Cross-scenario device fingerprint recognition method, device, electronic device and storage medium

Abstract

The embodiment of the present invention relates to a cross-scenario device fingerprint recognition method, apparatus, electronic device and storage medium, the method comprising: obtaining an event feature vector pair of an event to be matched, the event feature vector pair comprising an event feature vector of the event to be matched and an event feature vector of a known event in a device fingerprint library; inputting the event feature vector pair into an adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to the known event, wherein, in the training process of the adversarial triplet neural network model, event scenario parameters are used as adversarial data; determining the similarity between the first network feature and the second network feature; determining whether the event to be matched and the known event are from the same device based on the similarity, and when determining that the event to be matched and the known event are from the same device, assigning the device fingerprint of the known event to the event to be matched. Thus, cross-scenario device fingerprint recognition can be achieved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of computer technology, and in particular, to a cross-scenario device fingerprint recognition method, apparatus, electronic device, and storage medium. Background Art

[0002] At present, as people pay more attention to privacy and personal information, in the future, various equipment manufacturers are likely to prohibit the direct acquisition of users' mobile phone numbers and device numbers from mobile web pages, APP applications, mini-programs, etc. Therefore, in related technologies, the device fingerprint active matching method that uses the inherent device number of the mobile device (such as IDFA (Identifier for Advertising), IDFV (Identifier For Vendor), IMEI (International Mobile Equipment Identity) and other device identifiers) to determine whether it is the same device will not meet the needs of the industry.

[0003] The entrance to Internet products is often not limited to the APP port (such as using the APP application to enter). Users can also choose H5 landing pages (open web pages), WeChat applets and other scenarios as ways to log in to the product. For applications in the financial industry to combat black and gray industries such as anti-fraud and anti-black production, as well as common device fingerprint application scenarios such as evaluation and optimization of advertising effects, tracking and profiling of user trajectories, it is crucial to accurately associate events initiated by users in different scenarios (H5, applets, APP, etc.) through device identification technology. However, the products in the current related technologies can only achieve relatively accurate device matching in a single scenario, and the recognition effect across scenarios is poor. Summary of the invention

[0004] In view of this, in order to solve the above-mentioned technical problems, an embodiment of the present invention provides a cross-scenario device fingerprint recognition method, apparatus, electronic device and storage medium.

[0005] In a first aspect, an embodiment of the present invention provides a cross-scenario device fingerprint recognition method, the method comprising:

[0006] Acquire an event feature vector pair of the event to be matched, wherein the event feature vector pair includes an event feature vector corresponding to the event to be matched and an event feature vector of any known event in a device fingerprint library, wherein the known event refers to an event with a device fingerprint, and the device fingerprint library includes event feature vectors of multiple known events and corresponding relationships between device fingerprints;

[0007] Inputting the event feature vector pair into an adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to any known event, wherein in the training process of the adversarial triplet neural network model, event scene parameters are used as adversarial data;

[0008] determining a similarity between the first network feature and the second network feature;

[0009] It is determined whether the event to be matched and any known event come from the same device according to the similarity, and when it is determined that the event to be matched and any known event come from the same device, the device fingerprint of any known event is assigned to the event to be matched.

[0010] In a possible implementation manner, obtaining an event feature vector pair of an event to be matched includes:

[0011] Obtaining the event feature vector of the event to be matched;

[0012] Obtaining an event feature vector of any known event from a preset device fingerprint library;

[0013] The event feature vector of the event to be matched and the event feature vector of any known event are combined to form an event feature vector pair of the event to be matched.

[0014] In a possible implementation manner, determining the similarity between the first network feature and the second network feature includes:

[0015] determining a Euclidean distance between the first network feature and the second network feature;

[0016] The similarity between the first network feature and the second network feature is determined according to the Euclidean distance.

[0017] In a possible implementation manner, determining whether the event to be matched and any known event are from the same device according to the similarity includes:

[0018] comparing the similarity with a preset similarity threshold;

[0019] If the similarity is greater than or equal to the preset similarity threshold, it is determined that the event to be matched and any known event are from the same device;

[0020] If the similarity is less than the preset similarity threshold, it is determined that the event to be matched and any known event come from different devices.

[0021] In a possible implementation, the method further includes:

[0022] Acquire multiple pieces of sample data, the sample data including positive examples, negative examples, and anchor points, wherein the positive examples and the anchor points correspond to the same device, and the negative examples and the anchor points correspond to different devices;

[0023] The initial adversarial triplet neural network model is trained using the plurality of sample data to obtain a trained adversarial triplet neural network model.

[0024] In a possible implementation, the using of the plurality of sample data to train the initial adversarial triplet neural network model to obtain a trained adversarial triplet neural network model includes:

[0025] Inputting the positive examples, negative examples and anchor points in each of the sample data into three embedding layers in the initial adversarial triplet neural network model respectively, to obtain corresponding positive example network features, negative example network features and anchor point network features;

[0026] Inputting the positive network features, the negative network features, and the anchor network features into the trained device matching model and the scene classification model, respectively, to obtain a first distance and a second distance output by the device matching model, and a predicted scene probability distribution output by the scene classification model, wherein the first distance is used to characterize the distance between the anchor network features and the positive network features, and the second distance is used to characterize the distance between the anchor network features and the negative network features;

[0027] Determining a first loss value according to the first distance and the second distance, and determining a second loss value according to the predicted scene probability distribution;

[0028] When it is determined according to the first loss value and the second loss value that the current triplet neural network model does not meet the preset model convergence conditions, the model parameters of the current adversarial triplet neural network model are updated according to the first loss value and the second loss value and based on the adversarial learning mechanism, and the adversarial triplet neural network model after the updated model parameters is iteratively trained until it is determined that the current adversarial triplet neural network model meets the model convergence conditions.

[0029] In a possible implementation, the method further includes:

[0030] After the device fingerprint of any known event is assigned to the event to be matched, the event feature vector corresponding to the any known event in the device fingerprint library is updated to the event feature vector of the event to be matched.

[0031] In a second aspect, an embodiment of the present invention provides a cross-scenario device fingerprint recognition device, the device comprising:

[0032] An acquisition module, used to acquire an event feature vector pair of an event to be matched, wherein the event feature vector pair includes an event feature vector corresponding to the event to be matched and an event feature vector of any known event in a device fingerprint library, wherein the known event refers to an event with a device fingerprint, and the device fingerprint library includes event feature vectors of multiple known events and corresponding relationships between device fingerprints;

[0033] A feature extraction module, used for inputting the event feature vector pair into an adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to any known event, wherein in the training process of the adversarial triplet neural network model, event scene parameters are used as adversarial data;

[0034] a similarity determination module, configured to determine a similarity between the first network feature and the second network feature;

[0035] The identification module is used to determine whether the event to be matched and any known event come from the same device according to the similarity, and when it is determined that the event to be matched and any known event come from the same device, assign the device fingerprint of any known event to the event to be matched.

[0036] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a processor and a memory, wherein the processor is used to execute a cross-scenario device fingerprint recognition program stored in the memory to implement a cross-scenario device fingerprint recognition method described in any one of the first aspects.

[0037] In a fourth aspect, an embodiment of the present invention provides a storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement a cross-scenario device fingerprint recognition method as described in any one of the first aspects.

[0038] The technical solution provided by the embodiment of the present invention obtains an event feature vector pair of an event to be matched, the event feature vector pair includes an event feature vector corresponding to the event to be matched and an event feature vector of any known event in a device fingerprint library, where a known event refers to an event with a device fingerprint, and inputs the event feature vector pair into an adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to any known event, determine the similarity between the first network feature and the second network feature, determine whether the event to be matched and any known event are from the same device based on the similarity, and when it is determined that the event to be matched and the any known event are from the same device, assign the device fingerprint of any known event to the event to be matched. Since scene parameters are used as adversarial data in the training process of the adversarial triplet neural network model, the judgment of the scene by the network features extracted by the adversarial triplet neural network model becomes blurred, thereby improving the overall anti-scene change capability of the model and improving the accuracy of device fingerprint recognition in cross-scene applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of an embodiment of a cross-scenario device fingerprint recognition method according to an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of an embodiment of the present invention showing a process flow of an adversarial triplet neural network model training process;

[0041] Figure 3 A schematic diagram of a model architecture shown in an embodiment of the present invention;

[0042] Figure 4 A block diagram of an embodiment of a cross-scenario device fingerprint recognition device provided by an embodiment of the present invention;

[0043] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] At present, the entrance to Internet products is often not limited to the APP port, and users will also choose H5 landing pages, WeChat applets and other scenarios as ways to log in to the product. Therefore, common device fingerprint application scenarios such as real-time anti-fraud identification of devices, evaluation and optimization of advertising effects need to rely on the success rate of different scenarios (that is, the matching rate between devices across platforms). However, all kinds of device fingerprint products on the current market are unable to accurately associate H5, applets, APP and other scenarios. In the related technology, there is not much in-depth research on cross-scenario device fingerprint recognition methods. Some large manufacturers mainly use their huge databases for accurate matching (that is, recording the information of all mobile devices) rather than matching through algorithms. Therefore, this method can only be implemented by giant enterprises, and its versatility is poor. In addition, due to incomplete statistics, the matching rate does not exceed 50%.

[0046] In summary, the embodiments of the present invention provide a purely passive matching method that effectively solves the problem of cross-platform / cross-scenario device fingerprint recognition.

[0047] The following is a further explanation of the relevant method embodiments provided by the present invention with reference to specific embodiments in conjunction with the accompanying drawings, and the embodiments do not constitute a limitation on the embodiments of the present invention.

[0048] See also Figure 1 , is a schematic diagram of an embodiment of a cross-scenario device fingerprint recognition method according to an embodiment of the present invention. Figure 1 As shown, the following steps are included:

[0049] Step 101 : obtaining an event feature vector pair of an event to be matched, wherein the event feature vector pair includes an event feature vector corresponding to the event to be matched and an event feature vector of any known event in a device fingerprint library.

[0050] The above-mentioned device fingerprint library includes event feature vectors of multiple known events and corresponding relationships between device fingerprints. It can be seen that the known events here refer to events with device fingerprints. Preferably, the known events included in the device fingerprint library are from a recent period of time, such as events collected in the last 3 months.

[0051] Specifically, obtaining an event feature vector pair of an event to be matched includes: obtaining an event feature vector of the event to be matched, obtaining an event feature vector of any known event from a preset device fingerprint library, and combining the event feature vector of the event to be matched and the event feature vector of any known event to form an event feature vector pair of the event to be matched. Exemplarily, after determining the event feature vector of the matching event, the event feature vectors in the device fingerprint library can be paired with the event feature vector of the event to be matched in sequence according to the storage position order (e.g., queue form) of the event feature vectors of each known event in the device fingerprint library to form multiple event feature vector pairs of the event to be matched.

[0052] The event feature vector is obtained by feature encoding the event parameters. Here, the event parameters include at least scene parameters. Scene parameters refer to parameters (features) specific to each scene, such as browser features such as browser version and browser bit depth in the H5 scene, and WeChat version number, WeChat setting language and other mini-program features in the mini-program scene.

[0053] In addition, event parameters also include some common parameters (features). Common parameters (features) refer to parameters (features) that can be stably obtained across platforms / scenarios. For example, device software information, hardware information, network information, geographic information and other parameters.

[0054] The above-mentioned events are, for example, events that initiate a certain scenario, such as shopping events through a shopping app, transfer events through a financial app, and so on. In an embodiment of the present invention, by acquiring data packets (acquiring communication data packets during communication between the terminal device and the server), the OS, protocol stack, network status, events and other related feature sets of the terminal device (device to be identified) can be extracted from the OSI (Open System Interconnection Reference Model) seven-layer protocol of the data packet, and combined with a machine learning algorithm to identify and track specific terminal devices, thereby collecting the general parameters and scenario parameters of the terminal device (device to be identified). In summary, the information collection method of the embodiment of the present invention preferably adopts passive collection and acquisition, which can be independent of the dependence on the international mobile equipment identification and improve the overall stability.

[0055] Step 102: input the event feature vector pair into the adversarial triplet neural network model to extract the first network feature corresponding to the event to be matched and the second network feature corresponding to any known event.

[0056] First of all, it should be explained that the triplet neural network model is extended from the twin neural network model (Siamese Network). To this end, let’s briefly introduce the twin neural network: the twin neural network is a "conjoined neural network". The "conjoined" nature of the neural network is achieved by sharing weights. The twin neural network is a special type of neural network architecture that highlights its "ability" in tasks involving discovering similarities or relationships between two comparable things.

[0057] The triplet neural network is based on the twin neural network, with one more input item added, and the embedding layers corresponding to the three inputs of the network share weights.

[0058] Furthermore, the embodiment of the present invention adopts an adversarial learning mechanism based on the triplet neural network to train an adversarial triplet neural network model.

[0059] In an embodiment of the present invention, an event feature vector pair of an event to be matched is input into a trained adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to any known event.

[0060] See also Figure 2 , which is a schematic diagram of an embodiment of the present invention showing a process flow of an adversarial triplet neural network model training process; see Figure 3 , is a schematic diagram of a model architecture shown in an embodiment of the present invention. Figure 2 As shown, the following steps are included:

[0061] Step 201: Acquire multiple pieces of sample data, where the sample data includes positive examples, negative examples, and anchor points, wherein the positive examples and anchor points correspond to the same device, and the negative examples and anchor points correspond to different devices.

[0062] Similar to the event feature vector described in step 101 above, the positive examples, negative examples and anchor points mentioned above may also be collected from events triggered daily by known devices.

[0063] It should be noted that positive examples and anchor points correspond to the same device, but not the same event. For example, positive examples are collected from events triggered by known devices on the same day, while anchor points are collected from events triggered by known devices on the previous day.

[0064] Step 202: Use multiple sample data to train the initial adversarial triplet neural network model to obtain a trained adversarial triplet neural network model.

[0065] In some embodiments, training the initial adversarial triplet neural network model using multiple sample data to obtain the trained adversarial triplet neural network model includes the following steps a-d:

[0066] Step a: input the positive examples, negative examples and anchor points in each sample data into the three embedding layers in the initial adversarial triplet neural network model respectively to obtain the corresponding positive example network features, negative example network features and anchor point network features.

[0067] From the above description, it can be seen that the adversarial triplet neural network model has three embedding layers. Therefore, the positive examples, negative examples and anchor points in each sample data can be respectively input into the three embedding layers in the initial adversarial triplet neural network model to obtain the corresponding positive example network features, negative example network features and anchor point network features.

[0068] Step b: input the positive example network features, negative example network features and anchor network features into the trained device matching model and scene classification model respectively, and obtain the first distance and the second distance output by the device matching model, and the predicted scene probability distribution output by the scene classification model, wherein the first distance is used to characterize the distance between the anchor network features and the positive example network features, and the second distance is used to characterize the distance between the anchor network features and the negative example network features.

[0069] In some embodiments, the first distance and the second distance may be Euclidean distances.

[0070] In some embodiments, the predicted scene probability distribution includes the predicted scene probability distribution corresponding to the positive example, the predicted scene probability distribution corresponding to the negative example, and the predicted scene probability distribution corresponding to the anchor point. The predicted scene probability distribution includes the probability that the input object corresponds to multiple different scenes.

[0071] Step c: determining a first loss value according to the first distance and the second distance, and determining a second loss value according to the predicted scenario.

[0072] In some embodiments, the first distance and the second distance may be substituted into a TripletLoss function as shown in the following formula to obtain a first loss value Loss.

[0073] Loss = first distance - second distance + adjustment coefficient

[0074] Wherein, the adjustment coefficient is greater than 0. Preferably, the adjustment coefficient is 4.0.

[0075] The Triplet Loss function aims to reduce the distance between the anchor point and the positive example as much as possible while increasing the distance between the anchor point and the negative example. This enables the adversarial triplet neural network model to extract network features corresponding to events of the same device with a smaller distance between the two extracted network features, that is, a higher similarity. On the contrary, when extracting network features corresponding to events of different devices, the distance between the two extracted network features is larger, that is, a lower similarity. This enables the accurate determination of whether the two input event feature vectors are from the same device based on the network features extracted by the adversarial triplet neural network model.

[0076] In some embodiments, the predicted scene probability distribution may be substituted into a Cross Entropy loss function to obtain a second loss value.

[0077] Step d: When it is determined according to the first loss value and the second loss value that the current triplet neural network model does not meet the preset model convergence conditions, the model parameters of the current adversarial triplet neural network model are updated according to the first loss value and the second loss value and based on the adversarial learning mechanism, and the adversarial triplet neural network model after the updated model parameters is iteratively trained until it is determined that the current adversarial triplet neural network model meets the model convergence conditions.

[0078] From the description of step d, it can be seen that the scene classification model plays an adversarial learning role on the adversarial triplet neural network model. That is, in the training process of the adversarial triplet neural network model, the scene parameters are used as adversarial data, which can make the network features extracted by the adversarial triplet neural network model blur the judgment of the scene, thereby improving the overall model's ability to resist scene changes.

[0079] It can be seen that in the embodiment of the present invention, the application of the triplet neural network model can realize cross-scenario device fingerprint recognition compared to the application of the twin neural network model.

[0080] In some embodiments, the model convergence condition may include that the first loss value and the second loss value are less than a set loss value threshold. The loss value thresholds corresponding to the first loss value and the second loss value may be the same or different, and the embodiment of the present invention does not limit this.

[0081] Step 103: Determine the similarity between the first network feature and the second network feature.

[0082] In some embodiments, determining the similarity between the first network feature and the second network feature includes: determining the Euclidean distance between the first network feature and the second network feature, and determining the similarity between the first network feature and the second network feature according to the Euclidean distance. Exemplarily, the Euclidean distance between the first network feature and the second network feature is determined as the similarity between the first network feature and the second network feature.

[0083] Step 104, determining whether the event to be matched and any known event come from the same device based on the similarity, and when it is determined that the event to be matched and any known event come from the same device, assigning the device fingerprint of any known event to the event to be matched.

[0084] In some embodiments, it is determined based on the similarity whether the event to be matched and any known event come from the same device: the similarity between the first network feature and the second network feature is compared with a preset similarity threshold. If the similarity between the first network feature and the second network feature is greater than or equal to the preset similarity threshold, it means that the probability that the event to be matched and any known event come from the same device is relatively high, so it can be determined that the event to be matched and any known event come from the same device; if the similarity between the first network feature and the second network feature is less than the preset similarity threshold, it means that the probability that the event to be matched and any known event come from the same device is relatively low, so it can be determined that the event to be matched and any known event come from different devices.

[0085] Preferably, the preset similarity threshold is 0.8.

[0086] Furthermore, when it is determined that the event to be matched and any known event are from the same device, the device fingerprint of any known event may be assigned to the event to be matched.

[0087] When it is determined that the event to be matched and any known event come from different devices, the event feature vector of the event to be matched can be further paired with the event feature vector of another known event in the device fingerprint library to form an event feature vector pair, and then the relevant steps of step 101 to step 104 are executed for the event feature vector pair.

[0088] The technical solution provided by the embodiment of the present invention obtains an event feature vector pair of an event to be matched, the event feature vector pair includes an event feature vector corresponding to the event to be matched and an event feature vector of any known event in a device fingerprint library, where a known event refers to an event with a device fingerprint, and inputs the event feature vector pair into an adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to any known event, determine the similarity between the first network feature and the second network feature, determine whether the event to be matched and any known event are from the same device based on the similarity, and when it is determined that the event to be matched and the any known event are from the same device, assign the device fingerprint of any known event to the event to be matched. Since scene parameters are used as adversarial data in the training process of the adversarial triplet neural network model, the judgment of the scene by the network features extracted by the adversarial triplet neural network model becomes blurred, thereby improving the overall anti-scene change capability of the model and improving the accuracy of device fingerprint recognition in cross-scene applications.

[0089] In some embodiments, the cross-scenario device fingerprint recognition method provided by the embodiment of the present invention may also include the following steps (not shown in the figure):

[0090] Optimize the adversarial triplet neural network model when the recognition results are inaccurate.

[0091] Through this processing, when the recognition result is inaccurate, the adversarial triplet neural network model can be optimized to further improve the model accuracy.

[0092] Alternatively, in some other embodiments, the following steps may also be included:

[0093] Determine whether the recognition result is accurate, annotate the accuracy of the recognition result in real time, and add the annotated recognition result to the device fingerprint library.

[0094] Specifically, after assigning the device fingerprint of any known event to the event to be matched and determining the identification result is ready, the event feature vector corresponding to the known event in the device fingerprint library is updated to the event feature vector of the event to be matched. Through this processing, the latest event feature vector can be saved in the device fingerprint library.

[0095] In some embodiments, streaming real-time computing can be used, so that the computing granularity of the aforementioned embodiments reaches the event level. As a result, the method provided by the embodiments of the present invention, on the one hand, uses the trained model to perform real-time matching calculations on the event feature vectors to achieve the purpose of cross-scenario device fingerprint recognition; on the other hand, since we can obtain more data (such as mobile phone numbers, personal information, etc.) after the event is completed (for example, a purchase process is completed), the model can be transmitted back in real time to correct the erroneous recognition results. Through the aforementioned embodiments, the event data can be annotated in real time, and the device fingerprint library can be continuously corrected to reduce the drift rate and conflict rate.

[0096] See also Figure 4 , is a block diagram of an embodiment of a cross-scenario device fingerprint recognition device provided by an embodiment of the present invention. Figure 4 As shown, the device may include:

[0097] An acquisition module 41 is used to acquire an event feature vector pair of an event to be matched, wherein the event feature vector pair includes an event feature vector corresponding to the event to be matched and an event feature vector of any known event in a device fingerprint library, wherein the known event refers to an event with a device fingerprint, and the device fingerprint library includes event feature vectors of multiple known events and corresponding relationships between device fingerprints;

[0098] A feature extraction module 42 is used to input the event feature vector pair into an adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to any known event, wherein in the training process of the adversarial triplet neural network model, event scene parameters are used as adversarial data;

[0099] A similarity determination module 43, configured to determine the similarity between the first network feature and the second network feature;

[0100] The identification module 44 is used to determine whether the event to be matched and the any known event come from the same device according to the similarity, and when it is determined that the event to be matched and the any known event come from the same device, assign the device fingerprint of the any known event to the event to be matched.

[0101] In a possible implementation, the acquisition module 41 includes (not shown in the figure):

[0102] A first acquisition unit, used to acquire an event feature vector of an event to be matched;

[0103] A second acquisition unit, used to acquire an event feature vector of any known event from a preset device fingerprint library;

[0104] A combining unit is used to combine the event feature vector of the event to be matched and the event feature vector of any known event to form an event feature vector pair of the event to be matched.

[0105] In a possible implementation manner, the similarity determination module 43 is specifically used to:

[0106] Determine a Euclidean distance between the first network feature and the second network feature; and determine a similarity between the first network feature and the second network feature based on the Euclidean distance.

[0107] In a possible implementation manner, the identification module 44 is specifically used for:

[0108] The similarity is compared with a preset similarity threshold; if the similarity is greater than or equal to the preset similarity threshold, it is determined that the event to be matched and any known event are from the same device; if the similarity is less than the preset similarity threshold, it is determined that the event to be matched and any known event are from different devices.

[0109] In a possible implementation, the device further includes (not shown in the figure):

[0110] A sample acquisition module, used to acquire multiple sample data, wherein the sample data includes positive examples, negative examples, and anchor points, wherein the positive examples and the anchor points correspond to the same device, and the negative examples and the anchor points correspond to different devices;

[0111] The training module is used to train the initial adversarial triplet neural network model using the plurality of sample data to obtain a trained adversarial triplet neural network model.

[0112] In a possible implementation, the training module is specifically used for:

[0113] Inputting the positive examples, negative examples and anchor points in each of the sample data into three embedding layers in the initial adversarial triplet neural network model respectively, to obtain corresponding positive example network features, negative example network features and anchor point network features;

[0114] Inputting the positive network features, the negative network features, and the anchor network features into the trained device matching model and the scene classification model, respectively, to obtain a first distance and a second distance output by the device matching model, and a predicted scene probability distribution output by the scene classification model, wherein the first distance is used to characterize the distance between the anchor network features and the positive network features, and the second distance is used to characterize the distance between the anchor network features and the negative network features;

[0115] Determining a first loss value according to the first distance and the second distance, and determining a second loss value according to the predicted scene probability distribution;

[0116] When it is determined according to the first loss value and the second loss value that the current triplet neural network model does not meet the preset model convergence conditions, the model parameters of the current adversarial triplet neural network model are updated according to the first loss value and the second loss value and based on the adversarial learning mechanism, and the adversarial triplet neural network model after the updated model parameters is iteratively trained until it is determined that the current adversarial triplet neural network model meets the model convergence conditions.

[0117] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, Figure 5 The electronic device 500 shown includes: at least one processor 501, a memory 502, at least one network interface 504 and other user interfaces 503. The various components in the electronic device 500 are coupled together via a bus system 505. It is understood that the bus system 505 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 505 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 505 is not shown in FIG. Figure 5 Various buses are labeled as bus system 505 .

[0118] The user interface 503 may include a display, a keyboard or a pointing device (eg, a mouse, a trackball), a touch pad or a touch screen.

[0119] It can be understood that the memory 502 in the embodiment of the present invention can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct memory bus random access memory (DRRAM). The memory 502 described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0120] In some implementations, the memory 502 stores the following elements, executable units or data structures, or a subset thereof, or an extended set thereof: an operating system 5021 and an application program 5022 .

[0121] The operating system 5021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The application 5022 includes various application programs, such as a media player (MediaPlayer), a browser (Browser), etc., which are used to implement various application services. The program for implementing the method of the embodiment of the present invention can be included in the application 5022.

[0122] In the embodiment of the present invention, by calling the program or instruction stored in the memory 502, specifically, the program or instruction stored in the application 5022, the processor 501 is used to execute the method steps provided by each method embodiment, for example, including the aforementioned Figure 1 The steps associated with the method embodiment shown.

[0123] The method disclosed in the above embodiment of the present invention can be applied to the processor 501, or implemented by the processor 501. The processor 501 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 501 or the instruction in the form of software. The above processor 501 can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to execute, or the hardware and software units in the decoding processor can be executed. The software unit can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 502, and the processor 501 reads the information in the memory 502 and completes the steps of the above method in combination with its hardware.

[0124] It is understood that the embodiments described herein can be implemented by hardware, software, firmware, middleware, microcode or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPDevice, DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in the present application or a combination thereof.

[0125] For software implementation, the technology described herein can be implemented by a unit that performs the functions described herein. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or outside the processor.

[0126] The electronic device provided in this embodiment may be Figure 5The electronic device shown in FIG. 1 may perform the following steps: Figure 1 All steps of the device fingerprint recognition method across scenarios are implemented Figure 1 For details, please refer to the technical effect of the cross-scenario device fingerprint recognition method. Figure 1 For the sake of brevity, the relevant description is not repeated here.

[0127] The embodiment of the present invention also provides a storage medium (computer-readable storage medium). The storage medium here stores one or more programs. The storage medium may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a read-only memory, a flash memory, a hard disk or a solid-state drive; the memory may also include a combination of the above-mentioned types of memory.

[0128] When one or more programs in the storage medium can be executed by one or more processors, the cross-scenario device fingerprint recognition method executed on the electronic device side can be implemented.

[0129] The processor is used to execute the program of the cross-scenario device fingerprint recognition method stored in the memory, so as to implement the following steps on the electronic device side: Figure 1 The steps of the cross-scenario device fingerprint recognition method of the illustrated embodiment.

[0130] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0131] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0132] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A cross-scenario device fingerprint recognition method, characterized in that: The method comprises: Acquire an event feature vector pair of the event to be matched, wherein the event feature vector pair includes an event feature vector corresponding to the event to be matched and an event feature vector of any known event in a device fingerprint library, wherein the known event refers to an event with a device fingerprint, and the device fingerprint library includes event feature vectors of multiple known events and corresponding relationships between device fingerprints; Inputting the event feature vector pair into an adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to any known event, wherein in the training process of the adversarial triplet neural network model, event scene parameters are used as adversarial data; determining a similarity between the first network feature and the second network feature; Determining whether the event to be matched and any known event come from the same device according to the similarity, and when it is determined that the event to be matched and any known event come from the same device, assigning the device fingerprint of any known event to the event to be matched; Wherein, the method further comprises: Acquire multiple pieces of sample data, the sample data including positive examples, negative examples, and anchor points, wherein the positive examples and the anchor points correspond to the same device, and the negative examples and the anchor points correspond to different devices; Using the plurality of sample data to train the initial adversarial triplet neural network model to obtain a trained adversarial triplet neural network model; The method of training the initial adversarial triplet neural network model using the plurality of sample data to obtain a trained adversarial triplet neural network model includes: Inputting the positive examples, negative examples and anchor points in each of the sample data into three embedding layers in the initial adversarial triplet neural network model respectively, to obtain corresponding positive example network features, negative example network features and anchor point network features; Inputting the positive network features, the negative network features, and the anchor network features into the trained device matching model and the scene classification model, respectively, to obtain a first distance and a second distance output by the device matching model, and a predicted scene probability distribution output by the scene classification model, wherein the first distance is used to characterize the distance between the anchor network features and the positive network features, and the second distance is used to characterize the distance between the anchor network features and the negative network features; Determining a first loss value according to the first distance and the second distance, and determining a second loss value according to the predicted scene probability distribution; When it is determined according to the first loss value and the second loss value that the current triplet neural network model does not meet the preset model convergence conditions, the model parameters of the current adversarial triplet neural network model are updated according to the first loss value and the second loss value and based on the adversarial learning mechanism, and the adversarial triplet neural network model after the updated model parameters is iteratively trained until it is determined that the current adversarial triplet neural network model meets the model convergence conditions.

2. The method according to claim 1, characterized in that The step of obtaining an event feature vector pair of an event to be matched includes: Obtaining the event feature vector of the event to be matched; Obtaining an event feature vector of any known event from a preset device fingerprint library; The event feature vector of the event to be matched and the event feature vector of any known event are combined to form an event feature vector pair of the event to be matched.

3. The method according to claim 1, characterized in that: The determining the similarity between the first network feature and the second network feature includes: determining a Euclidean distance between the first network feature and the second network feature; The similarity between the first network feature and the second network feature is determined according to the Euclidean distance.

4. The method according to claim 1 or 3, characterized in that: The determining, according to the similarity, whether the event to be matched and any known event come from the same device includes: comparing the similarity with a preset similarity threshold; If the similarity is greater than or equal to the preset similarity threshold, it is determined that the event to be matched and any known event are from the same device; If the similarity is less than the preset similarity threshold, it is determined that the event to be matched and any known event come from different devices.

5. The method according to claim 1, characterized in that The method further comprises: After the device fingerprint of any known event is assigned to the event to be matched, the event feature vector corresponding to the any known event in the device fingerprint library is updated to the event feature vector of the event to be matched.

6. A cross-scenario device fingerprint recognition device, characterized in that: include: An acquisition module, used to acquire an event feature vector pair of an event to be matched, wherein the event feature vector pair includes an event feature vector corresponding to the event to be matched and an event feature vector of any known event in a device fingerprint library, wherein the known event refers to an event with a device fingerprint, and the device fingerprint library includes event feature vectors of multiple known events and corresponding relationships between device fingerprints; A feature extraction module, used for inputting the event feature vector pair into an adversarial triplet neural network model to extract a first network feature corresponding to the event to be matched and a second network feature corresponding to any known event, wherein in the training process of the adversarial triplet neural network model, event scene parameters are used as adversarial data; a similarity determination module, configured to determine a similarity between the first network feature and the second network feature; an identification module, configured to determine, based on the similarity, whether the event to be matched and any known event are from the same device, and when it is determined that the event to be matched and any known event are from the same device, assign a device fingerprint of any known event to the event to be matched; A sample acquisition module, used to acquire multiple sample data, wherein the sample data includes positive examples, negative examples, and anchor points, wherein the positive examples and the anchor points correspond to the same device, and the negative examples and the anchor points correspond to different devices; A training module, used to train the initial adversarial triplet neural network model using the plurality of sample data to obtain a trained adversarial triplet neural network model; Wherein, the training module is specifically used for: Inputting the positive examples, negative examples and anchor points in each of the sample data into three embedding layers in the initial adversarial triplet neural network model respectively, to obtain corresponding positive example network features, negative example network features and anchor point network features; Inputting the positive network features, the negative network features, and the anchor network features into the trained device matching model and the scene classification model, respectively, to obtain a first distance and a second distance output by the device matching model, and a predicted scene probability distribution output by the scene classification model, wherein the first distance is used to characterize the distance between the anchor network features and the positive network features, and the second distance is used to characterize the distance between the anchor network features and the negative network features; Determining a first loss value according to the first distance and the second distance, and determining a second loss value according to the predicted scene probability distribution; When it is determined according to the first loss value and the second loss value that the current triplet neural network model does not meet the preset model convergence conditions, the model parameters of the current adversarial triplet neural network model are updated according to the first loss value and the second loss value and based on the adversarial learning mechanism, and the adversarial triplet neural network model after the updated model parameters is iteratively trained until it is determined that the current adversarial triplet neural network model meets the model convergence conditions.

7. An electronic device, characterized in that: include: A processor and a memory, wherein the processor is used to execute a cross-scenario device fingerprint recognition program stored in the memory to implement the cross-scenario device fingerprint recognition method described in any one of claims 1 to 5.

8. A storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the cross-scenario device fingerprint recognition method described in any one of claims 1-5.

Images