An AI-based Internet property rights transaction risk assessment and decision-making system

Abstract

This invention provides an artificial intelligence-based internet property rights transaction risk assessment and decision-making system, relating to the field of data processing technology. The system includes: a data acquisition and preprocessing module for real-time acquisition of raw transaction data, followed by cleaning and standardization; a multi-dimensional feature construction module for modeling by time windows and constructing a multi-level interaction relationship graph to obtain time-series feature data and graph-related feature data; and for normalizing and weighting the features; a risk identification module for extracting suspicious transfer behavior chains and tracing the flow of benefits, performing composite risk discrimination; and an intelligent decision-making module for performing hierarchical adaptive discrimination and case migration comparison, summarizing and generating risk assessment results and pushing early warnings; and receiving feedback from the management terminal, which is then used to optimize identification and decision-making parameters. This invention improves the autonomy and accuracy of property rights transaction risk assessment and decision-making.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an artificial intelligence-based Internet property rights transaction risk assessment and decision-making system. Background Technology

[0002] In existing technologies, risk assessment for internet intellectual property transactions often employs a combination of rule-based risk models and manual review. Taking intellectual property trading platforms as an example, the system typically relies on historical data to formulate risk factors and uses manually set thresholds to screen transaction information. When the platform detects abnormal transaction behavior, such as large-scale transfers or frequent changes in holder information within a short period, it triggers a manual review process to determine the risk level of the transaction, thus achieving initial risk control.

[0003] However, the above methods suffer from insufficient accuracy when facing large-scale and complex transaction scenarios. Therefore, improving the intelligence and adaptability of risk assessment has become an urgent technical challenge. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based risk assessment and decision-making system for internet property rights transactions, aiming to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] An artificial intelligence-based risk assessment and decision-making system for internet property rights transactions, the system comprising:

[0007] The data acquisition and preprocessing module is used to collect raw transaction data from the Internet property rights trading platform in real time, and to clean and standardize the raw transaction data to obtain standardized transaction data. The raw transaction data includes information on the property rights, historical records of the trading parties, and information on related parties.

[0008] The multi-dimensional feature construction module is used to model and construct a multi-level interaction relationship graph based on standardized transaction data according to time windows, thereby obtaining time series feature data and graph-related feature data; and to normalize and weight the features to form fused feature data.

[0009] The risk identification module is used to extract suspicious transfer behavior chains and trace the flow of benefits based on fused feature data and multi-level interaction relationship graphs, perform composite risk judgment, and generate preliminary risk data.

[0010] The intelligent decision-making module is used to perform hierarchical adaptive discrimination and case migration comparison based on preliminary risk data, summarize and generate risk assessment results and push early warnings; it also receives feedback from the management terminal and uses it to optimize identification and decision parameters.

[0011] Preferably, the multidimensional feature construction module includes:

[0012] The time window modeling unit is used to group transaction sequences according to a set time window based on standardized transaction data, and to perform sliding statistics and trend analysis to obtain time series feature data.

[0013] The relationship graph generation unit is used to construct a multi-level interactive relationship graph with the property rights object as the core based on standardized transaction data. The nodes of the graph represent different participants, the edges represent transfer types and time tags, and the unit calculates the correlation between nodes to generate graph association feature data.

[0014] The feature fusion weighting unit is used to concatenate time series feature data with map-related feature data, and then weight the data by combining the feature importance weights of abnormal cases in the historical case library to obtain fused feature data.

[0015] Preferably, the risk identification module includes:

[0016] The suspicious behavior chain extraction unit is used to extract continuously occurring suspicious transfer behavior chains based on the fused feature data, using recursive traversal and sequence encoding methods, and generate behavior chain feature data.

[0017] The profit flow tracing unit is used to trace key nodes step by step based on behavioral chain characteristic data and multi-level interaction relationship graphs, mark suspected profit flows, and generate anomaly markers when transferring across domains or platforms to obtain tracing risk data.

[0018] The composite risk discrimination unit is used to logically combine behavioral chain feature data and source tracing risk data for discrimination. When multiple high-risk factors are met at the same time, a risk superposition and grading mechanism is executed to generate preliminary risk data.

[0019] Preferably, the intelligent decision-making module includes:

[0020] The hierarchical adaptive discrimination unit is used to calculate risk scores based on preliminary risk data, adopts an automatic matching processing strategy based on a hierarchical threshold system, and adaptively updates threshold parameters when a new abnormal pattern is detected to generate hierarchical risk results.

[0021] The case transfer learning unit is used to compare the hierarchical risk results with the historical case library one by one. If a high-confidence similar case exists, the corresponding treatment result is directly output. If no match is found, the current risk characteristics are stored in the case library, and an entry to be labeled is generated for the case library. When externally input labeled data is received, the case library is automatically updated to maintain accuracy and a case comparison result is generated.

[0022] The assessment summary and output unit is used to summarize and assess the stratified risk results and the case comparison results, generate risk assessment results, and push corresponding early warning information to the management terminal.

[0023] The feedback loop unit is used to receive the handling feedback from the management terminal and input the handling feedback back to the risk identification module and the intelligent decision-making module respectively, so as to continuously optimize the identification and decision parameters and form a closed-loop self-learning mechanism.

[0024] Preferably, the relation graph generation unit includes:

[0025] The entity standardization processing subunit is used to perform cross-platform entity identification and deduplication of participants based on standardized transaction data, and generate a unified set of nodes;

[0026] The temporal and domain labeling subunit is used to write timestamps and regional or platform source identifiers on the edges based on a unified set of nodes and transaction records, and to write role type and risk prior labels in the nodes to form an initial relationship graph with temporal and domain attributes.

[0027] The correlation calculation subunit is used to statistically analyze the interaction characteristics between nodes within a preset hop limit based on the initial relationship graph, calculate the number of interactions, freshness, role similarity and path credibility, and generate correlation scores for node pairs to form a weighted relationship graph.

[0028] The graph incremental update subunit is used to incrementally update the weighted relation graph based on newly arrived transaction data, generate a valid graph snapshot that conforms to the sliding time window, and invalidate edges that exceed the window period according to rules, so as to ensure the real-time performance and effectiveness of the graph.

[0029] Preferably, the feature fusion weighting unit includes:

[0030] The feature weight allocation subunit is used to assign learnable weights to time series feature data and graph-related feature data based on transaction type, transaction amount, data completeness and recent time features, and generate a weighted feature vector.

[0031] The case prior guidance subunit is used to compare the similarity of the weighted feature vector with the feature clusters of abnormal cases in the historical case library. When the similarity is higher than the threshold, the weight allocation result is biased and adjusted to generate a prior-guided feature vector.

[0032] The robust normalization processing subunit is used to perform median scaling and extreme value truncation on outliers in the feature vector guided by prior knowledge, so as to obtain normalized robust feature data.

[0033] The fusion calibration subunit is used to concatenate time series features and spectral correlation features based on the normalized robust feature data to obtain fused feature data.

[0034] Preferably, the suspicious behavior chain extraction unit includes:

[0035] The transaction sequence segmentation subunit is used to divide continuous transaction events into time sequences based on the fused feature data, and divide them into candidate subsequences according to changes in transfer amount, abnormal transaction frequency, and cross-domain occurrence conditions to obtain candidate behavior chain data;

[0036] The behavior chain feature encoding subunit is used to progressively encode the node role, transfer direction, transaction freshness and node hop length according to the candidate behavior chain data, and generate a multi-dimensional behavior chain feature vector.

[0037] The anomaly accumulation determination subunit is used to calculate the risk accumulation value of continuous transactions step by step based on the multi-dimensional behavior chain feature vector. When the risk accumulation value exceeds the dynamic threshold, the corresponding candidate behavior chain is marked as a suspicious behavior chain, and behavior chain feature data is generated.

[0038] Preferably, the benefit flow tracing unit includes:

[0039] The key node filtering subunit is used to identify nodes that participate in multiple transactions in the relationship graph based on behavioral chain feature data, and to filter them according to the concentration of transaction amount and the weight of node correlation to generate a set of key nodes.

[0040] The trajectory backtracking construction subunit is used to backtrack along the transaction link step by step based on the set of key nodes, starting from each key node, and record the node role type, interaction frequency and regional label to form the trajectory data of the flow of benefits.

[0041] The abnormal trajectory labeling subunit is used to generate abnormal labels for abnormal paths based on the profit flow trajectory data. When it is detected that the transfer path crosses different regions, different platforms or has multiple levels of concealed transfer, the abnormal path is used as source tracing risk data.

[0042] Preferably, the hierarchical adaptive discrimination unit includes:

[0043] The risk score generation subunit is used to calculate and output a risk score sequence based on preliminary risk data, the abnormality of transaction amount, the historical risk labels of participants, and time span factors.

[0044] The stratified threshold comparison subunit is used to compare risk scores layer by layer according to the risk score sequence, set threshold intervals according to at least three risk levels, and generate stratified discrimination results.

[0045] The threshold adaptive adjustment subunit is used to adjust the upper and lower limits of the threshold based on the stratification judgment results. When the risk score is detected to continuously deviate from the preset threshold range, the unit writes the updated threshold parameters into the decision parameter library to generate adaptively updated stratified risk results.

[0046] Preferably, the case transfer learning unit includes:

[0047] The feature similarity comparison subunit is used to extract key feature vectors based on the hierarchical risk results and compare them one by one with the feature set in the historical case library to form a candidate similar case set;

[0048] The confidence level comprehensive judgment subunit is used to calculate a comprehensive score based on the candidate similar case set, combined with the similarity score and the case handling accuracy. When the comprehensive score exceeds the preset threshold, the corresponding handling result is output and a case judgment result is formed.

[0049] The case library expansion and update subunit is used to write the current risk characteristic data into the case library and generate entries to be labeled when no high-confidence cases are matched based on the case judgment results. When externally input labeled data is received, the case library is expanded and corrected to generate case comparison results.

[0050] The above-described solution of the present invention has at least the following beneficial effects:

[0051] First, the data acquisition and preprocessing module can collect raw transaction data in real time from internet property rights trading platforms, and clean and standardize the data to obtain unified, standardized transaction data. Compared with existing technologies that rely on manually set thresholds and single-rule screening, this module can automatically process raw data from different platforms and in different formats, reducing data redundancy and noise interference, and ensuring that the input information for subsequent risk identification is more accurate and complete. For example, when the same transaction entity appears on different platforms with different names or registration codes, the system can automatically perform normalization processing to avoid risk omissions due to data inconsistencies.

[0052] Secondly, through a multi-dimensional feature construction module, time-series features and multi-level interaction relationship graphs can be built based on standardized transaction data, and these graphs are then normalized, weighted, and spliced ​​together to form fused feature data. This approach overcomes the limitations of existing technologies that rely solely on rule factors, reflecting not only the dynamic changes in transaction behavior over time but also revealing the complex relationships between different transaction entities. For example, when a patent is transferred across domains multiple times in a short period, the system can simultaneously detect frequency anomalies and cross-platform interactions, thereby enhancing its ability to characterize hidden interest chains.

[0053] Furthermore, through the risk identification module, based on fused feature data and interaction relationship graphs, suspicious transfer behavior chains can be automatically extracted and the flow of benefits traced. Combined with composite risk assessment, preliminary risk data is generated. Unlike existing methods that trigger manual review based on fixed rules, this invention can automatically identify potential benefit flows in transaction chains and, when high-risk factors are superimposed, form a higher level of risk assessment. For example, when transactions exhibit frequent cross-platform transfers accompanied by eventual return to the same affiliated group, the system can promptly generate preliminary risk data, achieving early warning.

[0054] Finally, the intelligent decision-making module enables hierarchical adaptive discrimination and case migration comparison based on preliminary risk data, summarizes and generates risk assessment results, pushes early warning information, and receives handling feedback from the management terminal to achieve parameter optimization and closed-loop self-learning. Compared with the manual review of existing technologies, this mechanism not only improves the automation level of assessment and decision-making, but also continuously optimizes the discrimination criteria and case library content as new risk patterns emerge. For example, when the system first identifies a new type of transfer evasion behavior and confirms it through manual annotation, it will automatically update the model parameters and case library, enabling similar risks to be quickly discovered subsequently.

[0055] Therefore, this invention can effectively overcome the shortcomings of existing technologies in terms of accuracy in large-scale and complex transaction scenarios, realize real-time monitoring and accurate risk assessment of Internet property rights transactions, improve the security and compliance of property rights transfer, and significantly enhance the intelligence and adaptability of risk assessment. Attached Figure Description

[0056] Figure 1 This is an architecture diagram of an internet property rights transaction risk assessment and decision-making system based on artificial intelligence, provided by an embodiment of the present invention. Detailed Implementation

[0057] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0058] like Figure 1 As shown, an embodiment of the present invention proposes an artificial intelligence-based internet property rights transaction risk assessment and decision-making system, the system comprising:

[0059] The data acquisition and preprocessing module is used to collect raw transaction data from the Internet property rights trading platform in real time, and to clean and standardize the raw transaction data to obtain standardized transaction data. The raw transaction data includes information on the property rights, historical records of the trading parties, and information on related parties.

[0060] The multi-dimensional feature construction module is used to model and construct a multi-level interaction relationship graph based on standardized transaction data according to time windows, thereby obtaining time series feature data and graph-related feature data; and to normalize and weight the features to form fused feature data.

[0061] The risk identification module is used to extract suspicious transfer behavior chains and trace the flow of benefits based on fused feature data and multi-level interaction relationship graphs, perform composite risk judgment, and generate preliminary risk data.

[0062] The intelligent decision-making module is used to perform hierarchical adaptive discrimination and case migration comparison based on preliminary risk data, summarize and generate risk assessment results and push early warnings; it also receives feedback from the management terminal and uses it to optimize identification and decision parameters.

[0063] In this embodiment of the invention, the Internet property rights transaction risk assessment and decision-making system can collect raw transaction data from the Internet property rights transaction platform in real time through a data acquisition and preprocessing module. The system then cleans and standardizes the collected property rights information, transaction history records of both parties, and related party data to obtain unified and standardized transaction data. This process eliminates data format differences between different platforms, reduces data redundancy and noise interference, and thus provides accurate and reliable basic data for subsequent risk identification. For example, when the same transaction entity uses different recording methods on different platforms, the system can automatically unify them, avoiding missed detections due to data inconsistency.

[0064] The multi-dimensional feature construction module can model standardized transaction data according to a set time window, and further construct a multi-level interaction relationship graph. This yields time-series feature data and graph-related feature data, which are then normalized and weighted and concatenated to form fused feature data. This processing method not only captures the continuity of transaction behavior over time but also depicts the complex relationships between transaction entities. For example, if an entity frequently transfers patent rights with multiple related companies within a short period, the system can identify the frequency anomalies through time-series features and reveal its potential interest connections with other nodes through graph features.

[0065] The risk identification module can automatically extract suspicious transfer behavior chains based on fused feature data and multi-level interaction relationship graphs, trace the flow of benefits, and generate preliminary risk data by combining composite risk discrimination. This function can overcome the limitations of traditional rule-based screening and has the ability to discover hidden abnormal patterns. For example, if a transaction chain shows multiple transfers across platforms, and the transferees all have hidden equity relationships with a specific group, the system can automatically identify the behavior chain and mark potential paths of concentrated benefits, thereby achieving early risk warning.

[0066] The intelligent decision-making module can perform tiered adaptive discrimination based on preliminary risk data and conduct migration comparisons with a historical case database, ultimately generating risk assessment results and pushing early warning information. This module can also receive feedback from the management terminal and input the feedback results back into the risk identification and decision-making modules, thereby achieving continuous parameter optimization and self-learning. Through this closed-loop mechanism, the system can continuously improve its ability to identify new risk patterns. For example, when managers confirm that a previously unseen evasion transfer pattern is an abnormal behavior, the system will automatically absorb this annotation information to adjust the discrimination threshold and case database content, thereby quickly identifying similar risks in subsequent transactions.

[0067] Through the synergistic effect of the above modules, the embodiments of the present invention can effectively improve the intelligence and adaptability of risk assessment for Internet property rights transactions, overcome the problem of insufficient identification of complex transaction scenarios by traditional rule-based methods, realize real-time monitoring and accurate early warning of large-scale, multi-platform property rights transactions, and improve the security and compliance of the property rights transfer process.

[0068] The data acquisition and preprocessing module specifically includes:

[0069] First, raw transaction data, including information on the property rights being traded, historical transaction records of both parties, and related party information, is collected in real time by calling the API of the internet property rights trading platform or through a web crawling mechanism. During the collection process, the data is cached in timestamp order to ensure the continuity of the transaction timeline for subsequent analysis.

[0070] Then, the collected raw data is cleaned to remove null values, outliers, and duplicate records. For example, if a transaction entry is missing a transaction amount or is recorded twice, the system will supplement or delete that data to ensure data integrity and uniqueness.

[0071] Next, the cleaned data undergoes standardization. This standardization process includes converting transaction amounts from different platforms into the same currency and unit, standardizing the format of participant names, and applying consistent storage rules to fields such as contact information and registration codes. For example, if a transaction amount is expressed in US dollars, it is converted to its equivalent in RMB before storage; if company names differ between "Limited Company" and "Limited Liability Company," they are uniformly grouped into a single description.

[0072] Finally, the cleaned and standardized data is stored as a standardized transaction dataset to serve as input for subsequent multidimensional feature building modules.

[0073] In a preferred embodiment of the present invention, the multidimensional feature construction module includes:

[0074] The time window modeling unit is used to group transaction sequences according to a set time window based on standardized transaction data, and to perform sliding statistics and trend analysis to obtain time series feature data.

[0075] The relationship graph generation unit is used to construct a multi-level interactive relationship graph with the property rights object as the core based on standardized transaction data. The nodes of the graph represent different participants, the edges represent transfer types and time tags, and the unit calculates the correlation between nodes to generate graph association feature data.

[0076] The feature fusion weighting unit is used to concatenate time series feature data with map-related feature data, and then weight the data by combining the feature importance weights of abnormal cases in the historical case library to obtain fused feature data.

[0077] In this embodiment of the invention, the multi-dimensional feature construction module can group standardized transaction data according to a set time window through a time window modeling unit, and perform sliding statistics and trend analysis within each group to generate time series feature data. In this way, the dynamic changes of transactions in the time dimension can be characterized. For example, when a patent is transferred more than three times in a short period of time, the system can highlight the abnormal frequency in the time series features to indicate potential risks.

[0078] The relationship graph generation unit can construct a multi-level interactive relationship graph centered on the property rights object based on standardized transaction data. Nodes represent transaction entities, and edges represent transfer relationships and time labels. By calculating the correlation between nodes, graph association feature data can be obtained. This process can intuitively present the transaction relationships and their strength between different entities. For example, when several nodes are frequently connected to each other through different types of transfer edges within a short period of time, the system can capture potential interest groups, thereby improving the visualization of implicit relationships.

[0079] The feature fusion weighting unit can concatenate time-series feature data with graph-related feature data and weight it according to the feature importance weights of abnormal cases in the historical case library to obtain fused feature data. This fusion process not only ensures the complementarity of time features and relational features, but also highlights features with higher risk correlation through weight adjustment. For example, if historical cases show that "cross-border transfer behavior" is a high-risk factor, the system will automatically increase the importance of such features during the fusion process, thereby improving the accuracy of risk identification.

[0080] Through the above processing, the embodiments of the present invention can characterize transaction behavior in multiple dimensions, retaining both time dynamic features and enhancing multi-party interaction features, thereby achieving a comprehensive quantitative description of complex transaction behavior and providing high-quality input data for subsequent risk identification.

[0081] The time window modeling unit specifically includes:

[0082] First, the standardized transaction data is sorted chronologically and then grouped according to a set time window. The time window can be set by day, week, or month, and the system can adjust it flexibly. For example, under a one-week time window, all transactions occurring within that time range will be grouped together.

[0083] Then, statistical analysis is performed within each time window, including the total number of transactions, the sum and average of transaction amounts, and the number of participants. These statistics help to identify the basic transaction characteristics within each window.

[0084] Next, a sliding statistics method is used to analyze the differences between adjacent time windows. For example, the number of transactions in the current window is compared with that in the previous window. If the increase exceeds a certain percentage of the historical average, it is marked as an abnormal transaction activity.

[0085] Furthermore, the system models the trend of characteristic changes within the time window. This trend modeling does not use formulas but is described in words: the system determines whether the changes in trading volume across multiple consecutive windows indicate an upward trend, a downward trend, or cyclical fluctuations, thereby generating trend labels.

[0086] Finally, statistical features and trend features are combined to output time series feature data, which serves as input for subsequent feature fusion and risk identification.

[0087] In a preferred embodiment of the present invention, the risk identification module includes:

[0088] The suspicious behavior chain extraction unit is used to extract continuously occurring suspicious transfer behavior chains based on the fused feature data, using recursive traversal and sequence encoding methods, and generate behavior chain feature data.

[0089] The profit flow tracing unit is used to trace key nodes step by step based on behavioral chain characteristic data and multi-level interaction relationship graphs, mark suspected profit flows, and generate anomaly markers when transferring across domains or platforms to obtain tracing risk data.

[0090] The composite risk discrimination unit is used to logically combine behavioral chain feature data and source tracing risk data for discrimination. When multiple high-risk factors are met at the same time, a risk superposition and grading mechanism is executed to generate preliminary risk data.

[0091] In this embodiment of the invention, the suspicious behavior chain extraction unit can extract continuously occurring suspicious transfer behavior chains based on fused feature data, using recursive traversal and sequence encoding, and generate behavior chain feature data. This mechanism automatically identifies long-term or high-frequency transaction chains, avoiding the inefficiency of manual screening. For example, if a patent is transferred to multiple subsidiaries within a month, the system can extract this series of transfer events into a single behavior chain and mark it as a suspicious pattern.

[0092] The profit flow tracing unit can trace key nodes step by step based on behavioral chain characteristic data and multi-level interaction relationship graphs, marking suspected profit flows and generating anomaly markers during cross-domain or cross-platform transfers, thereby obtaining source risk data. This function can reveal the flow of funds or rights hidden behind surface transactions. For example, when a patent is transferred multiple times across jurisdictions and ultimately flows back to an affiliated company of the original transferor, the system can mark this abnormal return path through profit tracing, thus assisting regulators in timely detection and prevention of circumvention.

[0093] The composite risk discrimination unit can logically combine behavioral chain feature data with source tracing risk data for discrimination. When multiple high-risk factors are simultaneously met, a risk superposition and grading mechanism is executed to generate preliminary risk data. Through this superposition and grading method, the system can accurately identify scenarios where the impact of a single factor is small but the combined presence of factors indicates significant risk. For example, a single "cross-platform transfer" may have a low risk, but if "high-frequency transfer" and "return path" are superimposed, the system will automatically increase the risk level, thereby improving the sensitivity of identification.

[0094] Through the above functions, the embodiments of the present invention can effectively identify complex chain transactions and hidden benefit transfer paths, significantly improve the ability to detect new circumvention methods, and enhance the depth and comprehensiveness of property rights transaction risk monitoring.

[0095] The composite risk assessment unit specifically includes:

[0096] First, it receives behavioral chain characteristic data, including the node roles, transfer direction, frequency, and amount of suspicious transaction chains. It also receives source tracing risk data, including cross-domain transfer markers, abnormal return path markers, and the concentration of key nodes.

[0097] Then, the behavioral chain characteristic data and the source tracing risk data are logically combined for judgment. The logical combination process is as follows: when a transaction chain is marked as a high-frequency abnormal transaction, and at the same time, a cross-domain transfer path is found in the source tracing results, the system will consider the risk level of the chain to be higher.

[0098] Next, a risk grading mechanism is implemented. The principle of risk grading is that the more high-risk factors are met, the higher the risk level becomes. For example, a single factor of "high-frequency transfer" may be classified as low risk; if the factors of "cross-domain transfer" and "related company repatriation" are combined, it will be upgraded to medium or high risk. The system will automatically output the corresponding risk level according to the preset grading rules.

[0099] Finally, the judgment results are used to generate preliminary risk data, including risk level, involved transaction chains and their triggered risk factors, so as to transmit them to the intelligent decision-making module for further hierarchical adaptive judgment and case migration comparison.

[0100] In a preferred embodiment of the present invention, the intelligent decision-making module includes:

[0101] The hierarchical adaptive discrimination unit is used to calculate risk scores based on preliminary risk data, adopts an automatic matching processing strategy based on a hierarchical threshold system, and adaptively updates threshold parameters when a new abnormal pattern is detected to generate hierarchical risk results.

[0102] The case transfer learning unit is used to compare the hierarchical risk results with the historical case library one by one. If a high-confidence similar case exists, the corresponding treatment result is directly output. If no match is found, the current risk characteristics are stored in the case library, and an entry to be labeled is generated for the case library. When externally input labeled data is received, the case library is automatically updated to maintain accuracy and a case comparison result is generated.

[0103] The assessment summary and output unit is used to summarize and assess the stratified risk results and the case comparison results, generate risk assessment results, and push corresponding early warning information to the management terminal.

[0104] The feedback loop unit is used to receive the handling feedback from the management terminal and input the handling feedback back to the risk identification module and the intelligent decision-making module respectively, so as to continuously optimize the identification and decision parameters and form a closed-loop self-learning mechanism.

[0105] In this embodiment of the invention, the hierarchical adaptive discrimination unit can calculate risk scores based on preliminary risk data and perform automatic matching processing using a hierarchical threshold system. When the system detects a new abnormal pattern, it can also automatically update the threshold parameters to form an adaptive hierarchical risk result. This dynamic update mechanism ensures that the risk discrimination criteria are continuously adjusted as the transaction pattern evolves. For example, when "cross-border short-term multiple transfers," which has never appeared in the past, becomes a new risk pattern, the system can automatically lower the discrimination threshold for this pattern, thereby capturing such risks more quickly.

[0106] The case transfer learning unit compares the stratified risk results with a historical case database one by one. If a high-confidence similar case is found, the corresponding handling result is directly output. If no match is found, the new risk characteristics are automatically stored in the case database, and the database is updated upon receiving labeled data from external input to maintain its completeness and accuracy. This transfer learning mechanism improves the system's response efficiency to known risk patterns while continuously expanding the case database to increase coverage of unknown risks. For example, when a new case emerges that is extremely similar to the historical "transfer of a false related company," the system can directly output a suggested handling action, saving time for manual judgment.

[0107] The assessment, aggregation, and output unit summarizes and evaluates the tiered risk results and case comparison results, generating the final risk assessment result and pushing it to the management terminal for rapid early warning and intervention. Simultaneously, the feedback loop unit receives feedback from the management terminal and feeds it back into the risk identification and intelligent decision-making modules, forming a closed-loop self-learning mechanism. This allows the system's parameter configuration and identification capabilities to continuously optimize with actual operation. For example, when management confirms that a low-risk marker is a misjudgment, the system automatically corrects the relevant parameters to avoid subsequent repeated misjudgments.

[0108] Through the above mechanism, the embodiments of the present invention can realize intelligent management of the entire process from risk discovery and risk assessment to handling feedback, which is both efficient and adaptive, and effectively improves the accuracy and real-time performance of risk prevention and control in Internet property rights transactions.

[0109] The evaluation summary and output unit specifically includes:

[0110] First, the system receives hierarchical risk results from the hierarchical adaptive discrimination unit, including scores for different risk levels and corresponding risk labels. Simultaneously, it receives case comparison results from the case transfer learning unit, including whether historical cases match and cited treatment recommendations.

[0111] Then, the stratified risk results and case comparison results are aggregated. The aggregation process is as follows: when the stratified risk results show a high risk level and no similar cases are found in the case comparison results, the system will increase the weight of manual review in the assessment; when the stratified risk results show a medium risk level and the case comparison results match high-confidence historical cases, the system will prioritize the handling methods of historical cases. In this way, the system can integrate the judgment results from different sources to form a more comprehensive risk assessment.

[0112] Next, the final risk assessment result is generated based on the summarized results. The generated content includes: risk level, triggering main risk factors, matching historical case numbers, and handling recommendations. If the result is high-risk, a priority label will be automatically set for rapid processing in the management terminal.

[0113] Finally, the generated risk assessment results are pushed to the management terminal via an interface and displayed visually, including text prompts, charts, and risk chain diagrams. This helps managers intuitively understand the sources and levels of risks, thereby improving the efficiency of early warning response.

[0114] The feedback closed-loop unit specifically includes:

[0115] First, the system receives processing feedback information from the management terminal. This feedback information includes the risk confirmation result (such as confirmation of anomaly or misjudgment), the processing measures taken (such as suspending the transaction, manual review, or releasing the transaction), and the feedback time.

[0116] Then, the feedback is transformed into structured feedback data. The structuring process includes: extracting the risk factors identified in the feedback, marking whether they are real risks, and recording the final handling results. For example, when the management terminal confirms that a certain transaction chain is a "fake transfer," the system will mark the chain as high-risk in the feedback data and add the handling measure "transaction suspension."

[0117] Next, the feedback data is fed back into the risk identification module to adjust the weights of risk factors and the identification strategy for abnormal links. For example, if a certain type of behavioral chain is marked as a misjudgment in multiple feedbacks, the system will reduce the sensitivity of that type of factor to reduce subsequent false alarms.

[0118] Furthermore, the feedback data is fed back into the intelligent decision-making module to optimize the stratification thresholds and case library content. For example, when a new risk pattern first appears and is confirmed by the management terminal, the system automatically writes its features into the case library and adds corresponding discrimination conditions to the stratification thresholds, enabling subsequent similar patterns to be quickly identified.

[0119] Finally, a closed-loop self-learning mechanism is established. This mechanism ensures that the system can continuously optimize its model and parameters based on actual feedback after each risk assessment and handling, thereby improving the system's adaptability and accuracy in new scenarios.

[0120] In a preferred embodiment of the present invention, the relationship graph generation unit includes:

[0121] The entity standardization processing subunit is used to perform cross-platform entity identification and deduplication of participants based on standardized transaction data, and generate a unified set of nodes;

[0122] The temporal and domain labeling subunit is used to write timestamps and regional or platform source identifiers on the edges based on a unified set of nodes and transaction records, and to write role type and risk prior labels in the nodes to form an initial relationship graph with temporal and domain attributes.

[0123] The correlation calculation subunit is used to statistically analyze the interaction characteristics between nodes within a preset hop limit based on the initial relationship graph, calculate the number of interactions, freshness, role similarity and path credibility, and generate correlation scores for node pairs to form a weighted relationship graph.

[0124] The graph incremental update subunit is used to incrementally update the weighted relation graph based on newly arrived transaction data, generate a valid graph snapshot that conforms to the sliding time window, and invalidate edges that exceed the window period according to rules, so as to ensure the real-time performance and effectiveness of the graph.

[0125] In this embodiment of the invention, the entity standardization processing subunit can perform cross-platform entity identification and deduplication of participants based on standardized transaction data, generating a unified node set. Through this processing, the system can eliminate duplicate records caused by differences in name spelling, registration codes, or contact information. For example, when the same company appears on different platforms as "XX Technology Co., Ltd." and "XX Technology Limited Liability Company," the system can identify it as the same entity through multi-feature comparison, thereby avoiding the problems of node dispersion and inaccurate identification during risk analysis.

[0126] The temporal and domain labeling subunit can write timestamps and regional or platform source identifiers onto the edges of the relationship graph based on a unified set of nodes and transaction records, and write role type and risk prior labels into the nodes, forming an initial relationship graph with temporal and domain attributes. In this way, the system can intuitively reflect the chronological order of transactions and their respective jurisdictions within the graph. For example, when the same patent is first transferred domestically and then transferred on an overseas platform, the system can clearly record the cross-regional transaction path, facilitating subsequent risk tracing.

[0127] The correlation calculation subunit can statistically analyze the interaction characteristics between nodes within a preset hop limit based on the initial relationship graph. This includes the number of interactions, the freshness of the most recent interaction, role similarity, and multi-hop path credibility. Based on this, it generates correlation scores for node pairs, thus forming a weighted relationship graph. Through this weighting mechanism, the system can more accurately reveal the closeness between trading entities. For example, even if two companies don't directly trade often, but frequently transact indirectly through multiple layers of intermediaries, the system will assign a higher correlation score based on path credibility calculation, thereby revealing hidden interests.

[0128] The incremental graph update subunit can incrementally update the weighted relation graph based on newly arrived transaction data and generate a valid graph snapshot that conforms to the sliding time window. Simultaneously, edges exceeding the window period are invalidated according to rules to ensure the real-time nature and effectiveness of the graph. In this way, the system avoids the inefficiency of rebuilding the entire graph every time new data arrives, enabling rapid response to real-time transaction dynamics. For example, on a large intellectual property trading platform, tens of thousands of new transfer records may be added daily. The system can update only the newly added portions and generate the latest snapshot, ensuring the timeliness and efficiency of risk assessment.

[0129] The entity standardization processing subunit specifically includes:

[0130] First, extract the participant information from the transaction data, including name, registration code, contact information, and registered address.

[0131] Then, to address the differences in records across different platforms, entity identification is performed using methods such as name fingerprint matching, character similarity comparison, and unified code verification. For example, when "XX Technology Co., Ltd." and "XX Technology Limited Liability Company" appear simultaneously, they can be identified as the same entity by removing redundant words and comparing registration codes.

[0132] Next, duplicate participants are deduplicated and assigned a unique identifier to each participant, forming consistent node information across platforms.

[0133] Finally, a unified set of nodes is generated and stored in the node library of the relation graph for subsequent annotation and calculation.

[0134] The temporal and domain labeling subunit specifically includes:

[0135] First, extract the time information of each transfer event from the transaction records and write it into the transaction edge as a timestamp label to reflect the order in which the transactions occurred.

[0136] Then, based on the platform or jurisdiction from which the transaction originates, a region or platform identifier is added to the transaction edge. For example, when a transfer occurs on a domestic platform, the edge is marked as "Domestic"; when it occurs on an overseas platform, it is marked as "Overseas".

[0137] Next, role types are written for participating nodes, such as "transferor", "transferee" and "intermediary". Based on historical data or existing tags, some nodes are assigned risk prior labels, such as "high-risk entity" or "abnormally frequent transaction party".

[0138] Finally, the graph with time attributes, geographical attributes, and node labels is output to form an initial relationship graph, providing input data for subsequent association calculations.

[0139] The correlation calculation subunit specifically includes:

[0140] First, select any two nodes in the graph, and within the set hop count range, count the number of direct interactions and the number of paths between them.

[0141] Then, the freshness of the interaction is determined based on the time interval between the most recent interaction and the current interaction. The shorter the time interval, the higher the interaction weight; the longer the interval, the lower the interaction weight.

[0142] Next, the role similarity between the two nodes is compared. For example, when both are transferors, the role similarity score is low; if one is a transferor and the other is a transferee, the role complementarity is strong and the score is high.

[0143] Furthermore, the reliability of multi-hop paths is analyzed. When there are multiple intermediate nodes between two nodes, the reliability decreases proportionally with each subsequent hop. For example, when two nodes are connected through three layers of intermediaries, the contribution of this path to the final association degree is significantly less than that of a directly connected path.

[0144] Finally, the weighted results of interaction count, freshness, role similarity and path credibility are normalized to generate a correlation score for node pairs, and a weighted relationship graph is formed accordingly.

[0145] The map incremental update subunit specifically includes:

[0146] First, when new transaction data arrives, it is mapped to the existing weighted relation graph. If new nodes are involved, new nodes are added; if existing nodes are involved, the interaction attributes of their edges are updated.

[0147] Then, based on the sliding time window mechanism, only transaction edges within the window range are retained. For example, when the time window is set to six months, the system will automatically delete transaction edges that have not occurred for more than six months.

[0148] Next, the retained edges are recalculated, including updating the interaction freshness, readjusting the path credibility, and generating new correlation scores to ensure that the scores always reflect the current transaction status.

[0149] Finally, the updated valid map snapshot is output as input data for subsequent risk identification and decision analysis, avoiding the need to rebuild the entire map with each data update and improving system efficiency and real-time performance.

[0150] In a preferred embodiment of the present invention, the feature fusion weighting unit includes:

[0151] The feature weight allocation subunit is used to assign learnable weights to time series feature data and graph-related feature data based on transaction type, transaction amount, data completeness and recent time features, and generate a weighted feature vector.

[0152] The case prior guidance subunit is used to compare the similarity of the weighted feature vector with the feature clusters of abnormal cases in the historical case library. When the similarity is higher than the threshold, the weight allocation result is biased and adjusted to generate a prior-guided feature vector.

[0153] The robust normalization processing subunit is used to perform median scaling and extreme value truncation on outliers in the feature vector guided by prior knowledge, so as to obtain normalized robust feature data.

[0154] The fusion calibration subunit is used to concatenate time series features and spectral correlation features based on the normalized robust feature data to obtain fused feature data.

[0155] In this embodiment of the invention, the feature weight allocation subunit can assign learnable weights to time series feature data and graph-related feature data based on transaction type, transaction amount, data completeness, and recent time characteristics, and generate a weighted feature vector. In this way, the system can automatically highlight features that have a greater impact on risk identification. For example, for cross-border transfers with particularly high amounts, the system can automatically assign higher weights to the amount feature, thereby improving risk perception capabilities.

[0156] The case-guided subunit compares the weighted feature vector with the feature clusters of abnormal cases in the historical case database. When the similarity exceeds a threshold, the weight allocation is adjusted to generate a priori-guided feature vector. Through this mechanism, the system can fully utilize historical knowledge and integrate empirical information into current feature processing. For example, when a certain combination of transaction features is highly similar to a historical case of "shell company patent transfer," the system automatically amplifies the impact of such features, thereby quickly alerting potential risks.

[0157] The robust normalization subunit can perform median scaling and extreme value truncation on outliers in the feature vector guided by prior knowledge, thereby obtaining normalized robust feature data. This processing can reduce the interference of noise or extreme data on the overall feature fusion result. For example, when a transaction amount is abnormally high but is only an isolated case, the system will weaken its excessive influence on the overall model through extreme value truncation, maintaining the stability of risk assessment.

[0158] The fusion calibration subunit can concatenate time-series features with spectral correlation features based on normalized robust feature data to obtain fused feature data. Simultaneously, it performs monotonic transformation calibration on the output distribution of the fused feature data, ensuring that the risk score remains consistent with the actual frequency of occurrence across different batches, and generates interpretable records containing feature contribution and influence direction. This mechanism not only improves model stability but also enhances the interpretability of the results. For example, when the system determines a transaction to be high-risk, it can simultaneously output "frequent cross-border transfers" and "high correlation with high-risk nodes" as primary evidence, making it easier for regulators to understand and accept.

[0159] Through the synergistic effect of the above sub-units, the embodiments of the present invention can achieve dynamic weight allocation, prior knowledge guidance, robust normalization and output calibration in the process of multi-dimensional feature fusion, which significantly improves the accuracy and interpretability of risk assessment.

[0160] The feature weight allocation subunit specifically includes:

[0161] First, the input time series feature data and graph-related feature data are collected and integrated to ensure that the time information, transaction amount, number of transaction entities, etc. corresponding to each transaction event are extracted.

[0162] Then, based on the preset learnable model parameters, the importance of different features is calculated sequentially. During the calculation process, transaction type (such as cross-border transfer, frequent transfer), transaction amount, data completeness, and time characteristics are used as the main factors and assigned different initial weights.

[0163] Next, during the training or operation phase, the system continuously adjusts the weight values ​​to make the weight allocation results more closely resemble historical risk identification experience. For example, for high-value cross-border transactions, the system will gradually learn to assign higher weights to transaction amount and geographical characteristics.

[0164] Finally, the assigned weights are multiplied by the original feature vector to obtain a new weighted feature vector, which is then output to the next sub-unit.

[0165] The case-based prior guidance subunit specifically includes:

[0166] First, retrieve the stored set of abnormal case features from the historical case database. Each case cluster contains a combination of important features that have shown high risk in past transactions.

[0167] Then, the weighted feature vector is compared with each of these historical case clusters to calculate the similarity. The calculation process uses a similarity measure based on feature dimensions, which can be described as follows: when the differences between several key features in the weighted vector and the corresponding features in the historical cases are small, the similarity score is higher.

[0168] Next, when the similarity score exceeds a preset threshold, the system will adjust the weighting of the current feature by referencing the importance ranking of high-risk features in the historical cases. For example, if the case shows that "multiple cross-platform transfers in a short period of time" is a high-risk factor, the system will increase the weight of this factor in the weighted vector.

[0169] Finally, the bias-adjusted prior guidance feature vector is output to play a stronger role in risk identification in subsequent processing stages.

[0170] The robust normalization processing subunit specifically includes:

[0171] First, collect all feature components in the current feature vector and calculate their overall distribution, focusing on analyzing the extreme values ​​and deviations.

[0172] Then, using the median of the feature components as the central benchmark, all feature values ​​are scaled to make the overall feature distribution more balanced and avoid the influence of individual outliers on the overall results.

[0173] Next, the extreme values ​​at the top and bottom of the distribution are identified and gently truncated. The truncation method can be described as follows: values ​​above the upper percentile are reduced to near the upper percentile, while values ​​below the lower percentile are increased to near the lower percentile, thereby limiting the interference of extreme values ​​on the overall vector.

[0174] Finally, the results after scaling and truncation are output to form normalized robust feature data, ensuring the stability and reliability of the subsequent risk assessment process.

[0175] The integrated calibration subunit specifically includes:

[0176] First, the normalized time series features and the normalized map association features are concatenated according to the feature dimensions to obtain a complete fused feature dataset.

[0177] Next, the distribution of the fused feature data is calibrated. The calibration method is as follows: compare the risk score distribution of the current batch of fused features with the actual risk frequency in historical batches. If there is a deviation, perform a monotonic transformation on the output distribution to make it more closely match the historical probability of actual risk occurrence.

[0178] Next, during the calibration process, the system generates an interpretability record, which includes the feature with the highest contribution and the direction of its impact on the risk assessment. For example, when the amount of cross-border transfers plays a major role in improving the risk outcome, the record will clearly indicate the contribution of that feature.

[0179] Finally, the calibrated fusion feature data is used as the final output and provided to the risk identification module and intelligent decision-making module to ensure that the risk assessment is both stable and interpretable.

[0180] In a preferred embodiment of the present invention, the suspicious behavior chain extraction unit includes:

[0181] The transaction sequence segmentation subunit is used to divide continuous transaction events into time sequences based on the fused feature data, and divide them into candidate subsequences according to changes in transfer amount, abnormal transaction frequency, and cross-domain occurrence conditions to obtain candidate behavior chain data;

[0182] The behavior chain feature encoding subunit is used to progressively encode the node role, transfer direction, transaction freshness and node hop length according to the candidate behavior chain data, and generate a multi-dimensional behavior chain feature vector.

[0183] The anomaly accumulation determination subunit is used to calculate the risk accumulation value of continuous transactions step by step based on the multi-dimensional behavior chain feature vector. When the risk accumulation value exceeds the dynamic threshold, the corresponding candidate behavior chain is marked as a suspicious behavior chain, and behavior chain feature data is generated.

[0184] In this embodiment of the invention, the transaction sequence segmentation subunit can segment continuous transaction events chronologically based on fused feature data, and divide them into candidate subsequences according to changes in transfer amount, abnormal transaction frequency, and cross-domain occurrence conditions, thus obtaining candidate behavioral chain data. In this way, the system can decompose complex transaction flows into several behavioral chain segments with clear characteristics, facilitating subsequent anomaly identification. For example, when a patent is transferred multiple times consecutively within a very short period, the system will automatically classify these transfers into a candidate chain segment as a potentially suspicious object.

[0185] The behavior chain feature encoding subunit can progressively encode node roles, transfer directions, transaction freshness, and node hop length based on candidate behavior chain data, generating a multi-dimensional behavior chain feature vector. This step-by-step encoding method can capture the complex relationships within the chain, reflecting not only single events but also chain-like transmission patterns. For example, the system can discover through the feature vector that a behavior chain exhibits a cyclical pattern of "transferring from the parent company to multiple subsidiaries and then flowing back."

[0186] The anomaly accumulation judgment subunit can progressively calculate the risk accumulation value of continuous transactions based on the multi-dimensional behavioral chain feature vector. When the risk accumulation value exceeds a dynamic threshold, the corresponding candidate behavioral chain is marked as a suspicious behavioral chain, and behavioral chain feature data is generated. This judgment mechanism can effectively detect chain risks caused by multiple factors. For example, a single cross-platform transfer may not be abnormal, but if it is accompanied by frequent transfers of high amounts and forms a backflow chain, the risk accumulation value will rise rapidly, triggering a high-risk mark, thereby achieving accurate early warning.

[0187] Through the above processing, the embodiments of the present invention can transform scattered transaction events into organic behavioral chains, and reveal potential hidden risk patterns through step-by-step encoding and cumulative judgment, significantly enhancing the ability to discover complex and abnormal transaction chains.

[0188] The transaction sequence segmentation subunit specifically includes:

[0189] First, the transaction events in the fused feature data are sorted by timestamp to ensure that the transaction events maintain temporal continuity when input.

[0190] Then, based on the set time window and event interval, consecutive transaction events are divided into different segments. For example, when multiple transaction events occur within a week and involve the same or related entities, the system will group these events into the same sequence.

[0191] Next, the segmented transaction sequences are examined according to the trend of changes in transfer amounts. If a significant increase or decrease in transaction amount is found in a short period of time, it is marked as an abnormal condition.

[0192] Furthermore, the frequency of events in the sequence is analyzed. When the transaction frequency exceeds a preset threshold of the historical average, the sequence is marked as a high-frequency anomaly. If cross-platform or cross-regional transfers occur within the same chain, they will also be recorded as cross-domain anomalies in the sequence features.

[0193] Finally, transaction sequences that meet the above conditions are extracted as candidate subsequences and uniformly stored as candidate behavior chain data for subsequent encoding sub-unit processing.

[0194] The behavior chain feature encoding subunit specifically includes:

[0195] First, extract the role type of each node in the candidate behavior chain, such as "transferor", "transferee" or "intermediary", and encode it as the corresponding role label.

[0196] Then, the direction of the transaction is coded. The direction is determined by the flow path of the property rights. For example, "flow from parent company to subsidiary" is coded as downward transfer, while "flow from subsidiary back to parent company" is coded as backflow transfer.

[0197] Next, the transaction freshness is calculated and encoded. In words, the system assigns a higher freshness value to more recent transactions and a lower value to more distant transactions based on the interval between the transaction's occurrence time and the current time, thus reflecting the timeliness of the transactions.

[0198] Furthermore, the node hop count length is encoded. The hop count refers to the number of intermediate nodes a transaction passes through from the starting node to the target node. The more hops, the more complex the path, and the lower the credibility score will be when the system encodes it.

[0199] Finally, the above-mentioned role encoding, direction encoding, freshness value and jump count features are concatenated into a multi-dimensional feature vector, and output as input data for subsequent decision sub-units.

[0200] The anomaly accumulation determination subunit specifically includes:

[0201] First, the system assigns a risk score to each of the input multi-dimensional behavioral chain feature vectors. During the scoring process, the system assigns a risk contribution value to high-value transfers, frequent transactions, and cross-domain paths.

[0202] Then, these risk contribution values ​​are accumulated sequentially according to the transaction time, forming a risk accumulation value sequence. In words, the system will gradually add the risk values ​​of earlier transactions to the risk values ​​of subsequent transactions to obtain a cumulative risk curve that changes over time.

[0203] Next, the cumulative risk value is compared with a dynamic threshold. The dynamic threshold is not a fixed value, but is adjusted in real time based on historical statistical distribution and the current trading environment. For example, during periods of unusually active market trading, the threshold will automatically increase to avoid false alarms.

[0204] Finally, when the accumulated risk value exceeds the dynamic threshold, the system will mark the corresponding candidate behavior chain as a suspicious behavior chain and generate structured behavior chain feature data, including the nodes in the chain, the path, the triggering high-risk factors, and their cumulative results. This data will be passed as input to the subsequent benefit flow tracing unit and composite risk discrimination unit.

[0205] In a preferred embodiment of the present invention, the benefit flow tracing unit includes:

[0206] The key node filtering subunit is used to identify nodes that participate in multiple transactions in the relationship graph based on behavioral chain feature data, and to filter them according to the concentration of transaction amount and the weight of node correlation to generate a set of key nodes.

[0207] The trajectory backtracking construction subunit is used to backtrack along the transaction link step by step based on the set of key nodes, starting from each key node, and record the node role type, interaction frequency and regional label to form the trajectory data of the flow of benefits.

[0208] The abnormal trajectory labeling subunit is used to generate abnormal labels for abnormal paths based on the profit flow trajectory data. When it is detected that the transfer path crosses different regions, different platforms or has multiple levels of concealed transfer, the abnormal path is used as source tracing risk data.

[0209] In this embodiment of the invention, the key node screening subunit can identify nodes participating in multiple transactions in the relationship graph based on behavioral chain feature data, and filter out key nodes as the starting and ending points of the profit flow by combining transaction amount concentration and node correlation weights. In this way, the system can quickly locate participants with controlling or core influence in the transaction network. For example, when a company repeatedly appears in multiple chains and occupies a major proportion of the transaction amount, the system will mark it as a key node for subsequent focused tracking.

[0210] The trajectory backtracking subunit can trace back along the transaction chain step by step, starting from each key node and recording the role type, interaction frequency, and geographical tags of each node, forming complete data on the flow of benefits. In this way, the system can reveal the flow paths of funds or property rights between different nodes and reconstruct potential benefit transfer patterns. For example, when a key node has continuous transfer relationships with multiple overseas companies, the system can clearly map the cross-border benefit transfer path.

[0211] The abnormal trajectory labeling subunit can generate anomaly markers for transfer paths that cross different regions, platforms, or involve multi-level concealed transfers, based on the profit flow trajectory data. The labeled trajectories are then output as source risk data. Through this mechanism, the system can automatically identify and label illogical or concealed transfer paths. For example, when a patent is transferred from a domestic company to an overseas affiliated company via multiple shell companies, the system can label this path as an abnormal trajectory and output it for regulatory or auditing personnel to focus on.

[0212] Through the above functions, the embodiments of the present invention can realize the whole process of tracing from key node identification to complete trajectory construction and abnormal path labeling, which significantly improves the system's ability to reveal hidden interest flows and cross-domain risks.

[0213] The key node screening sub-unit specifically includes:

[0214] First, receive behavioral chain feature data and map the transaction entities involved into a relationship graph.

[0215] Then, the frequency of each node's appearance in the transaction chain and the total transaction amount it participated in are counted. The more times a node appears, the more active it is in the chain; the higher its share of the transaction amount, the more important it is in the transfer of funds or property rights.

[0216] Next, the node's correlation score is calculated based on the strength of its interactions with other nodes. In simpler terms: a node has a higher correlation score if it has frequent and recent interactions with multiple other nodes.

[0217] Furthermore, a weighted combination of transaction amount concentration and node relevance is used to obtain a comprehensive score. Nodes with higher comprehensive scores are selected and included in the key node set.

[0218] Finally, output the set of key nodes, which will serve as the starting point set for subsequent trajectory backtracking.

[0219] The trajectory backtracking construction sub-unit specifically includes:

[0220] First, select a node from the set of key nodes as the starting point for backtracking.

[0221] Then, the historical transaction paths of each node are retrieved sequentially by tracing back along the transaction chain. For each node visited, its role type (e.g., transferor, transferee, or intermediary), the number of interactions with the previous node, and the region or platform information to which the node belongs are recorded.

[0222] Next, if new key nodes are encountered during the backtracking process, they are included in the path, and the backtracking scope is expanded until the set backtracking depth is reached or there is no more historical transaction data.

[0223] Furthermore, all path information is integrated to generate complete benefit flow trajectory data. This trajectory not only includes the path sequence but also carries the role information, frequency information, and regional tags of each node, thus forming a traceable chain record.

[0224] Finally, the generated trajectory data is output to provide input for subsequent anomaly labeling and risk identification.

[0225] The abnormal trajectory annotation sub-unit, specifically includes:

[0226] First, receive data on the trajectory of profit flow and then search for the transfer paths within each path.

[0227] Then, check the geographical tags of the path. If the path crosses different jurisdictions or different trading platforms, it is determined to be an abnormal situation of cross-domain or cross-platform transfer.

[0228] Next, the hierarchical structure of the path is analyzed. In textual terms: when a property transfer path passes through multiple intermediate nodes, and all intermediate nodes are inactive or shell companies, the path will be marked as a concealed transfer.

[0229] Furthermore, anomaly markers are added to paths identified as abnormal. The anomaly markers include the path anomaly type (cross-domain, cross-platform, or hidden) and the corresponding node sequence.

[0230] Finally, the marked complete trajectory is output as source tracing risk data for subsequent processing by the composite risk discrimination unit.

[0231] In a preferred embodiment of the present invention, the hierarchical adaptive discrimination unit includes:

[0232] The risk score generation subunit is used to calculate and output a risk score sequence based on preliminary risk data, the abnormality of transaction amount, the historical risk labels of participants, and time span factors.

[0233] The stratified threshold comparison subunit is used to compare risk scores layer by layer according to the risk score sequence, set threshold intervals according to at least three risk levels, and generate stratified discrimination results.

[0234] The threshold adaptive adjustment subunit is used to adjust the upper and lower limits of the threshold based on the stratification judgment results. When the risk score is detected to continuously deviate from the preset threshold range, the unit writes the updated threshold parameters into the decision parameter library to generate adaptively updated stratified risk results.

[0235] In this embodiment of the invention, the risk score generation subunit can calculate and output a risk score sequence based on preliminary risk data, comprehensively considering factors such as the anomaly of transaction amount, the historical risk labels of participants, and the time span. In this way, the system can quantify transaction risk, providing an objective basis for subsequent multi-level judgments. For example, when a transaction amount is significantly higher than the platform average and the participant has a history of high-risk behavior, its risk score will be significantly increased, facilitating priority handling.

[0236] The tiered threshold comparison subunit can set threshold ranges according to at least three risk levels based on the risk score sequence, and compare the risk scores layer by layer to generate tiered discrimination results. Through this mechanism, the system can automatically match differentiated processing strategies to transactions with different risk levels. For example, low-risk transactions only need to be recorded by the system, medium-risk transactions send alerts, and high-risk transactions directly enter the manual review process, thereby achieving tiered and efficient risk handling.

[0237] The threshold adaptive adjustment subunit can dynamically adjust the upper and lower limits of the risk threshold based on the stratified discrimination results. When a risk score is detected to continuously deviate from the preset threshold range, the subunit writes the updated threshold parameters into the decision parameter library, generating adaptively updated stratified risk results. This mechanism ensures that the discrimination criteria are continuously optimized as new trading patterns emerge. For example, if a new avoidance technique recently appears, causing frequent missed detections under the traditional threshold, the system will automatically lower the threshold for this type of pattern to facilitate timely identification, thereby improving its adaptability to new risk patterns.

[0238] Through the above functions, the embodiments of the present invention can realize a complete process of risk quantification, hierarchical identification and adaptive updating, so that risk assessment has both refined levels and the ability to dynamically adjust according to the scenario, which greatly improves the accuracy and flexibility of transaction risk prevention and control.

[0239] The risk score generation subunit specifically includes:

[0240] First, we receive preliminary risk data, which includes the transaction amount, time of occurrence, basic information of the participants, and their risk labels.

[0241] Next, the anomaly score is calculated for the transaction amount. In simpler terms, the system compares the current transaction amount with the average level of similar transactions. When the transaction amount is significantly higher or lower than the average, a higher anomaly score is assigned.

[0242] Next, a score is assigned based on the participant's historical risk labels. For example, if a trading entity has been marked as high-risk multiple times in past transactions, its risk label score will be higher; if it is a new entity appearing for the first time, its risk label score will be lower.

[0243] Furthermore, adjustments are made based on the time span of the transactions. When the time interval between consecutive transactions is too short, the system will increase the risk score to reflect the risk characteristics of high-frequency trading; when the interval is long, the score will decrease accordingly.

[0244] Finally, the risk scores of each of the above factors are weighted and summed to generate a risk score for a single transaction. The risk scores of consecutive transactions are then arranged sequentially to form a risk score sequence, which is output for subsequent judgment.

[0245] The hierarchical threshold comparison sub-unit specifically includes:

[0246] First, establish at least three risk levels, such as low risk, medium risk, and high risk, and assign a corresponding threshold range to each level. In written terms: low risk corresponds to a score below the first threshold, medium risk corresponds to a score between the first and second thresholds, and high risk corresponds to a score above the second threshold.

[0247] Then, each score in the risk score sequence is compared with the threshold interval one by one. If a score falls into the low-risk interval, it is marked as low risk; if it falls into the middle interval, it is marked as medium risk; if it exceeds the highest threshold, it is marked as high risk.

[0248] Next, all the marked results are integrated to form a hierarchical judgment result. The integration method is as follows: when a transaction chain contains multiple high-risk scores, the chain as a whole is identified as high-risk; if most scores fall within the low-risk range, the whole chain is identified as low-risk.

[0249] Finally, the generated stratified discrimination results will be output and used in subsequent combinations with the case comparison module to form a more accurate risk assessment.

[0250] The threshold adaptive adjustment subunit specifically includes:

[0251] First, the system receives the stratified discrimination results and statistically analyzes the scores that fall into the boundary region. The boundary region refers to the score range that is close to the upper and lower limits of the threshold.

[0252] Next, determine if there is a large concentration of risk scores near a certain boundary. If multiple consecutive batches of transactions show scores concentrated near the upper limit of the threshold, it indicates that the current threshold setting is too low; if they are concentrated near the lower limit, it indicates that the threshold setting is too high.

[0253] Next, the thresholds are adjusted based on the concentration of risk. In simpler terms: when risk scores frequently approach the high threshold, the system will appropriately raise the upper limit of the high threshold; when scores remain at the low threshold boundary for an extended period, the system will lower the lower limit of the low threshold, thus making risk stratification more consistent with reality.

[0254] Furthermore, the adjusted threshold parameters are written into the decision parameter library and used as a new threshold standard in subsequent risk assessment tasks.

[0255] Finally, an updated tiered risk result is generated to ensure that the risk level classification can be adaptively adjusted as market conditions and trading patterns change.

[0256] In a preferred embodiment of the present invention, the case transfer learning unit includes:

[0257] The feature similarity comparison subunit is used to extract key feature vectors based on the hierarchical risk results and compare them one by one with the feature set in the historical case library to form a candidate similar case set;

[0258] The confidence level comprehensive judgment subunit is used to calculate a comprehensive score based on the candidate similar case set, combined with the similarity score and the case handling accuracy. When the comprehensive score exceeds the preset threshold, the corresponding handling result is output and a case judgment result is formed.

[0259] The case library expansion and update subunit is used to write the current risk characteristic data into the case library and generate entries to be labeled when no high-confidence cases are matched based on the case judgment results. When externally input labeled data is received, the case library is expanded and corrected to generate case comparison results.

[0260] In this embodiment of the invention, the feature similarity comparison subunit can extract key feature vectors based on the hierarchical risk results and compare them one by one with the feature set in the historical case library to form a candidate similar case set. In this way, the system can quickly find historical cases similar to the current risk event, providing a reference for subsequent judgments. For example, when a transaction is highly consistent with typical past fraud cases in terms of amount, region, and associated entities, the system will automatically include the case in the candidate set.

[0261] The confidence level comprehensive judgment subunit can calculate a comprehensive score based on the candidate similar case set, combining similarity scores and the accuracy rate of handling historical cases. When the comprehensive score exceeds a preset threshold, the corresponding handling result is output, forming a case judgment result. In this way, the system can reuse historical experience with high confidence, accelerating the handling of similar risks. For example, when the comprehensive score shows that a transaction is highly consistent with a known "false related-party transfer" case, the system can directly output handling suggestions, shortening the review time.

[0262] The case library expansion and update subunit can, based on the case judgment results, write the current risk characteristic data into the case library when no high-confidence cases are matched, generating entries to be labeled. Upon receiving externally input labeled data, it expands and corrects the case library, generating case comparison results. In this way, the system can continuously supplement and amend the case library content, improving its coverage and accuracy. For example, when a new risk pattern first appears and is confirmed by manual labeling, the system automatically stores the pattern in the case library, enabling it to be quickly identified in subsequent transactions.

[0263] Through the above functions, the embodiments of the present invention can realize automatic retrieval of similar cases, high confidence determination, and self-expansion of the case library in risk assessment, which not only improves the system's response speed to known risks, but also enhances its learning and adaptation capabilities to emerging risks.

[0264] The feature similarity comparison subunit specifically includes:

[0265] First, key feature vectors for comparison are extracted from the stratified risk results. These features include transaction amount, risk labels of participants, transaction frequency, cross-platform transfer status, and complexity of the behavioral chain.

[0266] Then, retrieve the stored case feature sets from the historical case database, with each case set corresponding to one or more typical risk patterns.

[0267] Next, the extracted key feature vectors are compared one by one with the features in the case library. The comparison method is described in words as follows: if the key feature and the feature value in the case are small in numerical or attribute difference, the similarity score is high; if the difference is large, the similarity score is low.

[0268] Furthermore, based on the comparison results, cases with similarity scores exceeding a set threshold are selected and a candidate similar case set is formed.

[0269] Finally, the candidate similar case set is output for the confidence comprehensive judgment subunit to conduct subsequent analysis.

[0270] The confidence level comprehensive determination subunit specifically includes:

[0271] First, receive a set of candidate similar cases and obtain the similarity score and historical handling records for each candidate case.

[0272] Then, the accuracy rate of the handling results for each candidate case is statistically analyzed. In written terms: accuracy rate represents the proportion of cases that have been correctly identified and properly handled in past applications.

[0273] Next, the similarity score is weighted and combined with the historical processing accuracy rate to obtain a comprehensive score. The higher the similarity and the higher the processing accuracy rate, the higher the comprehensive score.

[0274] Furthermore, the comprehensive score is compared with a preset threshold. When the score exceeds the threshold, the system directly outputs the corresponding handling result for the case and generates a case judgment result; when the score does not reach the threshold, it proceeds to the next sub-unit for processing.

[0275] Finally, the case judgment results are used as output to assist in the final summary of risk assessment results.

[0276] The case library extension and update sub-unit specifically includes:

[0277] First, when the confidence level comprehensive judgment subunit fails to output high confidence level cases, the risk characteristic data of the current transaction is extracted and stored as an entry to be labeled.

[0278] Then, the entries to be labeled are added to the temporary storage area of ​​the case library, and the labeling information is awaited from administrators or external systems. The labeling information includes whether the risk event is abnormal and the final handling method.

[0279] Next, after receiving the labeled data from external input, the system will update the case library, convert the entry into a formal case, and compare it with the original cases to avoid duplicate recording.

[0280] Furthermore, existing cases in the case library are corrected. For example, if a case is repeatedly confirmed as a misjudgment, the system will reduce the weight of that case in the comparison process to ensure the accuracy of the case library.

[0281] Finally, updated case comparison results are generated and used as a new case set for subsequent risk event comparisons, thereby continuously improving the coverage and usability of the case library.

[0282] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based internet property rights transaction risk assessment and decision-making system, characterized in that, The system includes: The data acquisition and preprocessing module is used to collect raw transaction data from the Internet property rights trading platform in real time, and to clean and standardize the raw transaction data to obtain standardized transaction data. The raw transaction data includes information on the property rights, historical records of the trading parties, and information on related parties. The multi-dimensional feature construction module is used to model and construct a multi-level interaction relationship graph based on standardized transaction data according to time windows, thereby obtaining time series feature data and graph-related feature data; and to normalize and weight the features to form fused feature data. The risk identification module is used to extract suspicious transfer behavior chains and trace the flow of benefits based on fused feature data and multi-level interaction relationship graphs, perform composite risk judgment, and generate preliminary risk data. The intelligent decision-making module is used to perform hierarchical adaptive discrimination and case migration comparison based on preliminary risk data, summarize and generate risk assessment results and push early warnings; and receive feedback from the management terminal to optimize identification and decision parameters. The multidimensional feature construction module includes: The time window modeling unit is used to group transaction sequences according to a set time window based on standardized transaction data, and to perform sliding statistics and trend analysis to obtain time series feature data. The relationship graph generation unit is used to construct a multi-level interactive relationship graph with the property rights object as the core based on standardized transaction data. The nodes of the graph represent different participants, the edges represent transfer types and time tags, and the unit calculates the correlation between nodes to generate graph association feature data. The feature fusion weighting unit is used to concatenate time series feature data with map-related feature data, and then weight the data by combining the feature importance weights of abnormal cases in the historical case library to obtain fused feature data. The risk identification module includes: The suspicious behavior chain extraction unit is used to extract continuously occurring suspicious transfer behavior chains based on the fused feature data, using recursive traversal and sequence encoding methods, and generate behavior chain feature data. The profit flow tracing unit is used to trace key nodes step by step based on behavioral chain characteristic data and multi-level interaction relationship graphs, mark suspected profit flows, and generate anomaly markers when transferring across domains or platforms to obtain tracing risk data. The composite risk discrimination unit is used to logically combine behavioral chain feature data and source tracing risk data for discrimination. When multiple high-risk factors are met at the same time, a risk superposition and grading mechanism is executed to generate preliminary risk data. The intelligent decision-making module includes: The hierarchical adaptive discrimination unit is used to calculate risk scores based on preliminary risk data, adopts an automatic matching processing strategy based on a hierarchical threshold system, and adaptively updates threshold parameters when a new abnormal pattern is detected to generate hierarchical risk results. The case transfer learning unit is used to compare the hierarchical risk results with the historical case library one by one. If a high-confidence similar case exists, the corresponding treatment result is directly output. If no match is found, the current risk characteristics are stored in the case library, and an entry to be labeled is generated for the case library. When externally input labeled data is received, the case library is automatically updated to maintain accuracy and a case comparison result is generated. The assessment summary and output unit is used to summarize and assess the stratified risk results and the case comparison results, generate risk assessment results, and push corresponding early warning information to the management terminal. The feedback loop unit is used to receive the handling feedback from the management terminal and input the handling feedback back to the risk identification module and the intelligent decision-making module respectively, so as to continuously optimize the identification and decision parameters and form a closed-loop self-learning mechanism. The relation graph generation unit includes: The entity standardization processing subunit is used to perform cross-platform entity identification and deduplication of participants based on standardized transaction data, and generate a unified set of nodes; The temporal and domain labeling subunit is used to write timestamps and regional or platform source identifiers on the edges based on a unified set of nodes and transaction records, and to write role type and risk prior labels in the nodes to form an initial relationship graph with temporal and domain attributes. The correlation calculation subunit is used to statistically analyze the interaction characteristics between nodes within a preset hop limit based on the initial relationship graph, calculate the number of interactions, freshness, role similarity and path credibility, and generate correlation scores for node pairs to form a weighted relationship graph. The graph incremental update subunit is used to incrementally update the weighted relation graph based on newly arrived transaction data, generate a valid graph snapshot that conforms to the sliding time window, and invalidate edges that exceed the window period according to rules to ensure the real-time performance and effectiveness of the graph. The feature fusion weighting unit includes: The feature weight allocation subunit is used to assign learnable weights to time series feature data and graph-related feature data based on transaction type, transaction amount, data completeness and recent time features, and generate a weighted feature vector. The case prior guidance subunit is used to compare the similarity of the weighted feature vector with the feature clusters of abnormal cases in the historical case library. When the similarity is higher than the threshold, the weight allocation result is biased and adjusted to generate a prior-guided feature vector. The robust normalization processing subunit is used to perform median scaling and extreme value truncation on outliers in the feature vector guided by prior knowledge, so as to obtain normalized robust feature data. The fusion calibration subunit is used to concatenate time series features and spectral correlation features based on the normalized robust feature data to obtain fused feature data. The hierarchical adaptive discrimination unit includes: The risk score generation subunit is used to calculate and output a risk score sequence based on preliminary risk data, the abnormality of transaction amount, the historical risk labels of participants, and time span factors. The stratified threshold comparison subunit is used to compare risk scores layer by layer according to the risk score sequence, set threshold intervals according to at least three risk levels, and generate stratified discrimination results. The threshold adaptive adjustment subunit is used to adjust the upper and lower limits of the threshold based on the stratification judgment results. When the risk score is detected to continuously deviate from the preset threshold range, the unit writes the updated threshold parameters into the decision parameter library and generates adaptively updated stratified risk results. The case transfer learning unit includes: The feature similarity comparison subunit is used to extract key feature vectors based on the hierarchical risk results and compare them one by one with the feature set in the historical case library to form a candidate similar case set; The confidence level comprehensive judgment subunit is used to calculate a comprehensive score based on the candidate similar case set, combined with the similarity score and the case handling accuracy. When the comprehensive score exceeds the preset threshold, the corresponding handling result is output and a case judgment result is formed. The case library expansion and update subunit is used to write the current risk characteristic data into the case library and generate entries to be labeled when no high-confidence cases are matched based on the case judgment results. When externally input labeled data is received, the case library is expanded and corrected to generate case comparison results.

2. The Internet property rights transaction risk assessment and decision-making system based on artificial intelligence according to claim 1, characterized in that, The suspicious behavior chain extraction unit includes: The transaction sequence segmentation subunit is used to divide continuous transaction events into time sequences based on the fused feature data, and divide them into candidate subsequences according to changes in transfer amount, abnormal transaction frequency, and cross-domain occurrence conditions to obtain candidate behavior chain data; The behavior chain feature encoding subunit is used to progressively encode the node role, transfer direction, transaction freshness and node hop length according to the candidate behavior chain data, and generate a multi-dimensional behavior chain feature vector. The anomaly accumulation determination subunit is used to calculate the risk accumulation value of continuous transactions step by step based on the multi-dimensional behavior chain feature vector. When the risk accumulation value exceeds the dynamic threshold, the corresponding candidate behavior chain is marked as a suspicious behavior chain, and behavior chain feature data is generated.

3. The Internet property rights transaction risk assessment and decision-making system based on artificial intelligence according to claim 1, characterized in that, The benefit flow tracing unit includes: The key node filtering subunit is used to identify nodes that participate in multiple transactions in the relationship graph based on behavioral chain feature data, and to filter them according to the concentration of transaction amount and the weight of node correlation to generate a set of key nodes. The trajectory backtracking construction subunit is used to backtrack along the transaction link step by step based on the set of key nodes, starting from each key node, and record the node role type, interaction frequency and regional label to form the trajectory data of the flow of benefits. The abnormal trajectory labeling subunit is used to generate abnormal labels for abnormal paths based on the profit flow trajectory data. When it is detected that the transfer path crosses different regions, different platforms or has multiple levels of concealed transfer, the abnormal path is used as source tracing risk data.

Images