A cross-data word item standardization and mastery transfer method, system, storage medium and product program

Abstract

A method, system, storage medium, and product program for cross-database term standardization and mastery transfer are disclosed. In this method, features from multiple datasets are extracted for disambiguation to generate standard nodes and construct a multi-dimensional relationship graph. Next, historical state feature vectors are generated based on user historical interactions. Then, the migration state vector and migration confidence of the target node are calculated by combining graph mapping, matching weights, and time decay coefficients. Finally, dynamic decision-making is performed based on dual thresholds of confidence and mastery: high confidence and high mastery result in blocked push notifications; low confidence and low mastery result in push notifications; and low confidence results in the generation of a verification task to update the vector. This application aims to improve the accuracy of state data feature matching and transfer in multi-database scenarios, reduce the system's overhead in pushing and processing redundant data, and thus improve the system's data processing efficiency.

Description

Technical Field

[0001] This application belongs to the field of natural language processing and knowledge graph technology, and in particular relates to a method, system, storage medium and product program for cross-data term standardization and mastery transfer. Background Technology

[0002] In the field of digital language learning, early vocabulary learning products typically presented words and definitions to users in the form of static electronic word lists or basic flip-cards. However, this basic presentation method lacks dynamic tracking of the user's learning status and makes it difficult to conduct scientific review planning based on the user's memory patterns, which easily leads to an increased forgetting rate when users are faced with a massive amount of vocabulary.

[0003] The relevant technology typically employs a vocabulary learning system based on spaced repetition and knowledge tracking. It consists of a vocabulary management module, a flashcard display module, and a mastery calculation module. It treats vocabulary entries from specific textbooks or exam lists as independent learning objects, assigning each entry a unique entry ID. During the learning process, the system records the user's accuracy rate in answering questions for each entry, the number of reviews, and the review intervals. The mastery calculation module then calculates the user's mastery of each entry using either the word string or the vocabulary entry ID. Based on the memory decay curve, the system automatically schedules subsequent reviews for the user, thereby mitigating the forgetting problem.

[0004] When students complete their basic school textbook studies and then switch to preparing for advanced exams like IELTS and different vocabulary books, the meanings tested for the same word often differ between different levels of materials. Furthermore, when handling data synchronization across datasets, the coarse-grained data identification in related technologies reduces the accuracy of data feature matching during user state data migration. This coarse-grained data processing approach can easily lead to the system pushing redundant data to user terminals or erroneous data overwriting, thereby increasing the system's unnecessary computational overhead and data transmission burden. Summary of the Invention

[0005] This application provides a method, system, storage medium, and product program for cross-data term standardization and mastery transfer, which improves the accuracy of state data feature matching and transfer in multi-data source scenarios, reduces the system's overhead in pushing and processing redundant data, and thus improves the system's data processing efficiency.

[0006] In the first aspect, this application provides a cross-data term standardization and mastery transfer method to extract text features and context features of each original data entry in the first dataset and the second dataset;

[0007] Disambiguation is performed on the original data entries based on text features and context features to generate standard data nodes with unique semantic identifiers.

[0008] A multidimensional relationship graph is constructed based on standard data nodes and their corresponding semantic identifiers. The multidimensional relationship graph contains the association relationships between standard data nodes.

[0009] Acquire historical interaction data of the user terminal for the first standard data node in the first dataset. The historical interaction data includes interaction timestamps and multi-dimensional interaction results.

[0010] Based on the multi-dimensional interaction results, a historical state feature vector corresponding to the first standard data node is generated.

[0011] In response to the user terminal's trigger command for the second dataset, the mapping relationship and matching weight between the second standard data node and the first standard data node in the second dataset are determined based on the association relationship in the multidimensional relationship graph.

[0012] Determine the time decay coefficient based on the interactive timestamp;

[0013] Based on the time decay coefficient, mapping relationship, matching weight and historical state feature vector, calculate the migration state feature vector and migration confidence corresponding to the second standard data node. The migration state feature vector includes the mastery assessment value.

[0014] If the migration confidence level is greater than a preset threshold and the mastery assessment value is greater than a preset mastery threshold, then the push operation from the second standard data node to the user terminal will be blocked.

[0015] If the migration confidence level is less than or equal to the preset threshold and the mastery assessment value is less than or equal to the preset mastery threshold, then the push operation of the second standard data node to the user terminal is executed.

[0016] If the migration confidence is less than or equal to a preset threshold, a feature verification task for the second standard data node is generated and sent to the user terminal to update the migration state feature vector based on the feedback data from the user terminal regarding the feature verification task.

[0017] By employing the aforementioned technical solutions, textual and contextual features of the original data entries are extracted for disambiguation processing, and a multidimensional relationship graph is constructed. This unifies the data representation across different data sets and reduces semantic ambiguity during cross-dataset processing. Historical state feature vectors are generated based on multidimensional historical interaction results. These vectors are then combined with time decay coefficients, graph mapping relationships, and matching weights to calculate migration state feature vectors and migration confidence. This process fully considers the impact of time on user states and the tightness of topological connections between nodes, improving the accuracy and dynamic adaptability of cross-dataset state migration prediction. A dual-threshold joint decision is made based on the calculated migration confidence and mastery assessment values. Push notifications are blocked for nodes with high confidence and high mastery, while normal push notifications are performed for nodes with low confidence and low mastery. A verification task is generated for low-confidence nodes to update vectors in a closed loop. This differentiated processing mechanism reduces redundant data pushes and invalid interaction verifications, lowers system computational overhead and communication resource consumption, and improves the overall reliability of cross-dataset state migration.

[0018] In conjunction with some implementation methods of the first aspect, in some implementation methods, based on the multi-dimensional interaction results, a historical state feature vector corresponding to the first standard data node is generated, specifically including:

[0019] The multi-dimensional interaction results are decoupled according to the data type of the interaction, and the interaction feature sequences of multiple dimensions are extracted.

[0020] For each dimension of the interaction feature sequence, the feature fluctuation variance within a preset time window is calculated, and the reliability weight corresponding to each dimension is determined based on the feature fluctuation variance. The feature fluctuation variance and the reliability weight are negatively correlated.

[0021] Outlier suppression processing is performed on the corresponding interactive feature sequences using reliability weights to obtain denoised feature representations for each dimension.

[0022] Based on the reliability weight, the denoised feature representations corresponding to multiple dimensions are weighted and aggregated to generate the historical state feature vector corresponding to the first standard data node.

[0023] By employing the above technical solution, fine-grained analysis of interaction data is achieved by decoupling multi-dimensional interaction results into multi-dimensional interaction feature sequences according to the interaction data type. For each dimension's sequence, the feature fluctuation variance within a preset time window is calculated, and a negatively correlated reliability weight is determined accordingly, objectively reflecting the stability of interaction data across different dimensions. This reliability weight is used to perform outlier suppression processing on the corresponding interaction feature sequences, filtering out abnormal interaction data caused by random factors or system noise, resulting in a purer, denoised feature representation. Weighted aggregation of the denoised feature representation based on the reliability weight ensures that the final generated historical state feature vector relies more on stable and reliable interaction dimensions, reducing the interference of noisy data on state evaluation, thereby improving the accuracy and robustness of historical state feature vector generation.

[0024] In conjunction with some implementation methods of the first aspect, in some implementation methods, based on the multi-dimensional interaction results, a historical state feature vector corresponding to the first standard data node is generated, specifically including:

[0025] Based on the chronological order of the interactions, the multidimensional interaction results are reorganized into a temporal interaction trajectory containing multiple interaction nodes;

[0026] Extract the differences in interaction types and time intervals between adjacent interaction nodes in the temporal interaction trajectory;

[0027] Calculate the state transition gain value between adjacent interactive nodes based on the differences in interaction type and time interval;

[0028] Initialize the baseline state vector for the first standard data node;

[0029] Along the time-series interaction trajectory, the baseline state vector is iteratively updated level by level using the state transition gain value to obtain the final update vector;

[0030] The final update vector is used as the historical state feature vector corresponding to the first standard data node.

[0031] By employing the above technical solution, and reorganizing multidimensional interaction results into a temporal interaction trajectory according to their chronological order, the temporal evolution pattern of interaction behavior can be fully preserved. The differences in interaction types and time intervals between adjacent interaction nodes are extracted, and state transition gain values ​​are calculated accordingly. This quantifies the magnitude of dynamic state changes caused by adjacent interaction actions and the specific impact of time span on state transitions. After initializing the baseline state vector for standard data nodes, iterative updates are performed level by level along the temporal interaction trajectory using the state transition gain values. This realistically simulates the continuous evolution of states over time and the accumulation of interaction behavior, reducing the evaluation bias caused by treating historical interaction data as isolated static events. This improves the accuracy and timeliness of the historical state feature vector in depicting dynamic changes.

[0032] In conjunction with some embodiments of the first aspect, in some embodiments, the method further includes:

[0033] The intercepted second standard data node is marked as a verification-free trusted node, and the migration state feature vector corresponding to the verification-free trusted node is temporarily stored as the expected state.

[0034] In response to the user terminal's trigger command for the third standard data node in the second dataset, if the third standard data node triggers a feature verification task and the feedback data representation verification fails, then based on the multidimensional relationship graph, the preceding dependency links of the third standard data node in the second dataset are determined.

[0035] If there is at least one unverified trusted node in the preceding dependency link, it is determined that the unverified trusted node has implicitly falsified.

[0036] The credit backtracking penalty value is calculated based on the degree of verification failure of the third standard data node and the association depth of the preceding dependent links.

[0037] By using the credit backtracking penalty value, the expected state of the verification-free trust node is downgraded and updated, and the matching weight between the first standard data node and the verification-free trust node is reduced simultaneously.

[0038] By adopting the above technical solution, intercepted nodes are marked as verification-free trusted nodes and their expected states are temporarily stored, enabling the tracking and management of verification-free nodes. When the verification of the third standard data node fails, its preceding dependency links are determined based on the multi-dimensional relationship graph, which can accurately locate the deep-seated correlation factors that led to the verification failure. If there is a verification-free trusted node in the link, it is determined that implicit falsification has occurred, and a credit backtracking penalty value is calculated based on the degree of verification failure and the depth of correlation, ensuring that the penalty strength matches the actual error impact and structural distance. This penalty value is used to downgrade and update the expected state of the verification-free trusted node, and its matching weight with the first standard data node is reduced simultaneously, constructing a dynamic error correction mechanism based on downstream feedback. This mechanism can promptly correct erroneous state transitions caused by previous misjudgments of confidence, reducing the risk of cascading propagation of erroneous state assessments in the graph links, thereby improving the overall accuracy and adaptive error correction capability of the cross-data state transition system.

[0039] In conjunction with some implementation methods of the first aspect, in some implementation methods, a credit backtracking penalty value is calculated based on the degree of verification failure of the third standard data node and the association depth of the preceding dependent link, specifically including:

[0040] Based on the multidimensional relationship graph, at least one bypass dependency link of the third standard data node is extracted in addition to the preceding dependency link, and the current state features of the bypass preceding node on the bypass dependency link are obtained.

[0041] Aggregate the current state characteristics of the bypass preceding node and calculate the bypass health of the third standard data node;

[0042] Based on bypass health, the responsibility attribution weight for verification failure caused by the verification-free trusted node is determined, and bypass health is positively correlated with the responsibility attribution weight.

[0043] Calculate the initial penalty base based on the degree of failure of the third standard data node verification and the weight of responsibility attribution;

[0044] Based on the association depth of the preceding dependency links, the initial penalty base is decayed to obtain the credit backtracking penalty value.

[0045] By employing the aforementioned technical solution, the bypass dependency links of the third standard data node are extracted, and the current state characteristics of the bypass predecessor node are obtained. The bypass health score is then calculated to assess the comprehensive impact of other related nodes on the verification failure state. Determining the responsibility weight for verification failure caused by the exempt trust node based on the bypass health score avoids solely attributing the verification failure to the predecessor exempt trust node, making the responsibility allocation more aligned with the actual node association impact. The initial penalty base is calculated by combining the degree of verification failure with this responsibility weight, and further, the credit backtracking penalty value is obtained by using the association depth of the predecessor dependency link for attenuation calculation. This ensures that the final penalty intensity considers both the multi-dimensional state impact and the association distance between nodes, improving the accuracy and rationality of the credit backtracking penalty value determination, reducing the risk of excessive or misjudged degradation of the exempt trust node, and thus enhancing the reliability and accuracy of node state feature vector updates in the multi-dimensional relationship graph.

[0046] In conjunction with some implementations of the first aspect, in some implementations, after downgrading and updating the expected state of the authentication-free trusted node, the method further includes:

[0047] In a multidimensional relationship graph, find downstream derived nodes that are prerequisite dependent nodes of the verification trust node;

[0048] Suspend the verification-free status of downstream derivative nodes;

[0049] Disable directed edges from downstream derived nodes pointing to other nodes in the multidimensional relationship graph;

[0050] Retrieve other nodes in the multidimensional relationship graph that are currently in a verified state;

[0051] Determine whether other nodes meet the verification-free triggering conditions for downstream derived nodes;

[0052] If the conditions are met, the verification-free status of the downstream derived nodes is restored;

[0053] After restoring the verification-free status of downstream derived nodes, update the preceding dependent nodes of the downstream derived nodes in the multidimensional relationship graph to other nodes;

[0054] After updating the preceding dependent nodes, enable directed edges in the multidimensional relationship graph that point to other nodes in the downstream derived nodes;

[0055] If the conditions are not met, a downgrade update will be performed on the downstream derived nodes.

[0056] By adopting the above technical solution, after an unverified trust node is downgraded and updated, the unverified state of its downstream derived nodes can be found and suspended, and related directed edges can be disabled. This effectively blocks the cascading propagation of potential unreliable states or erroneous features in the multidimensional relationship graph. Furthermore, by acquiring other nodes currently in a verified state and determining whether they meet the unverification trigger conditions for downstream derived nodes, an opportunity is provided for affected downstream nodes to dynamically reconstruct trust links. When the conditions are met, the unverified state of the downstream derived nodes is promptly restored, and their preceding dependent nodes and directed edges are updated. This allows the system to quickly and adaptively adjust and repair the topology when local node states fail. When the conditions are not met, a downgrade update is performed, improving the fault tolerance and dynamic self-healing capabilities of the multidimensional relationship graph in the face of local node trust collapse. This reduces the risk of widespread erroneous state propagation, thereby improving the overall stability of the cross-data state migration system and the accuracy of data processing.

[0057] In conjunction with some implementation methods of the first aspect, in some implementation methods, determining whether other nodes meet the verification-free triggering conditions of downstream derived nodes specifically includes:

[0058] Obtain the most recent verification timestamp of other nodes and the shortest path hop count between other nodes and downstream derived nodes in the multidimensional relationship graph;

[0059] Calculate the time difference between the current system time and the most recent verification timestamp;

[0060] Calculate the trust decay coefficient of other nodes based on the time difference and the shortest path hop count;

[0061] Analyze the verification exemption trigger conditions to obtain the basic trust threshold;

[0062] The initial trust scores of other nodes are corrected using a trust decay coefficient to obtain the current effective trust score;

[0063] When the current valid trust score is greater than or equal to the basic trust threshold, other nodes are determined to meet the verification-free triggering conditions for downstream derived nodes.

[0064] When the current effective trust score is less than the basic trust threshold, it is determined that other nodes do not meet the verification-free triggering conditions for downstream derived nodes.

[0065] By adopting the above technical solution, when determining whether other nodes meet the verification-free triggering conditions, the time difference is calculated by obtaining the most recent verification timestamp of other nodes and the shortest path hop count with downstream derived nodes. Then, a trust decay coefficient is calculated by combining time and spatial topological distance. This calculation method fully considers the time-dependent decay of node trust over time, as well as the structural decay caused by the extension of the transmission path in the graph. The initial trust score of other nodes is dynamically corrected using this trust decay coefficient to obtain the current effective trust score, ensuring that the indicators for evaluating the trust status of nodes truly reflect their actual reliability at the current moment and under specific topological distances. Comparing this current effective trust score with the basic trust threshold to determine whether the triggering conditions are met improves the rigor and dynamic adaptability of the new pre-dependent node screening criteria, reduces the probability of trust link reconstruction failure due to the introduction of expired or distant weakly associated nodes, and further improves the accuracy and reliability of system topological relationship reconstruction.

[0066] Secondly, embodiments of this application provide a cross-data term standardization and mastery transfer system, which includes: one or more processors and a memory; the memory is coupled to one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and one or more processors call the computer instructions to cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0067] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a system, cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0068] Fourthly, embodiments of this application provide a computer program product that, when run on a system, causes the system to execute the method described in any possible implementation of the first aspect.

[0069] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0070] 1. This application provides a method for term standardization and mastery transfer across data sources. By extracting textual and contextual features from original data entries for disambiguation and constructing a multidimensional relationship graph, it unifies the data representation across different data sources and reduces semantic ambiguity during cross-dataset processing. Based on multidimensional historical interaction results, a historical state feature vector is generated. This vector is then combined with a time decay coefficient, graph mapping relationship, and matching weights to calculate the migration state feature vector and migration confidence. This process fully considers the impact of time on user states and the tightness of topological connections between nodes, improving the accuracy and dynamic adaptability of cross-data source state migration prediction. A dual-threshold joint decision is made based on the calculated migration confidence and mastery assessment values. Nodes with high confidence and high mastery are blocked from pushing data, while nodes with low confidence and low mastery are pushed normally. A verification task is generated for low-confidence nodes to update the vector in a closed loop. This differentiated processing mechanism reduces redundant data pushes and invalid interaction verifications, lowers the system's computational overhead and communication resource consumption, and improves the overall reliability of cross-data source state migration.

[0071] 2. This application provides a cross-data term standardization and mastery transfer method. By marking intercepted nodes as verification-free trusted nodes and temporarily storing their expected states, it achieves tracking and management of verification-free nodes. When the verification of a third standard data node fails, its preceding dependency links are determined based on a multi-dimensional relationship graph, which can accurately locate the deep-seated correlation factors that led to the verification failure. If there is a verification-free trusted node in the link, it is determined that implicit falsification has occurred, and a credit backtracking penalty value is calculated based on the degree of verification failure and the depth of association, ensuring that the penalty strength matches the actual error impact and structural distance. The expected state of the verification-free trusted node is downgraded and updated using this penalty value, and its matching weight with the first standard data node is reduced simultaneously, constructing a dynamic error correction mechanism based on downstream feedback. This mechanism can promptly correct erroneous state transfers caused by previous misjudgments of confidence, reducing the risk of cascading propagation of erroneous state assessments in the graph links, thereby improving the overall accuracy and adaptive error correction capability of the cross-data state transfer system.

[0072] 3. This application provides a cross-data term standardization and mastery transfer method. After a trust node that is exempt from verification is downgraded and updated, by finding and pausing the exemption status of its downstream derived nodes and disabling related directed edges, the cascading propagation of potential unreliable states or erroneous features in the multidimensional relationship graph can be effectively blocked. Based on this, by obtaining other nodes currently in a verified state and determining whether they meet the exemption trigger conditions of downstream derived nodes, an opportunity is provided for affected downstream nodes to dynamically reconstruct trust links. When the conditions are met, the exemption status of downstream derived nodes is promptly restored and their predecessor dependent nodes and directed edges are updated, enabling the system to quickly adaptively adjust and repair topological relationships when local node states fail. When the conditions are not met, a downgrade update is performed, improving the fault tolerance and dynamic self-healing ability of the multidimensional relationship graph in the face of local node trust collapse, reducing the risk of large-scale spread of erroneous states, and thus improving the overall stability and data processing accuracy of the cross-data state transfer system. Attached Figure Description

[0073] Figure 1 This is a flowchart illustrating a cross-data term standardization and mastery transfer method in an embodiment of this application.

[0074] Figure 2 This is another flowchart illustrating a cross-data term standardization and mastery transfer method in an embodiment of this application.

[0075] Figure 3 This is another flowchart illustrating a cross-data term standardization and mastery transfer method in the embodiments of this application.

[0076] Figure 4 This is a schematic diagram of the physical device structure of a cross-data term standardization and mastery transfer system provided in an embodiment of this application. Detailed Implementation

[0077] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0078] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0079] The following example is used in conjunction with Figure 1 This application describes a cross-data term standardization and mastery transfer method in its embodiments:

[0080] Please see Figure 1 This is a flowchart illustrating a cross-data term standardization and mastery transfer method in an embodiment of this application.

[0081] S101. Extract the text features and context features of each original data entry in the first and second datasets;

[0082] The first dataset typically refers to the source data set within the system that has accumulated a large amount of user historical behavior or basic information. It is not limited to text documents but can also include audio / video transcription records, historical question banks, etc. The second dataset refers to the target data set that needs to be processed across different data sets, transitioned between states, or newly introduced, such as entirely new teaching materials or a new question bank to be tested. Raw data entries refer to basic information units in these data sets that have not yet undergone in-depth processing. They are not limited to a single word but can also be a complete sentence, a paragraph, or a specific question. Text features refer to the inherent linguistic attributes extracted from the data entries themselves, such as word frequency, part of speech, syntactic structure, or semantic vector representation. Contextual features refer to the external contextual information surrounding the data entry, such as the topic of the chapter in which the entry is located, the semantics of adjacent sentences, and the document's metadata.

[0083] The system can implement this step by calling a pre-trained large natural language processing model. The large model is able to do this because its internal self-attention mechanism can pay attention to all positions in the input sequence at the same time, thereby effectively capturing word-level text features and long-distance dependent context features. The system inputs the original data entries and their paragraphs into the large model, extracts the output of specific hidden layers of the model as text feature vectors, and extracts the global output of the pooling layer as context feature vectors.

[0084] The system can also implement this step using traditional natural language processing pipeline techniques. The system uses the term frequency inverse document frequency algorithm combined with dependency parsing tools to parse the original data entries to extract text features. At the same time, the system uses document object model tree parsing technology to extract the parent headings, adjacent paragraph content, and hierarchical depth of the original data entries in the document structure, and encodes this structured information into contextual features.

[0085] S102. Based on text features and context features, disambiguate the original data entries to generate standard data nodes with unique semantic identifiers.

[0086] Disambiguation, in natural language processing, refers to the process of eliminating ambiguity in polysemous words or vague expressions based on context, ensuring that a word or phrase has only one definite meaning in a specific context. A unique semantic identifier is a globally unique identity code assigned by the system to each specific semantic term; it is not limited to a numerical sequence but can also be a hash value, a Unique Resource Identifier (URI), etc. A standard data node is a data entity that has been cleaned, disambiguated, and structurally encapsulated, representing a clear and unambiguous knowledge point or concept.

[0087] The system can implement this step using entity linking technology based on knowledge graphs. The system concatenates the extracted text features and context features into a query vector, retrieves candidate entities in a pre-built domain knowledge base, calculates the cosine similarity between the query vector and the candidate entity description vector, selects the candidate entity with the highest similarity as the target term, extracts the globally unique number of the entity as a unique term identifier, and encapsulates the original data entries and their features into standard data nodes.

[0088] The system can also implement this step using a density-based unsupervised clustering algorithm. The system fuses the text features and context features of all original data entries into a joint feature vector, which is then input into the density-based clustering algorithm for unsupervised clustering. Each cluster represents an independent semantic term. The system generates a random hash value for the center point of each cluster as a unique term identifier and maps all original data entries within the cluster to standard data nodes carrying the hash value.

[0089] S103. Construct a multidimensional relationship graph based on standard data nodes and their corresponding semantic identifiers;

[0090] The system constructs a multidimensional relationship graph based on standard data nodes and their corresponding semantic identifiers. This multidimensional relationship graph contains the associations between standard data nodes. A multidimensional relationship graph is a graph-structured data model used to represent entities and the complex network relationships between them. It is not limited to isomorphic graphs; it can also be a heterogeneous graph containing multiple node types and edge types. Associations refer to the edges connecting different standard data nodes in the graph, representing the connections between nodes in different dimensions, such as semantic similarity, logical order, inclusion, or co-occurrence.

[0091] The system can implement this step using rule-based graph database technology. The system uses a graph database system as a storage medium and mines node pairs in a corpus containing standard data nodes through preset syntactic pattern matching rules and co-occurrence frequency statistics rules. When two nodes meet a specific co-occurrence threshold or conform to a specific dependency syntax rule, the system creates these two nodes in the graph database and generates an association edge with a specific relation type label, thereby completing the construction of a multidimensional relation graph.

[0092] The system can also implement this step using a deep learning-based relation extraction model; the system inputs pairs of standard data nodes and their corresponding contextual features into a relation classification neural network; the neural network calculates the probability distribution of node pairs on different predefined relation categories using a multilayer perceptron; the system selects the relation category with the highest probability and connects the two nodes in a distributed graph computing framework to form a graph structure containing multidimensional relationships.

[0093] S104. Obtain historical interaction data of the user terminal for the first standard data node in the first dataset;

[0094] The system acquires historical interaction data from user terminals for the first standard data node in the first dataset. This historical interaction data includes interaction timestamps and multi-dimensional interaction results. A user terminal refers to the hardware device or software client through which the user interacts with the system; it is not limited to smartphones or personal computers, but can also be a tablet, virtual reality device, or in-vehicle terminal, etc. The first standard data node refers to a specific standardized conceptual entity in the source dataset that is currently locked by the system and serves as the starting point for state analysis. Historical interaction data is an objective record of the user's behavioral interactions with this node over a past period. The interaction timestamp is a time stamp that records the precise moment each interaction occurred. Multi-dimensional interaction results refer to various feedback indicators generated by user behavior; these are not limited to simple right or wrong states, but can also include multiple dimensions such as interaction dwell time, number of prompts used, and confidence scores.

[0095] The system can implement this step using event-driven streaming log collection technology. When a user performs an operation on the first standard data node on the user terminal, the terminal application will immediately generate an event payload containing an interaction timestamp and multi-dimensional interaction results, serialize it into a lightweight data exchange format, and send it to the data lake in the cloud in real time through a message queue. The system can obtain and parse these historical interaction data in real time by subscribing to specific topics in the message queue.

[0096] The system can also implement this step through a timed batch processing synchronization mechanism; the user terminal continuously records the interaction timestamps and multi-dimensional interaction results for the first standard data node in the local embedded database; during nighttime periods when the server load is low, the system sends a data retrieval command to the user terminal; after the terminal responds to the command, it packages and compresses the historical interaction data accumulated locally, and uploads it in batches to the system's distributed file system for the system to read via a secure hypertext transfer protocol.

[0097] S105. Based on the multi-dimensional interaction results, generate the historical state feature vector corresponding to the first standard data node;

[0098] Based on multi-dimensional interaction results, the system generates a historical state feature vector corresponding to the first standard data node. This historical state feature vector is a numerical array used to quantitatively characterize a user's overall mastery or behavioral pattern of a specific node at a historical stage in a multi-dimensional space. Interaction data type refers to the category to which the interaction result belongs, such as behavioral or cognitive categories. The interaction feature sequence is a set of features arranged according to certain rules. The preset time window is a time range defined by the system. Feature variance is a statistic that measures the dispersion of the data sequence, reflecting the stability of the interaction state. Reliability weight is a multiplier that measures the credibility of data in a certain dimension. Denoising features represent the data form after outlier removal or smoothing. The temporal interaction trajectory is a sequence of interaction nodes linked together chronologically. Interaction type difference refers to the distance or dissimilarity of different interaction behaviors in the feature space. The time interval is the length of time elapsed between two interactions. The state transition gain value refers to the positive or negative change in the user's state after an interaction. The baseline state vector is the initial starting point for iterative calculations. The final update vector is the final result obtained after a complete iteration. The system will deeply analyze the user's complex and diverse past interactions and feedback, and transform them into mathematical vectors that can be accurately calculated by computers.

[0099] This step can be achieved in at least two ways:

[0100] The system can decouple multi-dimensional interaction results according to the interaction data type, extracting multi-dimensional interaction feature sequences. For each dimension's interaction feature sequence, it calculates the feature fluctuation variance within a preset time window and determines the reliability weight corresponding to each dimension based on the feature fluctuation variance. The feature fluctuation variance and reliability weight are negatively correlated. Outlier suppression processing is applied to the corresponding interaction feature sequences using the reliability weights to obtain denoised feature representations for each dimension. Based on the reliability weights, the denoised feature representations for multiple dimensions are weighted and aggregated to generate the historical state feature vector corresponding to the first standard data node. The system can use a sliding window algorithm to extract multi-dimensional interaction results on the time axis, extracting interaction feature sequences of different interaction data types. The statistical variance of each sequence within the window is calculated as the feature fluctuation variance, and a negative exponential function is used to map this variance to a reliability weight between 0 and 1. Subsequently, the system uses this weight as a smoothing coefficient to apply an exponential moving average algorithm to the interaction feature sequences for outlier suppression, obtaining denoised feature representations. Finally, the system uses matrix multiplication to linearly weight and sum the denoised feature representations for each dimension using the reliability weights to generate the historical state feature vector.

[0101] The system can also be implemented using block statistics and the isolated forest algorithm. The system divides the multidimensional interaction results into sequences according to fixed time blocks (such as natural weeks). The absolute median difference of the sequence is calculated to characterize the feature fluctuation variance, and it is converted into discrete reliability weights through a step function. The system inputs the interaction feature sequence into the isolated forest algorithm, uses the reliability weights to adjust the outlier score threshold, and removes outliers that exceed the threshold to obtain a denoised feature representation. Finally, the system uses a multilayer perceptron neural network to concatenate the weighted denoised feature representations into the input and outputs the fused historical state feature vector.

[0102] The system can also reorganize multidimensional interaction results into a temporal interaction trajectory containing multiple interaction nodes according to the chronological order of the interaction; extract the interaction type differences and time intervals between adjacent interaction nodes in the temporal interaction trajectory; calculate the state transition gain value between adjacent interaction nodes based on the interaction type differences and time intervals; initialize the baseline state vector for the first standard data node; iteratively update the baseline state vector level by level along the temporal interaction trajectory using the state transition gain value to obtain the final update vector; and use the final update vector as the historical state feature vector corresponding to the first standard data node.

[0103] The system can implement the above scheme using a directed graph model and a recurrent neural network. The system constructs a temporal interaction trajectory in the form of a directed acyclic graph based on timestamps for the multi-dimensional interaction results; it calculates the interaction type difference between the feature vectors of adjacent nodes using Euclidean distance and directly calculates the timestamp difference as the time interval; the system uses a nonlinear activation function with a decay factor, combined with the difference and the interval, to calculate the state transition gain value; the system initializes the baseline state vector with the average state of all users; subsequently, the system inputs the trajectory into a long short-term memory neural network, updating the hidden layer state of the network at each time step using the state transition gain value. The hidden state output by the network at the last time step is the final update vector, serving as the historical state feature vector.

[0104] The system can also be implemented using discrete sequences and Kalman filtering techniques. The system arranges the multidimensional interaction results into a one-dimensional discrete sequence to form a temporal interaction trajectory by arranging them in chronological order. It obtains the interaction type differences between adjacent nodes by querying a preset penalty matrix and discretizes the time interval into days. The system calculates the state transition gain value using the state transition probability matrix of the Markov chain. The system initializes the baseline state vector with a zero vector. Then, the system uses the Kalman filtering algorithm, taking the baseline state vector as a prior estimate and the state transition gain value at each step as an observed variable. The system continuously calculates the Kalman gain to progressively correct the state estimate value. The posterior estimate value after the filtering process is completed is the final update vector.

[0105] S106. In response to the user terminal's trigger command for the second dataset, determine the mapping relationship and matching weight between the second standard data node and the first standard data node in the second dataset based on the association relationship in the multidimensional relationship graph.

[0106] A trigger command refers to an operation signal performed by the user on the terminal, such as clicking to open a new document or starting a new test task. This signal will wake up the system to execute subsequent processes. A second standard data node refers to an unknown state node in the target data set that the system currently needs to predict or evaluate. A mapping relationship refers to the correspondence between source nodes and target nodes in the graph space, such as direct equivalence, prerequisite dependencies, or similar concepts. Matching weight is a quantitative indicator used to measure the tightness of the association between two nodes or the strength of state transferability.

[0107] The system can implement this step using graph traversal algorithms and path decay calculations. Upon receiving a trigger command, the system starts from the second standard data node and performs a breadth-first search algorithm in the multidimensional relationship graph to find the shortest path to the first standard data node. The system defines the sequence of edge types contained in the shortest path as a mapping relationship. At the same time, the system extracts the preset weight values ​​of all edges on the path, multiplies these weight values ​​together, and the resulting product is the matching weight between the two nodes.

[0108] The system can also implement this step through graph neural networks and vector similarity calculation; after responding to the trigger command, the system inputs the entire multidimensional relationship graph into the graph convolutional neural network for feature aggregation to generate graph embedding vectors for all standard data nodes; the system determines the mapping relationship by comparing the dimensional features of the graph embedding vectors of the first standard data node and the second standard data node; subsequently, the system calculates the cosine similarity between the two graph embedding vectors and uses the similarity score directly as the matching weight.

[0109] S107. Determine the time decay coefficient based on the interactive timestamp;

[0110] The time decay factor is a dynamic multiplier used to adjust the influence of historical data. Its value gradually decreases over time, reflecting the forgetting curve of human memory or the objective fact that things age over time. The system accurately extracts the specific time points of user's historical interactions and compares them with the current time.

[0111] The system can implement this step using a continuous forgetting curve mathematical model; the system obtains the current system time of the server and calculates the time difference (in hours or days) between it and the interaction timestamp; the system substitutes this time difference into the negative exponential decay function of the Ebbinghaus forgetting curve for calculation and directly outputs a continuous floating-point number between 0 and 1 as the accurate time decay coefficient.

[0112] The system can also implement this step through discrete time window binning mapping technology; the system predefines multiple time span intervals (e.g., within 24 hours, 1 to 7 days, 7 to 30 days, etc.) and configures a fixed decay coefficient constant for each interval; the system calculates the difference between the current time and the interaction timestamp, determines which preset time span interval the difference falls into, and then directly obtains the fixed constant corresponding to the interval through a lookup table method, using it as the time decay coefficient.

[0113] S108. Based on the time decay coefficient, mapping relationship, matching weight, and historical state feature vector, calculate the migration state feature vector and migration confidence corresponding to the second standard data node.

[0114] The migration state feature vector is a numerical representation of the state or level of mastery that a user may have when facing a specific node in new data, as predicted by the system. The migration state feature vector includes a mastery assessment value.

[0115] Migration confidence is a quantified probability value used to measure the system's degree of certainty regarding the accuracy of the calculated migration state feature vector. It is not limited to a percentage value; it can also be a normalized floating-point number or the variance of a probability distribution. Mastery assessment is a scalar or low-dimensional vector extracted from the feature vector, used to intuitively reflect a user's familiarity with specific knowledge or data. It is not limited to a percentage score; it can also be a level label or a proficiency index. The system comprehensively considers the weakening effect of time on user memory or state, combining the topological connections and their closeness between source and target nodes in the graph, and performs a dual spatial and temporal transformation of the user's historical state at the source node.

[0116] The system can implement this step using a multilayer perceptron and Bayesian inference techniques. The system performs scalar multiplication on the historical state feature vector and the matching weights to achieve feature scaling. Then, the scaled vector is concatenated with the time decay coefficient and input into a pre-trained multilayer perceptron. The multilayer perceptron outputs a high-dimensional transfer state feature vector through a nonlinear activation function and extracts the mastery assessment value through a global average pooling layer. At the same time, the system uses a Bayesian inference network to calculate the posterior probability as the transfer confidence by using the sparse matrix of the mapping relationship as the prior distribution and combining it with the output variance of the multilayer perceptron.

[0117] The system can also implement this step using a graph attention mechanism and Monte Carlo random sampling technique. The system uses historical state feature vectors as initial node features, and utilizes a graph attention network in a multidimensional relationship graph to aggregate neighbor node information based on mapping relationships and matching weights. During the aggregation process, a time decay coefficient is introduced as an attention penalty term, thereby outputting the transition state feature vector and mastery assessment value. In the inference phase, the system activates a random deactivation layer in the network, performs multiple Monte Carlo forward propagation samplings on the same input, calculates the variance of multiple output results, and uses the reciprocal of the variance after nonlinear mapping as the transition confidence.

[0118] S109. If the migration confidence level is greater than the preset threshold and the mastery assessment value is greater than the preset mastery threshold, then the push operation from the second standard data node to the user terminal is blocked.

[0119] The preset threshold is a confidence threshold set by the system to determine whether the prediction result is sufficiently reliable. It is not limited to a globally fixed static constant, but can also be a floating-point value dynamically adjusted based on user activity. The preset mastery threshold is a standard line for measuring whether a user has fully mastered the content of that node. It is not limited to a single passing score, but can also be a tiered standard set based on the difficulty of different subjects. The preset threshold and the preset mastery threshold together constitute the system's dual interception judgment benchmark. The former determines the credibility of the data, and the latter determines the degree of achievement of the status. Logically, they are independent parallel conditions, originating from the output of the system's confidence calibration module and the achievement baseline set by teaching and research experts, respectively. The interception operation refers to the system actively cutting off the path for data content to enter the user terminal display queue. After obtaining the prediction result, the system performs a rigorous condition comparison. When the system is extremely confident in the authenticity of the prediction result and assesses that the user has already achieved a high level of mastery of the target node, the system will actively intercept the distribution of that node.

[0120] The system can implement this step using a rule-based logic gating engine. The system maintains a global configuration table in memory, from which it reads preset thresholds and preset mastery thresholds. The system inputs the calculated migration confidence and mastery assessment values ​​into two independent numerical comparators. When both comparators output a high-level signal, the logic AND gate is activated, triggering an interception command. After receiving the interception command, the system locates the second standard data node in the distribution queue using a unique identifier and calls the deletion method of the queue management interface to permanently remove the node from the list to be pushed.

[0121] The system can also implement this step using a streaming computing framework and a Bloom filter. The system encapsulates the real-time calculated confidence and mastery data into data stream events and sends them to the streaming computing engine. The engine is configured with a dynamic threshold-based filtering operator. When both indicators in an event exceed the corresponding threshold, the operator writes the hash value of the second standard data node into the distributed Bloom filter. At the final push distribution gateway, the gateway program queries the Bloom filter before sending any data. If a match is found, the data packet is discarded directly, thus achieving physical interception of the push operation at the network edge layer.

[0122] S110. If the migration confidence level is less than or equal to the preset threshold and the mastery assessment value is less than or equal to the preset mastery threshold, then the second standard data node is pushed to the user terminal.

[0123] Push notifications refer to the process by which the system sends filtered target data to the user's terminal via network transmission protocols and triggers front-end rendering and display. This is not limited to in-app pop-ups; it can also take the form of system-level notifications, silent background preloading, etc. When evaluating target nodes, if the system finds the prediction results unreliable and the predicted user mastery level does not meet the acceptable standard, the system will adopt a conservative yet proactive distribution strategy. The system believes that in cases lacking sufficient confidence and tending to assume the user has not mastered the data, the safest approach is to directly present the node's content to the user. Through the scheduling and distribution module, the system adds the node to the user's pending or learning list, ensuring the user can directly access this new data that may require additional learning or understanding, thus avoiding the omission of crucial information due to inaccurate system predictions.

[0124] The system can implement this step using asynchronous message queues and long-connection push technology. After the judgment condition is met, the system serializes the complete content payload of the second standard data node into a general data format and adds the user's device identifier to construct a push message. The system delivers the message to a specific topic in a high-throughput distributed message queue. The push microservice cluster continuously consumes the topic. When the message is read, it uses the long-connection of the Transmission Control Protocol maintained with the user terminal to send the data packet to the terminal in real time. The terminal receives, parses, and renders the data on the current interactive interface.

[0125] The system can also implement this step through timed polling and content delivery network preheating technology. After determining that a push is needed, the system writes the identifier and metadata of the second standard data node into the user's dedicated cloud synchronous database and marks it as pending retrieval. The background process of the user terminal initiates a Hypertext Transfer Protocol polling request to the cloud at preset time intervals. When the terminal finds a node to be pushed, it actively requests complete data from the system. In order to accelerate transmission, the system pushes the static media resources contained in the node to the edge nodes of the content delivery network in advance. When the terminal retrieves the data, it directly obtains the resources from the nearest edge node and completes the local assembly and display of the push operation.

[0126] S111. If the migration confidence is less than or equal to a preset threshold, a feature verification task for the second standard data node is generated and sent to the user terminal to update the migration state feature vector based on the feedback data from the user terminal regarding the feature verification task.

[0127] The feature verification task is an interactive probe dynamically generated by the system to detect the user's true state. It is not limited to multiple-choice questions; it can also be a fill-in-the-blank test, subjective question-and-answer session, interactive game level, or reading comprehension probe. Feedback data is a record of the user's behavior and results after completing the verification task. It is not limited to the correctness of answers but can also include fine-grained behavioral features such as answering time, modification trajectory, and hesitation / pause time. When the system detects a low confidence level in predicting the target node's state, it indicates that relying solely on historical data and graph extrapolation is insufficient to accurately determine the user's true situation. To eliminate this uncertainty, the system proactively intervenes, tailoring a lightweight probe task for that node and issuing it to the user. The system collects the user's actual response to the task, using this objective feedback data as high-value observation samples, and feeds it back into the state evaluation model. This performs closed-loop correction on the previously calculated, inaccurate transition state feature vector, thereby completing the adaptive calibration of the state data.

[0128] The system can implement this step using a template engine and a Kalman filter update mechanism. The system pre-stores structured templates for various interactive question types in the database. When verification is triggered, the system extracts the core attributes of the second standard data node, dynamically fills in the template engine to generate a feature verification task containing the question stem and options, and sends it to the user terminal through an interface. The terminal collects the accuracy and response time of the user's answers as feedback data and uploads it. The system converts the feedback data into an observation vector, uses the state update equation in the Kalman filter algorithm to calculate the Kalman gain, and uses this gain to fuse the observation vector with the original transition state feature vector, outputting the corrected latest feature vector.

[0129] The system can also achieve this step through large language model generation and reinforcement learning reward feedback technology. The system inputs the contextual features of the second standard data node into the large language model after instruction fine-tuning, prompting the model to generate a contextualized exploratory dialogue as a feature verification task and send it to the user. The large language model can achieve this function because it has learned rich language generation rules during the pre-training stage of massive corpora, and can generate logical and targeted probe text based on input prompts. The system receives the user's natural language response to the dialogue as feedback data, and uses a semantic similarity model to evaluate the quality score of the response. Subsequently, the system uses this score as a reward signal in the reinforcement learning environment, and updates the neural network parameters that generate the transfer state feature vector through backpropagation using the policy gradient algorithm, thereby indirectly and accurately updating the feature vector of the target node.

[0130] In the above embodiments, by extracting textual and contextual features from the original data entries for disambiguation processing and constructing a multidimensional relationship graph, the data representation format across different data sets is unified, reducing semantic ambiguity during cross-dataset processing. Historical state feature vectors are generated based on multidimensional historical interaction results, and migration state feature vectors and migration confidence are calculated by combining time decay coefficients, graph mapping relationships, and matching weights. This process fully considers the impact of time passage on user states and the tightness of topological connections between nodes, improving the accuracy and dynamic adaptability of cross-dataset state migration prediction. A dual-threshold joint decision is made based on the calculated migration confidence and mastery assessment values: nodes with high confidence and high mastery are blocked from pushing data, while nodes with low confidence and low mastery are pushed normally. A verification task is generated for low-confidence nodes to update vectors in a closed loop. This differentiated processing mechanism reduces redundant data pushes and invalid interaction verifications, lowers the system's computational overhead and communication resource consumption, and improves the overall reliability of cross-dataset state migration.

[0131] In the above embodiments, the system effectively reduces redundant data push and verification overhead by intercepting second-standard data nodes with high migration confidence. However, in complex cross-data scenarios, the initial confidence assessment may be biased due to the limitations of historical data, resulting in some intercepted nodes not actually reaching the expected state. To address this potential risk of misjudgment and ensure the accuracy of state transitions over long periods, the system needs to continuously track these nodes exempt from verification and dynamically correct errors when anomalies are detected. The following section combines... Figure 2 Another cross-data term standardization and mastery transfer method is described in the embodiments of this application:

[0132] Please see Figure 2 This is another flowchart illustrating a cross-data term standardization and mastery transfer method in an embodiment of this application.

[0133] S201. Mark the intercepted second standard data node as a verification-free trusted node, and temporarily store the migration state feature vector corresponding to the verification-free trusted node as the expected state.

[0134] A trust node without verification refers to a target data entity that, during cross-data state transitions, is exempt from explicit interactive verification because the system has a very high degree of confidence in predicting its state. It is not limited to a simple Boolean label; it can also be a composite state object with a timestamp and confidence score. An expected state, on the other hand, is a numerical representation derived by the system based on historical data, assuming the user has already reached a certain level of cognition or mastery. It is not limited to a static feature vector; it can also be a probability distribution model containing mean and variance.

[0135] The system can implement this step through a technology that combines distributed caching with relational databases. In the node metadata table of the relational database, the system updates the verification-free identifier field in the record representing the second standard data node to true, thereby marking it as a verification-free trusted node. At the same time, the system serializes the migration state feature vector corresponding to the node and writes the serialized vector data into a memory-based distributed cache system as the expected state, using the globally unique identifier of the node as the key, so as to facilitate high-speed reading in subsequent processes.

[0136] The system can also implement this step through the native attribute extension mechanism of the graph database; the system directly locates the intercepted second standard data node in the graph database that stores the multidimensional relation graph, and dynamically adds a specific verification-free trust label to the node; then, the system adds an attribute field to the node entity, and persists the migration state feature vector directly as the value of the attribute field, thereby directly completing the temporary storage of the expected state in the graph topology, ensuring the data consistency between the node state and the graph structure.

[0137] S202. In response to the user terminal's trigger command for the third standard data node in the second dataset, if the third standard data node triggers a feature verification task and the feedback data representation verification fails, then based on the multidimensional relationship graph, determine the preceding dependency links of the third standard data node in the second dataset.

[0138] A third-standard data node refers to a data entity in the target data set that is currently being interacted with or tested by the user in an unknown state. Verification failure means that after completing the feature verification task for that node, the user's feedback result fails to meet the system's preset passing standard or mastery threshold. This is not limited to simple incorrect answers; it can also include interaction timeouts, excessive requests for assistance, or chaotic operation trajectories. Prerequisite dependency links refer to the set of directed paths from basic concepts to higher-level concepts in a multi-dimensional relationship graph. They represent the prerequisite knowledge or preconditions necessary to master the current node. These are not limited to simple single-source, single-line paths but can also be complex tree structures with multiple sources and branches.

[0139] The system can implement this step using a depth-first search and edge type filtering algorithm. When a feedback signal indicating that the verification failed is received, the system starts a depth-first search traversal in the multidimensional relation graph, with the third standard data node as the starting vertex. During the traversal, the system strictly filters the associated edges in the graph, allowing only the traversal of directed edges marked as predecessor dependencies. The system records all traversal paths that meet the conditions until it reaches the root node without predecessor nodes or reaches the preset maximum traversal depth, and constructs these collected path sets into predecessor dependency links.

[0140] The system can also implement this step using adjacency matrix exponentiation and reachability analysis techniques. The system extracts subgraphs from the multidimensional relation graph that specifically represent dependencies and constructs corresponding directed graph adjacency matrices. The system performs continuous matrix multiplication exponentiation on the adjacency matrix to calculate the reachability of each node in the graph to the third standard data node at different step sizes. The system selects all source nodes that can reach the third standard data node with a step size within the effective range and extracts the connected subgraphs between these nodes, using them as the preceding dependency links of the third standard data node.

[0141] S203. If there is at least one unverified trusted node in the preceding dependency link, it is determined that the unverified trusted node has implicitly falsified.

[0142] Implicit falsification refers to a system indirectly overturning a previous optimistic assumption about the state of a node, not through direct, targeted testing, but through logical reasoning and corroborating evidence. It is not limited to absolute state negation; it can also involve a sharp drop in confidence or a significant shift in the state vector. Once the system identifies the preceding dependency links that led to the current verification failure, it verifies the identity of all nodes on that link. If, within this essential foundational link, the system discovers one or more nodes previously considered mastered and exempt from verification, the system will perform rigorous logical deduction: since the high-level concept verification failed, there must be a weak link in the foundational concept, and this exempted node happens to be located within that foundational link.

[0143] The system can implement this step through set intersection operation and event triggering mechanism; the system extracts the unique identifiers of all nodes in the preceding dependency link to form a dependency node set; at the same time, the system pulls all node identifiers with verification-free trust flags from the global state registry to form a verification-free node set; the system performs an intersection operation on these two sets, and if the intersection returned by the operation is not empty, the system generates an implicit disproving event for each node in the intersection and pushes it into the system's exception handling message queue.

[0144] The system can also implement this step through the message passing mechanism of a graph neural network; the system encodes the verification failure status of the third standard data node into a high-intensity error signal vector; under the framework of the graph neural network, the system uses the message passing mechanism to broadcast the error signal vector layer by layer along the reverse direction of the preceding dependency link; when the unverified trust node on the link receives the aggregated error signal, its internal activation function will evaluate the strength of the signal. If the signal strength exceeds the preset perception threshold, the system will update the state mark of the unverified trust node in memory to implicit falsification.

[0145] S204. Calculate the credit backtracking penalty value based on the degree of failure of the third standard data node verification and the association depth of the preceding dependent link.

[0146] The system calculates a credit backtracking penalty value based on the verification failure rate of the third-standard data node and the association depth of the preceding dependent links. The verification failure rate is a quantitative indicator measuring the user's poor performance in the interactive task; it is not limited to a single error rate but can also be a penalty integral for semantic distance from the standard answer or operation time. Association depth refers to the structural distance or logical tightness between the preceding node and the current node in the graph topology; it is not limited to a simple hop count but can also be the cumulative value of all edge weights on the path. The credit backtracking penalty value is a negative correction amount calculated by the system to correct previous over-trust; it is not limited to a scalar value but can also be a multi-dimensional penalty vector. Bypass dependent links refer to independent knowledge paths that, besides the main link containing the unverified trusted node, also lead to the third-standard data node. Bypass preceding nodes are the basic conceptual entities on these independent paths. Bypass health is an indicator that comprehensively evaluates the mastery of nodes on these independent paths. Responsibility attribution weight is a proportional coefficient used to define how much responsibility the unverified trusted node should bear for this verification failure. The initial penalty base is the original penalty amount before distance decay.

[0147] This step can be implemented as follows: Based on the multidimensional relationship graph, extract at least one bypass dependency link from the third standard data node, excluding the preceding dependency link, and obtain the current state features of the bypass preceding nodes on the bypass dependency link; aggregate the current state features of the bypass preceding nodes and calculate the bypass health of the third standard data node; based on the bypass health, determine the responsibility attribution weight for verification failure caused by the verification-free trust node, as the bypass health is positively correlated with the responsibility attribution weight; calculate the initial penalty base according to the degree of verification failure of the third standard data node and the responsibility attribution weight; based on the association depth of the preceding dependency link, perform attenuation calculation on the initial penalty base to obtain the credit backtracking penalty value.

[0148] The system uses the shortest path algorithm to search for alternative paths in the graph that avoid the unverified trust nodes as bypass dependency links, and reads the current state feature vectors of the bypass predecessor nodes on these links. The system calculates the scalar mean of these feature vectors as the bypass health score. Subsequently, the system uses the logistic function to map the bypass health score to a responsibility attribution weight between zero and one. The system multiplies the verification failure degree (e.g., error rate score) of the third standard data node with the responsibility attribution weight to obtain the initial penalty base. Finally, the system extracts the network hop count of the predecessor dependency link as the association depth, substitutes it into the exponential decay function, and uses the decay coefficient to scale the initial penalty base to obtain the final credit backtracking penalty value.

[0149] The system can also implement this refined technical solution through multilayer perceptron and graph embedding distance calculation. The system uses a random walk algorithm to sample bypass dependency links and inputs the state features of the bypass predecessor nodes into the pooling layer for feature fusion, outputting a continuous scalar as the bypass health score. The system inputs this health score into a pre-trained multilayer perceptron regression model, which directly outputs the responsibility attribution weights. Then, the system calculates the dot product of the verification failure degree vector and the responsibility attribution weights to generate the initial penalty base. Finally, the system calculates the cosine distance between the verification-free trust node and the third standard data node in the graph embedding space as the association depth, uses this distance to construct a logarithmic decay factor, multiplies it by the initial penalty base, and thus calculates the credit backtracking penalty value.

[0150] S205. Using the credit backtracking penalty value, the expected state of the verification-free trust node is downgraded and updated, and the matching weight between the first standard data node and the verification-free trust node is reduced simultaneously.

[0151] Downgrade updates refer to the system's correction operation of adjusting or resetting previously inflated state predictions using newly acquired negative evidence. This is not limited to simple numerical subtraction; it can also involve vector space shifts or variance expansion of probability distributions. Matching weights are the core parameter for measuring the strength of state transferability between nodes in the source and target datasets.

[0152] The system can implement this step through vector scalar operations and dynamic decay factor adjustment. The system broadcasts the credit backtracking penalty value as a penalty vector with the same dimension as the expected state feature vector. Then, it uses vector subtraction to deduct the penalty vector from the temporarily stored expected state, completing the state downgrade update and ensuring that the updated values ​​of each dimension are not lower than the preset minimum state limit. At the same time, the system locates the mapping edge between the first standard data node and the verification-free trust node in the edge attributes of the multidimensional relationship graph. It constructs a decay factor less than one using the credit backtracking penalty value, multiplies the original matching weight with the decay factor, and thus synchronously reduces the matching weight between the two.

[0153] The system can also implement this step through Bayesian posterior update and gradient descent optimization algorithms. The system treats the expected state as a prior probability distribution, transforms the credit backtracking penalty value into a negative log-likelihood loss, and uses Bayes' theorem to calculate the mean-decreasing posterior probability distribution, which is then used as the new state after the downgrade update. At the same time, the system uses the matching weights as learnable parameters, introduces the credit backtracking penalty value into the loss function of the transfer model, uses the gradient descent optimization algorithm to calculate the gradient of the matching weights with respect to the loss function, and fine-tunes the matching weights along the opposite direction of the gradient to achieve synchronous reduction of the weights.

[0154] In the above embodiments, by marking intercepted nodes as verification-free trusted nodes and temporarily storing their expected states, tracking and management of verification-free nodes are achieved. When the verification of the third standard data node fails, its preceding dependency links are determined based on the multi-dimensional relationship graph, which can accurately locate the deep-seated correlation factors that led to the verification failure. If there are verification-free trusted nodes in the link, it is determined that implicit falsification has occurred, and a credit backtracking penalty value is calculated based on the degree of verification failure and the depth of correlation, ensuring that the penalty intensity matches the actual error impact and structural distance. The expected state of the verification-free trusted node is downgraded and updated using this penalty value, and its matching weight with the first standard data node is reduced simultaneously, constructing a dynamic error correction mechanism based on downstream feedback. This mechanism can promptly correct erroneous state transitions caused by previous misjudgments of confidence, reduce the risk of cascading propagation of erroneous state assessments in the graph link, thereby improving the overall accuracy and adaptive error correction capability of the cross-data state transition system.

[0155] In the above embodiments, the system successfully implemented the downgrade update of the unverified trust node that underwent implicit falsification through a credit backtracking mechanism. However, since all standard data nodes are closely connected in a multidimensional relationship graph, the credit downgrade of a single unverified trust node will inevitably affect downstream nodes that depend on it. If only the current node is penalized without control, erroneous state assessments can easily cascade and propagate through the graph links. To effectively block this risk and repair broken trust links for affected downstream nodes as much as possible, the system needs to have the ability to dynamically adjust and self-heal its topology. The following section combines... Figure 3 This application describes yet another method for cross-data term standardization and mastery transfer in its embodiments:

[0156] Please see Figure 3 This is another flowchart illustrating a cross-data term standardization and mastery transfer method in the embodiments of this application.

[0157] S301. In the multidimensional relationship graph, find the downstream derived nodes of the unverified trust node as the preceding dependent node;

[0158] Downstream derived nodes refer to data entities in the topological structure of a multidimensional relational graph whose cognitive state or mastery assessment logically depends on the current unverified trusted node as a prerequisite. Such nodes are not limited to direct child nodes in a simple tree structure; they can also be multi-hop successor nodes spanning multiple levels, complex composite skill nodes integrating multiple basic concepts, or conceptual entities with implicit semantic dependencies across data domains.

[0159] The system can implement this step by using a breadth-first search algorithm combined with an edge type filtering mechanism. The system takes the trusted node that has undergone implicit falsification as the starting vertex, pushes the globally unique identifier of the node into the search queue in memory, and strictly checks the attribute labels of all adjacent edges connected to the current node during the cyclic execution of the dequeue traversal operation. The algorithm is only allowed to expand outward along the directed edges that represent the preceding dependencies, and collects the identifiers of all traversed nodes into a deduplicated set until the search queue is empty or the preset maximum number of network hops is reached, thereby accurately locating all downstream derived nodes.

[0160] The system can also use the native path matching query language of the graph database to implement this step. The system submits a complex query statement with a variable depth path pattern to the underlying distributed graph database engine. In the statement, the starting node is specified as the target unverified trusted node, and the traversal direction of the path is strictly constrained to be a dependency pointer. After receiving the instruction, the graph database engine uses its highly optimized graph traversal index and concurrent query mechanism to quickly retrieve all terminal nodes that match the topology pattern in milliseconds, and returns the data entities of these nodes as downstream derived nodes to the system's business processing memory.

[0161] S302. Suspend the verification-free status of downstream derivative nodes;

[0162] Pausing refers to the system temporarily revoking or freezing the exemption from interactive verification privileges currently enjoyed by the target node. This operation is not limited to a simple Boolean switch toggling; it can also trigger a state isolation mechanism, move the node into a pending queue for re-evaluation, or dynamically tighten the node's access control permissions.

[0163] The system can utilize the key-value update and automatic expiration mechanism of the distributed in-memory database to implement this step. The system iterates through the unique identifiers of all downstream derived nodes found, accurately locates the corresponding status records of these nodes in the high-speed distributed in-memory database, and uses atomic update instructions to change the verification-free privilege field in the record from active to suspended. At the same time, a short automatic expiration timestamp is set for the status record to ensure that before the subsequent topology repair or re-verification process is completed, any downstream microservice attempting to read the status of this node will immediately recognize that it has been suspended, and thus refuse to execute the trust inference based on this node.

[0164] The system can also utilize state machine transition technology under an event-driven architecture to achieve this step. The system generates a high-priority state change event for each affected downstream derived node and publishes these events in batches to a high-throughput distributed message middleware. The consumer service responsible for global node state management subscribes to and pulls these events in real time, triggering the state transition logic of its internal complex state machine. This safely and smoothly transitions the node from an authentication-free trust state to a frozen isolation state, and locks the relevant feature vectors in memory to prevent any concurrent data tampering during this period.

[0165] S303. Disable directed edges from downstream derived nodes pointing to other nodes in the multidimensional relationship graph;

[0166] Disabling refers to cutting off the data channel through which a target node outputs trust or dependency relationships at the logical computation or physical storage level. A directed edge is a connection line in a multidimensional relationship graph that explicitly indicates the direction of knowledge flow or the order of logical dependencies. Disabling is not limited to permanently deleting the connection line; it can also be marking the state of the edge entity as inactive, setting its traversal weight in graph computation to an absolute zero value, or applying a layer of logical interception mask.

[0167] The system can achieve this step by modifying the attribute labels of edge entities in the graph database. The system initiates a batch update transaction in the graph database, uses the graph query language to locate all directed edges that originate from downstream derived nodes and point to other nodes, and adds a system-level failure flag attribute to these edge entities, or forces the original active status attribute to false. This ensures that any subsequent graph traversal algorithm based on the graph database's native application interface will automatically filter out these edges with failure flags at the underlying driver level when scanning adjacent nodes, thus achieving a break in the physical storage layer.

[0168] The system can also achieve this step by constructing a dynamic adjacency matrix mask in the in-memory computing framework. When loading the multidimensional relation graph into a large amount of memory for global topology calculation, the system dynamically generates a binary mask matrix with the same dimensions as the original global adjacency matrix for all out-degree edges of downstream derived nodes. In this mask matrix, the positions pointing to other nodes in the corresponding rows of downstream derived nodes are precisely set to zero, and all other positions are set to one. When performing subsequent graph matrix multiplication or feature aggregation calculations, the system performs a Hadamard product operation between the original adjacency matrix and this mask matrix, thereby logically cutting off these directed edges at the pure mathematical calculation level without changing the original structure of the underlying database.

[0169] S304. Obtain other nodes in the multidimensional relationship graph that are currently in a verified state;

[0170] A verified state refers to a reliable state in which a data node's mastery or cognitive level has been definitively confirmed after undergoing explicit user interaction testing, historical behavior analysis, or multiple cross-validations within the system. This state is not limited to a question node that a user just answered correctly in a recent session; it can also be a stable node that has maintained an extremely high accuracy rate historically, or a core conceptual entity verified through a cross-data domain consensus algorithm. Other nodes refer to candidate data entities in the multidimensional relationship graph that are outside the currently damaged links and have the potential to become new foundational support.

[0171] The system can achieve this step through conditional index queries in relational databases and graph identifier mapping. In the relational database responsible for persistently storing node metadata, the system uses a pre-built, efficient B-tree index for state fields to quickly filter out all node records whose states are marked as verified and whose internal confidence scores far exceed the system security baseline. The system extracts globally unique identifiers from these records. Then, through a cross-database data bus, the system uses these identifiers to perform precise mapping in the multidimensional relational graph, instantiating these discrete database records into a set of candidate node objects in the graph memory space.

[0172] The system can also achieve this step through node embedding vector similarity retrieval in graph neural networks. To avoid the huge computational consumption caused by global blind search, the system first extracts the high-dimensional semantic feature vectors of downstream derived nodes. Then, it uses a high-dimensional vector retrieval engine based on locality-sensitive hashing to perform an approximate nearest neighbor search in the pre-computed pool of verified nodes in the graph. It calculates the cosine similarity between each verified node and the downstream derived node in the continuous embedding space, thereby filtering out other nodes that are not only absolutely reliable in state, but also highly related to the downstream derived nodes in semantic logic and knowledge system.

[0173] S305. Determine whether other nodes meet the verification-free triggering conditions for downstream derived nodes;

[0174] The system determines whether other nodes meet the verification-free triggering conditions for downstream derived nodes. Specifically, it obtains the most recent verification timestamp of other nodes and the shortest path hop count between other nodes and downstream derived nodes in the multidimensional relationship graph; calculates the time difference between the current system time and the most recent verification timestamp; calculates the trust decay coefficient of other nodes based on the time difference and the shortest path hop count; parses the verification-free triggering conditions to obtain the basic trust threshold; corrects the initial trust score of other nodes using the trust decay coefficient to obtain the current effective trust score; if the current effective trust score is greater than or equal to the basic trust threshold, it determines that other nodes meet the verification-free triggering conditions for downstream derived nodes; if the current effective trust score is less than the basic trust threshold, it determines that other nodes do not meet the verification-free triggering conditions for downstream derived nodes.

[0175] The most recent verification timestamp refers to the time stamp recording the last successful reliability check of a node. It is not limited to a single absolute time point; it can also be a time series vector containing multiple verification frequencies or an effective time window weighted by time decay. The shortest path hop count refers to the minimum number of edges traversed between two nodes in the graph topology. It is not limited to pure structural distance; it can also be semantic distance incorporating edge weights or information transmission resistance values. The trust decay coefficient is a dynamic penalty factor used to quantify the degree of trust loss over time and spatial distance. The basic trust threshold is the minimum passing grade for the system to determine whether a node is qualified to support a node; it originates from system configuration or the inherent difficulty of the node. The initial trust score is the absolute reliability indicator of the candidate node itself. The current effective trust score is the actual usable trust remaining after both temporal and spatial penalties.

[0176] The system can implement this refined technical solution by combining classical graph theory algorithms with algebraic operations. The system uses Dijkstra's algorithm to search for the shortest path from other nodes to downstream derived nodes in the multidimensional relationship graph, accurately obtaining the shortest path hop count, and reading the most recent verification timestamp of other nodes from the distributed cache. The system subtracts this timestamp from the current system time to obtain a time difference accurate to milliseconds. This time difference and the shortest path hop count are then substituted into a pre-defined negative exponential decay function. The outputs of the two functions are multiplied to obtain a trust decay coefficient between zero and one. The system parses the metadata configuration file of the downstream derived nodes to extract a basic trust threshold customized for the node's difficulty. Subsequently, the system multiplies the initial trust scores of other nodes by the trust decay coefficient to calculate the current effective trust score. Finally, a numerical comparator determines whether this score is greater than or equal to the basic trust threshold, thus outputting a Boolean-type judgment result.

[0177] The system can also implement this refinement technique using a spatiotemporal graph convolutional network and a multilayer perceptron model. The spatiotemporal graph convolutional network can effectively capture the temporal features and spatial topology of nodes. The system inputs the historical verification time series of other nodes as temporal features and their locally connected subgraphs in the graph as spatial features into the network. The network automatically fuses the implicit representations of time difference and shortest path hop count through multilayer convolution operations, outputting a comprehensive trust decay coefficient vector. The system concatenates this vector with the initial trust score and inputs it into a pre-trained multilayer perceptron, which directly regresses and outputs the current effective trust score. At the same time, the system uses a natural language processing model to parse the text description of the exemption trigger condition and dynamically generate a basic trust threshold. Finally, the system compares these two values ​​to determine whether the condition is met.

[0178] S306. Restore the verification-free status of downstream derived nodes;

[0179] If the conditions are met, the downstream derived nodes are restored to their authentication-free state. Restoration means that the system removes the isolation restrictions previously imposed on the node and re-grants it the privileged state of being exempt from interactive verification. This restoration is not limited to simply rolling the state back to the exact same historical origin; it can also involve attaching a completely new, more convergent confidence interval to the node, labeling it with a verification tag, or placing it in a trial period active state with silent background monitoring.

[0180] The system can achieve this step through database transaction rollback and status field overwriting. The system initiates a database transaction with atomicity, consistency, isolation, and durability characteristics, accurately locates the status record of the downstream derived node in the relational database, overwrites the previously urgently modified suspended privilege field to the active state, and completely clears the previously set automatic expiration timestamp. At the same time, it inserts a status recovery record with the new supporting node identifier into the system's global audit log table. Finally, the transaction is committed to ensure the consistency of the underlying data, so that other computing modules within the system can immediately perceive the full recovery of the node's privileges through polling or triggers.

[0181] The system can also implement this step through the node data update mechanism of the distributed coordination service. The system connects to the distributed coordination service cluster used to manage the state of the microservice cluster via remote procedure call, finds the temporary data node representing the downstream derived node, and uses the atomic update interface to change the enumeration state stored in the node from the frozen isolation state to the verification-free recovery state. This state change will immediately trigger the event listening mechanism of the underlying distributed coordination service, automatically push the state update notification to all business microservices that have subscribed to the state of the node. After receiving the network notification, each microservice synchronously updates the node state in its local memory cache, thereby instantly completing the state recovery in the entire distributed architecture.

[0182] S307. After restoring the verification-free status of the downstream derived nodes, update the predecessor dependent nodes of the downstream derived nodes in the multidimensional relationship graph to other nodes.

[0183] Updates refer to the substantial rerouting and rebinding of dependency connections between nodes in the underlying topology of a multidimensional relational graph. Such updates are not limited to simple one-to-one edge replacements; they can also involve recalculating the weight coefficients of new connections, constructing composite support arrays composed of multiple new nodes, or deeply reconstructing the topological form of local subgraphs.

[0184] The system can implement this step through the edge redirection application interface of the graph database. The system sends a compound operation instruction to the graph database. First, it completely deletes the invalid directed edges that originally connected the unverified trusted nodes and downstream derived nodes based on the edge identifiers, releasing the relevant storage space. Then, it uses the identifiers of newly selected other nodes as the starting point and the identifiers of downstream derived nodes as the ending point to create a brand new forward dependency directed edge in the graph database. It also assigns an initial connection weight and a specific type of dependency label to the new edge, thereby completing the physical reconstruction of the topology at the persistent storage level.

[0185] The system can also implement this step by replacing the adjacency list pointer in the in-memory computing framework. In large-scale graph computing scenarios based on memory, the system locates the array of predecessor node pointers maintained in memory by the downstream derived nodes, and uses the underlying atomic comparison and swap operation to instantly replace the memory address pointers in the array that point to the original unverified trusted nodes with memory address pointers that point to other newly selected nodes. At the same time, the system updates the out-degree list of the new nodes and adds the downstream derived nodes to it. This pure memory-level pointer swap greatly improves the execution efficiency of topology updates and ensures the real-time consistency of the graph structure under extremely high concurrency.

[0186] S308. After updating the preceding dependent nodes, enable the directed edges of downstream derived nodes pointing to other nodes in the multidimensional relationship graph;

[0187] Enabling refers to removing the blockade on a node's outward output channels, restoring its ability to transmit information and trust within the graph network. This enabling is not limited to simply removing the logical mask; it can also involve the gradual restoration of edge weights, triggering a local trust propagation wave, or activating a re-evaluation mechanism in deeper downstream subgraphs.

[0188] The system can achieve this step through the reverse update operation of graph database attributes; the system executes efficient batch update instructions in the graph database again, locates all out-degree directed edges that start from downstream derived nodes and were previously marked with failure or set to false state, completely removes the failure marks on these edge entities, or sets the active state attribute back to true, so that the underlying graph traversal engine can re-identify and successfully traverse these edges in subsequent query tasks, and restores the connectivity of the graph from the physical storage level.

[0189] The system can also achieve this step by revoking the memory dynamic adjacency matrix mask; in the memory computing engine, the system locates the binary mask matrix that was previously generated specifically for downstream derived nodes, restores all the zero elements in the corresponding row positions of the mask matrix to one, stops the Hadamard product operation between the original global adjacency matrix and the mask matrix, thereby restoring the connectivity weight of the original matrix.

[0190] S309. Perform a downgrade update on the downstream derived nodes.

[0191] If the conditions are not met, a downgrade update is performed on the downstream derived nodes. A downgrade update refers to the system's forced adjustment or reset of the target node's state assessment after all repair methods have failed. This downgrade is not limited to simply deducting a value from the trust score; it can also trigger an interactive re-verification task that forces user participation, significantly shift the node's vector in a multi-dimensional feature space, or increase the variance in its probability distribution model to represent extreme uncertainty.

[0192] The system can implement this step through vector subtraction and forced state reset. The system generates a standard penalty vector with the same dimension as the state feature vector of the downstream derived node based on the hierarchical depth of the downstream derived node in the graph. Using high-dimensional vector subtraction, the system subtracts the penalty vector from the current state feature vector of the downstream derived node, thus completing the numerical-level depth reduction. At the same time, in the relational database, the system forcibly updates the state enumeration value of the node from the unverified state to the unverified state, and pushes a forced verification task for the node to the user terminal's interactive interface through a message queue, ensuring that its true state can be re-evaluated in the future.

[0193] The system can also implement this step using the penalty function of the reinforcement learning agent and the translation of the graph embedding space. The system treats the downstream derived node as a state in the reinforcement learning environment. When the degradation condition is triggered, the environment feeds back a large negative reward signal to the agent. The agent uses this negative signal to update its policy network through backpropagation and calculates a translation vector in the graph embedding space. The system adds the embedding vector of the downstream derived node to the translation vector, so that it moves rapidly away from the cluster center of the mastered node in the multidimensional semantic space and moves significantly towards the cluster center of the unmastered or unknown state. This achieves degradation update at the deep semantic representation level. The reinforcement learning model can accurately grasp the degradation range that different nodes should have when they lose support through continuous trial and error.

[0194] In the above embodiments, after an unverified trust node is downgraded and updated, by finding and pausing the unverified state of its downstream derived nodes and disabling related directed edges, the cascading propagation of potential unreliable states or erroneous features in the multidimensional relationship graph can be effectively blocked. Based on this, by obtaining other nodes currently in a verified state and determining whether they meet the unverification triggering conditions of downstream derived nodes, an opportunity is provided for affected downstream nodes to dynamically reconstruct trust links. When the conditions are met, the unverified state of the downstream derived nodes is promptly restored and its preceding dependent nodes and directed edges are updated, enabling the system to quickly adaptively adjust and repair the topology when local node states fail. When the conditions are not met, a downgrade update is performed, improving the fault tolerance and dynamic self-healing ability of the multidimensional relationship graph in the face of local node trust collapse, reducing the risk of widespread erroneous state propagation, and thus improving the overall stability and data processing accuracy of the cross-data state migration system.

[0195] The system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference needed]. Figure 4 This is a schematic diagram of the physical device structure of a cross-data term standardization and mastery transfer system provided in an embodiment of this application.

[0196] It should be noted that, Figure 4 The structure of the system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0197] like Figure 4 As shown, the system includes a Central Processing Unit (CPU) 401, which can perform various appropriate actions and processes based on a program stored in Read-Only Memory (ROM) 402 or a program loaded from storage portion 408 into Random Access Memory (RAM) 403, such as executing the methods described in the above embodiments. The RAM 403 also stores various programs and data required for system operation. The CPU 401, ROM 402, and RAM 403 are interconnected via a bus 404. An Input / Output (I / O) interface 405 is also connected to the bus 404.

[0198] The following components are connected to I / O interface 405: input section 406 including a camera, infrared sensor, etc.; output section 407 including a liquid crystal display (LCD) and speakers, etc.; storage section 408 including a hard disk, etc.; and communication section 409 including a network interface card such as a LAN (Local Area Network) card and a modem, etc. Communication section 409 performs communication processing via a network such as the Internet. Drive 410 is also connected to I / O interface 405 as needed. Removable media 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 410 as needed so that computer programs read from it can be installed into storage section 408 as needed.

[0199] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit (CPU) 401, it performs the various functions defined in the present invention.

[0200] It should be noted that the computer-readable medium shown in the embodiments of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. The transmitted data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof.

[0201] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0202] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the system described in the above embodiments; or it may exist independently and not assembled into the system. The storage medium carries one or more computer programs that, when executed by a processor of a system, cause the system to implement the methods provided in the above embodiments.

[0203] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0204] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0205] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0206] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for cross-database term standardization and mastery transfer, characterized in that, include: Extract text features and context features from each original data entry in the first and second datasets; Based on the text features and the context features, the original data entries are disambiguated to generate standard data nodes with unique semantic identifiers. A multidimensional relationship graph is constructed based on the standard data nodes and their corresponding semantic identifiers, and the multidimensional relationship graph contains the association relationships between the standard data nodes. Acquire historical interaction data of the user terminal for the first standard data node in the first dataset, wherein the historical interaction data includes interaction timestamps and multi-dimensional interaction results; Based on the multi-dimensional interaction results, a historical state feature vector corresponding to the first standard data node is generated. In response to the user terminal's trigger command for the second dataset, based on the association relationships in the multidimensional relationship graph, the mapping relationship and matching weight between the second standard data node and the first standard data node in the second dataset are determined; The time decay coefficient is determined based on the interaction timestamp; Based on the time decay coefficient, the mapping relationship, the matching weight, and the historical state feature vector, calculate the migration state feature vector and migration confidence corresponding to the second standard data node. The migration state feature vector includes a mastery assessment value. If the migration confidence is greater than a preset threshold and the mastery assessment value is greater than a preset mastery threshold, then the push operation from the second standard data node to the user terminal is blocked. If the migration confidence is less than or equal to the preset threshold and the mastery assessment value is less than or equal to the preset mastery threshold, then the second standard data node is pushed to the user terminal. If the migration confidence is less than or equal to the preset threshold, a feature verification task for the second standard data node is generated and sent to the user terminal to update the migration state feature vector based on the feedback data from the user terminal for the feature verification task.

2. The method according to claim 1, characterized in that, The step of generating the historical state feature vector corresponding to the first standard data node based on the multi-dimensional interaction results specifically includes: The multidimensional interaction results are decoupled according to the interaction data type, and multidimensional interaction feature sequences are extracted. For each dimension of the interaction feature sequence, the feature fluctuation variance within a preset time window is calculated, and the reliability weight corresponding to each dimension is determined based on the feature fluctuation variance. The feature fluctuation variance is negatively correlated with the reliability weight. The reliability weights are used to perform outlier suppression processing on the corresponding interactive feature sequences to obtain denoised feature representations for each dimension. Based on the reliability weights, the denoised feature representations corresponding to multiple dimensions are weighted and aggregated to generate the historical state feature vector corresponding to the first standard data node.

3. The method according to claim 1, characterized in that, The step of generating the historical state feature vector corresponding to the first standard data node based on the multi-dimensional interaction results specifically includes: According to the chronological order of the interactions, the multidimensional interaction results are reorganized into a temporal interaction trajectory containing multiple interaction nodes; Extract the differences in interaction types and time intervals between adjacent interaction nodes in the temporal interaction trajectory; Calculate the state transition gain value between adjacent interactive nodes based on the difference in interaction type and the time interval; Initialize the baseline state vector for the first standard data node; Along the time-series interaction trajectory, the reference state vector is iteratively updated step by step using the state transition gain value to obtain the final update vector; The final update vector is used as the historical state feature vector corresponding to the first standard data node.

4. The method according to claim 1, characterized in that, The method further includes: The intercepted second standard data node is marked as a verification-free trusted node, and the migration state feature vector corresponding to the verification-free trusted node is temporarily stored as the expected state. In response to the user terminal's trigger command for the third standard data node in the second dataset, if the third standard data node triggers the feature verification task and the feedback data representation verification fails, then based on the multidimensional relationship graph, the preceding dependency link of the third standard data node in the second dataset is determined. If at least one of the verification-free trusted nodes exists in the preceding dependency link, then the verification-free trusted node is determined to have undergone implicit falsification. The credit backtracking penalty value is calculated based on the degree of verification failure of the third standard data node and the association depth of the preceding dependency link. Using the credit backtracking penalty value, the expected state of the verification-free trust node is downgraded and updated, and the matching weight between the first standard data node and the verification-free trust node is reduced simultaneously.

5. The method according to claim 4, characterized in that, The calculation of the credit backtracking penalty value based on the degree of verification failure of the third standard data node and the association depth of the preceding dependency link specifically includes: Based on the multidimensional relationship graph, at least one bypass dependency link of the third standard data node is extracted, excluding the preceding dependency link, and the current state characteristics of the bypass preceding node on the bypass dependency link are obtained. Aggregate the current state features of the bypass front node to calculate the bypass health of the third standard data node; Based on the bypass health, the responsibility attribution weight for the failure of the verification caused by the verification-free trusted node is determined, and the bypass health is positively correlated with the responsibility attribution weight. Calculate the initial penalty base number based on the degree of failure of the verification of the third standard data node and the responsibility attribution weight; Based on the association depth of the preceding dependent link, the initial penalty base is decayed to obtain the credit backtracking penalty value.

6. The method according to claim 4, characterized in that, After performing a downgrade update on the expected state of the authentication-free trusted node, the method further includes: In the multidimensional relationship graph, find the downstream derived nodes that use the verification-free trust node as a preceding dependent node; Suspend the verification-free status of the downstream derived nodes; Disable the directed edges that the downstream derived nodes point to other nodes in the multidimensional relationship graph; Obtain other nodes in the multidimensional relationship graph that are currently in the verified state; Determine whether the other nodes meet the verification-free triggering conditions of the downstream derived nodes; If the conditions are met, the verification-free status of the downstream derived node is restored; After restoring the verification-free status of the downstream derived node, the preceding dependent nodes of the downstream derived node in the multidimensional relationship graph are updated to the other nodes; After updating the preceding dependent node, enable the directed edges of the downstream derived node pointing to other nodes in the multidimensional relationship graph; If the conditions are not met, a downgrade update is performed on the downstream derived nodes.

7. The method according to claim 6, characterized in that, The determination of whether the other nodes meet the verification-free triggering conditions of the downstream derived nodes specifically includes: Obtain the most recent verification timestamp of the other nodes and the shortest path hop count between the other nodes and the downstream derived nodes in the multidimensional relationship graph; Calculate the time difference between the current system time and the most recent verification timestamp; Based on the time difference and the shortest path hop count, calculate the trust decay coefficient of the other nodes; The basic trust threshold is obtained by analyzing the authentication exemption trigger conditions. The initial trust scores of the other nodes are corrected using the trust decay coefficient to obtain the current effective trust scores; When the current valid trust score is greater than or equal to the basic trust threshold, it is determined that the other nodes meet the verification-free triggering conditions of the downstream derived nodes; When the current valid trust score is less than the basic trust threshold, it is determined that the other nodes do not meet the verification-free triggering conditions of the downstream derived nodes.

8. A cross-database term standardization and mastery transfer system, characterized in that, The system includes: One or more processors and a memory; the memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the system, the system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the system, the system performs the method as described in any one of claims 1-7.

Images