An artificial intelligence assisted authentication method and system for data element right confirmation

Abstract

The application discloses an artificial intelligence auxiliary authentication method and system for data element right confirmation, and relates to the technical field of data element management and credible circulation. The method comprises the following steps: data acquisition: acquiring a data product, a right ownership declaration, contract terms, and calculating an initial digital fingerprint of the data product; right ownership identification: based on a multi-modal intelligent analysis model, extracting core value elements and characteristics of the data product, quantifying value contribution weights of each link to the data product, and automatically generating a data element circulation graph. The application reduces the dependence on pre-defined rules by constructing a multi-modal intelligent analysis model to automatically scan and analyze the data product in depth, realizes the intellectualization and automation of right ownership identification, realizes real-time monitoring and comparison of data operation behavior by constructing an intelligent compliance verification model, realizes substantive compliance control of the whole process of data use, and establishes a credible use closed loop.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data element governance and trusted circulation, in particular to an artificial intelligence assisted authentication method and system for data element right confirmation. BACKGROUND

[0002] With the rapid development of digital economy, data has been ranked as the fifth production factor along with land, labor, and is involved in the intersection of multiple interests such as privacy and trade secrets. Data element right confirmation is a basic prerequisite for the effective operation of the data element market, and is crucial to releasing data value and promoting the healthy development of the digital economy. However, there are many difficulties in data element right confirmation at present, such as the intangible and replicable nature of data, the involvement of multiple interest subjects, and the different logic of right confirmation from traditional property. Existing right confirmation theories are difficult to be put into practice and to respond to the needs of different stages of data transaction.

[0003] For example, the patent with the publication number CN116167752A discloses a data flow process element right confirmation method and system. The element right confirmation method includes: establishing the correspondence between the element subject, the subject characteristics, and the element subject rights in different data circulation stages, and generating mapping data; in the data flow process, the data circulation stage is judged, and the element subject, the subject characteristics, and the element subject rights in the data circulation stage are determined according to the mapping data. The invention clearly defines the responsibilities, rights, and interests of each party in data transaction, and effectively prevents tampering and consensus mechanism by using blockchain technology, solving the problem of right confirmation in the process of data element transaction.

[0004] However, the above-mentioned similar technical solutions lack dynamic intelligent analysis capability, resulting in that the data element right confirmation process relies on pre-defined rules and cannot handle complex data composition, causing application rigidity; and lack of real-time compliance monitoring throughout the process, resulting in only post-event evidence storage, and inability to substantially constrain the data use process. SUMMARY

[0005] The purpose of the present application is to provide an artificial intelligence assisted authentication method and system for data element right confirmation to solve the problems raised in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution: an artificial intelligence assisted authentication method for data element right confirmation, comprising:

[0007] Data acquisition: acquiring data products, ownership declaration, contract terms, and calculating the initial digital fingerprint of the data products;

[0008] Ownership identification: based on a multi-modal intelligent analysis model, extracting the core value elements and characteristics of the data products, quantifying the value contribution weight of each link to the data products, and automatically generating a data element circulation map;

[0009] Real-time behavior monitoring and compliance verification: Based on the intelligent compliance verification model, compile the contract terms into a set of behavior rules, monitor all data operation behaviors in real time, and compare them with the behavior rule set to obtain compliance verification results;

[0010] Value assessment: Based on the value contribution weight, combined with multi-dimensional influencing factors, calculate the value proportion of each contributor, and obtain the value assessment result;

[0011] Certificate generation: Integrate data element circulation map, compliance verification results and value assessment results to generate digital rights certificate;

[0012] On-chain storage: Calculate the digital fingerprints of data element circulation map, digital rights certificate, compliance verification results and value assessment results and store them on the blockchain to obtain the on-chain storage result;

[0013] Dynamic traceability: Deep scanning of suspected infringing data products to locate the source authorization link to obtain the infringement traceability certification result.

[0014] Further, the construction method of the multi-modal intelligent analysis model comprises:

[0015] Multi-modal data content analysis: Deep scanning and analysis of data products to identify and extract core value elements and features; extract key information from the ownership declaration and structure it;

[0016] Cross-validation: Compare and cross-validate the core value elements and features with the key information in the ownership declaration to check consistency and authenticity, and mark consistent, doubtful and conflicting items to obtain the verification result;

[0017] Value contribution quantification: Decompose the formation process of the data product into independent contribution nodes, extract the features of each contribution node and generate digital representation; based on the attention mechanism, calculate the relevance score of the data product and each contribution node, and after normalization, obtain the quantified value contribution weight; through the value contribution weight, trace back to the original data source to generate a value contribution weight table;

[0018] Circulation map construction: According to the value contribution weight table, automatically generate a data element circulation map in a visual form to show the flow path of data from the original source to the final product, and the map content includes the contribution degree and ownership relationship of each link;

[0019] Result output: Output the ownership identification result and data element circulation map, and provide interpretable analysis results, including ownership verification conclusion, value contribution distribution and detailed explanation of the map.

[0020] Further, the construction method of the data element circulation map comprises:

[0021] Node identification and extraction: identify all entities involved in the formation of data products from the value contribution weight table, and classify them into different types of nodes; the different types include data source nodes, processing link nodes, and data product nodes;

[0022] Relationship definition and connection: define the directed relationship between nodes according to the actual flow and processing sequence of data, and connect them to form a directed network;

[0023] Attribute assignment and graph formation: assign key attribute information to each node and relationship edge to generate a data element circulation graph;

[0024] Dynamic update and maintenance: when the data product source, processing flow and ownership change, update and version manage the graph according to the new analysis results.

[0025] Further, the construction method of the intelligent compliance verification model comprises:

[0026] Rule compilation: read and parse the contract terms, identify the key compliance elements, convert them into structured machine-readable rules, and combine them into a complete set of behavior rules, and bind them with the basic information of this authorization;

[0027] Digital authentication credential generation: generate digital authentication credentials based on the behavior rule set and the basic information of the authorization;

[0028] Behavior monitoring and identifier embedding: capture and record various operation behaviors on data products in real time, and generate behavior logs; when the authorized party processes the data product based on the data product and generates a derivative data product, write the unique identifier of the generated digital authentication credential into the derivative data product;

[0029] Compliance verification and result output: extract key operation features from behavior logs and compare them with behavior rule sets to determine whether the current captured data operation behavior is allowed, and output compliance verification results.

[0030] Further, the method of behavior monitoring and identifier embedding comprises:

[0031] Behavior monitoring: intercept and review operation requests in real time, capture key information of operation behavior according to predefined rules, and perform format processing to obtain behavior log entries, and compare them with behavior rule sets, if illegal operation behavior is found, trigger alarm and execute response strategy immediately;

[0032] Identifier embedding: In the process of derivative data product generation, a unique identifier associated with this derivative behavior log record is created and written inside the derivative data product; a new node representing this derivative data product is created in the data element circulation graph, a generation relationship edge from the data product to the derivative data product is established, and the behavior log of this derivative behavior is associated with the new node.

[0033] Further, the value evaluation method comprises:

[0034] Basic contribution weight extraction: Extract the value contribution weight of each node to the data product from the data element circulation graph to obtain the basic contribution weight;

[0035] Determine multi-dimensional influencing factors: Identify and define other factors that affect value allocation in addition to contribution degree, and develop quantifiable indicators;

[0036] Comprehensive value contribution calculation: Combine the basic contribution weight with the multi-dimensional influencing factors, and assign adjustment coefficients to the multi-dimensional influencing factors to calculate the adjusted comprehensive value contribution of each node;

[0037] Aggregation and distribution: Map each node in the data element circulation graph to a contributor, and integrate the comprehensive value contribution of all nodes belonging to the same contributor to obtain the total value contribution of the contributor; normalize the total value contribution of each contributor to calculate the value allocation proportion of each contributor.

[0038] Further, the dynamic tracing method comprises:

[0039] Fingerprint extraction and comparison: Calculate the digital fingerprint of the suspected infringing data product and compare it with the initial digital fingerprint to obtain the global fingerprint comparison result; if they match completely, the dynamic tracing result is generated directly; if they do not match, proceed to the next step;

[0040] Scan and analyze identifier: Deeply scan the content of the suspected infringing product and analyze its internal identifier;

[0041] Certificate and graph association verification: Cross-verify the parsed identifier with the digital rights certificate and the identifier information recorded in the data element circulation graph;

[0042] Graph traceability positioning: According to the matched identifier, perform reverse traceability on the data element circulation graph to locate the initial authorized node of the data component and obtain the graph traceability result;

[0043] On-chain evidence audit: Based on the graph traceability result, query the behavior log of the node stored on the blockchain to obtain the on-chain authorization result;

[0044] Dynamic traceability result generation: comprehensive global fingerprint comparison results, atlas traceability results and on-chain authorization results are obtained to obtain infringement traceability authentication results.

[0045] An artificial intelligence assisted authentication system for data element right confirmation, comprising:

[0046] A data acquisition module: for acquiring data products, right declaration, contract terms, and calculating product digital fingerprints;

[0047] An intelligent right confirmation module: for building a multi-modal intelligent analysis model, extracting the core value elements and characteristics of the data product, quantifying the value contribution weight of each link to the data product, and automatically generating a data element circulation atlas; and for calculating the value proportion of each contributor based on the value contribution weight and combining multi-dimensional influencing factors to obtain a value assessment result; and for generating a digital right certificate by comprehensively integrating the data element circulation atlas, compliance verification results and value assessment results;

[0048] A dynamic authentication module: for building an intelligent compliance verification model, compiling contract terms into a behavior rule set, monitoring all data operation behaviors in real time, and comparing with the behavior rule set to obtain compliance verification results;

[0049] A traceability evidence module: for calculating the digital fingerprints of the data element circulation atlas, digital right certificate, compliance verification results and value assessment results and notarizing them on the blockchain to obtain on-chain notarization results; and for deep scanning of suspected infringing data products, locating the source authorization link to obtain infringement traceability authentication results.

[0050] Compared with the prior art, the beneficial effects of the present application are:

[0051] An artificial intelligence assisted authentication method and system for data element right confirmation, by building a data element circulation atlas, clearly mapping the flow path, processing link and right relationship of data from source to end, and quantifying the value contribution weight of each node based on attention mechanism, providing an objective basis for fair value distribution; by establishing a dynamic traceability method, deep scanning the data content itself and cross verifying with the atlas and on-chain records, realizing accurate traceability and responsibility identification.

[0052] At the same time, by building a multi-modal intelligent analysis model, the data product is automatically scanned and analyzed, reducing the dependence on pre-defined rules, realizing the intelligentization and automation of right identification; by building an intelligent compliance verification model, the data operation behavior is monitored and compared in real time, realizing the substantive compliance control of the whole process of data use, and establishing a credible use closed loop. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1A schematic diagram of an artificial intelligence assisted authentication method for data element right confirmation of the present application;

[0054] Figure 2 A schematic diagram of a multi-modal intelligent analysis model construction method of the present application;

[0055] Figure 3 A schematic diagram of a data element circulation map construction method of the present application;

[0056] Figure 4 A schematic diagram of an intelligent compliance verification model construction method of the present application;

[0057] Figure 5 A schematic diagram of a value evaluation method of the present application;

[0058] Figure 6 A schematic diagram of a dynamic tracing method of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0060] As shown in Figure 1 , the present application provides a technical solution: an artificial intelligence assisted authentication method for data element right confirmation, comprising:

[0061] Data acquisition: acquire data products, right ownership declaration, contract terms, and calculate the initial digital fingerprint of the data products.

[0062] It should be noted that for data products, a dedicated database connector (such as JDBC) can be used to obtain data product files or data sets from a specified storage location; for right ownership declaration and contract term files, they can be obtained through RESTful API calls, email parsing (such as MAP / POP3 protocol) or specified directory monitoring. In the data transmission process, checksum or hash value comparison is used to verify whether the file is completely consistent before and after transmission to prevent data damage due to network problems; if the verification fails, the retransmission mechanism is automatically triggered to ensure the integrity of the input data.

[0063] The data product is standardized and pretreated, and then a cryptographic hash function (such as SHA-256) is applied for calculation to generate a fixed-length hash value, that is, an initial digital fingerprint of the data product. The initial digital fingerprint is associated with the meta-information of the data product and the storage path of the corresponding ownership declaration, contract clause file and the like and is stored into a fingerprint database.

[0064] Ownership identification: based on a multi-modal intelligent analysis model, core value elements and characteristics of the data product are extracted, value contribution weights of each link to the formation of the data product are quantified, and a data element circulation map is automatically generated.

[0065] As shown in Figures 2-3 , the present application provides a multi-modal intelligent analysis model construction method;

[0066] Specifically:

[0067] Multi-modal data content analysis: the data product is deeply scanned and analyzed, core value elements and characteristics are identified and extracted, and key information is extracted from the ownership declaration and structured;

[0068] Cross-validation: the core value elements and characteristics are compared and cross-validated with the key information in the ownership declaration, consistency and authenticity are checked, consistent, doubtful and conflicting items are marked, and a verification result is obtained;

[0069] Value contribution quantification: the formation process of the data product is decomposed into independent contribution nodes, features of each contribution node are extracted and digitalized representation is generated; based on an attention mechanism, a correlation score of the data product and each contribution node is calculated, and after normalization, a quantized value contribution weight is obtained; through the value contribution weight, the original data source is traced back to generate a value contribution weight table;

[0070] Circulation map construction: according to the value contribution weight table, a data element circulation map is automatically generated to display the flow path of data from the original source to the final product in a visual form, and the map content includes the contribution degree and ownership relationship of each link;

[0071] Result output: the ownership identification result and the data element circulation map are output, and an interpretable analysis result is provided, including the ownership verification conclusion, the value contribution distribution and the detailed description of the map.

[0072] The construction method of the data element circulation map comprises:

[0073] Node identification and extraction: all entities participating in the formation of the data product are identified from the value contribution weight table, and are classified into different types of nodes; the different types include data source nodes, processing link nodes and data product nodes;

[0074] Relationship definition and connection: Define the directed relationship between nodes according to the actual flow and processing sequence of data, and connect them to form a directed network;

[0075] Attribute assignment and graph formation: Assign key attribute information to each node and relationship edge to generate a data element circulation graph;

[0076] Dynamic update and maintenance: When the data product source, processing flow and ownership change, update and version manage the graph according to the new analysis results.

[0077] It should be noted that the first step is multi-modal data content analysis. Through file format parser (such as ApacheTika) to identify the overall format of data product, and then use pre-defined rules and machine learning model to automatically identify the data modalities contained in the data product, such as structured table data, unstructured text, image / video, audio, etc. The machine learning model includes but is not limited to convolutional neural network CNN (for image recognition), Transformer model (for text recognition), open source library Librosa (for audio recognition).

[0078] For structured data, use statistical feature extraction (such as mean, variance, quantile) and model-based feature importance analysis (such as using XGBoost model to calculate feature importance score) to identify key data fields and dimensions; for text data, use NLP technology to find the core theme of the text; for image / video data, use target detection model (such as YOLO) for feature extraction and target recognition, get image classification label, salient object feature vector, etc. For audio data, perform acoustic feature extraction (such as Mel-frequency cepstral coefficients MFCCs), and apply speech-to-text model (such as Whisper) to convert it to text for text analysis. Convert all extracted features into a unified feature vector representation to get the core value element feature vector set.

[0079] Use optical character recognition technology to scan the electronic copy of the paper file of the ownership declaration, and use NLP model based on sequence labeling or template filling to extract structured key information from unstructured ownership declaration, form the ownership declaration list, including declaration owner, data source, creation / modification time, license agreement type, usage constraints, original contributor list, etc. Field.

[0080] Second step, cross-validation, to check the consistency of the actual content and the ownership statement of the data product, find potential contradictions or doubts, and enhance the feasibility of ownership authentication. Timestamp verification, check whether the maximum and minimum timestamps extracted from the data product (such as the recording time in log data) are within the data generation time range claimed by the ownership statement; geographic information verification, check whether the geographic location information extracted from the data product (such as the places identified by NER, EXIF information of pictures) is within the range; content source verification, match the data source and original contributor claimed in the ownership statement with the entities extracted from the data content (such as the names of institutions identified by NER, database watermark information) by string similarity; license constraint verification, check whether the data content contains prohibited content by the license in the ownership statement, such as the license declared as "no personally identifiable information PII", but the identity card number, phone number, etc. are identified in the text by NER.

[0081] Calculate a credibility score for each comparison result, which can be weighted based on the confidence of the matching algorithm, the strength of the evidence (such as EXIF information is strong evidence, NER identified information is weak evidence), etc. Pre-set score threshold, when the credibility score > upper limit of score threshold, mark as consistent; when the credibility score falls within the score threshold, mark as suspicious, such as partial information matching or low matching confidence; when the credibility score is lower than the lower limit of the score threshold or explicit evidence is found, such as the timestamp is completely out of range, mark as conflict. Record all comparison results and marked states in the cross-validation result table.

[0082] Third step, value contribution quantification, to quantify the contribution of each link and each source in the formation process of the data product. Through data provenance technology (such as PROV data model), track its formation process, and decompose it into multiple contribution nodes, including original data source, data cleaning program, feature engineering algorithm, data fusion module, analysis model, etc. Each node is considered as an independent contribution entity. Generate a digital representation for each node, methods include: extract the metadata of the node, such as the name and parameters of the algorithm, the description of the original data source, if the node itself is data, call the method in the multi-modal data content analysis step to generate the content feature vector of the data node. Fuse metadata information and data content feature vector to form a comprehensive feature representation; then use embedding techniques such as graph neural network or document embedding model (such as Doc2Vec) to encode the above comprehensive feature representation into a fixed, low-dimensional node embedding vector.

[0083] The overall features of the data product are represented as a query vector, and the embedding vectors of all contributing nodes are represented as key vectors. The query vector and each key vector are input into an attention mechanism layer, such as the scaled dot-product attention mechanism in the Transformer, to calculate the similarity between the query vector and each key vector and output a set of attention weights. The set of attention weights directly reflects the relevance strength of the final data product to each contributing node, i.e., the value contribution of the node. The obtained attention weights are normalized by a Softmax function to obtain the quantitative value contribution weight of each contributing node. The weights are then propagated back from the final digital product to the original data source until the upstream original data provider is traced. Finally, a value contribution weight table is generated, including fields such as contributing node ID, node description, and contribution weight.

[0084] In the fourth step, the circulation map is constructed. First, node identification and extraction are performed. Rule-based pattern matching and named entity recognition are used to automatically extract entity names, such as database table names, algorithm model names, and original data files, from the node description field. Based on the extracted entity names and metadata information, a node classifier is established. The classifier classifies each entity into the following three core categories based on a predefined rule set and a machine learning classification model (such as a decision tree for classifying node description text): data source node, representing the original source of data; processing link node, representing the process of value-added operation on data; and data product node, representing the final output of the circulation chain, i.e., the data product to be righted itself; and a type label is assigned to each node.

[0085] To prevent the same entity from being repeatedly created as multiple nodes at different links, the system maintains a global node registration table. Before creating a new node, the node registration table is compared to calculate the similarity of node key information (such as ID, name, and hash value). If the similarity exceeds a preset threshold, it is determined that the nodes are the same and are merged.

[0086] Next, the relationship is defined and connected. According to the actual processing logic and flow order of the data, the relationship between the nodes is defined, and all the nodes are connected into a directed acyclic graph to form the skeleton of the map. The data flow analysis tool (such as Apache Atlas) is used to obtain the flow relationship between the data in each node; the exact upstream source and downstream target correspondence relationship is extracted from these information to form the original edge set. A directed edge is created for each pair of correspondence relationship, and the direction of the edge represents the direction of data dependency and value flow; and each edge is assigned a descriptive relationship type attribute, such as derived from, processed how, input to, generated, etc. To prevent logical errors in data flow, graph algorithms (such as depth-first search) are used to detect loops in the constructed directed acyclic graph; if a loop is detected, the subgraph is marked as an abnormal structure, triggering an alarm and manual review.

[0087] Then, attribute assignment and graph formation. Fill in the attributes for each node, the key attribute information includes: value contribution weight, obtained from the value contribution weight table; ownership verification state, obtained from the cross-validation result table; metadata, such as node creation time, owner, version number, description information, etc.; original data sample hash, used for unique identification and integrity check. Fill in the attributes for each edge to describe the details of the relationship between nodes, the key attribute information includes: processing operation, describing what kind of processing occurs in the relationship represented by this edge, such as data cleaning, feature merging, model inference; data flow, the amount of data flowing through this relationship; execution timestamp, the time when the processing operation occurs. After the attributes of nodes and edges are filled in, the data element flow graph is obtained, which is stored in a graph database (such as Neo4j).

[0088] Finally, dynamic update and maintenance. By regularly executing scheduling tasks, monitor whether the data product source, processing code library, ownership declaration file, etc. have changed; once a change is detected, a new round of ownership identification process is automatically triggered to generate new analysis results and value contribution weight table. Compare the new value contribution weight table with the current information stored in the data element flow graph, and calculate the difference; then use the incremental graph update algorithm to modify only the changed part, such as adding new nodes, deleting edges, updating weight attributes. Each time the graph is updated, a graph version snapshot is automatically generated and stored using the backup function of the graph database. All snapshots are indexed through a version management table, which records the version number, creation timestamp, and change summary of each version.

[0089] Fifth step, result output. Use a templated report generation engine (such as Apache POI to operate word documents) to automatically integrate the cross-validation result table, value contribution weight table, and data element flow graph into a structured report, including textual interpretation and risk prompts for key nodes, complex relationships, and suspicious / conflicting items found in the graph. The generated report can be stored in a database and assigned a unique searchable digital fingerprint (such as a hash value); at the same time, publish the authentication results to external systems (such as data trading platforms, monitoring platforms) through RESTful API interfaces, supporting on-demand query and verification.

[0090] Real-time behavior monitoring and compliance verification: based on the intelligent compliance verification model, compile the contract clauses into a set of behavior rules, monitor all data operation behaviors in real time, and compare them with the set of behavior rules to obtain compliance verification results.

[0091] As shown in Figure 4 , the present application provides a method for constructing an intelligent compliance verification model;

[0092] Specifically:

[0093] Rule Compilation: Read and parse the contract terms, identify key compliance elements, convert into structured machine-readable rules, and combine into a complete set of behavioral rules, bound to the underlying information of this authorization;

[0094] Digital Authentication Credential Generation: Based on the set of behavioral rules and the underlying information of the authorization, generate a digital authentication credential;

[0095] Behavior Monitoring and Identifier Embedding: Capture and record various operational behaviors on the data product in real time, and generate a behavior log; when the authorized party processes the data product based on compliance and generates a derivative data product, write the unique identifier of the generated digital authentication credential into the derivative data product;

[0096] Compliance Verification and Result Output: Extract key operational features from the behavior log and compare them with the set of behavioral rules to determine whether the current captured data operation behavior is allowed, and output the compliance verification result.

[0097] It should be noted that the first step, rule compilation. Obtain the contract terms text through optical character recognition technology; then use NLP technology to analyze the contract terms in depth: through named entity recognition model, the model uses a deep learning architecture that integrates pre-trained language models, bidirectional sequence modeling and global optimization techniques to accurately extract key compliance elements such as authorized subject, data object, purpose of use, authorization period, geographic restrictions, prohibited behaviors, etc.; and use the relationship extraction model to identify the constraints between these entities. Map the identified key compliance elements to a pre-defined rule template to generate a set of behavioral rules; the template uses a declarative rule language (such as the Drools rule language) to ensure that the rules are unambiguous and executable. Bind the generated set of behavioral rules to the unique authorization ID, data product hash, authorized party, authorized party, etc. underlying information of this data authorization; load the rule set into memory through the rule management engine (such as Drools Fusion) to prepare for real-time stream processing.

[0098] The second step is to generate a digital authentication credential. Calculate the hash value of the set of behavioral rules using the SHA-256 algorithm; the authorized party signs the hash value of the credential content using its private key to generate a digital signature. The credential content includes the authorization ID, data product hash, behavioral rule set hash, authorized party digital signature, effective / inactive timestamp. Serialize the credential content and digital signature into a standard format to generate the final digital authentication credential. This digital authentication credential can be distributed to the authorized party separately, or directly packaged with the data product in a secure environment.

[0099] Third step, behavior monitoring and identifier embedding. All operation behaviors on data products are intercepted in real time, operation behaviors are parsed, key information of operation behaviors is extracted, key information includes operation subject, operation type, operation time, access source IP, involved data range and other key fields, and the key information is formatted into structured behavior log entries. The behavior log entries are compared with the behavior rule set in the memory in real time; once a violation operation is found, the following responses are triggered immediately: real-time alarm information is sent to the administrator; the pre-defined response strategy is automatically executed, such as blocking the current operation, terminating the session, and returning the result blurred. When it is monitored that the authorized party processes the authorized data product in compliance and generates a new derivative data product, a globally unique identifier is created for the derivative behavior, and the globally unique identifier is associated with the complete log record of the derivative behavior. Then, the globally unique identifier is written into the derivative data product through metadata field injection or digital watermarking technology. Then, a new node representing the derivative data product is created in the data element circulation graph, and a new directed edge from the original data product node to the new node is established, and the relationship type is generated or derived from; the behavior log of the derivative behavior is associated with the new node as audit evidence of the generation process. The real-time behavior log stream, real-time response to violations, derivative data products embedded with unique identifiers, and updated data element circulation graph are output.

[0100] Fourth step, compliance verification and result output. The real-time generated behavior log entries are stored in a time series database (such as InfluDB), and batch processing tasks (such as using Spark SQL) are periodically executed to analyze a large number of historical behavior logs to find operation habits, behavior sequences or statistical rules, and extract key operation features such as daily access volume of a user, abnormal time access frequency, and sensitive data access heat map. The aggregated log features are compared with the behavior rule set for offline deep analysis to generate compliance verification results, including overall compliance rate, violation event list, high-risk operation analysis, and potential risk prompt. Finally, the compliance verification results are displayed through a visualization dashboard (such as Grafana), and RESTful API interfaces are provided to allow other systems to query or obtain detailed compliance verification results.

[0101] Value evaluation: based on the value contribution weight, combined with multi-dimensional influence factors, the value proportion of each contribution party is calculated to obtain the value evaluation result.

[0102] As shown in Figure 5 , the present application provides a value evaluation method;

[0103] Specifically:

[0104] Basic contribution weight extraction: the value contribution weight of each node to the formation of a data product is extracted from the data element circulation graph to obtain the basic contribution weight;

[0105] Determine multi-dimensional influencing factors: Identify and define factors other than contribution that affect value allocation and develop quantifiable indicators;

[0106] Calculate comprehensive value contribution: Combine the base contribution weight with multi-dimensional influencing factors and assign adjustment coefficients to the multi-dimensional influencing factors to calculate the adjusted comprehensive value contribution of each node.

[0107] Aggregation and distribution: Map each node in the data element circulation map to the contributor and integrate the comprehensive value contribution of all nodes belonging to the same contributor to obtain the total value contribution of the contributor. Normalize the total value contribution of each contributor to calculate the value allocation proportion of each contributor.

[0108] It should be noted that the first step is to extract the base contribution weight. Through the graph query language (such as Cypher in Neo4j graph database), a traversal query is executed. The query starts from the node representing the final data product and traverses all directed edges pointing to it in reverse until the most upstream data source node, thus covering all contributing nodes. Read the value contribution weight attribute stored on each node during traversal. Convert the query results into a structured list, where each entry contains a node's unique identifier (node ID) and its corresponding value contribution weight value (i.e., base contribution weight). Output a base contribution weight set containing key-value pairs of node ID and base contribution weight.

[0109] Second step, determine multi-dimensional influencing factors. Based on economic principles and industry practice, a set of multi-dimensional factor system affecting data value is defined in advance, including: data quality factor (Q), which measures the accuracy, completeness, consistency, timeliness and uniqueness of data; data scarcity (S), which measures the scarcity of data in the market or whether it is exclusive authorization; cost input factor (C), which measures the direct and indirect costs generated by the node in the process of data collection, processing, storage and maintenance; risk bearing factor (R), which measures the legal compliance risk, data security risk and market risk borne by the node; market demand factor (M), which measures the market demand of the data content provided by the node. Then, design quantifiable calculation indicators and data collection methods for each influencing factor: data quality factor, run a series of data quality checking rules (such as checking null value range, value range compliance, format consistency, logical consistency) through the pre-set data quality analysis framework (such as Apache Griffin) and calculate the comprehensive score; data scarcity factor, which can be assigned according to the exclusivity clause of the contract terms, such as exclusive authorization = 1.5, non-exclusive = 1.0; cost input factor, pull the calculation, storage, bandwidth and human cost data related to the node from the enterprise's financial management system or resource monitoring system through API interface and perform normalization processing; risk bearing factor based on the sensitivity level of the node data (such as PII quantity), the strictness of the supervision of the region to which it belongs, etc., calculated through a weighted scorecard model. Finally, a multi-dimensional influencing factor vector is calculated for each node.

[0110] Third step, comprehensive value contribution calculation. Assign a global adjustment coefficient to each multi-dimensional influencing factor, representing the relative importance of the factor in the overall value assessment. These coefficients can be determined by domain experts through the analytic hierarchy process or optimized through machine learning models. Normalize and standardize the values of each influencing factor for each node to eliminate dimensional effects. Then calculate the comprehensive value contribution of each node, i.e. Σ (normalized and standardized influencing factor values x adjustment coefficient).

[0111] Fourth step, aggregation and distribution. In the data element circulation process, each node has an owner and a contributor attribute; according to the attribute, all nodes are grouped, and each group contains all nodes belonging to the same contributor. For each contributor, the total value contribution of all nodes under its name is summed up to obtain the total value contribution of the contributor. Normalize all total value contributions of the contributors, and calculate the value proportion of each contributor, that is, the total value contribution of each contributor / the sum of the total value contributions of all contributors, to obtain the value proportion of each contributor, which is the value distribution weight of the contributor to the final data product. Format the calculation result as a value distribution table to obtain the value evaluation result. The fields of the value distribution table include contributor ID, contributor name, total value contribution, and value distribution proportion.

[0112] Certificate generation: generate a digital right certificate by integrating the data element circulation graph, compliance verification result and value evaluation result.

[0113] It should be noted that the following core elements are automatically extracted from the output of each step: the globally unique identifier of the data element circulation graph and the hash value of the graph, ensuring a strong binding relationship between the certificate and the graph; the behavior rule set signed by the digital signature in the compliance verification result, and the unique authorization ID of this authorization or transaction; the value distribution table in the value evaluation result, which clearly shows the value proportion of each contributor; the metadata of the data product, the identity of the authorized party and the authorized party, the authorization period, the effective conditions and other basic information.

[0114] All the extracted information above is obtained through a pre-defined data model to obtain a structured certificate declaration set containing all right confirmation information. The data model adopts an internationally recognized standard format, such as W3C verifiable certificate data model.

[0115] A unique, fixed-length digest information is generated by using a cryptographic hash function (such as SHA-256) to calculate the generated certificate declaration set. The certificate issuer uses a private key to digitally sign the hash digest through a digital signature algorithm (such as ECDSA). The digital signature, the private key identifier used for signing, and the signature algorithm type are combined with the certificate declaration set to form the final digital right certificate, which is returned to the authorized applicant through a secure API interface. The digital right certificate is stored in a publicly accessible or authorized access certificate library, and a verifiable certificate parsing service is provided.

[0116] On-chain storage: calculate the digital fingerprints of the data element circulation graph, digital right certificate, compliance verification result and value evaluation result and store them on the blockchain to obtain the on-chain storage result.

[0117] It should be noted that the data element circulation map, digital right certificate, compliance verification result and value assessment result to be notarized are converted into a standard and determined serialization format. For data structure, it is converted into a standard format such as JSON, and the ordering of key-value pairs is ensured to be consistent to prevent different hash values caused by serialization differences; for files or certificates, the binary content or serialized string thereof is directly read. For each item of serialized content, a corresponding hash value is generated by using an encryption-level hash function such as an SHA-256 algorithm. The calculated hash value and necessary metadata are combined into a structured notarization data package, including a notarization content hash value, a timestamp of notarization initiation, an identifier of an application notarization party and a corresponding digital right certificate ID. A transaction is constructed using a blockchain account address controlled by the notarization party; for a blockchain supporting a smart contract, a storeHash function in a pre-deployed notarization smart contract is called, and the notarization data package is taken as transaction input data; for a blockchain taking data storage as a core, the notarization data package is directly written into a storage field of the transaction; for a blockchain such as Bitcoin, an OP_RETURN operation code is used, and the hash value of the notarization data package is written into an input script to realize information chaining. The transaction is digitally signed using a private key of the notarization party, and the authorized and paid notarization operation is implemented; the signed transaction is broadcast to the corresponding blockchain network through a P2P network. After consensus confirmation of the transaction by the blockchain network, a chain-based notarization result with legal and technical effectiveness is generated by parsing a returned certificate of the blockchain.

[0118] Dynamic tracing: deep scanning of suspected infringing data products, locating the source authorization link to obtain the infringement traceability certification result.

[0119] As shown in Figure 6 , the present application provides a dynamic tracing method;

[0120] Specifically:

[0121] Fingerprint extraction and comparison: calculating the digital fingerprint of the suspected infringing data product, and comparing it with the initial digital fingerprint to obtain the global fingerprint comparison result; if it is completely matched, the dynamic tracing result is directly generated; if it is not matched, the next step is entered;

[0122] Scanning and analyzing the identifier: deep scanning of the content itself of the suspected infringing product to analyze the internal identifier;

[0123] License and map association verification: cross-verification of the analyzed identifier with the identifier information recorded in the digital right certificate and the data element circulation map;

[0124] Map trace positioning: According to the matched identifier, reverse trace on the data element circulation map to locate the original authorized node of the data component, and get the map trace result;

[0125] On-chain storage audit: Based on the map trace result, query the behavior log of the node stored on the blockchain to get the on-chain authorization result;

[0126] Dynamic trace result generation: Comprehensive global fingerprint comparison result, map trace result and on-chain authorization result to get the infringement trace authentication result.

[0127] It should be noted that the first step is fingerprint extraction and comparison. The global digital fingerprint of the suspected infringing data product is calculated using a cryptographic hash algorithm (SHA-256 algorithm). The calculated global digital fingerprint is matched with the pre-stored authorized product fingerprint library, which stores the initial digital fingerprints of all data products authorized by the system and their associated authorization information. If there is a certain initial fingerprint that matches it in the fingerprint library, the process terminates, and a dynamic trace result is automatically generated, directly outputting the authentication conclusion "the product is completely identical to the authorized product [product ID], confirming that it is an unauthorized copy", and attaching the corresponding authorization credential information; If not matched, it is determined that the product may be a modified derivative or a partial infringement penalty, and the next step of deep scanning process is performed.

[0128] The second step is to scan and analyze the identifier. The suspected infringing data product is analyzed in multiple modalities to identify its internal structure. According to different data modalities, corresponding technologies are used to scan all possible locations of hidden identifiers: for file metadata, parse the file header and customize metadata fields; for structured data, scan specific data columns, redundant bits or pseudo records that may contain watermarks; for unstructured data, call digital watermark detection algorithms or steganalysis tools to detect whether there are watermark signals in the data content. Extract the potential identifier code from the scanning results, and use the predetermined decoding algorithm (such as watermark decoding key) to decode the extracted code to restore the original unique identifier. If one or more valid identifiers are successfully extracted, proceed to the next step; if no identifier is extracted, automatically start the degradation trace mechanism: re-input the suspected infringing data product into the multi-modal intelligent analysis model in the ownership identification process to extract the feature vector of its core value elements, and calculate the similarity with the feature vectors of all authorized data products in the database to lock the high similarity candidate source. At the same time, query the data platform access audit log to find out which users or system accounts have accessed data products similar to the above candidate source in the relevant time period. Finally, generate an inferential trace report, clearly indicating the absence of identifier warning, and providing high-value investigation clues such as candidate data list and suspicious entity behavior record.

[0129] Third step, verification of the association between the warrant and the map. Use the parsed identifier as a key query condition to search in the database; query the digital rights warrant library to check if there is a warrant containing the identifier, thereby confirming whether the derivative behavior is within the authorized scope and generating a compliant warrant; query the data element circulation map to check if there is a node representing the derivative data product whose attributes contain the identifier, thereby confirming that the identifier has been officially recorded in the map. If the same record is found in both the warrant and the map, the identifier is verified as real and valid; if only one of them is found or not at all, it is marked as “invalid identifier / unauthorized” and the corresponding warning information is generated.

[0130] Fourth step, map traceability positioning. Locate the node of the derivative data product in the data element circulation map, and from this node, traverse all “generated” or “derived from” relationship edges pointing to it in reverse. Recursively access its upstream parent nodes until it cannot continue to trace; this process is automatically performed using a graph query language. Record all nodes and edges on the entire traversal path to form a complete traceability path. Identify the initial authorized node on the path, which is the first node that contributes data and has passed the ownership verification.

[0131] Fifth step, on-chain evidence audit. Find the source node ID located in the previous step in the database to find the on-chain evidence record corresponding to the node, and obtain its evidence transaction hash. Through the PRC interface of the blockchain node, query the detailed information of the transaction according to the transaction hash, including the block height, block timestamp, and transaction content. Parse the transaction content to read the behavior log hash of the node stored therein. Calculate the hash value of the original behavior log of the node and compare it with the hash value read on the chain; if they are consistent, it proves that the log has not been tampered with since the evidence was stored, and the authorized behavior recorded is real and valid.

[0132] Sixth step, dynamic traceability result generation. Automatically integrate the traceability results of all previous steps: global fingerprint comparison results to prove whether the product as a whole has been copied; parsed identifiers and their associated verification results to prove the association between the infringing product and a specific derivative behavior; map traceability path and source node information to show the complete flow path of the infringing data and locate the source; on-chain authorization results to provide tamper-proof proof of the authorized behavior of the source node. Based on these traceability results, generate the final authentication conclusion. Use a templated report engine to integrate the authentication conclusion, key evidence, and complete data flow traceability map into a detailed infringement traceability and authentication report. Store the report and generate a unique access link; through the API interface, provide the report to the data market regulator, judicial authorities, or rights holders.

[0133] The application discloses an artificial intelligence auxiliary authentication system for data element right confirmation.

[0134] While the embodiments of the application have been illustrated and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the spirit and scope of the application, which is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence assisted authentication method for data element right, characterized in that, Comprise: Data acquisition: acquire data products, ownership declaration, contract terms, and calculate the initial digital fingerprint of the data product; Ownership identification: based on a multi-modal intelligent analysis model, extract the core value elements and characteristics of the data product, quantify the value contribution weight of each link to the data product, and automatically generate a data element circulation map based on the value contribution weight; Real-time behavior monitoring and compliance verification: based on an intelligent compliance verification model, compile the contract terms into a behavior rule set, monitor all data operation behaviors in real time, and compare them with the behavior rule set to obtain compliance verification results; Value assessment: based on the value contribution weight, combined with multi-dimensional influencing factors, calculate the value proportion of each contributor, and obtain the value assessment result; Right certificate generation: generate a digital right certificate by integrating the data element circulation map, compliance verification result and value assessment result; On-chain storage: calculate the digital fingerprint of the data element circulation map, digital right certificate, compliance verification result and value assessment result and store it on the blockchain to obtain the on-chain storage result; Dynamic traceability: deep scanning of suspected infringing data products, locating the source authorization link, and obtaining the infringement traceability certification result; The construction method of the multi-modal intelligent analysis model comprises: S1: multi-modal data content analysis: deep scanning and analysis of data products to identify and extract core value elements and characteristics; extract key information from the ownership declaration and structure it; S2: cross-validation: compare and cross-validate the core value elements and characteristics with the key information in the ownership declaration, check consistency and authenticity, and mark consistent, doubtful and conflicting items to obtain the verification result; S3: value contribution quantification: decompose the formation process of the data product into independent contribution nodes, extract the characteristics of each contribution node and generate a digital representation; calculate the relevance score of the data product and each contribution node based on the attention mechanism, and obtain the quantized value contribution weight after normalization; through the value contribution weight, the original data source is traced back to generate a value contribution weight table; S4: circulation map construction: automatically generate a data element circulation map in a visual form showing the flow path of data from the original source to the final product based on the value contribution weight table, and the map content includes the contribution degree and ownership relationship of each link; S5: result output: output the ownership identification result and the data element circulation map, and provide an interpretable analysis result including the ownership verification conclusion, value contribution distribution and detailed description of the map; The construction method of the data element circulation map comprises: M1: node identification and extraction: identify all entities participating in the formation of the data product from the value contribution weight table and classify them into different types of nodes; the different types include data source nodes, processing link nodes and data product nodes; M2: relationship definition and connection: define the directed relationship between nodes according to the actual flow and processing order of data, and connect them to form a directed network; M3: attribute assignment and map formation: assign key attribute information to each node and relationship edge to generate a data element circulation map; M4: Dynamic update and maintenance: When the data product source, processing flow and ownership change, update and version manage the graph according to the new analysis results.

2. The artificial intelligence assisted authentication method of data element right according to claim 1, characterized in that: The method for constructing the intelligent compliance verification model comprises: N1: Rule compilation: read and parse the contract terms, identify the key compliance elements, convert them into structured machine-readable rules, and combine them into a complete set of behavior rules, which are bound to the basic information of this authorization; N2: Digital authentication credential generation: based on the behavior rule set and the basic information of the authorization, generate a digital authentication credential; N3: Behavior monitoring and identifier embedding: capture and record various operation behaviors on the data product in real time, and generate a behavior log; when the authorized party processes the data product based on compliance and generates a derivative data product, write the unique identifier of the generated digital authentication credential into the derivative data product; N4: Compliance verification and result output: extract key operation features from the behavior log and compare them with the behavior rule set to determine whether the current captured data operation behavior is allowed, and output the compliance verification result.

3. The artificial intelligence assisted authentication method of data element ownership according to claim 2, characterized in that: The method for behavior monitoring and identifier embedding comprises: Behavior monitoring: intercept and review operation requests in real time, capture key information of operation behavior according to predefined rules, format the information to obtain behavior log entries, and compare them with the behavior rule set; if a violation of operation behavior is found, an alarm is triggered immediately and a response strategy is executed; Identifier embedding: create a unique identifier associated with the current derivative behavior log record during the generation of a derivative data product, and write it into the derivative data product; create a new node representing the derivative data product in the data element circulation graph, establish a generation relationship edge from the data product to the derivative data product, and associate the behavior log of the current derivative behavior with the new node.

4. The artificial intelligence aided authentication method of data element right according to claim 1, characterized in that: The method for value assessment comprises: P1: Basic contribution weight extraction: extract the value contribution weight of each node to the data product from the data element circulation graph to obtain the basic contribution weight; P2: Determine multi-dimensional influencing factors: identify and define other factors that affect value allocation in addition to contribution, and develop quantifiable indicators; P3: Comprehensive value contribution calculation: combine the basic contribution weight with the multi-dimensional influencing factors, assign adjustment coefficients to the multi-dimensional influencing factors, and calculate the adjusted comprehensive value contribution of each node; P4: Aggregation and distribution: map each node in the data element circulation graph to a contributor, and integrate the comprehensive value contribution of all nodes belonging to the same contributor to obtain the total value contribution of the contributor; Normalize the total value contribution of each contributor to calculate the value allocation proportion of each contributor.

5. The artificial intelligence aided authentication method of data element right according to claim 1, characterized in that: The method for dynamic tracing comprises: Q1: Fingerprint extraction and comparison: calculate the digital fingerprint of the suspected infringing data product, and compare it with the initial digital fingerprint to obtain the global fingerprint comparison result; if they are completely matched, the dynamic tracing result is generated directly; if they are not matched, go to the next step; Q2: Scan and analyze identifier: deeply scan the content of the suspected infringing product and analyze its internal identifier; Q3: Verify the association between the warrant and the graph by cross-verifying the extracted identifier with the identification information recorded in the digital rights certificate and the data element circulation graph; Q4: Trace the graph: Based on the matched identifier, perform reverse tracing on the data element circulation graph to locate the original authorized node of the data component and obtain the graph trace result; Q5: On-chain storage audit: Based on the graph trace result, query the behavior logs of the nodes stored on the blockchain to obtain the on-chain authorization result; Q6: Generate dynamic tracing results: Combine the global fingerprint comparison results, graph trace results, and on-chain authorization results to obtain the infringement trace authentication results.

6. An artificial intelligence assisted authentication system for data element provenance, characterized in that, It includes: Data collection module: used to obtain data products, ownership declaration, contract terms, and calculate product digital fingerprints; Intelligent right confirmation module: used to build a multi-modal intelligent analysis model, extract the core value elements and characteristics of data products, quantify the value contribution weight of each link to the formation of data products, and automatically generate a data element circulation graph based on the value contribution weight; and based on the value contribution weight, combined with multi-dimensional influencing factors, calculate the value proportion of each contributor, and obtain the value evaluation result; And generate a digital rights certificate by integrating the data element circulation graph, compliance verification result, and value evaluation result; Dynamic authentication module: used to build an intelligent compliance verification model, compile contract terms into a behavior rule set, monitor all data operation behaviors in real time, and compare them with the behavior rule set to obtain a compliance verification result; Trace evidence module: used to calculate the digital fingerprints of data element circulation graph, digital rights certificate, compliance verification result, and value evaluation result and store them on the blockchain to obtain the on-chain storage result; And for suspected infringing data products, locate the source authorization link to obtain the infringement trace authentication result; The construction method of the multi-modal intelligent analysis model includes: S1: Multi-modal data content analysis: deeply scan and analyze the data products, identify and extract the core value elements and characteristics; extract key information from the ownership declaration and structure it; S2: Cross-verification: compare and cross-verify the core value elements and characteristics with the key information in the ownership declaration, check consistency and authenticity, and mark consistent, doubtful and conflicting items to obtain verification results; S3: Value contribution quantification: decompose the formation process of data products into independent contribution nodes, extract the characteristics of each contribution node and generate digital representation; calculate the relevance score of data products and each contribution node based on the attention mechanism, and obtain the quantized value contribution weight after normalization; through the value contribution weight, trace back to the original data source to generate a value contribution weight table; S4: Circulation graph construction: automatically generate a data element circulation graph that visually displays the flow path of data from the original source to the final product based on the value contribution weight table, and the graph content includes the contribution degree and ownership relationship of each link; S5: Result output: output the ownership identification result and data element circulation graph, and provide interpretable analysis results, including ownership verification conclusion, value contribution distribution, and detailed explanation of the graph; The method for constructing the data element flow map comprises the following steps: M1: node identification and extraction: identify all entities participating in the formation of data products from the value contribution weight table, and classify them into different types of nodes; the different types include data source nodes, processing link nodes, and data product nodes; M2: relationship definition and connection: define the directed relationships between nodes according to the actual data flow and processing sequence, and connect them to form a directed network; M3: attribute assignment and map formation: assign key attribute information to each node and relationship edge to generate a data element flow map; M4: dynamic updating and maintenance: when the data product source, processing flow, and ownership change, update and version manage the map according to the new analysis results.

Images