and the query semantic vector on the accuracy of the answer and user satisfaction in a large number of knowledge question-and-answer scenarios. It can effectively balance the efficiency and accuracy of answer generation and trigger data exchange between nodes. Each node shares globally consistent features that are not integrated into the framework, such as "Quick reset: SMS verification is more time-saving than face recognition". These features are gradually integrated according to the number of iterations until the matching degree reaches the standard. The integrated features are then organized to form an updated feature set.

[0144] For example, the original framework and the query vector have a matching degree of 0.75 (below the threshold of 0.8). The node "ND04" shares the feature "SMS verification saves more time". After fusion, the matching degree is improved to 0.83. The updated feature set record is "Updated feature 1: Fast verification method - SMS verification takes priority (improves matching degree to 0.83); Updated feature 2: Operation efficiency - SMS verification time < 1 minute".

[0145] In step S52, the answer candidates corresponding to the updated feature set are distributed to each distributed node according to a preset distribution rule, the consistency between the data of each node and the updated feature set is verified, and a refined answer set is output.

[0146] It should be noted that candidate answers are generated based on updated features, such as "Quickly reset broadband password: Log in to the APP → Select 'Password Reset' → Prioritize SMS verification → Complete within 1 minute", and distributed to each node according to the "nearest node priority" rule; each node compares the semantic consistency between its local answer and the candidate (vector similarity exceeding 0.9 is considered acceptable), filters the acceptable answers and marks the node confirmation results, and integrates the confirmed answers to form a refined answer set.

[0147] For example, the answer candidates are distributed to “ND01-ND05”, and all of them report a similarity of 0.92-0.95 (meeting the standard). The refined answer set is recorded as “Refined answer 1: quick reset - APP + SMS verification (ND01-ND05 all confirmed); refined answer 2: alternative - customer service reset, takes 5-10 minutes, ND01-ND05 all confirmed”.

[0148] In step S53, the refined answer set is iteratively encoded in accordance with the knowledge base version control requirements to eliminate remaining semantic deviation patterns and output the optimized refined answer.

[0149] It should be noted that the knowledge base version corresponding to the refined answer should be checked (it must be the latest V2.0) to check for any minor deviations, such as whether the expression "SMS verification" is consistent with "SMS verification code verification"; according to the iterative coding rules, the deviation should be checked after each encoding, and the semantic vector and expression of the answer should be adjusted until there are no deviations; finally, the output should be organized according to "Refined Answer - Corresponding Version - Deviation Elimination Instructions".

[0150] For example, the original refined answer "SMS verification" had a wording error, with some nodes recorded as "SMS code verification". After iterative encoding, it was uniformly expressed as "SMS verification code verification". The vector was adjusted synchronously and recorded as "Optimized refined answer 1: Quick reset - Log in to the APP → Select 'Password reset' → Prioritize SMS verification code verification", which is the optimized refined answer.

[0151] In step S6, the refined answer is distributed to each distributed node and the processing node receives and confirms the receipt, resulting in a preliminary globally consistent output, including:

[0152] S61, Distribute the optimized and refined answer to all nodes in the distributed system, perform hash verification on the received results of each node, and generate a set of node reception confirmation identifiers;

[0153] S62, based on the set of confirmation identifiers received by the nodes, after confirming that all nodes have received the complete and refined answer, record the answer distribution process and results to obtain a globally consistent output result;

[0154] S63, extract the answer version tag used to identify the answer version from the globally consistent output result, and output the answer version tag.

[0155] In step S61, the optimized and refined answer is distributed to all nodes in the distributed system, and the received results of each node are hash-verified to generate a set of node reception confirmation identifiers.

[0156] It should be noted that, taking "Optimized Refined Answer 1: Quick Reset - Log in to the APP → Select 'Password Reset' → Prioritize SMS Verification Code Verification" as an example, it is distributed to all nodes "ND01-ND05" through the broadcast protocol of the distributed system; a unique hash value for the refined answer is generated for each node, such as "Hash: a3f2d1e5". After receiving it, the node calculates its local hash value and returns the comparison result. If they match, it marks "Receipt Successful"; otherwise, it marks "Receipt Failed". The node reception confirmation identifier set is formed by "Node Number - Refined Answer Hash Value - Receipt Status".

[0157] For example, the set records are as follows: "ND01: Hash = a3f2d1e5, received successfully; ND02: Hash = a3f2d1e5, received successfully; ND03: Hash = a3f2d1e5, received successfully; ND04: Hash = a3f2d1e5, received successfully; ND05: Hash = a3f2d1e5, received successfully."

[0158] In step S62, based on the set of confirmation identifiers received by the nodes, after confirming that all nodes have received the complete and refined answer, the answer distribution process and result are recorded to obtain a globally consistent output result.

[0159] It should be noted that the set of verification node confirmation identifiers is checked. If all nodes mark "received successfully", it is confirmed that the refined answer has been fully synchronized. The distribution time, such as "2025-09-01 10:30:00", the content of the distributed refined answer, the number of participating nodes (e.g., 5), and the result of no receiving anomalies are recorded and organized in the format of "distribution time-refined answer-number of nodes-receiving status" to form a globally consistent output result.

[0160] For example, the result is recorded as "Distribution time: 2025-09-01 10:30:00; Refined answer: Quick reset - Log in to the APP → Select 'Password reset' → Number of priority SMS verification code verification nodes: 5 (ND01-ND05); Receiving status: All successful, no abnormalities".

[0161] In step S63, an answer version tag for identifying the answer version is extracted from the globally consistent output result, and the answer version tag is output.

[0162] It should be noted that, from the refined answer content of the globally consistent output results, the pre-embedded version identification information is extracted, including the version number (e.g., V2.0), the version update time (e.g., "2025-08-30"), and the core update points corresponding to the version (e.g., "adding SMS verification code verification method"). This information is then organized in the format of "version number-update time-core update points" to form the answer version tag.

[0163] For example, the extracted result is "Answer version tag: Version number V2.0, Update time 2025-08-30, Core update point: Added SMS verification code verification method", which is the output answer version tag.

[0164] In step S7, the answer version tag is extracted from the preliminary globally consistent output result and the distributed node is updated to obtain the optimized system state, including:

[0165] S71, Based on the answer version tag, the output results of each distributed node are aggregated and merged to generate a globally consistent answer record;

[0166] S72, based on the globally consistent answer record, compare the answer version marker with the preset version threshold, detect semantic consistency deviation, and output the semantic consistency deviation detection result;

[0167] S73, Based on the semantic consistency deviation detection result, adjust the state parameters of the distributed nodes to update the nodes, determine whether the semantic deviation has been eliminated, and output the optimized system state;

[0168] S74, in the optimized system state, integrate the additional query data synchronously collected between nodes to provide data support for the potential iterative coding process and output iterative coding preparation data.

[0169] In step S71, the output results of each distributed node are aggregated and merged according to the answer version tag to generate a globally consistent answer record.

[0170] It should be noted that, based on the answer version marker, version number V2.0, and the core update point "adding SMS verification code verification", the answer results output by each node (ND01-ND05) based on this version are collected; the content in the node output that is consistent with the core information of the version marker is filtered out, and aggregated according to "answer topic - core steps - version matching degree" to ensure that the record covers the consensus content of all nodes and forms a globally consistent answer record.

[0171] For example, the aggregated record is as follows: "Answer Topic: Broadband Password Reset; Core Steps: 1. Log in to the APP → 2. Select 'Password Reset' → 3. Prioritize SMS verification code verification → 4. Alternative Solutions: Customer Service Reset, Offline Reset; Version Matching: 100% (ND01-ND05 all comply with V2.0)".

[0172] In step S72, based on the globally consistent answer record, the answer version marker is compared with a preset version threshold to detect semantic consistency deviation and output the semantic consistency deviation detection result.

[0173] It should be noted that the preset version threshold is "must match the latest knowledge base version V2.0, semantic deviation rate ≤ 5%"; compare the core information of the answer record with the version mark, such as whether "SMS verification code verification" is missing or whether the step order is consistent, calculate the semantic deviation rate, if the node output has no deviation, the deviation rate is 0%; if the deviation rate is ≤ the threshold, it is determined that "there is no semantic consistency deviation", otherwise mark the deviation point to form the detection result.

[0174] For example, the answer record fully includes the "SMS verification code verification" steps with version tags, the step order is consistent with the version requirements, the deviation rate is 0% ≤ 5%, and the detection result is recorded as "Semantic consistency deviation detection result: No deviation (meets the preset version threshold V2.0, deviation rate 0%)".

[0175] In step S73, based on the semantic consistency deviation detection result, the state parameters of the distributed nodes are adjusted to update the nodes, it is determined whether the semantic deviation has been eliminated, and the optimized system state is output.

[0176] It should be noted that if the detection result is "no deviation", adjust the node status parameters, such as setting "version adaptation status" to "V2.0 synchronized" and "deviation warning threshold" to 5%; if there is a deviation, first correct the node output, such as adding the missing "SMS verification" step, and then update the parameters; after updating, check whether all nodes are without deviation. If there is no deviation, output "Optimized system status: nodes are fully synchronized to V2.0, semantic deviation has been eliminated, and queries can be responded to normally".

[0177] For example, all ND01-ND05 have completed parameter updates without deviation, and the optimized system status is recorded as "Nodes ND01-ND05: Version adapted to V2.0, semantic deviation rate 0%, response latency ≤ 1 second, system status optimization completed".

[0178] In step S74, in the optimized system state, additional query data synchronously collected between nodes is integrated to provide data support for the potential iterative coding process, and iterative coding preparation data is output.

[0179] It should be noted that additional query data collected synchronously by the collection nodes, such as "Does the APP password reset support remote operation?" and "How long is the SMS verification code valid?", are filtered to be related to the topic of "broadband password reset". The data is organized according to "additional query content - associated answer record entries - data source nodes" to ensure that the data can supplement the details of existing answer records, provide new semantic information for subsequent iterative coding, and form preparation data for iterative coding.

[0180] For example, the data record is prepared as follows: "Additional query content 1: Does APP reset support remote operation; Related item: Core step 1 (login to APP); Data source: ND02, ND04; Additional query content 2: SMS verification code validity period; Related item: Core step 3 (SMS verification); Data source: ND01, ND05".

[0181] In step S8, if the optimized system status display deviation pattern still exists, additional query data is integrated and iterated from the semantic encoding step to update the unified answer framework, obtaining the final globally consistent output result, including:

[0182] S81, if the optimized system status display still shows a semantic deviation pattern, additional query data related to the deviation pattern is collected synchronously between nodes to determine the deviation pattern dataset;

[0183] S82, based on the deviation pattern dataset, extract related data from the knowledge base, update the entries in the knowledge base corresponding to the deviation patterns, and output the updated knowledge base dataset;

[0184] S83, based on the updated knowledge base dataset, iteratively encode and update the unified answer framework, adjust the semantic vector representation in the framework by combining additional query data, and output the updated unified answer framework;

[0185] S84. Based on the updated unified answer framework, optimize the semantic processing logic of each distributed node to ensure that the semantics of the node output are consistent, and obtain the final optimized system.

[0186] In step S81, if the optimized system status display still shows a semantic deviation pattern, additional query data related to the deviation pattern is collected synchronously between nodes to determine the deviation pattern dataset.

[0187] It should be noted that if the optimization system detects a deviation pattern where "some nodes have inconsistent statements about 'resetting passwords in different locations via the APP'", such as ND03 mentioning "supporting different locations" while ND05 is not explicit, the nodes will be triggered to synchronously collect additional query data containing the keyword "reset in different locations", such as "Can I reset my broadband password across provinces via the APP?" and "Does resetting the password via the APP from a different location require additional verification?" The data will be organized according to "deviation type - additional query content - data source node" to form a deviation pattern dataset.

[0188] For example, the dataset is recorded as "Deviation type: Inconsistent description of resetting in a different location; Additional query 1: Can broadband password be reset via APP across provinces (source ND02); Additional query 2: Does resetting via APP in a different location require additional verification (source ND05)".

[0189] In step S82, based on the deviation pattern dataset, related data is extracted from the knowledge base, the entries in the knowledge base corresponding to the deviation patterns are updated, and the updated knowledge base dataset is output.

[0190] It should be noted that the original entries related to "remote operation permissions" were extracted from the latest knowledge base (V2.0), and combined with the query requirements in the deviation pattern dataset, the rule "supports nationwide remote APP reset, remote login requires additional device verification (such as SMS + device SN code)" was added; the updated knowledge base dataset was formed by organizing it according to "original entry content - updated content - associated deviation type".

[0191] For example, the dataset record is as follows: "Original entry: APP password reset requires login verification; Updated content: APP reset supports nationwide remote operation, remote login requires additional verification of device SN code; Correlation deviation type: Inconsistent description of remote reset".

[0192] In step S83, the unified answer framework is iteratively encoded and updated based on the updated knowledge base dataset. The semantic vector representation in the framework is adjusted in combination with additional query data, and the updated unified answer framework is output.

[0193] It should be noted that the updated knowledge base content is integrated into the original unified answer framework, and "Login from a different location requires additional device SN code verification" is added after "Core Step 3 (SMS Verification)"; the semantic vector of the framework is recalculated, such as increasing the weight of the related dimensions of "different location" and "device SN code" from 0.1 to 0.3 to ensure that the vector can accurately reflect the new information; and the output is in the format of "updated framework content - adjusted semantic vector - update instructions".

[0194] For example, the updated framework is "Core solution: Log in to APP → 'Password reset' → SMS verification → Set new password", the vector is adjusted to "[0.16,0.37,...,0.32]", and the record is "Update description: Supplement the verification requirements for resetting devices in different locations, and adjust the weight of the associated semantic dimension".

[0195] In step S84, based on the updated unified answer framework, the semantic processing logic of each distributed node is optimized to ensure that the semantics of the node output are consistent, thus obtaining the final optimized system.

[0196] It should be noted that the updated unified answer framework and semantic vector are pushed to all nodes, and the semantic processing logic of the nodes is adjusted, such as adding a "remote scene recognition module" to automatically match the device verification process; the output results of each node are verified, such as querying "cross-province APP password reset", all nodes must include "device SN code verification", if they are all consistent, the system optimization is completed, and the final optimized system status is output.

[0197] For example, the final optimized system record is as follows: "Nodes ND01-ND05: Semantic processing logic is updated synchronously, all outputs conform to the 'remote reset + device verification' rule, semantic deviation rate is 0%, response latency is ≤0.8 seconds, system optimization is complete".

[0198] Reference Figure 2 The second embodiment of the present invention provides a knowledge question-answering data processing device based on artificial intelligence, comprising:

[0199] The data acquisition and encoding module is used to collect user query data and local knowledge base information through distributed nodes, and to perform semantic encoding on the user query data to obtain a semantic vector representation.

[0200] The cluster generation module is used to obtain preliminary answer results based on the semantic vector representation, group the preliminary answer results to determine similar answer clusters, and detect whether the similar answer clusters involve knowledge base version deviations.

[0201] The knowledge correction module is used to share cluster center information and determine the answer inconsistency deviation pattern if the number of similar answer clusters exceeds a preset threshold, so as to obtain the corrected knowledge representation.

[0202] The framework construction module is used to extract and fuse globally consistent features from the corrected knowledge representation to determine a unified answer framework.

[0203] The refined answer module is used to iteratively fuse globally consistent features and verify node consistency to obtain a refined answer if the matching degree between the unified answer framework and the semantic vector representation is lower than a preset matching degree threshold.

[0204] The global output module is used to distribute the refined answer to each distributed node and process the node's confirmation of receipt to obtain a preliminary globally consistent output result;

[0205] The system update module is used to extract the answer version marker from the preliminary globally consistent output result and update the distributed nodes to obtain an optimized system state;

[0206] An enhancement and optimization module is used to integrate additional query data and iterate from the semantic encoding steps if the deviation mode of the optimized system status still exists, update the unified answer framework, and obtain the final globally consistent output result.

[0207] It should be noted that the knowledge question-and-answer data processing device based on artificial intelligence provided in this embodiment of the invention is used to execute all the process steps of the knowledge question-and-answer data processing method based on artificial intelligence in the first embodiment described above, such as all operations such as "data acquisition and encoding", "answer cluster generation", and "knowledge correction" in steps S1 to S8; the working principle of each module in the device corresponds one-to-one with the implementation logic of the corresponding method steps, and the beneficial effects of each module are also completely matched with the beneficial effects of the corresponding steps in the method, so they will not be described in detail here.

[0208] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an artificial intelligence-based knowledge question-and-answer data processing program. When the processor executes the computer program, it implements the steps in various embodiments of the artificial intelligence-based knowledge question-and-answer data processing method described in the first embodiment above, for example... Figure 1The steps S11, S22, etc., are shown. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiment. For example, the data acquisition and encoding module performs the function of "acquiring data and semantically encoding to obtain a semantic vector representation," and the enhancement and optimization module performs the function of "iteratively encoding and updating the unified answer framework to obtain the final optimized system." Exemplarily, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device.

[0209] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0210] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0211] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0212] Wherein, if the modules / units integrated in the electronic device 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, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0213] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0214] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A knowledge question-answering data processing method based on artificial intelligence, characterized in that, include: By collecting user query data and local knowledge base information through distributed nodes, semantic encoding is performed on the user query data to obtain a semantic vector representation; Based on the semantic vector representation, preliminary answer results are obtained from each distributed node and grouped to determine similar answer clusters. This includes: obtaining preliminary answer data that matches the semantics from the distributed nodes based on the semantic vector representation, calculating the distance between semantic vectors to filter related data, and outputting a preliminary answer set; grouping the preliminary answer results according to semantic similarity based on the preliminary answer set to form similar answer clusters, and outputting an answer cluster set; detecting the semantic consistency of similar answer clusters in the answer cluster set, determining whether there is a deviation due to differences in knowledge base versions, and outputting the deviation judgment result. If the number of similar answer clusters exceeds a preset threshold, the cluster center information is shared and the answer inconsistency deviation pattern is determined to obtain a corrected knowledge representation. This includes: if the number of similar answer clusters in the answer cluster set exceeds a preset threshold, the cluster center information of each similar answer cluster is shared through inter-node broadcasting to determine the cluster center information sharing range; based on the cluster center information sharing range, combined with the comparison of different versions of the knowledge base content, the deviation pattern of answer inconsistency at each node is detected, and a deviation pattern set is output; deviation features are extracted from the deviation pattern set, the initial knowledge representation is corrected, and the corrected knowledge representation is output; the corrected knowledge representation is associated with additional query data, the iterative encoding parameters and process are adjusted, the iterative encoding process is updated, and the updated encoded data is output. Globally consistent features are extracted and fused from the corrected knowledge representation to determine a unified answer framework; If the matching degree between the unified answer framework and the semantic vector representation is lower than a preset matching degree threshold, then the globally consistent features are iteratively fused and node consistency is verified to obtain a refined answer. This includes: if the matching degree between the unified answer framework and the semantic vector representation is lower than a preset matching degree threshold, globally consistent features are iteratively fused to supplement semantic information through an inter-node data exchange protocol, and an updated feature set is output; the answer candidates corresponding to the updated feature set are distributed to each distributed node according to a preset distribution rule, the consistency between the data of each node and the updated feature set is verified, semantically consistent answers are selected, and a refined answer set is output; the refined answer set is iteratively encoded in accordance with knowledge base version control requirements to eliminate remaining semantic deviation patterns, and the optimized refined answer is output. The refined answer is distributed to each distributed node and the node receives and confirms the receipt, resulting in a preliminary globally consistent output result. Extract the answer version marker from the preliminary globally consistent output and update the distributed nodes to obtain the optimized system state; If the deviation pattern in the optimized system status display still exists, then additional query data is integrated and iterated from the semantic encoding step to update the unified answer framework and obtain the final globally consistent output result.

2. The method according to claim 1, characterized in that, The process of collecting user query data and local knowledge base information through distributed nodes, and semantically encoding the user query data to obtain a semantic vector representation includes: The system collects user query data and local knowledge base information through distributed nodes, performs consistent allocation, and outputs an initial data set. Based on the initial data set, semantic encoding is performed on the user query data to generate a semantic vector representation, and semantic encoded data containing the semantic vector representation is output; Based on the semantically encoded data containing semantic vector representation, data between nodes is synchronized, deviations caused by rare queries are identified and processed in deviation detection, and usable semantically encoded data containing semantic vector representation is output.

3. The method according to claim 1, characterized in that, The step of extracting and fusing globally consistent features from the corrected knowledge representation to determine a unified answer framework includes: From the corrected knowledge representation, extract globally consistent features that are common across nodes and output a set of globally consistent features; Based on the globally consistent feature set, calculate the distance between the semantic vectors corresponding to each feature, fuse the globally consistent features by similarity weighting, and output the fused feature representation; Based on the fused feature representation, a structured unified answer framework is constructed and output; The unified answer framework is supplemented with additional query data collected synchronously between nodes to supplement semantic information, enhance the accuracy of deviation pattern detection, and output an optimized unified answer framework.

4. The method according to claim 1, characterized in that, The process of distributing the refined answer to each distributed node and processing the node's confirmation to obtain a preliminary globally consistent output result includes: The refined answer is distributed to all nodes in the distributed system, and the received results of each node are hash-verified to generate a set of node reception confirmation identifiers. Based on the set of confirmation identifiers received by the nodes, after confirming that all nodes have received the complete and refined answer, the answer distribution process and results are recorded to obtain a globally consistent output result; Extract the answer version tag used to identify the answer version from the globally consistent output results, and output the answer version tag.

5. The method according to claim 1, characterized in that, The step of extracting the answer version flag from the preliminary globally consistent output result and updating the distributed nodes to obtain the optimized system state includes: Based on the answer version tag, the output results of each distributed node are aggregated and merged to generate a globally consistent answer record; Based on the globally consistent answer record, the answer version marker is compared with a preset version threshold to detect semantic consistency deviation and output the semantic consistency deviation detection result. Based on the semantic consistency deviation detection results, adjust the state parameters of the distributed nodes to update the nodes, determine whether the semantic deviation has been eliminated, and output the optimized system state. In the optimized system state, additional query data collected synchronously between nodes is integrated to provide data support for potential iterative coding processes and output iterative coding preparation data.

6. The method according to claim 1, characterized in that, If the optimized system status display deviation pattern still exists, then additional query data is integrated and iterated from the semantic encoding step to update the unified answer framework, obtaining the final globally consistent output result, including: If the optimized system status display still shows a semantic deviation pattern, additional query data related to the deviation pattern is collected synchronously between nodes to determine the deviation pattern dataset; Based on the deviation pattern dataset, extract related data from the knowledge base, update the entries in the knowledge base corresponding to the deviation patterns, and output the updated knowledge base dataset. Based on the updated knowledge base dataset, the unified answer framework is iteratively encoded and updated, and the semantic vector representation in the framework is adjusted in combination with additional query data to output the updated unified answer framework. Based on the updated unified answer framework, the semantic processing logic of each distributed node is optimized to obtain a final optimized system with semantic consistency and deviation elimination.

7. A knowledge question-answering data processing system based on artificial intelligence, characterized in that, For implementing the method as described in any one of claims 1-6, comprising: The data acquisition and encoding module collects user query data and local knowledge base information through distributed nodes, performs semantic encoding on the user query data, and obtains a semantic vector representation. Cluster generation module: Based on the semantic vector representation, it obtains preliminary answer results from each distributed node, groups them, and determines similar answer clusters; Knowledge correction module: If the number of similar answer clusters exceeds a preset threshold, the cluster center information is shared and the answer inconsistency deviation pattern is determined to obtain the corrected knowledge representation; Framework construction module: Extracts and fuses globally consistent features from the corrected knowledge representation to determine a unified answer framework; Refined Answer Module: If the matching degree between the unified answer framework and the semantic vector representation is lower than the preset matching degree threshold, then the globally consistent features are iteratively fused and the node consistency is verified to obtain the refined answer; Global output module: Distributes the refined answer to each distributed node and processes the node's confirmation of receipt to obtain a preliminary globally consistent output result; System update module: Extracts the answer version tag from the preliminary globally consistent output result and updates the distributed nodes to obtain the optimized system state; Enhanced optimization module: If the deviation mode of the optimized system status display still exists, the module integrates additional query data and iterates from the semantic encoding steps to update the unified answer framework and obtain the final globally consistent output result.

Images