User risk assessment method and system based on user address aggregation

Abstract

The invention provides a user risk assessment method and system based on user address aggregation. The method comprises the following steps: acquiring a user address information group and a user associated information group; performing aggregation association mining on the user address information group, dividing user address groups with adjacent geographic positions and associated attributes in combination with attribute information groups in the user association information group, and generating an address aggregation association feature group; inputting the address aggregation association feature group and the user association information group into a risk assessment AI model to generate a risk association conduction feature group; generating a user risk assessment feature group based on the risk association conduction feature group; and feeding back the user risk assessment feature group to an aggregation association mining link, adjusting association parameters of next aggregation mining, and outputting a risk assessment report including the user risk assessment feature group and the address aggregation association feature group.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a user risk assessment method and system based on user address aggregation. Background Technology

[0002] Traditional user risk assessment methods primarily focus on individual user attributes and behavioral information, such as analyzing data on a user's occupation, income, spending habits, and transaction records to assess their risk level. However, this approach has significant limitations.

[0003] On the one hand, viewing each user in isolation ignores the relationships between users and the impact of their geographical environment on their risk. In reality, geographically adjacent user groups often exhibit similar risk patterns due to shared regional characteristics and social environments, and behavioral interactions between users can also trigger the transmission and spread of risk. On the other hand, existing assessment methods struggle to dynamically adapt to changes in risk conditions and lack mechanisms for continuous optimization of risk assessment models and parameters, thus affecting the accuracy and timeliness of assessment results. Summary of the Invention

[0004] The purpose of this invention is to provide a user risk assessment method and system based on user address aggregation, which aims to comprehensively consider user address aggregation characteristics, user association information and risk transmission mechanisms, and to achieve a dynamically optimized user risk assessment method.

[0005] To achieve the above objectives, a first aspect of this disclosure provides a user risk assessment method based on user address aggregation, comprising: Obtain user address information group and user association information group, wherein the user address information group includes the actual residential address information of multiple users, and the user association information group includes user attribute information group and behavior information group; The user address information group is subjected to cluster association mining. Combined with the attribute information group in the user association information group, user address groups with adjacent geographical locations and related attributes are divided to generate address cluster association feature groups. The address cluster association feature group and the behavior information group in the user association information group are input into the pre-trained risk assessment AI model. Through the risk transmission link analysis of the risk assessment AI model, risk association transmission feature group is generated. Generate a user risk assessment feature set based on the aforementioned risk correlation and transmission feature set; The user risk assessment feature group is fed back to the aggregation and association mining stage to adjust the association parameters for the next aggregation mining and output a risk assessment report including the user risk assessment feature group and the address aggregation and association feature group. The association parameters include the set range of geographical proximity distance and the set conditions for the existence of attribute association.

[0006] In a preferred embodiment, the step of performing cluster association mining on the user address information group, combining the attribute information group in the user association information group, and dividing the user address groups that are geographically adjacent and have related attributes to generate address cluster association feature groups includes: Parse the geographic coordinate information in the user address information group, establish a mapping relationship group between user address and geographic coordinate, and form an address coordinate list group; Identify the adjacency relationship between any two user addresses in the address coordinate list group on the geographic coordinates, and filter out user address pairs whose adjacent distance is within a set range; The degree of association of the attribute information of the user address pair group is determined based on the attribute information group in the user association information group. User addresses that are adjacent to each other within a set range and whose attribute information correlation meets the set conditions are aggregated to form an initial address group, wherein the user addresses in the initial address group are geographically adjacent and have related attributes. The initial address group is expanded to include other user addresses whose adjacent distance to the edge user addresses of the initial group is within a set range and whose attribute information correlation meets the standard, forming a target user address group. For each target user address group, determine the group's geographical boundary, attribute association pattern, and user distribution density, and integrate them to generate the address cluster association feature group.

[0007] In a preferred embodiment, determining the attribute information association degree of the user address pair group based on the attribute information group in the user association information group includes: The attribute information groups in the user-related information group are classified and coded, and the non-numerical attributes of occupation information, family composition information and consumption characteristics information are converted into numerical codes to form attribute code vector groups; Each attribute encoding vector group is normalized so that the encoded values ​​of different attributes are in the same numerical range. Calculate the association similarity between the attribute encoding vector groups of the user address pair groups, and use the association similarity as the initial attribute information association degree; The initial attribute information correlation is adjusted by assigning weights based on the importance of different attributes in risk assessment. The weighted correlation is the sum of the products of the correlation similarity of each attribute and its corresponding weight. The weighted attribute information correlation is filtered by a threshold, and user address pairs whose attribute information correlation exceeds the set threshold are retained to obtain the attribute information correlation of the user address pairs.

[0008] In a preferred embodiment, the step of inputting the address cluster association feature group and the behavioral information group in the user association information group into a pre-trained risk assessment AI model, and generating a risk association transmission feature group through risk transmission link analysis of the risk assessment AI model, includes: The address clustering and association feature group and the behavioral information group in the user association information group are input into the risk assessment AI model to generate a geographic-behavioral interaction feature group. The risk assessment AI model uses an intra-group transmission analysis module to mine intra-group associations of the geographic-behavioral interaction feature group, identify the risk mutual influence path between any two user addresses within the group, and form an intra-group risk transmission path. The cross-group diffusion analysis module of the risk assessment AI model is used to compare and analyze the geographic-behavioral interaction feature groups of different user address groups, identify group pairs with similar features, analyze the risk transmission path between group pairs, and generate cross-group risk diffusion path. The coupling path analysis module of the risk assessment AI model is used to perform a fusion analysis of risk transmission paths within a group and risk diffusion paths across groups. Combined with the attribute association patterns in the address clustering association feature group, a behavior-address risk coupling path is generated. The risk transmission path within the group, the risk diffusion path across groups, and the behavior-address risk coupling path are integrated to form the risk association transmission characteristic group.

[0009] In a preferred embodiment, the step of inputting the address clustering association feature group and the behavioral information group in the user association information group into the risk assessment AI model to generate a geographic-behavioral interaction feature group includes: The address clustering and association feature group is input into the geographic feature processing module of the risk assessment AI model, and a geographic feature vector group is obtained by feature embedding to retain geospatial association information. The behavioral information group in the user-associated information group is input into the behavioral feature processing module of the risk assessment AI model, and a behavioral feature vector group is obtained through time-series encoding conversion. The geographic feature vector group and the behavioral feature vector group are input into the risk association mining module of the risk assessment AI model. The association weights of the two features are analyzed through a two-way interaction mechanism, and the geographic-behavioral interaction feature group is generated based on the association weights.

[0010] In a preferred embodiment, the intra-group transmission analysis module of the risk assessment AI model performs intra-group correlation mining on the geographic-behavioral interaction feature group to identify the risk mutual influence path between any two user addresses within the group, forming an intra-group risk transmission path, including: The risk assessment AI model uses an intra-group transmission analysis module to uniquely identify each user address in the geographic-behavioral interaction feature group and establish an intra-group user address index group. Extract the feature sub-vectors corresponding to each user address from the geographic-behavioral interaction feature group, and construct a feature matrix group within the group, where the rows of the matrix correspond to the user address index and the columns of the matrix correspond to the feature dimensions. The correlation strength between any two feature vectors in the feature matrix group within the population is analyzed through the correlation identification submodule of the intra-population transmission analysis module. Based on the association strength, an intra-group association network is constructed, where nodes in the network represent user addresses and edge weights represent association strength. The path mining submodule of the intra-group transmission analysis module identifies the critical paths of risk transmission in the associated network. The critical paths are connection paths whose association strength exceeds a set threshold. The identified critical paths are analyzed for path features, and features such as path length, distribution of association strength on the path, and number of user addresses involved in the path are extracted and integrated to form the risk transmission path within the group.

[0011] In a preferred embodiment, the cross-group diffusion analysis module of the risk assessment AI model compares and analyzes the geographic-behavioral interaction feature groups of different user address groups, identifies group pairs with similar features, analyzes the risk transmission path between group pairs, and generates cross-group risk diffusion paths, including: Geographic-behavioral interaction feature groups of all address clusters are obtained from the risk association mining module of the risk assessment AI model, and a cluster-feature mapping library is established. Analyze the feature similarity between the geographic-behavioral interaction feature groups of any two address clusters. The feature similarity is used to reflect the degree of similarity between the two sets of features. The more significant the feature similarity, the more significant the degree of similarity. Filter out address clusters whose feature similarity is within a set range and mark them as similar clusters; We extract behavioral association features from the geographic-behavioral interaction features of similar clustered groups and analyze the frequency of cross-association between the two groups in transaction information and social interaction information. Based on the cross-correlation frequency and feature similarity analysis, the probability of risk transmission between groups is analyzed, and the probability of risk transmission is adjusted by time decay analysis and combined with the time interval of historical risk events between the two groups. The adjusted risk transmission probability is correlated with the geographical distance between similar cluster groups to generate the cross-group risk diffusion path, which includes probability value, geographical distance, and feature similarity.

[0012] In a preferred embodiment, generating a user risk assessment feature set based on the risk correlation and transmission feature set includes: Extract the risk occurrence trend, risk impact range, and risk persistence characteristics from the risk correlation and transmission feature group, and integrate them to form an overall risk pattern of the address cluster group; Extract local association features related to each user address within the group from the risk association transmission feature group. The local association features include the position of the user address in the risk transmission path within the group and its contribution in the behavior-address risk coupling path. Analyze the correlation between the local association characteristics of each user address and the overall risk pattern of the address cluster, and generate the individual risk involvement degree; The individual risk involvement degree is extended by feature expansion, and combined with the abnormal behavior features in the behavior information group corresponding to the user address, an individual risk association pattern of each user address in the group is generated. The overall risk model and the individual risk correlation model are integrated to form the user risk assessment feature group.

[0013] In a preferred embodiment, the step of extracting the risk occurrence trend, risk impact range, and risk persistence characteristics from the risk correlation and transmission feature group, and integrating them to form an overall risk pattern for the address cluster, includes: A dimensional importance analysis is performed on the overall risk feature vector in the aforementioned risk correlation and transmission feature group to identify the salient feature dimension that contributes most significantly to the risk pattern; By learning the correlation with historical risk events, a first feature component related to the risk occurrence trend is extracted from the salient feature dimension. The first feature component is used to reflect the changing trend of the probability of risk occurrence in the address cluster group. Based on the geographical boundaries of address clusters and user distribution density, a second feature component related to the scope of risk impact is generated. The second feature component is used to reflect the range of user addresses and geographical areas that may be affected after the risk occurs. By combining the attribute association patterns of address clusters and historical risk persistence data, a third feature component related to the risk persistence characteristics is generated. The third feature component is used to reflect the duration and intensity decay trend after the risk occurs. The first feature component, the second feature component, and the third feature component are correlated and verified to check the logical consistency between the features and integrate them to form the overall risk pattern of the address cluster.

[0014] A second aspect of this disclosure provides a user risk assessment system based on user address aggregation, the system comprising: A processor and a memory, wherein the memory is connected to the processor; The memory is used to store programs, instructions, or code; The processor is used to execute programs, instructions, or code in the memory to implement the user risk assessment method based on user address aggregation as described in any one of the first aspects.

[0015] This invention provides a user risk assessment method and system based on user address aggregation. Compared with existing technologies, it has the following advantages: By acquiring user address information groups containing actual residential addresses and user association information groups covering attribute and behavioral information, and conducting comprehensive analysis, the method performs aggregation and association mining on the user address information groups. Combined with attribute information groups, it can accurately segment user address groups that are geographically adjacent and have related attributes, generating address aggregation and association feature groups. This process overcomes the limitations of traditional single-dimensional segmentation, making group segmentation more realistic, laying a solid foundation for subsequent risk assessment, and greatly improving the targeting of risk assessment.

[0016] Address clustering and association feature groups, along with behavioral information groups, are input into a pre-trained risk assessment AI model. Leveraging the model's powerful risk transmission link analysis capabilities, potential risk correlations can be deeply mined, generating risk correlation transmission feature groups, and further generating user risk assessment feature groups. The application of the AI ​​model achieves intelligent and precise risk assessment, enabling rapid processing of massive amounts of data, accurate capture of risk signals, effectively avoiding the subjectivity and errors of manual assessment, and significantly improving assessment efficiency and accuracy.

[0017] User risk assessment feature groups are fed back to the clustering and association mining stage to adjust the association parameters for the next mining iteration, such as the range of geographical proximity distances and attribute association conditions. Adjustments can be made based on real-time risk assessment results to adapt to constantly changing user group characteristics and risk patterns, ensuring consistently high accuracy and timeliness in risk assessment. The output risk assessment report covers user risk assessment feature groups and address clustering and association feature groups, providing users with a comprehensive and intuitive display of risk information. This forms a dynamically optimized risk assessment chain, enabling timely adaptation to changes in risk conditions and improving the accuracy and timeliness of risk assessment.

[0018] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a user risk assessment method based on user address aggregation, according to an embodiment of the specification.

[0020] Figure 2 This is a block diagram of a user risk assessment system based on user address aggregation, as shown in the embodiment of the specification. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0023] This disclosure provides a user risk assessment method based on user address aggregation. Figure 1 This is a block diagram illustrating a user risk assessment method based on user address aggregation according to an embodiment. Specifically, the user risk assessment method based on user address aggregation includes: In step S11, user address information group and user association information group are obtained. The user address information group includes the actual residential address information of multiple users, and the user association information group includes the user attribute information group and behavior information group.

[0024] The user address information group can be a collection of multiple users' actual residential address information, used to locate the user's geographical location. In this embodiment, when obtaining the user address information group, the actual residential address information of multiple users is collected through official channels authorized by the users. This address information is anonymized and only retained at the street or community level, such as "a certain street in a certain city and district," avoiding sensitive content such as specific house numbers. Simultaneously, the collected address information is verified in multiple dimensions to ensure its authenticity and validity, eliminating erroneous or invalid address information. For example, regular expressions and rule bases are used to verify whether the components of the collected address information conform to official format specifications. For instance, verification is performed by checking whether the postal code matches the mailing address, whether the house number format is correct, and the completeness of the city / county mailing address. Address information with inconsistent postal codes and mailing addresses, address information with incorrect house number formats, and incomplete city / county mailing addresses are eliminated.

[0025] User-related information groups include user attribute information groups and behavioral information groups. Attribute information reflects the user's basic characteristics, such as age, gender, occupation, family structure, and consumption characteristics. Occupation information includes the user's industry and position; family structure information includes the number of family members and marital status; and consumption characteristics reflect the user's consumption habits and consumption level. Behavioral information records various user activities and behavioral patterns, such as transaction information, social interaction information, and activity trajectory information. Transaction information includes transaction type, transaction partner, and transaction frequency; social interaction information involves social partners, interaction frequency, and interaction duration; and activity trajectory information reflects the user's activity range and activity patterns.

[0026] In this embodiment, user-related information is obtained through multiple data collection channels. For the user address information group, the user's actual residential address information can be collected from user registration information, logistics delivery records, location services, etc. The attribute information group within the user association information group can be obtained through user registration information, questionnaires, third-party data interfaces, etc., covering basic user attributes such as age, gender, occupation, and income level. The behavioral information group is obtained by analyzing user operation records, consumption records, and social interaction records on applications and websites, such as the user's browsing history, types and frequency of purchased goods, and social circles. The collected address information and association information are organized and stored separately to form the user address information group and the user association information group.

[0027] In step S12, cluster association mining is performed on the user address information group. Combined with the attribute information group in the user association information group, user address groups with adjacent geographical locations and related attributes are divided to generate address cluster association feature groups.

[0028] The address cluster association feature group includes group geographic boundaries, attribute association patterns, and user distribution density. The attribute information group in the user association information group is analyzed to obtain the attribute information association degree, which is used to reflect the tightness of attribute association.

[0029] The address cluster association feature group is a data set generated by performing cluster association mining on user address information groups and combining it with user attribute information groups. It is used to describe the characteristics of user address groups that are geographically adjacent and have related attributes, including the geographical location range and attribute characteristics of the group.

[0030] In this embodiment of the disclosure, in financial credit risk assessment, users with similar attributes within the same area often exhibit a certain correlation in terms of credit risk. For example, users in a certain community who work in the same industry and have similar family structures may have similar credit repayment abilities and risk preferences. By performing clustering and association mining on user address information groups, these potentially related user addresses can be grouped together.

[0031] In this embodiment, spatial clustering algorithms (such as DBSCAN algorithm, K-Means algorithm, etc.) are used to process user address information groups. These algorithms, based on the geographic coordinates of user addresses and according to a certain distance metric (such as Euclidean distance), group geographically adjacent user addresses into the same cluster, forming preliminary geographic location clusters. For example, density clustering based on the DBSCAN algorithm separates low-density regions from high-density regions. A minimum number of neighbors (min_samples) is formed for each core point based on a neighborhood distance threshold used to define the adjacent radius. A circle is drawn with an unvisited point as the center and a neighborhood distance threshold as the radius. If the number of points within the circle exceeds the minimum neighbor (min_samples), a cluster is formed, and this point is called the core point. Then, all density-reachable points of this core point are recursively grouped into the same cluster, forming multiple preliminary geographic location clusters.

[0032] For example, consider centroid clustering based on the K-Means algorithm. Based on a pre-set number of clusters, points are iteratively assigned to the nearest centroid (cluster center) to form spherical clusters. For instance, K centroids are randomly initialized. Then, the point corresponding to each address is assigned to the cluster containing the nearest centroid. The Euclidean distance between points corresponding to addresses covered in each cluster is then calculated to obtain a new centroid (new cluster center). If the original cluster center and the new cluster center do not meet preset conditions (e.g., they do not overlap, the distance is greater than a preset threshold, or the variance is greater than a preset threshold), the process of assigning the point corresponding to each address to the cluster containing the nearest centroid and calculating the Euclidean distance between points corresponding to addresses covered in each cluster is repeated until the preset conditions are met.

[0033] Then, combining the attribute information groups within the user association information groups, the user attributes within each geographic location cluster are analyzed. By setting judgment conditions for attribute association (such as the same occupation, similar income levels, etc.), users with associated attributes are filtered out, further refining the user address groups. Finally, features are extracted from the identified geographically adjacent user address groups with associated attributes, including the group's central geographic location, coverage area, and main attribute characteristics, generating address cluster association feature groups.

[0034] In step S13, the address aggregation association feature group and the behavior information group in the user association information group are input into the pre-trained risk assessment AI model. Through the risk transmission link analysis of the risk assessment AI model, a risk association transmission feature group is generated.

[0035] Among them, the risk assessment AI model is used to learn risk patterns and regularities from large amounts of data through pre-training, enabling it to assess and analyze input data and generate risk-related features and results. Risk transmission link analysis is used to study the propagation paths and impact levels of risks among different factors or groups, revealing the correlation and transmission mechanisms between risks. The risk correlation transmission feature set is generated by the risk assessment AI model after analyzing address cluster correlation feature sets and user behavior information sets. It is a dataset reflecting the correlation and transmission characteristics of risks among different user groups and factors, and can include intra-group risk transmission paths, cross-group risk diffusion paths, and behavior-address risk coupling paths.

[0036] In this embodiment, the pre-trained risk assessment AI model typically employs a deep learning architecture (such as a neural network model). During the training phase, the model uses a large amount of labeled historical data to learn the complex relationships between different user characteristics (including address clustering features, attribute features, and behavioral features) and risk. When inputting address clustering-related feature groups and user behavior information groups, the model first performs feature encoding and preprocessing on these input data, converting them into numerical vector forms that the model can process. Then, the model performs complex calculations and feature extraction through its internal neural layers to simulate the transmission process of risk between different user groups and factors. By analyzing the correlations and interactions between input features, the model identifies potential risk transmission links and generates risk-related transmission feature groups. For example, the inputs received by the model include: address clustering-related feature groups (such as the user's cluster being "XX Industrial Zone migrant worker cluster area," which has a high historical average delinquency rate) and user behavior information groups (such as a surge in "recent multiple loan applications" for user A). The model's internal calculations revealed that when the characteristics of "high historical delinquency rate within the cluster" and "individual multiple borrowing behavior" coexist, they have a strong synergistic amplification effect on the final risk score, far exceeding the simple sum of their independent effects. Therefore, a risk transmission pattern can exist: individuals in high-risk environments are more susceptible to the influence of a negative credit atmosphere within the group, thus imitating high-risk borrowing behaviors, leading to risk transmission within the group. Consequently, the model can generate a transmission feature set called "Behavioral Imitation Risk in High-Risk Environments," which comprehensively reflects the coupling state between group risk and individual behavior.

[0037] In this embodiment of the disclosure, in financial credit risk assessment, the address clustering and association feature group reflects the geographical and attribute association characteristics of user address groups, while the behavioral information group reflects user behavior patterns. This information is crucial for analyzing risk transmission. By inputting these two sets of features into a pre-trained risk assessment AI model, the model, through its internal risk transmission link analysis mechanism, can deeply explore the transmission path of risk within a group, the diffusion path between different groups, and the risk coupling path between behavior and address.

[0038] For example, frequent borrowing and lending activities among users within a certain address group may lead to the rapid transmission of risk within the group, while financial transactions between users in different address groups may cause the risk to spread across groups. The model generates corresponding risk correlation and transmission feature groups through comprehensive analysis of this information.

[0039] In step S14, a user risk assessment feature group is generated based on the risk association transmission feature group.

[0040] Among them, the user risk assessment feature group is a set of feature data generated based on the risk correlation and transmission feature group, used to comprehensively assess the user's risk status. The user risk assessment feature group includes the overall risk pattern of each address cluster and the individual risk correlation pattern of each user address within the cluster.

[0041] In this embodiment, the risk correlation and transmission feature group is further analyzed and integrated. First, the weights of each feature in the risk correlation and transmission feature group are evaluated, assigning different weights based on their impact on user risk. For example, certain risk transmission links may have a greater impact on user risk, and their corresponding feature weights are higher. Then, the risk correlation and transmission features are weighted and summed or other fusion methods are used to integrate multiple risk correlation and transmission features into a comprehensive user risk assessment index. Simultaneously, combined with the user's basic attributes and behavioral characteristics, user risk is classified and refined, generating a user risk assessment feature group that includes various risk types (such as credit risk, fraud risk, market risk, etc.) and risk levels.

[0042] In this embodiment of the disclosure, in financial credit risk assessment, based on the risk correlation and transmission characteristic group, the risk situation can be assessed from both the overall group and individual levels. The overall risk model reflects the risk level and transmission characteristics of the entire address cluster, while the individual risk correlation model reflects the impact of group risk on each user address within the group and its correlation with the risks of other user addresses.

[0043] In step S15, the user risk assessment feature group is fed back to the aggregation and association mining stage, the association parameters for the next aggregation mining are adjusted, and a risk assessment report including the user risk assessment feature group and the address aggregation and association feature group is output. The association parameters include the set range of geographical adjacent distance and the set conditions for the existence of attribute association.

[0044] In this embodiment, the generated user risk assessment feature group is fed back to the aggregation and association mining stage as a basis for adjusting the association parameters. Based on the user risk assessment results, the relationship between the user address groups segmented under the current association parameters and their risks is analyzed. If it is found that the user groups segmented under certain geographical proximity distance settings or attribute association settings have higher or lower risks than expected, these association parameters are adjusted.

[0045] For example, if setting the geographical proximity distance too large leads to inaccurate risk transmission analysis, the range of the geographical proximity distance setting should be appropriately narrowed; if the setting conditions for certain attribute associations are too lenient, causing high-risk users to be incorrectly classified into low-risk groups, these attribute association setting conditions should be tightened. After adjusting the association parameters, the cluster association mining and subsequent risk assessment processes are re-performed. Finally, the user risk assessment feature groups and address cluster association feature groups are organized and visualized to generate a risk assessment report, providing users with a comprehensive and intuitive display of risk information.

[0046] In this embodiment of the disclosure, by feeding back the user risk assessment feature group to the early aggregation and association mining stage, the association parameters can be dynamically adjusted according to the risk assessment results, thereby improving the accuracy of subsequent risk assessments and outputting a detailed risk assessment report.

[0047] In one embodiment of this disclosure, step S151: analyze the overall risk pattern in the user risk assessment feature group, identify address clusters with significant risk patterns, and mark the key parameters in the address cluster association feature group corresponding to the address cluster.

[0048] The overall risk patterns within the user risk assessment feature groups are analyzed to identify address clusters with significant risk occurrence trends, large impact ranges, or long durations. For these clusters, key parameters are extracted from their corresponding address cluster association feature groups, such as the actual range of geographical proximity and the actual achievement value of attribute information correlation. These parameters are the key factors in the formation of this cluster.

[0049] Step S152: Input the marked key parameters into the parameter adjustment module, and determine the parameter adjustment range based on the significance of the risk pattern. The associated parameters include the setting range of geographical adjacent distance and the setting conditions of attribute information correlation.

[0050] After receiving the marked key parameters, the parameter adjustment module analyzes the relationship between the risk characteristics of address clusters with significant risk patterns and the key parameters. If a group has a significant risk pattern but its actual geographical proximity range is small, it indicates that the current geographical proximity range may be too large, and the adjustment range needs to be reduced. If the actual attainable value of attribute information correlation is high, but the risk pattern is significant, it may be necessary to increase the setting conditions for attribute information correlation to more accurately screen groups. Based on the significance of the risk pattern (such as the rate of increase of the risk occurrence trend, the size of the impact range, etc.), the specific adjustment range for the geographical proximity range and the attribute information correlation setting conditions is determined. The more significant the risk pattern, the larger the adjustment range may be to enhance the targeting of subsequent cluster association mining.

[0051] Step S153: Update the adjusted association parameters to the parameter configuration of the aggregation association mining stage, so that they can be used for the next aggregation association mining of user address information groups.

[0052] In one embodiment of this disclosure, the adjusted association parameters (including the new geographical proximity distance setting range and attribute information correlation setting conditions) determined in step S152 are updated in the parameter configuration file of the aggregation association mining stage, overriding the original parameter settings. When the next aggregation association mining of user address information groups is performed, the system will automatically adopt these adjusted parameters, making the mined user address groups more consistent with actual risk assessment needs, thereby improving the accuracy and effectiveness of risk assessment. For example, if the adjusted geographical proximity distance setting range is narrowed, the next mining will only group closer user addresses into the same group, reducing interference from irrelevant user addresses.

[0053] Step S154: Establish a parameter adjustment record, recording the time of each parameter adjustment, the related parameter values ​​before and after the adjustment, and the changes in the corresponding user risk assessment feature groups, forming a parameter optimization trajectory.

[0054] In this embodiment of the disclosure, a dedicated parameter adjustment record is established to track the impact of parameter adjustments on risk assessment results. Each time a parameter is adjusted, the adjustment time, the value of the associated parameter before the adjustment, the value of the associated parameter after the adjustment, and the changes in the user's risk assessment feature groups after the adjustment (such as changes in the overall risk pattern, differences in individual risk association patterns, etc.) are recorded in detail. These records are arranged in chronological order to form a parameter optimization trajectory. By analyzing this trajectory, the patterns of parameter adjustments can be summarized.

[0055] Step S155: Associate the overall risk pattern in the user risk assessment feature group with the corresponding address clustering feature group to establish an overall risk-clustering feature association table.

[0056] In this embodiment, the overall risk pattern of each address cluster is mapped one-to-one with features such as the geographical boundary, attribute association pattern, and user distribution density in its corresponding address cluster association feature group, forming an overall risk-cluster feature association table. This association table clearly shows the correspondence between different address cluster features and the overall risk pattern. For example, a cluster with a large geographical boundary and dense user distribution may correspond to a larger risk impact range, while a cluster with a tight attribute association pattern may have a more significant risk occurrence trend. Through this association table, the impact of address cluster features on the risk pattern can be analyzed intuitively.

[0057] Step S156: Extract individual risk association patterns from the user risk assessment feature group, and label the risk level and association path of each user address in the address cluster.

[0058] In this embodiment, based on the individual risk association patterns in the user risk assessment feature group, each user address is labeled with a corresponding risk involvement level. The level classification can be determined according to the magnitude of the individual risk involvement, such as high, medium, and low levels, with higher risk involvement resulting in a higher level. Simultaneously, the risk involvement path of each user address within the group is labeled, indicating which other user addresses the user address is affected by and which user addresses it may affect. For example, the risk involvement path of user address A is that A is affected by B, and also affects C. Through these labels, the position and role of each user address in the group risk can be clearly understood.

[0059] Step S157: Set up a pattern-feature correlation analysis section in the risk assessment report to analyze the correlation between the clustering correlation features of different addresses and the overall risk pattern, and identify the common attributes of the significant clustering features of the risk pattern.

[0060] In this embodiment of the disclosure, the pattern-feature correlation analysis section of the risk assessment report, based on the overall risk-cluster feature correlation table, deeply analyzes the correlation between different address clustering correlation features (such as the size of the group's geographical boundary, the type of attribute correlation pattern, and the density of user distribution) and the overall risk pattern. Through statistical analysis and comparative studies, common clustering characteristic attributes of address clusters with significant risk patterns are identified. For example, these groups may generally share common attributes such as tight attribute correlation patterns and high user distribution density. Identifying these common attributes helps to provide early warning of potentially high-risk address clusters.

[0061] Step S158: Convert the geographic boundaries of the address cluster association feature group into a visual map layer, embed the map display section in the report, and use different graphics to identify address clusters with different overall risk patterns.

[0062] In this embodiment of the disclosure, geographic information visualization technology is used to convert the geographic boundaries of address clusters in the associated feature groups into a visualized map layer. Within the map layer, different graphics (such as polygons of different colors and border styles) are used to identify address clusters with different overall risk patterns. For example, clusters with significant risk patterns are identified by solid red polygons, clusters with moderate risk patterns are identified by hollow yellow polygons, and clusters with low risk patterns are identified by dashed green polygons. This visualization allows for an intuitive presentation of the geographical locations and risk status of different clusters, enabling report users to quickly understand the spatial distribution of risk.

[0063] Step S159: Add an individual risk association analysis section to the report to show the relationship between the individual risk association patterns of user addresses within the group and the risk transmission path within the group, explaining how individual risk is affected by other user addresses through group association.

[0064] In this embodiment, the individual risk association analysis section details the relationship between the individual risk association patterns of each user address within a group and the risk transmission path within the group. Using a combination of charts and graphs, it illustrates how individual risk is influenced by other user addresses through group-wide associations (such as geographical proximity, attribute association, and behavioral interactions). For example, user address B is located on the critical risk transmission path ABC within the group. Its individual risk association pattern shows a significant impact from the risk of user address A, while also having a certain transmission effect on the risk of user address C. This aligns with the characteristics of risk transmission paths within the group. This analysis provides a deeper understanding of the sources and propagation pathways of individual risk.

[0065] Step S1510: Integrate and format the association mapping table, individual risk labeling, pattern-feature association analysis, map display section and individual risk association analysis section according to the set report structure, generate the final risk assessment report and output it.

[0066] In this embodiment, the content, including the association mapping table, individual risk labeling, pattern-feature association analysis, map display, and individual risk association analysis, is integrated and formatted according to a preset report structure (such as cover, table of contents, executive summary, overall risk analysis, individual risk analysis, risk map, conclusions and recommendations). During the formatting process, the content is ensured to be logically clear, hierarchically structured, and with standardized charts, facilitating reading and understanding by report users. The final risk assessment report can be output in electronic or paper form and provided to relevant decision-makers, offering important reference for credit approval, risk monitoring, and other related work.

[0067] The aforementioned technical solution acquires user address information groups containing actual residential addresses and user association information groups encompassing attribute and behavioral information, and then conducts comprehensive analysis on them. By performing clustering and association mining on the user address information groups, and combining this with the attribute information groups, it can accurately segment user address groups that are geographically adjacent and have related attributes, generating address clustering and association feature groups. This process overcomes the limitations of traditional single-dimensional segmentation, making group segmentation more realistic, laying a solid foundation for subsequent risk assessment, and greatly improving the targeting of risk assessment.

[0068] Address clustering and association feature groups, along with behavioral information groups, are input into a pre-trained risk assessment AI model. Leveraging the model's powerful risk transmission link analysis capabilities, potential risk correlations can be deeply mined, generating risk correlation transmission feature groups, and further generating user risk assessment feature groups. The application of the AI ​​model achieves intelligent and precise risk assessment, enabling rapid processing of massive amounts of data, accurate capture of risk signals, effectively avoiding the subjectivity and errors of manual assessment, and significantly improving assessment efficiency and accuracy.

[0069] User risk assessment feature groups are fed back to the clustering and association mining stage to adjust the association parameters for the next mining iteration, such as the range of geographical proximity distances and attribute association conditions. Adjustments can be made based on real-time risk assessment results to adapt to constantly changing user group characteristics and risk patterns, ensuring consistently high accuracy and timeliness in risk assessment. The output risk assessment report covers user risk assessment feature groups and address clustering and association feature groups, providing users with a comprehensive and intuitive display of risk information. This forms a dynamically optimized risk assessment chain, enabling timely adaptation to changes in risk conditions and improving the accuracy and timeliness of risk assessment.

[0070] In a preferred embodiment, step S12, which involves performing clustering and association mining on the user address information group, and combining the attribute information group in the user association information group to segment user address groups that are geographically adjacent and have related attributes, thereby generating address clustering and association feature groups, includes: In step S121, the geographic coordinate information in the user address information group is parsed, a mapping relationship group between user address and geographic coordinate is established, and an address coordinate list group is formed.

[0071] In this embodiment, the collected user's actual residential address information is converted into corresponding geographic coordinate information using address resolution services provided by a professional Geographic Information System (GIS). For example, "a certain city, a certain district, a certain street, a certain community" is resolved into specific latitude and longitude coordinates. During the resolution process, for possible address ambiguities or vagueness, precise matching is performed by combining information such as surrounding landmarks and roads to ensure the accuracy of the resolution results.

[0072] In this embodiment of the disclosure, a one-to-one mapping relationship is established between each user address and its corresponding geographic coordinates, forming an address coordinate list group. Each record in the address coordinate list group contains a user identifier, a user address, and corresponding geographic coordinates, such as (User Identifier A, a certain city, a certain district, a certain street, a certain community, latitude and longitude coordinates A), (User Identifier B, a certain city, a certain district, a certain street, a certain community, latitude and longitude coordinates B), etc.

[0073] In step S122, the adjacency relationship between any two user addresses in the address coordinate list group on the geographic coordinates is identified, and user address pairs whose adjacent distance is within a set range are filtered out.

[0074] In this embodiment of the disclosure, the actual distance between any two user addresses is calculated based on the geographic coordinate information in the address coordinate list group. During the calculation, a distance calculation method suitable for latitude and longitude coordinates is used to ensure the accuracy of the distance results.

[0075] In this embodiment, an adjacent distance range is set, which can be adjusted according to the actual application scenario and analysis needs. For example, in urban areas, the adjacent distance range can be set to a smaller value, while in suburban or rural areas, the range can be appropriately expanded. User address pairs whose calculated distances fall within this set range are filtered out to form user address pair groups. Through this step, user addresses that are geographically close can be initially filtered out.

[0076] In step S123, the association degree of the attribute information of the user address pair group is determined based on the attribute information group in the user association information group.

[0077] Among them, the attribute information correlation degree is used to reflect the tightness of attribute association.

[0078] In this embodiment of the disclosure, attribute information groups are accurately extracted from user association information groups, and user attribute information corresponding to each pair of user addresses in the user address pair group is analyzed in depth to determine the degree of correlation between their attribute information, thereby judging the closeness of the association between users in terms of attributes.

[0079] In step S124, user address pairs whose adjacent distances are within a set range and whose attribute information correlation meets the set conditions are aggregated to form an initial address group, wherein the user addresses in the initial address group are geographically adjacent and have related attributes.

[0080] In this embodiment, user address pairs whose adjacent distances are within a set range and user address pairs whose attribute information correlation meets set conditions are combined, and user address pairs that simultaneously meet both conditions are aggregated. During the aggregation process, a suitable clustering method is used to gradually merge related user addresses together to form initial address groups. User addresses within each initial address group are not only geographically adjacent but also have strong correlation in attribute information.

[0081] In step S125, the initial address group is expanded to include other user addresses whose adjacent distance to the edge user addresses of the initial group is within a set range and whose attribute information correlation meets the standard, thus forming a target user address group.

[0082] In this embodiment of the disclosure, to more comprehensively cover associated user addresses, the initial address group needs to be expanded. First, the edge user addresses of the initial group are identified; these addresses are typically located on the periphery of the group. Other user addresses whose adjacent distance to these edge user addresses is within a set range and whose attribute information correlation meets the standard are calculated. These addresses are then included in the initial address group, forming the final user address group. This expansion allows the resulting user address group to more completely reflect the overall integrity of geographical location and attribute correlation.

[0083] In step S126, for each target user address group, the geographical boundary, attribute association pattern and user distribution density of the group are determined, and the address cluster association feature group is generated by integration.

[0084] Specifically, the address clustering association feature group is generated by integrating the group's geographical boundary (based on the geographical coordinates of the farthest address within the group), attribute association pattern (the distribution characteristics of the association degree of user attribute information within the group), and user distribution density (the number of user addresses within a unit geographical area).

[0085] In this embodiment of the disclosure, for each final user address group, its geographical boundary is first determined. By finding the extreme values ​​of latitude and longitude coordinates among all user addresses within the group, i.e., the easternmost, westernmost, northernmost, and southernmost coordinates, the area enclosed by these four coordinates is the geographical boundary of the group, used to clearly define the geographical location range of the group.

[0086] In this embodiment of the disclosure, the attribute association pattern is determined by analyzing the distribution of the attribute information association degree of all user address pairs within the group, including statistical characteristics such as the central tendency and dispersion of the association degree, so as to reflect the overall characteristics and distribution pattern of attribute association within the group.

[0087] In this embodiment, the calculation of user distribution density first involves calculating the area of ​​the region based on the geographical boundary of the group, and then dividing the number of user addresses within the group by the area to obtain the number of user addresses per unit geographical area, i.e., the user distribution density, which reflects the density of user addresses within the group. The geographical boundary of the group, attribute association patterns, and user distribution density are integrated to form the address aggregation association feature group of the user address group.

[0088] In a preferred embodiment, step S123, determining the attribute information association degree of the user address pair group based on the attribute information group in the user association information group, includes: In step S1231, the attribute information group in the user-associated information group is classified and encoded, and the non-numerical attributes of occupational information, family composition information and consumption characteristic information are converted into numerical codes to form an attribute code vector group.

[0089] In this embodiment of the disclosure, since the attribute information is mostly non-numerical data, it needs to be classified and encoded to facilitate subsequent calculations and analysis. For occupational information, different occupational categories are assigned specific codes, such as encoding "teacher" with a specific value and "doctor" with another specific value, to ensure that each occupational category has a unique corresponding code.

[0090] In this embodiment of the disclosure, regarding family composition information, different family compositions are coded, such as "single," "married without children," and "married with children," each corresponding to a different code. Consumption characteristic information is coded according to consumption habits, consumption capacity, etc., such as "frugal consumption," "moderate consumption," and "high-end consumption," each assigned a different code value.

[0091] In this embodiment, the codes for each user's occupation, family structure, and consumption characteristics are combined to form the user's attribute coding vector. For example, if a user's occupation is teacher, their family structure is married with children, and their consumption characteristic is moderate, their attribute coding vector consists of the codes corresponding to these three attributes. The attribute coding vectors of all users together constitute an attribute coding vector group, realizing the numerical conversion of attribute information.

[0092] In step S1232, each attribute encoding vector group is normalized so that the encoded values ​​of different attributes are in the same numerical range.

[0093] In this embodiment of the disclosure, since the range of encoded values ​​for occupation, family composition, and consumption characteristics may differ, in order to eliminate the influence of different dimensions between attributes and ensure the rationality and accuracy of subsequent calculations, it is necessary to normalize each attribute encoding vector group.

[0094] In this embodiment, normalization maps the encoded values ​​of different attributes to the same numerical range, such as [0,1]. During the processing, the maximum and minimum values ​​are determined based on the range of encoded values ​​for each attribute, and each encoded value is converted to the target range using appropriate conversion methods. For example, for occupational codes, the encoded values ​​corresponding to its maximum and minimum values ​​are converted to 1 and 0 respectively, and other encoded values ​​are converted to the [0,1] range proportionally. After normalization, the encoded values ​​of different attributes are comparable, facilitating the calculation of attribute correlation.

[0095] In step S1233, the association similarity between the attribute encoding vector groups of the user address pair groups is calculated, and the association similarity is used as the initial attribute information association degree.

[0096] In this embodiment of the disclosure, for each pair of users in the user address pair group, the association similarity between the two is calculated based on their normalized attribute encoding vectors. A method suitable for vector similarity calculation is used to obtain the association similarity value by comparing the overall similarity of the two vectors. This association similarity value reflects the degree of similarity between the two users in their attribute information; the larger the value, the stronger the attribute association between the two. This association similarity is used as the initial attribute information association degree.

[0097] In step S1234, weights are set according to the importance of different attributes in risk assessment, and the initial attribute information correlation degree is adjusted by weighting. The weighted attribute information correlation degree is the sum of the products of the correlation similarity of each attribute and its corresponding weight.

[0098] In this process, weights are assigned based on the importance of different attributes in risk assessment, with occupational information taking precedence over family composition information, and family composition information taking precedence over consumption characteristics information.

[0099] In this embodiment of the disclosure, different attribute information has varying degrees of impact on risk assessment in financial credit risk assessment. Generally, occupational information is more important than family composition information and consumption characteristics information because occupation is directly related to the user's income source and stability; family composition information is next, as it affects the user's expenditure structure and repayment pressure; consumption characteristics information is relatively less important.

[0100] In this embodiment, based on this importance ranking, corresponding weights are assigned to occupational information, family composition information, and consumption characteristic information, with occupational information having the highest weight, followed by family composition information, and consumption characteristic information having the lowest weight. The initial attribute information correlation is then weighted and adjusted according to the weights of each attribute to more accurately reflect the contribution of different attributes to the correlation. The weighted and adjusted attribute information correlation is used to reflect the degree of association between the attribute information of user address pairs.

[0101] In this embodiment, the weighted attribute information correlation degree is obtained by multiplying the similarity of each attribute association by its corresponding weight and then summing these products. This attribute information correlation degree comprehensively considers the importance of different attributes and can more objectively reflect the degree of correlation between the attribute information of user address pairs.

[0102] For example, if the products of the similarity of occupational information association with its corresponding weight, the similarity of family composition information association with its corresponding weight, and the similarity of consumption characteristic information association with its corresponding weight are all fixed values, then adding these three values ​​together yields the weighted adjusted association degree of the attribute information. The higher this association degree, the stronger the association between the user address pairs in terms of attribute information.

[0103] In step S1235, the weighted attribute information correlation degree is subjected to threshold filtering, and user address pairs whose attribute information correlation degree exceeds the set threshold are retained to obtain the attribute information correlation degree of the user address pairs.

[0104] In this embodiment, an attribute information correlation threshold is set, which can be adjusted according to the accuracy requirements of the actual risk assessment and the characteristics of the data. User address pairs with a weighted adjusted attribute information correlation exceeding the set threshold are retained, while those below or equal to the threshold are discarded. Through threshold filtering, user address pairs with closely related attributes can be further selected, ensuring that the resulting user address groups have strong attribute correlation.

[0105] In a preferred embodiment, step S13, which involves inputting the address aggregation association feature group and the behavioral information group from the user association information group into a pre-trained risk assessment AI model, and generating a risk association transmission feature group through risk transmission link analysis of the risk assessment AI model, includes: In step S131, the address clustering association feature group and the behavior information group in the user association information group are input into the risk assessment AI model to generate a geographic-behavioral interaction feature group.

[0106] In a preferred embodiment, step S131, which involves inputting the address clustering association feature group and the behavioral information group from the user association information group into the risk assessment AI model to generate a geographic-behavioral interaction feature group, includes: The address clustering and association feature group is input into the geographic feature processing module of the risk assessment AI model, and geographic feature vector group is obtained by feature embedding to retain geospatial association information.

[0107] In this embodiment of the disclosure, the geographic feature processing module of the risk assessment AI model includes a dedicated feature embedding layer for processing address clustered and associated feature groups. This geographic feature processing module receives features such as group geographic boundaries, attribute association patterns, and user distribution density, and converts these features into fixed-dimensional geographic feature vector groups through the embedding layer.

[0108] In this embodiment of the disclosure, a specific embedding algorithm is employed during the conversion process to ensure that geospatial association information is preserved. For example, geographically adjacent groups have high similarity in their geographic feature vectors, while geographically distant groups have significantly different geographic feature vectors. Each vector in the geographic feature vector group contains condensed information on the geographic and attribute association features of the corresponding group.

[0109] The behavioral information group in the user-associated information group is input into the behavioral feature processing module of the risk assessment AI model, and the behavioral feature vector group is obtained through time-series encoding conversion.

[0110] The behavioral information group includes user transaction information, social interaction information, and activity trajectory information, while the behavioral feature vector group includes behavioral frequency features, behavioral type features, and behavioral time features.

[0111] In this embodiment, the behavior feature processing module is responsible for processing and transforming user behavior information. This module first preprocesses transaction information, social interaction information, and activity trajectory information, including cleaning abnormal data and filling in missing values, to ensure data quality.

[0112] Then, this behavioral information is converted into behavioral feature vector sets through temporal encoding. For transaction information, features such as transaction frequency, transaction type distribution, and transaction time patterns are extracted; for social interaction information, features such as interaction frequency, interaction object characteristics, and interaction time distribution are extracted; for activity trajectory information, features such as activity range characteristics and activity time patterns are extracted. These features are encoded into fixed-dimensional vectors, forming behavioral feature vector sets, where each vector contains information such as the corresponding user's behavioral frequency characteristics, behavioral type characteristics, and behavioral time characteristics.

[0113] The geographic feature vector group and the behavioral feature vector group are input into the risk association mining module of the risk assessment AI model. The association weights of the two features are analyzed through a two-way interaction mechanism, and the geographic-behavioral interaction feature group is generated based on the association weights.

[0114] In this embodiment, the risk association mining module employs a two-way interaction mechanism to analyze the association between the geographic feature vector group and the behavioral feature vector group. This two-way interaction mechanism allows the geographic feature vector to focus on the related parts of the behavioral feature vector, while the behavioral feature vector can also focus on the related parts of the geographic feature vector. Through this two-way focus, the association weight between the two is calculated.

[0115] Based on these association weights, the geographic feature vector group and the behavioral feature vector group are fused to generate a geographic-behavioral interaction feature group. During the fusion process, geographic features and behavioral features are combined accordingly according to the magnitude of the association weights, so that the interaction features contain both geographic and behavioral information, while highlighting the relationship between the two.

[0116] In step S132, the intra-group transmission analysis module of the risk assessment AI model is used to mine the intra-group association of the geographic-behavioral interaction feature group, identify the risk mutual influence path between any two user addresses in the group, and form the intra-group risk transmission path.

[0117] In this embodiment, the intra-group transmission analysis module is specifically designed to analyze the risk transmission within a single user address group. Based on geographic-behavioral interaction feature sets, this module delves into the relationships between users within the group, identifying the mutual influence paths of risks within the group.

[0118] In step S133, the cross-group diffusion analysis module of the risk assessment AI model is used to compare and analyze the geographic-behavioral interaction feature groups of different user address groups, identify group pairs with similar features, analyze the risk transmission path between group pairs, and generate cross-group risk diffusion paths.

[0119] In this embodiment of the disclosure, the cross-group diffusion analysis module can focus on the possibility of risk transmission between different user address groups, and by comparing and analyzing the geographic-behavioral interaction characteristics of different groups, it can explore potential cross-group risk diffusion paths.

[0120] In step S134, the risk assessment AI model's coupling path analysis module performs a fusion analysis of intra-group risk transmission paths and cross-group risk diffusion paths, and generates behavior-address risk coupling paths by combining attribute association patterns in the address cluster association feature group.

[0121] In this embodiment, the coupling path analysis module uses intra-group risk transmission paths and cross-group risk diffusion paths as the basis for analysis. It introduces attribute association patterns from address clustering and association feature groups to analyze the interaction between behavioral features and address features in the risk transmission and diffusion process. For example, attribute association patterns among users within a group (such as shared occupations) can strengthen risk transmission through behavioral interactions (such as intra-industry transactions), while geographical proximity across groups, along with inter-group social interactions, can jointly promote risk diffusion. Through this fusion analysis, risk paths where behavioral features and address features are coupled and influenced are identified—that is, behavioral-address risk coupling paths. These address risk coupling paths encompass both the driving force of user behavior on risk transmission and the influence of address clustering features on risk diffusion.

[0122] In step S135, the risk transmission path within the group, the risk diffusion path across groups, and the behavior-address risk coupling path are integrated to form the risk association transmission feature group.

[0123] Among them, the risk association transmission characteristic group comprehensively covers the various transmission path characteristics of risk within groups, between groups, and under the coupling effect of behavior and address.

[0124] In a preferred embodiment, in step S132, the intra-group transmission analysis module of the risk assessment AI model performs intra-group correlation mining on the geographic-behavioral interaction feature group to identify the risk mutual influence path between any two user addresses within the group, forming an intra-group risk transmission path, including: In step S1321, the intra-group transmission analysis module of the risk assessment AI model is used to uniquely identify each user address in the geographic-behavioral interaction feature group and establish an intra-group user address index group.

[0125] In this embodiment of the disclosure, to facilitate the management and analysis of user addresses within a group, a unique identifier is assigned to each user address within the group, such as a numerical number or a string identifier. A user address index group is established within the group, which associates each unique identifier with the corresponding user address, such as (identifier 1, user address A), (identifier 2, user address B), etc.

[0126] In step S1322, feature sub-vectors corresponding to each user address are extracted from the geographic-behavioral interaction feature group to construct a feature matrix group within the group, wherein the rows of the matrix correspond to the user address index and the columns of the matrix correspond to the feature dimensions.

[0127] In this embodiment of the disclosure, feature sub-vectors corresponding to each user address are extracted from the geographic-behavioral interaction feature group based on the user address index group within the group. Each feature sub-vector contains the geographic-behavioral interaction feature information of that user address.

[0128] In this embodiment of the disclosure, these feature sub-vectors are arranged in rows to construct a feature matrix group within the group. The number of rows in this feature matrix group is the same as the number of user addresses within the group, with each row corresponding to the index of a user address; the number of columns is the same as the feature dimension, with each column corresponding to a specific feature.

[0129] In this embodiment, the feature dimension covers various attributes resulting from the interaction of geographical and behavioral features, such as the combined feature of geographical proximity and transaction frequency, and the combined feature of attribute association patterns and social interaction types. For example, for a group containing 5 user addresses, the feature matrix within the group will be presented as a matrix with 5 rows and multiple columns, with each row corresponding to a feature sub-vector of a user address, thus fully preserving the user's geographical-behavioral interaction features within the group.

[0130] In step S1323, the correlation strength between any two feature vectors in the group feature matrix group is analyzed through the correlation identification submodule of the intra-group transmission analysis module.

[0131] Among them, the association strength is used to reflect the degree of association between two user addresses in terms of geographic-behavioral characteristics.

[0132] In this embodiment, the association identification submodule employs vector similarity analysis to compare and analyze any two rows (i.e., feature subvectors of any two user addresses) in the feature matrix group within the group. By calculating the similarity between the two vectors in the multidimensional feature space, an association strength value is obtained. This association strength value directly reflects the degree of association between the two user addresses in terms of geographic-behavioral characteristics. A higher association strength indicates a stronger correlation between the two in terms of geographical location and behavioral patterns, and a greater likelihood of mutual risk influence. For example, if two user addresses are not only geographically close but also have highly overlapping transaction types and social contacts, their association strength value will be relatively high.

[0133] In step S1324, an intra-group association network is constructed based on the association strength, where nodes in the network represent user addresses and edge weights represent association strength.

[0134] In this embodiment of the disclosure, each user address within the group is treated as an independent node, and the association strength calculated in step S1343 is used as the weight of the edge connecting two nodes to construct an intra-group association network. In this network, the thicker the edge between nodes (the greater the weight), the stronger the geographic-behavioral characteristic association between the two user addresses. Through this network structure, the association relationship between user addresses within the group can be intuitively displayed. For example, in an intra-group association network, if the edge weight between node A and node B, and node B and node C is large, it indicates that the association between A, B, and C is relatively strong, which may form a potential risk transmission chain.

[0135] In step S1325, the path mining submodule of the intra-group transmission analysis module identifies the critical path of risk transmission in the association network. The critical path is the connection path whose association strength exceeds a set threshold.

[0136] In this embodiment, the path mining submodule first sets a correlation strength threshold, which is determined based on the overall correlation level of the network within the group and the risk assessment requirements. Then, it traverses all possible connection paths within the network, filtering out paths where the weight (correlation strength) of all edges exceeds the set threshold, and identifies these paths as critical paths for risk propagation. Identifying critical paths allows focusing on user address chains with a high probability of risk propagation within the group. For example, in the aforementioned network, if the correlation strength of AB and BC both exceed the threshold, then ABC constitutes a critical path, indicating that risk may propagate from A through B to C.

[0137] In step S1326, path feature analysis is performed on the identified critical paths to extract features such as path length, distribution of association strength on the path, and number of user addresses involved in the path, and these features are integrated to form the risk transmission path within the group.

[0138] In this embodiment, a detailed feature analysis is performed on each critical path. Path length refers to the number of user addresses contained in the path minus one; for example, the path length of ABC is 2. The association strength distribution on the path includes the association strength values ​​of each edge on the path and their changing trends; for example, the association strength of AB is 0.8, and the association strength of BC is 0.7, showing a gradually decreasing trend. The number of user addresses involved in the path is the total number of nodes contained in the path; for example, ABC involves 3 user addresses. These features are combined with the critical path itself to form a risk transmission path within the group, fully describing the possible transmission direction of risk within the group, the nodes it passes through, and the association strength characteristics.

[0139] In a preferred embodiment, in step S133, the cross-group diffusion analysis module of the risk assessment AI model compares and analyzes the geographic-behavioral interaction feature groups of different user address groups, identifies group pairs with similar features, analyzes the risk transmission path between group pairs, and generates cross-group risk diffusion paths, including: In step S1331, the geographic-behavioral interaction feature groups of all address clusters are obtained from the risk association mining module of the risk assessment AI model, and a cluster-feature mapping library is established.

[0140] In this embodiment of the disclosure, the geographic-behavioral interaction feature group corresponding to each address cluster is extracted from the risk association mining module, and the cluster identifier is associated with the feature group for storage, forming a cluster-feature mapping library. In this feature mapping library, each entry contains a unique identifier for the cluster and its corresponding geographic-behavioral interaction feature group. The feature group covers the cluster's geographic boundary features, attribute associations, and behavioral combination features, etc.

[0141] In step S1332, the feature similarity between the geographic-behavioral interaction feature groups of any two address clusters is analyzed. The feature similarity is used to reflect the degree of similarity between the two sets of features. The more significant the feature similarity, the more significant the degree of similarity.

[0142] In this embodiment of the disclosure, the geographic-behavioral interaction feature groups of any two groups in the cluster-feature mapping library are compared and analyzed. The feature vector similarity calculation method is used to measure the similarity between the two sets of features as a whole, and a feature similarity value is obtained. The larger the feature similarity value, the higher the similarity between the two groups in geographic-behavioral features. For example, if the geographic boundaries of two groups are adjacent, their user attribute association patterns are similar, and their behavioral features (such as transaction frequency and social scope) are similar, their feature similarity value will be higher.

[0143] In step S1333, address cluster pairs with feature similarity within a set range are filtered and marked as similar cluster pairs.

[0144] In this embodiment of the disclosure, a feature similarity range is defined, which is determined based on the overall distribution of feature similarity among all groups. Address clusters with feature similarity values ​​within this defined range are selected and marked as similar clusters. These clusters exhibit high similarity in geographic-behavioral characteristics and are key targets for analysis regarding potential cross-group diffusion of risks.

[0145] In step S1334, behavioral association features are extracted from the geographic-behavioral interaction feature groups of similar clustered groups, and the cross-association frequency of the two groups in transaction information and social interaction information is analyzed.

[0146] In this embodiment, features related to behavioral associations are extracted from the geographic-behavioral interaction feature set of similar clustered groups, with a focus on analyzing the cross-correlation of transaction information and social interaction information between users within two groups. The cross-correlation frequency includes the number of transactions and the frequency of social interactions between users in group A and users in group B. A higher frequency of social interaction indicates a stronger behavioral interaction between users in the two groups, and a higher probability of risk transmission through behavioral interaction. For example, if multiple users in group A have frequent financial transactions and social interactions with multiple users in group B, their cross-correlation frequency will be relatively high.

[0147] In step S1335, the probability of risk transmission between groups is analyzed based on the cross-association frequency and feature similarity. The probability of risk transmission is then analyzed by time decay, and the probability value is adjusted by combining the time interval between the two groups of historical risk events.

[0148] Among these, the more significant the frequency of cross-association and the more significant the feature similarity, the more significant the likelihood of risk transmission. The longer the time interval, the more significant the decay in the likelihood of risk transmission.

[0149] In this embodiment, the probability of risk transmission between groups is assessed by comprehensively considering two factors: cross-association frequency and feature similarity. Cross-association frequency and feature similarity are assigned corresponding weights, and a weighted calculation is used to obtain the risk transmission probability value. The higher the cross-association frequency and the more significant the feature similarity, the greater the risk transmission probability value, indicating a higher likelihood of risk transmission between the two groups. For example, the risk transmission probability value of two groups with highly similar features and frequent user interaction will be significantly higher than that of a group with large feature differences and sparse interaction.

[0150] In this embodiment, considering the timeliness of risk transmission, a time decay analysis is performed on the probability value of risk transmission. Historical records of risk events occurring between the two groups are collected, and the time interval between the most recent risk event is calculated. The probability value of risk transmission is adjusted based on the length of the time interval. A longer time interval indicates weaker timeliness of risk transmission, and a greater decay in the probability value; conversely, a shorter time interval results in a smaller decay. This adjustment makes the assessment of the probability of risk transmission more consistent with actual time effects.

[0151] In step S1336, the adjusted risk transmission probability is correlated with the geographical distance of similar cluster groups to generate the cross-group risk diffusion path, which includes probability value, geographical distance and feature similarity.

[0152] In this embodiment of the disclosure, the adjusted risk transmission probability is further analyzed by combining the geographical distance between similar clusters (i.e., the shortest distance between the geographical boundaries of two groups). The closer the geographical distance, the smaller the physical barrier to risk diffusion across groups, and the higher the diffusion probability. By integrating the adjusted risk transmission probability value, geographical distance, and characteristic similarity, a cross-group risk diffusion path is formed, which fully describes the probability of risk diffusion between different groups, the geographical distance factor, and the basis of group characteristic similarity.

[0153] In a preferred embodiment, step S14, generating a user risk assessment feature set based on the risk correlation transmission feature set, includes: Extract the risk occurrence trend, risk impact range, and risk persistence characteristics from the risk correlation and transmission feature group, and integrate them to form an overall risk pattern of the address cluster group; In this embodiment, the risk correlation and transmission feature set is input into the risk pattern recognition module of the risk assessment AI model, and a comprehensive risk feature vector for each address cluster is generated through deep feature interaction. The risk pattern recognition module receives the risk correlation and transmission feature set and performs deep feature interaction processing using a deep neural network. Each layer of the network deeply fuses various features (such as path length, correlation strength, geographical distance, etc.) in the risk transmission path through nonlinear transformations and feature combinations, ultimately generating a comprehensive risk feature vector for each address cluster. This comprehensive risk feature vector encapsulates the overall risk transmission characteristics of the cluster, such as the speed of risk propagation within the cluster and the potential range of cross-cluster diffusion.

[0154] In this embodiment, pattern parsing is performed on the overall risk feature vector to extract core feature components. These core feature components include risk occurrence trends, risk impact ranges, and risk persistence characteristics, which are then integrated to form an overall risk pattern for address clusters. Pattern parsing of the overall risk feature vector is performed, and feature importance analysis identifies the core feature components that contribute the most to the overall risk pattern.

[0155] Extract local association features related to each user address within the group from the risk association transmission feature group. The local association features include the position of the user address in the risk transmission path within the group and its contribution in the behavior-address risk coupling path. In this embodiment of the disclosure, within the risk-related transmission feature group, local correlation features are extracted for each user address within the group. The position of a user address in the risk transmission path within the group includes whether it is the starting point, intermediate node, or ending point of the path. Different positions indicate different roles in risk transmission (such as risk source or transmission intermediary). The contribution in the behavior-address risk coupling path reflects the degree to which the behavioral and address features of the user address promote risk transmission in the coupling path. The higher the contribution, the more critical the role of the user address in the coupling path.

[0156] Analyze the correlation between the local association characteristics of each user address and the overall risk pattern of the address cluster, and generate the individual risk involvement degree; Among them, the individual risk impact degree is used to reflect the extent to which a user's address is affected by the overall risk of the group.

[0157] In this embodiment, the local correlation characteristics of each user address are analyzed in relation to the overall risk pattern of the group to which it belongs, and the degree of correlation between the two is calculated to obtain the individual risk involvement degree. The higher the individual risk involvement degree, the greater the impact of the overall risk of the group on the user address. For example, user addresses located on the critical path of risk transmission within the group usually have a higher individual risk involvement degree.

[0158] The individual risk involvement degree is extended by feature expansion, and combined with the abnormal behavior features in the behavior information group corresponding to the user address, an individual risk association pattern of each user address in the group is generated. In this embodiment, based on the individual risk involvement degree, the individual risk involvement degree is extended by combining abnormal behavioral characteristics related to risk (such as frequent large transactions, social interactions with high-risk users, etc.) in the behavioral information group corresponding to the user address. By analyzing the relationship between abnormal behavioral characteristics and individual risk involvement degree, an individual risk association pattern is generated. This risk association pattern includes not only the degree to which the user address is affected by group risk, but also the risk brought about by its own abnormal behavior and the risk association with other user addresses.

[0159] The overall risk model and the individual risk correlation model are integrated to form the user risk assessment feature group.

[0160] In this embodiment of the disclosure, the overall risk pattern of each address cluster and the individual risk association pattern of each user address within the cluster are integrated to form a user risk assessment feature group, which comprehensively reflects the risk status at both the group and individual levels.

[0161] In a preferred embodiment, the step of extracting the risk occurrence trend, risk impact range, and risk persistence characteristics from the risk correlation and transmission feature group, and integrating them to form an overall risk pattern for the address cluster, includes: A dimensional importance analysis is performed on the overall risk feature vector in the aforementioned risk correlation and transmission feature group to identify the salient feature dimension that contributes most significantly to the risk pattern; In this embodiment, the pattern parsing submodule of the risk assessment AI model is invoked to perform dimensional importance analysis on the overall risk feature vector, identifying the feature dimensions that contribute most significantly to the risk pattern. The pattern parsing submodule employs a feature importance assessment algorithm to analyze each dimension of the overall risk feature vector, calculates the contribution of each dimension in determining the overall risk pattern, and selects the feature dimensions with the highest contribution to obtain the significant feature dimensions.

[0162] By learning the correlation with historical risk events, a first feature component related to the risk occurrence trend is extracted from the salient feature dimension. The first feature component is used to reflect the changing trend of the probability of risk occurrence in the address cluster group. In this embodiment of the disclosure, feature components related to the risk occurrence trend are extracted from the identified salient feature dimensions. These feature components, learned by the model from historical risk event data during the training phase, can reflect the increasing, decreasing, or stable trend of the probability of risk occurrence within the address cluster over a future period. For example, if the frequency of abnormal transactions within the cluster has recently increased and is similar to characteristics prior to historical risk events, this feature component will indicate an increasing risk occurrence trend.

[0163] Based on the geographical boundaries of address clusters and user distribution density, a second feature component related to the scope of risk impact is generated. The second feature component is used to reflect the range of user addresses and geographical areas that may be affected after the risk occurs. In this embodiment of the disclosure, feature components related to the scope of risk impact are extracted from the salient feature dimension by combining the geographical boundaries of the address cluster and the user distribution density. The geographical boundaries determine the geographical area affected by the risk, while the user distribution density affects the number of affected user addresses within that area. Together, these factors enable the feature component to accurately reflect the potential scope of the risk after it occurs. For example, a cluster with a large geographical boundary and dense user distribution will show a larger impact range in its feature components related to the scope of risk impact.

[0164] By combining the attribute association patterns of address clusters and historical risk persistence data, a third feature component related to the risk persistence characteristics is generated. The third feature component is used to reflect the duration and intensity decay trend after the risk occurs. In this embodiment of the disclosure, feature components related to the persistence of risk are extracted from the salient feature dimension by referencing the attribute association patterns (such as the degree of correlation of user attributes) of the clustered addresses and the continuous data of similar risk events in history. These feature components can reflect the duration of the risk after it occurs and the rate at which the risk intensity decays over time. For example, for clusters with closely correlated attributes, the risk duration may be longer and the intensity may decay more slowly.

[0165] The first feature component, the second feature component, and the third feature component are correlated and verified to check the logical consistency between the features and integrate them to form the overall risk pattern of the address cluster.

[0166] In this embodiment, the extracted risk occurrence trend, risk impact range, and risk persistence characteristics are correlated and verified to check the logical consistency between features. If inconsistencies exist, the weight allocation of feature extraction is readjusted. Correlation verification is performed on the three core feature components to analyze the rationality of their logical relationships. For example, an upward risk occurrence trend should generally match a larger risk impact range and a longer duration. If contradictions occur (such as an upward trend but a very small impact range), the weight allocation of each feature extraction is readjusted until the features are logically consistent.

[0167] In this embodiment, the verified risk occurrence trend, risk impact scope, and risk persistence characteristics are integrated according to a set structure to form an overall risk model of the address cluster. The verified risk occurrence trend, risk impact scope, and risk persistence characteristics are integrated according to a preset structure (such as priority ranking or logical hierarchy) to form an overall risk model of the address cluster, fully describing the overall risk status of the cluster.

[0168] This disclosure also provides a user risk assessment system based on user address aggregation. The system includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code. The processor is used to execute the programs, instructions, or code in the memory to implement the user risk assessment method based on user address aggregation described in any of the foregoing embodiments.

[0169] Figure 2 The user risk assessment system 100 based on user address aggregation shown can be configured as a cloud environment as described in any of the foregoing embodiments. The user risk assessment system 100 based on user address aggregation includes a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, for example, via a bus 1002. Optionally, the user risk assessment system 100 based on user address aggregation may further include a communication component 1004, which can be used for data interaction between the device 100 and other devices, such as sending and / or receiving data. It should be noted that in actual scheduling, the communication component 1004 is not limited to one, and the structure of this user risk assessment system 100 based on user address aggregation does not constitute a limitation on the embodiments of this application.

[0170] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0171] Bus 1002 may include a pathway for transmitting information between the aforementioned components. Bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 1002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 2 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0172] The memory 1003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing program code and capable of being read by a computer, without limitation herein.

[0173] The memory 1003 is used to store program code for executing embodiments of the present disclosure, and its execution is controlled by the processor 1001. The processor 1001 is used to execute the program code stored in the memory 1003 to implement the steps shown in the aforementioned user risk assessment system embodiment based on user address aggregation.

[0174] This disclosure also provides a computer-readable storage medium storing program code, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned user risk assessment system embodiment based on user address aggregation.

[0175] The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present disclosure, various changes, modifications, substitutions and variations can be made to these embodiments, and all such changes, modifications, substitutions and variations fall within the protection scope of the present disclosure.

[0176] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction, and such combinations should also be considered as part of this disclosure. To avoid unnecessary repetition, this disclosure will not further describe the various possible combinations. The technical scope of this application is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A user risk assessment method based on user address aggregation, characterized in that, The method comprises: obtaining a user address information set and a user association information set, the user address information set comprising actual residence address information of multiple users, and the user association information set comprising attribute information and behavior information of the users; conducting aggregated association mining on the user address information set, combining the attribute information in the user association information set, dividing out a user address group which is adjacent in geographical position and has attribute association, and generating an address aggregated association feature set; inputting the address aggregated association feature set and the behavior information in the user association information set into a pre-trained risk assessment AI model, generating a risk association conduction feature set through risk conduction link analysis of the risk assessment AI model; generating a user risk assessment feature set based on the risk association conduction feature set; feeding back the user risk assessment feature set to the aggregated association mining link, adjusting the association parameters of the next aggregated mining, and outputting a risk assessment report comprising the user risk assessment feature set and the address aggregated association feature set, wherein the association parameters comprise a set range of geographical adjacent distance and a set condition of attribute association.

2. The user risk assessment method based on user address aggregation according to claim 1, characterized in that, The aggregated association mining on the user address information set, combining the attribute information in the user association information set, dividing out a user address group which is adjacent in geographical position and has attribute association, and generating an address aggregated association feature set, comprises: analyzing geographical coordinate information in the user address information set, establishing a mapping relationship set of user addresses and geographical coordinates, and forming an address coordinate list set; identifying the adjacent relationship of any two user addresses in the address coordinate list set in geographical coordinates, and screening out a user address pair set with adjacent distance within a set range; determining the attribute information association degree of the user address pair set according to the attribute information in the user association information set; aggregating the user address pair set with adjacent distance within a set range and attribute information association degree meeting a set condition, forming an initial address group set, wherein the user addresses in the initial address group set are adjacent in geographical position and have attribute association; extending the range of the initial address group set, including other user addresses with adjacent distance within a set range to the edge user addresses of the initial group and meeting the attribute information association degree, forming a target user address group set; determining the geographical boundary, attribute association mode and user distribution density of each target user address group set, and integrating to generate the address aggregated association feature set.

3. The user risk assessment method based on user address clustering according to claim 2, characterized in that, The determination of the attribute information association degree of the user address pair set according to the attribute information in the user association information set comprises: classifying and coding the attribute information in the user association information set, converting non-numerical attributes of occupation information, family composition information and consumption characteristic information into numerical coding, and forming an attribute coding vector set; normalizing each attribute coding vector set to make the coding values of different attributes in the same numerical interval; calculating the association similarity between the attribute coding vector sets of the user address pair set, and taking the association similarity as the initial attribute information association degree; According to the importance of different attributes in risk assessment, weights are set, and the initial attribute information correlation degree is adjusted by weighting, wherein the adjusted attribute information correlation degree is the sum of the product of each attribute correlation similarity and the corresponding weight; The adjusted attribute information correlation degree is threshold filtered, and the user address pair group with an attribute information correlation degree exceeding a set threshold is retained to obtain the attribute information correlation degree of the user address pair group. 4.The user risk assessment method based on user address aggregation of claim 1, wherein, The address aggregation association feature group and the behavior information group in the user association information group are input into the pre-trained risk assessment AI model, and a risk association transmission feature group is generated through risk transmission link analysis of the risk assessment AI model, including: The address aggregation association feature group and the behavior information group in the user association information group are input into the risk assessment AI model to generate a geographic-behavior interaction feature group; Through the intra-group transmission analysis module of the risk assessment AI model, intra-group correlation mining is performed on the geographic-behavior interaction feature group to identify the risk mutual influence path of any two user addresses in the group, and an intra-group risk transmission path is formed; Through the cross-group diffusion analysis module of the risk assessment AI model, the geographic-behavior interaction feature groups of different user address groups are compared and analyzed to identify feature similar group pairs, analyze the risk transmission path between the group pairs, and generate a cross-group risk diffusion path; Through the coupling path analysis module of the risk assessment AI model, the intra-group risk transmission path and the cross-group risk diffusion path are fused and analyzed, and the attribute association mode in the address aggregation association feature group is combined to generate a behavior-address risk coupling path; The intra-group risk transmission path, the cross-group risk diffusion path, and the behavior-address risk coupling path are integrated to form the risk association transmission feature group.

5. The user risk assessment method based on user address clustering according to claim 4, characterized in that, The address aggregation association feature group and the behavior information group in the user association information group are input into the risk assessment AI model to generate a geographic-behavior interaction feature group, including: The address aggregation association feature group is input into the geographic feature processing module of the risk assessment AI model, and the geographic feature vector group is obtained by converting the geographic spatial association information through feature embedding; The behavior information group in the user association information group is input into the behavior feature processing module of the risk assessment AI model, and the behavior feature vector group is obtained by converting the time sequence coding; The geographic feature vector group and the behavior feature vector group are input into the risk association mining module of the risk assessment AI model, the correlation weight of the two features is analyzed through the bidirectional interaction mechanism, and the geographic-behavior interaction feature group is generated based on the correlation weight.

6. The user risk assessment method based on user address clustering according to claim 4, wherein, Through the intra-group transmission analysis module of the risk assessment AI model, the intra-group correlation mining is performed on the geographic-behavior interaction feature group to identify the risk mutual influence path of any two user addresses in the group, and an intra-group risk transmission path is formed, including: Through the intra-group transmission analysis module of the risk assessment AI model, each user address in the geographic-behavior interaction feature group is uniquely identified, and a group of intra-group user address indexes is established; extracting a feature sub-vector corresponding to each user address from the geographical-behavioral interaction feature set, and constructing an intra-group feature matrix set, wherein the rows of the matrix correspond to user address indexes, and the columns of the matrix correspond to feature dimensions; analyzing the correlation strength of any two feature sub-vectors in the intra-group feature matrix set through the correlation identification submodule of the intra-group conduction analysis module; constructing an intra-group correlation network based on the correlation strength, wherein the nodes in the network represent user addresses, and the weights of the edges represent the correlation strength; identifying a key path of risk conduction in the correlation network through the path mining submodule of the intra-group conduction analysis module, wherein the key path is a connection path with a correlation strength exceeding a set threshold; performing path feature analysis on the identified key path, extracting the path length, the correlation strength distribution on the path, and the number of user addresses involved in the path, and integrating to form the intra-group risk conduction path.

7. The user risk assessment method based on user address clustering according to claim 4, characterized in that, comparing and analyzing the geographical-behavioral interaction feature sets of different user address groups through the cross-group diffusion analysis module of the risk assessment AI model, identifying feature-similar group pairs, analyzing the risk transmission paths between the group pairs, and generating cross-group risk diffusion paths, including: obtaining the geographical-behavioral interaction feature sets of all address aggregation groups from the risk correlation mining module of the risk assessment AI model, and establishing an aggregation group-feature mapping library; analyzing the feature similarity between the geographical-behavioral interaction feature sets of any two address aggregation groups, wherein the feature similarity reflects the degree of similarity between the two feature sets, and the more significant the feature similarity, the more significant the degree of similarity; screening address aggregation group pairs with feature similarity within a set range, and marking them as similar aggregation group pairs; extracting behavioral correlation features in the geographical-behavioral interaction feature sets of the similar aggregation group pairs, and analyzing the cross-correlation frequency of the two groups in transaction information and social interaction information; analyzing the risk transmission possibility between the group pairs based on the cross-correlation frequency and the feature similarity, performing time decay analysis on the risk transmission possibility, and adjusting the possibility value in combination with the time interval of historical risk events of the two groups; performing correlation analysis on the adjusted risk transmission possibility and the geographical distance of the similar aggregation group pairs, and generating the cross-group risk diffusion path including the possibility value, the geographical distance, and the feature similarity.

8. The user risk assessment method based on user address clustering according to any one of claims 1-7, characterized in that, generating a user risk assessment feature set based on the risk correlation conduction feature set, including: extracting risk occurrence trends, risk impact ranges, and risk duration characteristics from the risk correlation conduction feature set, and integrating to form the overall risk pattern of the address aggregation group; extracting local correlation features related to each user address in the address aggregation group from the risk correlation conduction feature set, wherein the local correlation features include the position of the user address in the intra-group risk conduction path and the contribution degree of the user address in the behavior-address risk coupling path; analyzing the correlation degree of the local correlation features of each user address and the overall risk pattern of the address aggregation group, and generating an individual risk involvement degree. The individual risk involvement degree is characterized, combined with the abnormal behavior features in the behavior information group corresponding to the user address, to generate an individual risk association mode of each user address in the group; The overall risk mode and the individual risk association mode are integrated to form the user risk assessment feature group.

9. The user risk assessment method based on user address clustering according to claim 8, characterized in that, The risk occurrence trend, risk impact range and risk duration characteristics in the risk association conduction feature group are extracted to form an overall risk mode of the address aggregation group, including: The dimension importance analysis is performed on the overall risk feature vector in the risk association conduction feature group to identify the most significant feature dimension that contributes most to the risk mode; The first feature component related to the risk occurrence trend is extracted from the significant feature dimension through the association learning with historical risk events, and the first feature component is used to reflect the possibility change trend of the risk occurrence of the address aggregation group; Based on the geographical boundary and the user distribution density of the address aggregation group, the second feature component related to the risk impact range is generated, and the second feature component is used to reflect the user address range and geographical area that may be affected after the risk occurrence; The third feature component related to the risk duration characteristics is generated in combination with the attribute association mode of the address aggregation group and the historical risk duration data, and the third feature component is used to reflect the duration and intensity attenuation trend after the risk occurrence; The first feature component, the second feature component and the third feature component are associated and verified to check the logical consistency between the features, and the overall risk mode of the address aggregation group is integrated.

10. A user risk assessment system based on user address aggregation, characterized in that, The system comprises: a processor and a memory connected with the processor; the memory is used to store programs, instructions or codes; the processor is used to execute the programs, instructions or codes in the memory to realize the user risk assessment method based on user address aggregation in any one of claims 1 to 9.

Images