An adaptive probabilistic identity recognition system and method

Abstract

The present application relates to the field of data analysis, and more particularly to a self-adaptive probabilistic identity recognition system and method, wherein the system first collects multi-source interaction data by using a context perception module, realizes the same analysis of the multi-source interaction data, and thus overcomes the problem of "data island"; then, the type of context information is recognized by using a strategy scheduling center, and an identification activation instruction is distributed to multiple recognition engine modules in parallel, so as to drive the corresponding recognition engine modules to realize identity recognition and output one or more preliminary recognition results; then, the one or more preliminary recognition results are fused to obtain an identity recognition result used for binding original attribution data. Compared with the prior art, on the basis of overcoming the problem of "data island", the system of the present application integrates multiple sets of recognition engine modules, supports multi-dimensional / multi-scene entity identity recognition, and is more suitable for solving multi-dimensional / multi-scene data attribution problems.

Description

Technical Field

[0001] This invention relates to the field of data analysis. More specifically, this invention relates to an adaptive probabilistic identity recognition system and method. Background Technology

[0002] In the process of attribution analysis of a certain entity (including users, devices or network sessions, etc.), it is necessary to continuously identify whether the entity that triggers / issues the data source is the same, so as to ensure that the attribution analysis results are strongly correlated with the entity, thereby avoiding major decision-making errors caused by poor attribution results.

[0003] To achieve the aforementioned entity identity recognition function, various identity recognition technologies have been developed in this field, namely, device-based recognition, user behavior pattern-based recognition, and network environment-based recognition. These recognition functions face two major drawbacks in complex data attribution scenarios, specifically: The problem of data silos exists, as the aforementioned identification technologies are often deployed in independent and unrelated systems. For a single entity, such as a user browsing information across multiple devices, software applications, and browsers, the complex data attribution scenario may require continuous tracking of both single-device / cross-device paths of the data source and cross-application / cross-browser paths. Simply using the aforementioned technologies for multi-dimensional / multi-scenario data attribution is nearly impossible, easily leading to data source breaks in the attribution chain when the entity crosses the identification domain.

[0004] Therefore, existing technologies are not suitable for multi-dimensional / multi-scenario data attribution due to the data silo problem. Summary of the Invention

[0005] To address the problem that existing technologies are not applicable to multi-dimensional / multi-scenario data attribution, this invention discloses an adaptive probabilistic identity recognition system and method.

[0006] In a first aspect, the present invention discloses an adaptive probabilistic identity recognition system, the data output end of which is connected to the data source receiving end of an attribution analysis system. The system of the present invention includes: The context-aware module is used to receive interaction data from the target entity and parse the interaction data to obtain context information to support identity recognition. The policy scheduling center, connected to the context-aware module, is used to dynamically match and activate one or more identification activation instructions from a preset policy library based on the classification results of context information. The recognition engine module is connected to the policy scheduling center. Each recognition engine module encapsulates a corresponding identity feature analysis algorithm. The recognition engine module is configured to: when receiving an activation command, input the interaction data and / or context information into the corresponding identity feature analysis algorithm and output one or more preliminary recognition results. The decision fusion module is used to input one or more preliminary identification results into preset fusion rules to generate an identity recognition result with a confidence score; The information binding module is used to bind the identity recognition results with the raw attribution data supplied to the attribution analysis system.

[0007] Beneficial Effects: This invention constructs an adaptive probabilistic identity recognition system. Its data output end is connected to the data source receiving end of an attribution analysis system, approximating the multi-source data interface of the attribution analysis system, enabling the binding of original attribution data with identity recognition results. Specifically, the system first utilizes a context-aware module to collect multi-source interaction data, achieving unified parsing of multi-source interaction data, thereby overcoming the "data silo" problem. Then, a strategy scheduling center identifies the type of context information and distributes recognition activation instructions in parallel to multiple recognition engine modules to drive the corresponding recognition engine modules to perform identity recognition and output one or more preliminary recognition results. Next, the one or more preliminary recognition results are fused to obtain the identity recognition result used to bind the original attribution data. Compared to existing technologies, this invention, while overcoming the "data silo" problem, integrates multiple recognition engine modules, supports multi-dimensional / multi-scenario entity identity recognition, and is more suitable for solving multi-dimensional / multi-scenario data attribution problems.

[0008] Preferably, the interaction data includes user behavior data and environmental pattern data; the identity feature analysis algorithm includes a similarity comparison algorithm, and the recognition engine module includes at least one user behavior recognition engine, which is configured as follows: Based on the environmental pattern data, the corresponding similarity comparison algorithm is invoked; Based on the similarity comparison algorithm, similarity calculations are performed on user behavior data to generate preliminary recognition results.

[0009] Preferably, the context information includes hardware and software features; the identity feature analysis algorithm includes a device fingerprint recognition algorithm; the recognition engine module includes at least one device fingerprint recognition engine, which is configured as follows: The hardware and software characteristics are input into the built-in device fingerprint recognition algorithm to generate preliminary recognition results.

[0010] Secondly, the present invention also discloses an adaptive probabilistic identity recognition method for use in the adaptive probabilistic identity recognition system described in the first aspect. The method of the present invention includes: Receive interaction data from the target entity and parse the interaction data to obtain contextual information to support identity recognition; Based on the classification results of contextual information, dynamically match and activate one or more recognition activation instructions from a preset strategy library; In response to the activation command, the interaction data and / or context information are input into the corresponding identity feature analysis algorithm, and one or more preliminary identification results are output. One or more preliminary identification results are input into a preset fusion rule to generate an identity recognition result with a confidence score; The identification results are linked to the raw attribution data supplied to the attribution analysis system.

[0011] Preferably, the interaction data includes user behavior data and environmental pattern data; the identity feature analysis algorithm includes a similarity comparison algorithm; the interaction data is input into the corresponding identity feature analysis algorithm, and one or more preliminary identification results are output, including: Based on the environmental pattern data, the corresponding similarity comparison algorithm is invoked; Based on the similarity comparison algorithm, similarity calculations are performed on user behavior data to generate preliminary recognition results.

[0012] Preferably, the preliminary identification results include user similarity; based on a similarity comparison algorithm, similarity calculations are performed on user behavior data to generate preliminary identification results, specifically: Based on user behavior data from two different data sources, user profile vectors for the first target entity and the second target entity are established. The similarity between two user profile vectors is calculated based on the similarity comparison algorithm.

[0013] Preferably, the user profile vector includes at least a shared network environment profile vector, a geographic location trajectory profile vector, or a user activity time profile vector.

[0014] Preferably, the context information includes hardware and software features; the identity feature analysis algorithm includes a device fingerprint recognition algorithm; the interaction data is input into the corresponding identity feature analysis algorithm, and one or more preliminary identification results are output, including: The hardware and software characteristics are input into the built-in device fingerprint recognition algorithm to generate preliminary recognition results.

[0015] Preferably, the device fingerprinting algorithm includes a hash algorithm, and the preliminary identification result includes a fingerprint digest. Hardware and software features are input into the built-in device fingerprinting algorithm to generate the preliminary identification result, specifically: Assign weight coefficients based on the stability of each feature in the software and hardware features; A hash algorithm is used to combine weighted features to generate a fingerprint digest.

[0016] Preferably, after binding the identity recognition results with the raw attribution data supplied to the attribution analysis system, the method of the present invention further includes: An interactive verification sampling method was used to verify the stability and accuracy of the identity recognition results.

[0017] The beneficial effects of this invention are as follows: (1) Compared with the prior art, the system of the present invention overcomes the problem of "data silos" and integrates multiple recognition engine modules, supports multi-dimensional / multi-scenario entity identity recognition, and is more suitable for solving multi-dimensional / multi-scenario data attribution problems.

[0018] (2) Compared with the prior art, the method of the present invention is installed in the system of the present invention and can support the system of the present invention to realize multi-dimensional / multi-scenario entity identity recognition function.

[0019] (3) Compared with the prior art, the method of the present invention proposes a "weighted + fault-tolerant hashing" method to generate fingerprint digests, and the generated fingerprint digests are more reliable and accurate. Attached Figure Description

[0020] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a schematic diagram of the adaptive probabilistic identity recognition system in Embodiment 1 of the present invention; Figure 2 This is a flowchart of the adaptive probabilistic identity recognition method in Embodiment 2 of the present invention. 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, 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.

[0022] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0023] Example 1 like Figure 1As shown in the figure, this embodiment discloses an adaptive probabilistic identity recognition system, whose data output end is connected to the data source receiving end of the attribution analysis system, and is used to supply the attribution analysis system with raw attribution data containing identity recognition results.

[0024] The system includes a context-aware module, a policy scheduling center, multiple recognition engine modules, a decision fusion module, and an industry information binding module.

[0025] Furthermore, the context-aware module receives and parses the interaction data of the target entity to obtain contextual information supporting identity recognition. In this embodiment, the context-aware module integrates multiple APIs to support the access of various types of interaction data. The target entity can be an individual user or a single device. The interaction data is mainly divided into several types: user behavior data, environmental pattern data, the aforementioned raw attribution data, and device interaction data. The contextual information is mainly divided into two categories: user preference features and hardware / software features.

[0026] It should be noted that prior authorization from the user or device is required before obtaining the aforementioned data.

[0027] Furthermore, the aforementioned user behavior data includes user interaction time, user browsing data, user click data, and / or user purchase data. The aforementioned environmental pattern data includes user ID data, user login IP address, user login device related data, and / or GPS location. The aforementioned raw attribution data overlaps with the aforementioned user behavior data, primarily including user click data and user purchase data. Device interaction data mainly includes device ID data, device IP data, and device interaction time.

[0028] Furthermore, the policy scheduling center is connected to the context-aware module, which dynamically matches and activates one or more recognition activation commands from a preset policy library based on the classification results of the context information. The classification results of the aforementioned context information are software and hardware features. It should be noted that these software and hardware features are primarily used to characterize the technical application scenarios and can be analyzed using rule engines or artificial intelligence models.

[0029] It should be explained that the aforementioned policy library is essentially a set of matching rules. This set of matching rules is stored in the policy scheduling center in the form of a data table, which can retrieve the corresponding activation command based on the characteristics of the software and hardware.

[0030] For example, the above strategy library can be shown in Table 1: Table 1

[0031] It should be explained that the matching dependencies in Table 1 serve as the basis for the rule engine or artificial intelligence model to evaluate software and hardware features. Hardware features in Table 1 refer to device model, screen resolution, battery capacity, and other hardware information. System features include system version, system initialization time, startup time, and other information.

[0032] Compared to existing technologies, the system of this invention, through the design of the aforementioned strategy scheduling center, supports concurrent activation command sending to drive multiple recognition engine modules to execute their internal functional logic within the same time period, thereby improving the data processing efficiency of the system. The aforementioned strategy library is configured with same-device activation commands, first cross-device activation commands, same-device same-browser activation commands, same-device cross-app activation commands, same-device cross-browser activation commands, same-device cross-platform activation commands, and second cross-device activation commands to adapt to the recognition and scheduling of recognition functions in seven complex scenarios. This effectively overcomes the problem that traditional technologies cannot adapt to complex attribution scenarios.

[0033] Furthermore, multiple recognition engine modules are connected to the policy scheduling center, and each recognition engine module encapsulates a corresponding identity feature analysis algorithm.

[0034] In this embodiment, the recognition engine module is configured as follows: Upon receiving an activation command, the interaction data and / or context information are input into the corresponding identity feature analysis algorithm, which outputs one or more preliminary identification results.

[0035] It should be noted that the identity feature analysis algorithms used for the seven different scenarios mentioned above also differ. Therefore, a "pre-encapsulation, adaptation activation" approach can be adopted to adapt to identity recognition in the seven different scenarios.

[0036] In this embodiment, the aforementioned identity feature analysis algorithm includes a weighted analysis method and a similarity comparison algorithm. Specifically, the weighted analysis method calculates a weighted sum of scores from multiple data items to serve as the confidence level of the preliminary identification result.

[0037] For example, for scenarios involving conversions within the same device, the corresponding data tables for data item scores and their weight allocation in the OS and Android operating systems can be: Table 2

[0038] Table 3

[0039] In Tables 2 and 3, data items are scored using two dimensions: "stability" and "uniqueness." Stability indicates the consistency of the data over a certain period. If the data remains relatively consistent without significant change during this period, its stability is high and is rated as "1"; otherwise, it is rated as "0." Uniqueness indicates the ability of a device fingerprint to distinguish itself from other devices. The lower the similarity of the data item to other similar data items, or the more unique it is, the higher the score is, and otherwise, it is rated as "0."

[0040] Similarly, for identity recognition in the other six different scenarios, it is also necessary to configure the corresponding data item weight allocation table, and then rely on the weighted analysis method to calculate the weighted sum of the scores of multiple data items as the confidence level of the preliminary recognition result.

[0041] Furthermore, in addition to calculating the confidence level of the preliminary identification results, it is even more important to determine whether the target entity is identical in the seven different scenarios. Therefore, the identification engine module includes at least one user behavior identification engine, which is configured as follows: Based on the environmental pattern data, the corresponding similarity comparison algorithm is invoked.

[0042] Based on the similarity comparison algorithm, similarity calculations are performed on user behavior data to generate preliminary recognition results.

[0043] This technical solution primarily determines whether the raw attribution data generated in different scenarios originated from the same "person." As shown in Table 1, in the "same device conversion" scenario, the raw attribution data output needs to identify GAID (Google Advertising ID) and IDFA (Identifier For Advertising), device identifiers used for ad tracking and user identification. Both are non-permanent identifiers, meaning that a strong device-user binding has been achieved in this scenario, and the device is user-specific. Therefore, the above solution can be disregarded, and the confidence level can be used as the preliminary identification result. As for the other six scenarios, these include complex scenarios such as "the same user using the same software on multiple devices" and "the same user using different software / browsers on the same device." For these complex situations, a similarity comparison algorithm can be used to calculate the similarity between multiple sets of user behavior data as part of the preliminary identification result.

[0044] Conversely, if the "device" is used as the identification subject, then the identification engine module includes at least one device fingerprint recognition engine, which is configured as follows: The hardware and software characteristics are input into the built-in device fingerprint recognition algorithm to generate preliminary recognition results.

[0045] It should be noted that in this embodiment, the device fingerprint recognition algorithm is mainly a hash algorithm, and the preliminary recognition result also includes a fingerprint digest. For example, similar to Tables 2 and 3, a hash algorithm is used to combine weighted features to generate a fingerprint digest. Unlike the examples in Tables 2 and 3, the features described in this solution mainly revolve around the device's hardware and software specifications. These features include information such as device model, screen resolution, number of CPU cores, operating system version, and / or the list of installed applications.

[0046] Furthermore, the aforementioned decision fusion module is used to input one or more preliminary identification results into preset fusion rules to generate an identity recognition result with a confidence score.

[0047] For example, using the above technical solution and fusion rules, taking a specific "user" as the subject of analysis, assuming the existing original attribution data are A, B, C, and D, which are output by the "user" in different scenarios, there are a total of 6 pairwise combinations that need to be identified, namely AB, AC, AD, BC, BD, and CD. Correspondingly, the identification results can be obtained as shown in Table 4: Table 4

[0048] Therefore, it can be concluded that the original attribution data A and B were sent by the same user "Wang" through the same browser on the same device.

[0049] It should be noted that the above examples are only for the purpose of helping those skilled in the art to understand the solution of this application. The actual data decision fusion and evaluation process is more complex, and users may have the same name. Their user information may include IP address.

[0050] Furthermore, the information binding module is used to bind the identity recognition results with the raw attribution data supplied to the attribution analysis system.

[0051] Based on the case in Table 4, we can see that the original attribution data A and B were sent by the same user "Wang" through the same browser on the same device. Therefore, the original attribution data A and B can be assigned the identity binding label "Wang".

[0052] Based on the above technical solution, this embodiment proposes an adaptive probabilistic identity recognition system. Its data output end is connected to the data source receiving end of the attribution analysis system, enabling the addition of identity recognition results to the original attribution data. The system integrates multi-source interactive data through a context-aware module, achieving unified data parsing to break down "data silos." Then, the strategy scheduling center synchronously sends activation commands to multiple recognition engine modules based on the context information type, prompting the corresponding modules to execute identity recognition tasks and generate one or more preliminary results. Finally, by integrating these preliminary results, an identity recognition conclusion bound to the original attribution data is formed.

[0053] Compared to existing technologies, the system in this embodiment not only solves the problem of data silos, but also integrates multiple recognition engines, supports entity identification across multiple dimensions and scenarios, and more effectively addresses complex data attribution needs.

[0054] Example 2 like Figure 2 As shown, based on Embodiment 1, this embodiment also discloses an adaptive probabilistic identity recognition method for the adaptive probabilistic identity recognition system described in Embodiment 1. Specifically, the method of this embodiment includes: S10: Receive the interaction data from the target entity and parse the interaction data to obtain contextual information used to support identity recognition.

[0055] In this embodiment, step S10 relies on the context-aware module described in Embodiment 1. The relevant explanations of the target entity, interaction data and context information have been clearly explained in the preceding text and will not be repeated here.

[0056] S20: Based on the classification results of the context information, dynamically match and activate one or more recognition activation instructions from a preset policy library.

[0057] In this embodiment, the classification results correspond to the seven scenarios mentioned above, and there are also seven activation commands. If all seven recognition engine modules are equipped or enabled, concurrent action commands for the seven recognition engine modules to be executed simultaneously can be issued through the policy scheduling center to achieve comprehensive and accurate data traceability.

[0058] S30: In response to the activation command, input the interaction data and / or context information into the corresponding identity feature analysis algorithm and output one or more preliminary recognition results.

[0059] Regarding data tracing for "people," step S30 above includes: S31: Based on the environmental pattern data, call the corresponding similarity comparison algorithm.

[0060] S32: Based on the similarity comparison algorithm, perform similarity calculation on the user behavior data to generate preliminary recognition results.

[0061] Furthermore, step S32 above specifically includes: S321: Based on user behavior data from two different data sources, establish user profile vectors for the first target entity and the second target entity; S322: Calculate the user similarity between two user profile vectors based on the similarity comparison algorithm.

[0062] Among them, the user profile vector includes at least the shared network environment profile vector, the geographic location trajectory profile vector, or the user's active time profile vector.

[0063] Regarding data tracing for "hardware", step S30 above includes: The hardware and software characteristics are input into the built-in device fingerprint recognition algorithm to generate preliminary recognition results.

[0064] More specifically, first, weight coefficients based on their stability are assigned to each feature in the software and hardware features, and then a hash algorithm is used to combine the weighted features to generate a fingerprint digest.

[0065] Through the above step S30, the method of this embodiment supports data tracing in multiple scenarios and can meet the attribution requirements of complex data.

[0066] S40: Input one or more preliminary identification results into a preset fusion rule to generate an identity recognition result with a confidence score.

[0067] It should be explained that the above fusion rule is to list the data analysis results of all scenarios into the same data table, and obtain the above identity recognition results by selecting the scenario with the highest confidence and similarity.

[0068] S50: Bind the identity recognition results with the raw attribution data supplied to the attribution analysis system.

[0069] It should be noted that the above-mentioned attribution analysis systems are mainly advertising data attribution systems, e-commerce data attribution systems, or game data attribution systems.

[0070] Through the above steps S10-S50, the method of this embodiment can support the system of the present invention to realize multi-dimensional / multi-scenario entity identity recognition function.

[0071] Preferably, after step S50 above, the method of this embodiment further includes: An interactive verification sampling method was used to verify the stability and accuracy of the identity recognition results.

[0072] It should be explained that the above-mentioned interactive verification sampling method is a verification sampling method that supports human interaction. It uses a human interaction window to sample and check the stability and accuracy of identity recognition results, thereby improving the reliability of identity recognition.

[0073] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0074] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. An adaptive probabilistic identity recognition system, characterized in that, Its data output terminal is connected to the data source receiving terminal of the attribution analysis system, the system comprising: The context-aware module is used to receive interaction data from the target entity and parse the interaction data to obtain context information for supporting identity recognition. The policy scheduling center, connected to the context-aware module, is used to dynamically match and activate multiple recognition activation instructions from a preset policy library based on the classification results of the context information. The data is stored in the policy scheduling center in the form of a data table, which retrieves the corresponding activation command based on the hardware and software characteristics and matching dependencies. Multiple recognition engine modules are connected to the policy scheduling center, wherein each recognition engine module encapsulates a corresponding identity feature analysis algorithm; the recognition engine module is configured to: when receiving the activation instruction, input the interaction data and the context information into the corresponding identity feature analysis algorithm, and output multiple preliminary recognition results; The decision fusion module is used to input multiple preliminary identification results into preset fusion rules to generate an identity recognition result with a confidence score; The information binding module is used to bind the identity recognition result to the raw attribution data of the attribution analysis system for information binding. Data tracing for individuals: The interaction data includes user behavior data and environmental pattern data; the identity feature analysis algorithm includes a similarity comparison algorithm; the recognition engine module includes at least one user behavior recognition engine, which is configured as follows: Based on the environmental pattern data, the corresponding similarity comparison algorithm is invoked; Based on the similarity comparison algorithm, a similarity calculation is performed on the user behavior data to generate the preliminary identification result; Data traceability for hardware: The context information includes hardware and software features; the identity feature analysis algorithm includes a device fingerprint recognition algorithm; the recognition engine module includes at least one device fingerprint recognition engine, which is configured as follows: The hardware and software features are input into the built-in device fingerprint recognition algorithm to generate the preliminary recognition result; Device fingerprinting algorithms include hash algorithms, and preliminary identification results include fingerprint digests. Hardware and software features are input into the built-in device fingerprinting algorithm to generate preliminary identification results, specifically: Assign weight coefficients based on the stability of each feature in the software and hardware features; A hash algorithm is used to combine weighted features to generate a fingerprint digest; The strategy library is configured with same-device activation instructions, first cross-device activation instructions, same-device same-browser activation instructions, same-device cross-app activation instructions, same-device cross-browser activation instructions, same-device cross-platform activation instructions, and second cross-device activation instructions.

2. An adaptive probabilistic identity recognition method, characterized in that, In the adaptive probabilistic identity recognition system of claim 1, the method includes: Receive interaction data from the target entity and parse the interaction data to obtain contextual information for supporting identity recognition; Based on the identification context information, multiple identification activation instructions are dynamically matched and activated from a preset strategy library; In response to the activation command, the interaction data and / or the context information are input into the corresponding identity feature analysis algorithm, and multiple preliminary identification results are output. Multiple preliminary identification results are input into a preset fusion rule to generate an identity recognition result with a confidence score; The identity recognition results are linked with the raw attribution data supplied to the attribution analysis system.

3. The adaptive probabilistic identity recognition method according to claim 2, characterized in that, The interaction data includes user behavior data and environmental pattern data; the identity feature analysis algorithm includes a similarity comparison algorithm; The interactive data is input into the corresponding identity feature analysis algorithm, which outputs multiple preliminary identification results, including: Based on the environmental pattern data, the corresponding similarity comparison algorithm is invoked; Based on the similarity comparison algorithm, the user behavior data is subjected to similarity calculation to generate the preliminary identification result.

4. The adaptive probabilistic identity recognition method according to claim 3, characterized in that, The preliminary identification result includes user similarity; based on the similarity comparison algorithm, similarity calculation is performed on the user behavior data to generate the preliminary identification result, specifically as follows: Based on user behavior data from two different data sources, user profile vectors for the first target entity and the second target entity are established. The similarity between two user profile vectors is calculated based on the similarity comparison algorithm.

5. The adaptive probabilistic identity recognition method according to claim 4, characterized in that, The user profile vector includes at least a shared network environment profile vector, a geographic location trajectory profile vector, or a user activity time profile vector.

6. The adaptive probabilistic identity recognition method according to claim 2, characterized in that, The context information includes hardware and software features; the identity feature analysis algorithm includes a device fingerprint recognition algorithm. The interactive data is input into the corresponding identity feature analysis algorithm, which outputs multiple preliminary identification results, including: The hardware and software features are input into the built-in device fingerprint recognition algorithm to generate the preliminary recognition result.

7. The adaptive probabilistic identity recognition method according to claim 6, characterized in that, The device fingerprint recognition algorithm includes a hash algorithm, and the preliminary recognition result includes a fingerprint digest. The hardware and software features are input into the built-in device fingerprint recognition algorithm to generate the preliminary recognition result, specifically as follows: Assign weight coefficients based on their stability to each feature in the software and hardware features; The fingerprint digest is generated by combining the weighted features using the hash algorithm.

8. The adaptive probabilistic identity recognition method according to claim 2, characterized in that, After binding the identity recognition result with the raw attribution data supplied to the attribution analysis system, the method further includes: An interactive verification sampling method was used to verify the stability and accuracy of the identity recognition results.

Images