A method and system for realizing unified user identification based on graph computing

Abstract

The application discloses a kind of based on graph computing to realize unified user identification method and system, the method includes the following steps: S1: data collection: collect enterprise or organization multi-source data, and the feature existing in data is divided into priority group and confidence setting;S2: data processing: the multi-source data of S1 access is merged, generates user register, and the point and edge file for graph computing are output;S3: after reading point and edge data, obtain the connected graph structure of all data;S4: result post-processing: the result table of graph computing is combined with user register and exports OneID result table.The application is divided by the priority of feature, and optimization merging rule is effectively distinguished whether OneID sequence is merged.

Description

Technical Field

[0001] This invention relates to the fields of computer networks and data processing technology and ID-Mapping, and in particular to a method and system for implementing unified user identification (OneID) based on graph computing. Background Technology

[0002] In the era of data explosion, the amount and types of data generated daily are growing exponentially. For users, creating and managing multiple accounts and passwords across different applications and services is not only cumbersome but also increases the likelihood of forgotten passwords and security risks. For most enterprises, their customer data is scattered across various systems and channels, lacking a unified customer data asset and failing to manage and extract value from the data. This makes it difficult for enterprises to generate effective metrics or tags to support business value realization and optimize customer experience. Furthermore, the simultaneous existence of user information on multiple digital products or devices undoubtedly increases the difficulty of accurately identifying users and hinders the advancement of digital marketing.

[0003] Accurate user identification is key to generating user profiles and implementing personalized user operations. In practice, all data is fragmented and lacks unique identifiers. How to reassemble and organize these data pieces and eliminate data silos has always been an urgent problem to be solved in the process of enterprise digitalization and intelligentization. The emergence of OneID is precisely to solve this problem. OneID is a concept of Single Identity. Its goal is to provide a unified identity identifier through various technical means, so that user data from different sources and in different forms can be associated and integrated based on it. Users can also achieve seamless identity verification and data sharing between multiple applications and services. Through OneID, enterprises can achieve full-domain management of user identity and full-link observation of user journey, thereby gaining a more comprehensive understanding of user behavior information and attribute characteristics, and more accurately evaluating user value, laying the foundation for refined operations. Compared with traditional user identification methods based on cookies and device fingerprints, OneID has the following advantages: (1) Uniqueness: Each real user corresponds to only one OneID, avoiding duplicate counting and missed counting. (2) Stability: OneID does not change with users changing devices or deleting cookies, and can continuously identify users. (3) Global reach: OneID can track user behavior across various digital products, channels, and devices to achieve a comprehensive customer view. (4) Scalability: OneID has a massive scale and can identify hundreds of millions of users, providing a foundation for user identification in a big data environment.

[0004] Currently, OneID is mainly implemented through streaming computing and batch computing. Streaming computing offers strong real-time performance, but its initialization cost is high and it is not stable enough for large datasets. Batch computing can balance resource and time costs, but it can only achieve simple rule matching, and its performance is poor when dealing with complex rule matching of large datasets. Graph computing technology effectively solves the problems encountered by the above two technologies. It has strong scalability and stability and can handle complex matching rules for large datasets. Graph computing has been widely used in fields such as healthcare, finance, social analytics, natural sciences, and transportation. The most basic data structure in a graph computing system consists of three factors: nodes V (vertex), edges E (edge), and weights D (data), i.e., G = (V, E, D). Nodes represent objects in a certain event, while edges describe the relationships between different objects. In the OneID scenario, nodes are used to describe user characteristics, edges are used to describe the relationships between characteristics, and the weights of the edges are used to describe the confidence of the relationships between characteristics. Among them, the WCC (Weakly Connected Components) algorithm based on graph computing is a commonly used algorithm for user identification. It can describe the connectivity of a graph, distinguish connected subgraphs, and achieve the result of distinguishing user features. However, in real-world customer data, some features may belong to multiple users or be features of agents. In the graph structure, this is represented by different users being interconnected based on this feature. If OneID is set based on the WCC calculation result, it is impossible to distinguish different users, and the OneID result is inaccurate. Therefore, existing graph-based user identification algorithms cannot well meet the needs of complex user association scenarios. Summary of the Invention

[0005] To address the problems existing in current technologies, the present invention aims to provide a method for implementing Unified User Identification (OneID) based on graph computing. By prioritizing features and optimizing merging rules, it effectively distinguishes whether OneID sequences have been merged. Furthermore, a new feature confidence score is added, characterizing the degree of credibility of a feature attached to a particular OneID, which can assist marketers in decision-making.

[0006] To achieve the above objectives, the present invention provides a method for unified user identification based on graph computing, the method comprising the following steps:

[0007] S1: Data Collection: Collect multi-source data from enterprises or organizations, and prioritize and set confidence levels for the features present in the data;

[0008] S2: Data Processing: Merge multi-source data accessed by S1, generate user registry, and produce point and edge files for graph computation;

[0009] S3: After reading the point and edge data, obtain the connected graph structure of all data;

[0010] S4: Post-processing of results: Combine the graph calculation result table with the user registry to output the OneID result table.

[0011] Furthermore, step S2 includes:

[0012] S21: Extract all feature data from all data sources connected to S1 and merge it into the user registry; add an auto-incrementing "index" field to the user registry as the primary key and also as the tag for each data entry.

[0013] S22: Add a "Temporary OneID" field to the user registry obtained in S21. The value of this field is determined according to the priority classification table in S1: the identifier of the highest priority group feature of each data entry plus this feature value is the temporary OneID.

[0014] S23: Extract the index and each feature field from the user registry obtained in S22 and remove duplicates to obtain six CSV point files: index, ID number, unionid, mobile phone number, device number, and cookie; extract the fields "index-ID number-ID number confidence", "index-unionid-unionid confidence", "index-mobile phone number-mobile phone number confidence", "index-device number-device number confidence", and "index-cookie-cookie confidence" from the user registry obtained in S22 to obtain five CSV side files.

[0015] S24: Perform anomaly checks on the point and edge files generated in S23; if the point and edge files do not exist or there are abnormal features in the edge files, the data processing in step S2 needs to be repeated to generate new point and edge files.

[0016] Furthermore, in step S3, the graph distinguishes between index nodes and feature nodes. In the edge file, index-feature node pairs constitute edges, and the feature confidence is the attribute value of the edge.

[0017] Furthermore, step S3 also includes implementing the OneID algorithm, which features distributed computing and easy deployment, based on the Pregel computing model and the Kubernetes platform.

[0018] Furthermore, the calculation process in step S3 consists of several supersteps separated by global synchronization points. The system moves from one "superstep" to the next "superstep" until the iterative calculation ends. Whether the calculation ends depends on the state of the nodes in the graph and whether there is information transmission.

[0019] Furthermore, in the first superstep, every node is in an active state, and active nodes participate in the computation and information transmission of this round of supersteps; a node can vote to become inactive, and in subsequent supersteps, the node will no longer participate in the relevant computations unless it receives information from other nodes, at which point its state will become active again.

[0020] Furthermore, when all nodes in the graph are inactive and there is no information transmission, the termination condition is met, the iterative calculation ends, and the graph calculation result file is output.

[0021] Furthermore, step S3 includes:

[0022] S31: Initialize the node attribute values ​​in the graph structure;

[0023] S32: After initialization, the index node converts the attribute value into a message and sends it to the adjacent feature node, and sets its state to inactive;

[0024] S33: The feature node receives and processes the message passed from the index node in S32, and updates the feature node attribute value;

[0025] S34: The index node receives and processes the message passed from the feature node in S33, and updates the index node attribute value;

[0026] S35: Repeat steps S33 and S34, with all nodes continuously updating attribute values ​​and sending messages until all nodes are inactive and no messages are being sent. Then, terminate the OneID algorithm and output the attribute values ​​of all feature nodes to obtain the graph calculation result table.

[0027] Furthermore, in step S35, the attribute value of each feature includes the index value and confidence level of the OneID to which the feature belongs.

[0028] On the other hand, the present invention provides a system for implementing unified user identification based on graph computing, the system being applied to implement the method for implementing unified user identification based on graph computing according to the present invention.

[0029] Beneficial effects:

[0030] This invention addresses the contradiction between existing technologies and the OneID algorithm's inability to effectively balance computational time and cost for large datasets. Furthermore, considering the high connectivity of user features and complex differentiation logic in real-world unified user identification scenarios, this invention proposes a graph-based method for unified user identification (OneID). By prioritizing features and optimizing merging rules, data with different high-priority feature groups cannot be associated or merged into OneIDs through low-priority feature groups, thus being considered different individuals. Data associated through high-priority feature groups are merged into OneIDs, with all features attached to the same OneID, and are considered the same individual. This effectively distinguishes whether OneID sequences have been merged. Additionally, a new feature confidence score is added, representing the credibility of a feature attached to a particular OneID, which can assist marketers in decision-making. In the graph, this is represented as the product of the attribute values ​​of the edges traversed on the path from a given feature to a OneID feature. Attached Figure Description

[0031] Figure 1 A flowchart of the anti-tampering method based on the SHA-256 hash algorithm interface for adding signatures according to the present invention is shown;

[0032] Figure 2 A flowchart illustrating the process of a client requesting a server using an interface carrying a signature, according to an embodiment of the present invention, is shown. Detailed Implementation

[0033] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0034] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0035] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0036] The following is in conjunction with the appendix Figure 1-2 Specific embodiments of the present invention will be described in detail below. It should be understood that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the present invention.

[0037] To achieve the above objectives, such as Figure 1 As shown, the method for implementing unified user identification (OneID) based on graph computation according to the present invention includes the following steps:

[0038] S1: Data Collection: Collect multi-source data from a company or organization, and prioritize and set confidence levels for the features present in the data.

[0039] S2: Data Processing: Merges multi-source data accessed by S1, generates a user registry, and produces point and edge files for graph computation.

[0040] S3: After reading the point and edge data, obtain the connected graph structure of all data.

[0041] S4: Post-processing of results: Combine the graph calculation result table with the user registry to output the OneID result table.

[0042] Specifically, step S1: Data Collection: Collect multi-source data from a company or organization, and prioritize and set confidence levels for the features present in the data. This invention uses ID numbers, unionids, mobile phone numbers, device numbers, and data stored on user terminals (cookies) as examples, but is not limited to these. Details of the priority division are shown in Table 1 (the specific features and division can be flexibly changed according to the company's requirements, and have strong scalability).

[0043] Table 1 Priority Division Table

[0044]

[0045]

[0046] S2: Data Processing: Merging multi-source data from S1, generating a user registry, and producing vertex and edge files for graph computation. The specific implementation is as follows:

[0047] S21: Extract all feature data from all data sources connected to S1 and merge it into the user registry; add an auto-incrementing "index" field to the user registry as the primary key and also as the tag for each data entry.

[0048] S22: Add a "Temporary OneID" field to the user registry obtained in S21. The value of this field is determined based on the priority classification table (Table 1) of S1: the identifier of the highest priority group feature of each data entry plus that feature value is the temporary OneID. For example, for the ID number "c001", the temporary OneID value is "c_c001", and so on. The priority classification table can be flexibly set according to the company's requirements.

[0049] S23: Extract the index and each feature field from the user registry obtained in S22, and remove duplicates to obtain six CSV point files: index, ID number, unionid, mobile phone number, device number, and cookie. CSV is a comma-separated value format. Extract the fields "index-ID number-ID number confidence score", "index-unionid-unionid confidence score", "index-mobile phone number-mobile phone number confidence score", "index-device number-device number confidence score", and "index-cookie-cookie confidence score" from the user registry obtained in S22 to obtain five CSV edge files. The edges represent the association information between features.

[0050] S24: Perform anomaly detection on the point and edge files generated in S23. Anomaly detection includes the following two aspects: ① Analyze the target folder sequentially using the file judgment function to see if the 6 CSV point files and 5 CSV edge files defined in S23 exist. If they do not exist, an anomaly is raised; ② Determine if there are any abnormal feature values; sequentially read the 5 edge files mentioned in S23 and count the frequency of feature fields appearing in the files. If the frequency of a feature exceeds a set threshold, it is judged as an abnormal feature value and removed. If an anomaly is found in this step, the data processing in step S2 needs to be repeated to generate new point and edge files. If there are no anomalies, proceed to step S3.

[0051] S3: After reading the point and edge data, obtain the connected graph structure of all data. Distinguish between index nodes and feature nodes in the connected graph. Connect the index nodes and feature nodes from the edge file to form edges, with the feature confidence being the attribute value of the edge. Then, based on the Pregel computation model and the Kubernetes platform, implement distributed computing and the OneID algorithm to effectively support large-scale data computation and improve computational efficiency. The entire computation process is as follows: Figure 2As shown. Here, "superstep" refers to each round of iterative computation. Each iteration must wait for all nodes to complete their computations before the generated message is passed to the next "superstep". Figure 1 The diagram illustrates nodes 1, 2, and 3 in a graph structure, connected by directed edges. Node 1 can send messages to nodes 2 and 3, and node 3 can send messages to nodes 1 and 2. When nodes finish computation, they generate messages and pass them to their neighbors via edges. In superstep 0, node 1 generates `msg_1`, which can be passed to nodes 2 and 3 via directed edges. Node 3 generates `msg_2`, which can be passed to nodes 1 and 2 via directed edges. Node 2 has no edges pointing to nodes 1 and 3, so it does not pass any messages. In superstep 1, nodes 1, 2, and 3 receive messages from superstep 0, process the messages, and update their nodes. Finally, similar to superstep 0, nodes 1 and 3 generate new messages `msg_3` and `msg_4`, respectively, and pass them to their neighbors via directed edges. The entire process consists of several supersteps separated by global synchronization points. The system moves from one superstep to the next until the iterative computation ends. Whether the computation ends depends on the state of the nodes in the graph and whether any messages have been passed. Node states are categorized as active and inactive. An active node participates in the node value update calculation and message sending for the current superstep. An inactive node does not participate in the node value update calculation and message sending for the current superstep if it has not received any new messages. A node can vote to become inactive. In the first superstep, every node is active and performs message passing. In subsequent supersteps, an inactive node no longer participates in the calculation unless it receives a message from another node, at which point it becomes active again and participates in the superstep calculation. When all nodes are inactive and no nodes perform message passing in a superstep, the termination condition is met, the iteration ends, and the graph calculation result file is output. This calculation process is determined by the OneID algorithm, whose specific steps are as follows:

[0052] S31: Initialize the node attribute values ​​in the graph structure. Attribute values ​​consist of four sets of values ​​separated by vertical bars, for example, "2|1|1|0.5". Attribute values ​​are explained as follows: ① For the index node, the first value records the priority group identifier of the highest priority feature initially contained in this data (Table 1). During algorithm iteration, this index node only receives information from feature nodes with this priority group identifier. For feature nodes, the first value records the priority group identifier of that feature. This value remains unchanged for all nodes throughout the entire algorithm iteration; ② The second value records the feature ranking of the highest priority feature that the node can be associated with in the current superstep; ③ The third value records the index value of the highest priority feature associated with the node in the current superstep; ④ The fourth value records the confidence that the node belongs to the highest priority feature in the current superstep, i.e., the product of the confidence of the path (edge) from the node to the associated highest priority feature.

[0053] S32: After initialization, the index node converts its attribute values ​​into messages and sends them to adjacent feature nodes, then sets its state to inactive. Since the confidence level of node information decreases as the transmission path increases when node attribute values ​​are passed to other nodes through the graph structure, the fourth value, the confidence level, needs to be changed when converting attribute values ​​into messages. Specifically, the confidence level in the attribute value is multiplied by the attribute value of the edge traversed during message transmission, thus achieving real-time updates to the confidence level of node information during transmission. Simultaneously, this invention also adds a confidence threshold. If the confidence level in a message is lower than the set threshold, the message is not transmitted and does not participate in the calculation of the next superstep. Setting a confidence threshold effectively prevents low-confidence information from being transmitted and attached to other nodes, ensuring the accuracy of OneID identification.

[0054] S33: The feature node receives and processes the message passed from the index node in S32, and updates the feature node attribute values. The message processing and node attribute value update steps are as follows:

[0055] S331. Compare the message's second attribute value with the second value of the feature's own attribute value. If the message's second value is smaller and the feature's ranking is higher, then all attribute values ​​of the feature node except the first value are updated to the corresponding values ​​in the message. For example, if the feature node's attribute value is "2|2|2|0.5" and the received message's attribute value is "1|1|1|0.8", then the feature node's attribute value is updated to "2|1|1|0.8".

[0056] S332. When the message and the feature node have the same feature order, the attribute value and the message's third value, i.e., the index value of the highest priority feature to which they belong, are deduplicated and merged, separated by ":". The confidence value also corresponds one-to-one with the merged index. For example, if the feature node's attribute value is "2|1|2|0.5" and the received message's attribute value is "1|1|1|0.8", then the feature node's attribute value is updated to "2|1|1:2|0.8:0.5".

[0057] S333. When the message's feature ranking is low, the feature node attribute value is not updated. For example, if the feature node attribute value is "2|1|2|0.5" and the received message attribute value is "2|2|1|0.8", the feature node attribute value will still be "2|1|2|0.5" after this round of calculation.

[0058] S334. After message processing is complete, if the attribute value of a feature node changes, message passing will occur. The feature node will convert the attribute value into a message and send it to the adjacent index nodes, and set the state of all nodes to inactive. The message conversion and sending process is the same as described in S32.

[0059] S34: The index node receives and processes the message passed from the feature node in S33, and updates the index node's attribute values. The message processing and attribute value update steps are as follows:

[0060] S341. First, message filtering is performed. Only messages with the same initial highest feature priority group as the index node can participate in attribute value update calculations. For example, if the attribute value of the index node is "2|1|1|0.8", and the attribute value of the received message is "2|1|2|0.8", the first value is equal, and the message participates in subsequent calculations. However, if the attribute value of the received message is "3|1|3|0.8", the first value is not equal, and the message does not participate in subsequent node attribute value calculations. This step can effectively distinguish whether multiple connected data messages belong to the same OneID (same natural person) based on feature priority rules.

[0061] S342. After message filtering, update the attribute values ​​of feature nodes in the same way as described in S33.

[0062] S343. After message processing and attribute value updates are completed, if the attribute value of the index node changes, a message will be passed.

[0063] The S345.index node converts the attribute value into a message and sends it to neighboring feature nodes, then sets the state of all nodes to inactive. The message conversion and sending process is the same as described in S32.

[0064] S35: Repeat steps S33 and S34, with all nodes continuously updating attribute values ​​and sending messages until all nodes are inactive and no messages are being sent. Then, terminate the OneID algorithm and output the attribute values ​​of all feature nodes to obtain the graph calculation result table. At this point, the attribute value of each feature includes the index value and confidence score of the OneID to which that feature belongs.

[0065] S4: Post-processing of results: Combine the graph calculation result table obtained in S35 with the user registry obtained in S22 to output the OneID result table. Map the index value in each feature attribute value to the row in the user registry where that index value is located, and form the temporary OneID field of this row as the OneID of that feature.

[0066] This invention provides a system for implementing unified user identification based on graph computing, the system being applied to implement the method for implementing unified user identification based on graph computing according to the present invention.

[0067] To verify the effectiveness of this invention, the accuracy and performance of the OneID algorithm of this invention were verified and tested using user test data of different orders of magnitude.

[0068] This invention utilizes 40 collected user data entries. The data encompasses all possible scenarios involving multi-level associations of user characteristics in real-world situations, used to verify the accuracy of the OneID algorithm and compare it with the WCC algorithm. Additionally, 166,707,183 samples were collected and randomly divided into 8 parts using a stepped data volume method to verify the computational performance of the OneID algorithm.

[0069] The comparison method is as follows:

[0070] WCC Algorithm: The core of the WCC algorithm is to find a connected group of nodes in an undirected graph, where all nodes in the same group form a component that points to the same OneID, meaning that the node group is identified as a characteristic of the same user.

[0071] The experimental results are as follows:

[0072] The WCC algorithm and the OneID algorithm designed in this invention were used to perform unified user identification on the 40 user data points in the dataset. The results showed that the OneID identification accuracy of the WCC algorithm was 73.33% for the 40 data points (containing 60 features), while the accuracy of the method proposed in this invention was 100%.

[0073] Next, the performance of the OneID algorithm was verified. This experiment used 10 servers with 8 cores and 64GB of memory, but this invention is not limited to this; equipment can be added or removed according to actual needs, demonstrating strong scalability. The performance results of the OneID algorithm are shown in Table 2. When the sample size was 10 million and the number of nodes was 24.48 million, the memory usage per node was 2.52GB, and the execution time was only 221 seconds. As the sample size and number of nodes increased, both the peak memory usage and execution time showed a linear increase. When the sample size reached 166 million and the number of nodes exceeded 300 million, the memory usage per node was approximately 23GB, and the execution time was 793 seconds.

[0074] Test results show that the graph-based OneID algorithm designed in this invention has high accuracy, stability, scalability and execution efficiency, and has high practical value for achieving unified user identification.

[0075] Table 2 Performance Verification of OneID Algorithm

[0076]

[0077] This invention proposes a graph-based method for unified user identification (OneID). By prioritizing features and optimizing merging rules, it improves identification accuracy. It exhibits strong scalability and stability, capable of handling complex matching rules with large datasets, and effectively addresses the problems encountered in streaming and batch computing. Furthermore, the graph-based OneID algorithm designed in this invention only needs to focus on processing graph nodes during computation, resulting in clear logic and high parallel computing efficiency.

[0078] In the description of this specification, references to terms such as "embodiment," "example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, those skilled in the art can combine or combine the different embodiments or examples described in this specification and the features therein without causing contradiction.

[0079] While embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and alterations to the above embodiments within the scope of the present invention.

Claims

1. A method for unified user identification based on graph computation, characterized in that, The method includes the following steps: S1: Data Collection: Collect multi-source data from enterprises or organizations, and prioritize and set confidence levels for the features present in the data; S2: Data Processing: Merge multi-source data accessed by S1, generate user registry, and produce point and edge files for graph computation; S3: After reading the point and edge data, obtain the connected graph structure of all data; S4: Post-processing of results: Combine the graph calculation result table with the user registry to output the OneID result table; Step S2 includes: S21: Extract all feature data from all data sources connected to S1 and merge it into the user registry; add an auto-incrementing "index" field to the user registry as the primary key and also as the tag for each data entry; S22: Add a "Temporary OneID" field to the user registry obtained in S21. The value of this field is determined according to the priority division table in S1: the identifier of the highest priority group feature of each data entry plus the feature data is the temporary OneID. S23: Extract the index and each feature field from the user registry obtained in S22 and remove duplicates to obtain six CSV point files: index, ID number, unionid, mobile phone number, device number, and cookie; extract the fields "index-ID number-ID number confidence", "index-unionid-unionid confidence", "index-mobile phone number-mobile phone number confidence", "index-device number-device number confidence", and "index-cookie-cookie confidence" from the user registry obtained in S22 to obtain five CSV side files. S24: Perform anomaly checks on the point and edge files generated in S23; if the point and edge files do not exist or there are abnormal features in the edge files, the data processing in step S2 needs to be repeated to generate new point and edge files.

2. The method for unified user identification based on graph computation according to claim 1, characterized in that, In step S3, the graph distinguishes between index nodes and feature nodes. In the edge file, index-feature node pairs constitute edges, and the feature confidence is the attribute value of the edge.

3. The method for unified user identification based on graph computation according to claim 2, characterized in that, Step S3 also includes implementing the OneID algorithm, which features distributed computing and easy deployment, based on the Pregel computing model and the Kubernetes platform.

4. The method for unified user identification based on graph computation according to claim 3, characterized in that, The calculation process in step S3 consists of several supersteps separated by global synchronization points. The system moves from one "superstep" to the next "superstep" until the iterative calculation ends. Whether the calculation ends depends on the state of the nodes in the graph and whether there is information transmission.

5. The method for unified user identification based on graph computation according to claim 4, characterized in that, In the first superstep, every node is in an active state, and active nodes participate in the computation and information transmission of this round of superstep. A node can vote to become inactive, and in subsequent superstep computations, the node will no longer participate in the relevant computations unless it receives information from other nodes, at which point its state will become active again.

6. The method for unified user identification based on graph computation according to claim 5, characterized in that, When all nodes in the graph are inactive and there is no information transmission, the termination condition is met, the iterative calculation ends, and the graph calculation result file is output.

7. The method for unified user identification based on graph computation according to claim 6, characterized in that, Step S3 includes: S31: Initialize the node attribute values ​​in the graph structure; S32: After initialization, the index node converts the attribute value into a message and sends it to the adjacent feature node, and sets its state to inactive; S33: The feature node receives and processes the message passed from the index node in S32, and updates the feature node attribute value; S34: The index node receives and processes the message passed from the feature node in S33, and updates the index node attribute value; S35: Repeat steps S33 and S34, with all nodes continuously updating attribute values ​​and sending messages until all nodes are inactive and no messages are being sent. Then, terminate the OneID algorithm and output the attribute values ​​of all feature nodes to obtain the graph calculation result table.

8. The method for unified user identification based on graph computation according to claim 7, characterized in that, In step S35, the attribute value of each feature includes the index value and confidence level of the OneID to which the feature belongs.

9. A system for unified user identification based on graph computing, characterized in that, The system is used to implement the method for unified user identification based on graph computing according to any one of claims 1-8.

Images