A method and apparatus for identifying a continuation of life, an electronic device, and a storage medium thereof

Abstract

The application provides a method and device for identity continuation, electronic equipment and a storage medium thereof. The method comprises: obtaining a rated number of paired training samples, obtaining a similarity training model by deep learning training according to the paired training samples, receiving a sample to be continued, obtaining a plurality of historical samples of different sources in historical data according to the sample to be continued, calculating a target historical sample by the similarity training model, the target historical sample being a sample with the highest similarity, and mapping the identity to be continued to the target historical sample to achieve identity continuation. The data in the new user identity and the original user identity are associated in a deep learning manner, which is used for synchronization or retrieval, and helps to solve the technical problem that the new user identity cannot be associated with the data under the original user identity after the id-Mapping in the prior art is replaced by the device or modified by the device parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computers, and in particular to a method and device for identifying a user, an electronic device and a storage medium thereof. BACKGROUND

[0002] At present, user tracking technology is widely used. For example, a user has a tablet computer A and then purchases a mobile phone A. Since the user is the same, the user hopes to associate the data of the two devices (i.e., the tablet computer A and the mobile phone A) so that the tablet computer A and the mobile phone A can synchronize or access data. How to achieve the function in the above scenario? The existing technology in the industry is id-Mapping. Id-Mapping is to map and aggregate different sources of identification (such as device identification, user identification, etc.), and ultimately link fragmented data, eliminate data islands, and provide a complete information graph of a user. After having a complete information graph, the tablet computer A or the mobile phone A can synchronize or access data in other devices according to the information graph to achieve the purpose of user tracking.

[0003] However, the id-Mapping in the prior art can not trace the historical identification information after a change in a device parameter due to user device parameters, APP product iteration, etc. This is a common problem in the industry. For example, since some parameters in the mobile phone A are different from the corresponding parameters in the tablet computer A, the tablet computer A and the mobile phone A will determine that they do not belong to the same user, resulting in the id-Mapping being unable to correspond the user identification in the mobile phone A to the information graph, and thus being unable to map the user identification to the historical identification (i.e., associate with the original identification). Therefore, the tablet computer A and the mobile phone A will be unable to synchronize or access data, thereby hindering the application of the user identification.

[0004] Therefore, it is necessary to associate a new user identification to data under an original user identification after the user replaces a device or modifies a device parameter, which helps to solve the technical problem that the new user identification cannot be associated to data under the original user identification after the user replaces a device or modifies a device parameter in the id-Mapping of the prior art. SUMMARY

[0005] The present application provides a method for identifying a user, which aims to associate a new user identification with data in an original user identification after the user replaces a device or modifies a device parameter in a deep learning manner, for synchronization or access, which helps to solve the technical problem that the new user identification cannot be associated to data under the original user identification after the user replaces a device or modifies a device parameter in the id-Mapping of the prior art.

[0006] The method comprises:

[0007] obtaining a rated number of paired training samples, wherein the paired training samples are two training samples whose similarity has been determined, and the training samples comprise training sample identifiers and training sample feature data corresponding to the training sample identifiers;

[0008] training a similarity training model according to the paired training samples through deep learning, wherein the similarity training model is a model for calculating similarity;

[0009] receiving a sample to be continued, wherein the sample to be continued comprises a to-be-continued identifier and a continued feature data;

[0010] obtaining a plurality of historical samples of different sources in historical data according to the sample to be continued, wherein the historical samples comprise historical sample identifiers and historical sample feature data, and the historical samples are samples corresponding to the continued feature data and the historical sample feature data;

[0011] calculating a target historical sample according to the sample to be continued and the plurality of historical samples through the similarity training model, wherein the target historical sample is a sample with the highest similarity to the to-be-continued identifier among the historical samples;

[0012] mapping the to-be-continued identifier to the target historical sample in the historical data to realize identifier continuation.

[0013] In an embodiment, the training of the similarity training model according to the paired training samples through deep learning, wherein the similarity training model is a model for calculating similarity, comprises:

[0014] encoding the training sample feature data of two samples in the paired training samples into first fixed-dimension vectors, respectively;

[0015] splicing the first fixed-dimension vectors of the same paired training sample according to the determined similarity through deep learning to obtain a similarity neural network;

[0016] performing full-network splicing on the similarity neural networks of different paired training samples to obtain the similarity training model.

[0017] In an embodiment, the full-network splicing of the similarity neural networks of different paired training samples to obtain the similarity training model comprises:

[0018] performing full-network splicing on the similarity neural networks of different paired training samples to obtain a to-be-detected similarity training model;

[0019] substitute the pair of training samples into the training model to be detected to calculate a detection similarity;

[0020] calculate a difference value as an error value according to the similarity of the pair of training samples and the detection similarity;

[0021] determine whether the error value is less than a predetermined error threshold,

[0022] if the error value is less than the predetermined error threshold, use the training model to be detected as the similarity training model.

[0023] In an embodiment, the determination of whether the error value is less than a predetermined error threshold, the method further comprises:

[0024] if the error value is greater than or equal to the predetermined error threshold, adjust the parameters in the training model to be detected to make the error value less than the predetermined error threshold, and then use the training model to be detected as the similarity training model.

[0025] In an embodiment, the target historical sample is calculated according to the sample to be continued and the plurality of historical samples by the similarity training model, wherein the target historical sample is the sample with the highest similarity to the sample to be continued in the historical samples.

[0026] encode the continuation feature data and the historical sample feature data into a second fixed dimension vector;

[0027] obtain a target historical sample by the similarity training model according to the second fixed dimension vector, wherein the target historical sample is the sample with the highest similarity to the sample to be continued in the historical samples.

[0028] In an embodiment, the step of obtaining a predetermined number of pair of training samples, wherein the pair of training samples are two training samples with a determined similarity, and the training sample comprises a training sample identifier and training sample feature data corresponding to the training sample identifier, the method further comprises:

[0029] clean the training sample feature data in the pair of training samples to obtain data type and category type training sample feature data.

[0030] In an embodiment, the plurality of historical samples of different sources in the historical data are obtained according to the sample to be continued, wherein the historical sample comprises a historical sample identifier and historical sample feature data, and the historical sample is the sample corresponding to the continuation feature data and the historical sample feature data, the method comprises:

[0031] determining whether historical sample feature data of a historical sample in a data source corresponds to the continuation feature data, and if the historical sample feature data corresponds to the continuation feature data, performing a subsequent step.

[0032] In an embodiment, the application further provides an identification continuation device, comprising:

[0033] an acquisition module configured to acquire a rated number of paired training samples, wherein the paired training samples are two training samples with a determined similarity, and the training samples comprise training sample identifiers and training sample feature data corresponding to the training sample identifiers;

[0034] a training module configured to train a similarity training model through deep learning according to the paired training samples, wherein the similarity training model is a model for calculating similarity;

[0035] a receiving module configured to receive a sample to be continued, wherein the sample to be continued comprises a continuation identifier and continuation feature data;

[0036] The acquisition module is further configured to acquire a plurality of historical samples of different sources in historical data according to the sample to be continued, wherein the historical samples comprise historical sample identifiers and historical sample feature data, and the historical samples are samples with the historical sample feature data corresponding to the continuation feature data;

[0037] a calculation module configured to calculate a target historical sample through the similarity training model according to the sample to be continued and the plurality of historical samples, wherein the target historical sample is a sample with the highest similarity to the continuation identifier among the historical samples;

[0038] a mapping module configured to map the continuation identifier to the target historical sample in the historical data to realize identification continuation.

[0039] In an embodiment, the device further comprises:

[0040] an encoding module configured to encode the training sample feature data of two samples in the paired training samples into first fixed-dimension vectors, respectively;

[0041] a concatenation module configured to concatenate the training sample feature data of two samples in the paired training samples into first fixed-dimension vectors, respectively;

[0042] and further configured to concatenate the two first fixed-dimension vectors of the same paired training sample into a similarity neural network through deep learning according to the determined similarity;

[0043] The similarity neural networks of different pairs of training samples are fully network spliced to obtain the similarity training model.

[0044] In an embodiment, the splicing module is further configured to fully network splice the similarity neural networks of different pairs of training samples to obtain a to-be-detected similarity training model.

[0045] The apparatus further includes:

[0046] The substitution module is configured to substitute the pairs of training samples into the to-be-detected similarity training model to calculate a detected similarity.

[0047] The calculation module is further configured to calculate a difference value as an error value according to the similarity determined for the pairs of training samples and the detected similarity.

[0048] The judgment module is configured to judge whether the error value is less than a predetermined error threshold value, and if the error value is less than the predetermined error threshold value, to take the to-be-detected similarity training model as the similarity training model.

[0049] In an embodiment, the judgment module is further configured to, if the error value is greater than or equal to the predetermined error threshold value, adjust parameters in the to-be-detected similarity training model to make the error value less than the predetermined error threshold value, and then take the to-be-detected similarity training model as the similarity training model.

[0050] In an embodiment, the encoding module is configured to encode the to-be-continued feature data and the historical sample feature data into a second fixed-dimension vector.

[0051] The calculation module is configured to obtain a target historical sample through the similarity training model from the second fixed-dimension vector, where the target historical sample is a sample in the historical samples that has the highest similarity with the to-be-continued identifier.

[0052] In an embodiment, the apparatus further includes:

[0053] The cleaning module is configured to clean the training sample feature data in the pairs of training samples to obtain data-type and category-type training sample feature data.

[0054] In an embodiment, the judgment module is configured to judge whether historical sample feature data of a historical sample in a data source corresponds to the to-be-continued feature data, and if the historical sample feature data corresponds to the to-be-continued feature data, to execute a subsequent step.

[0055] In an embodiment, the present application provides an electronic device, which includes a processor and a memory.

[0056] The memory stores an application executable by the processor, for causing the processor to execute the steps of the method.

[0057] In an embodiment, the present application provides a computer readable storage medium, which stores a computer program, characterized in that the program is executed by a processor to implement the steps of the method.

[0058] As can be seen from the above, based on the above embodiment, the present application provides a method for identity continuation, which can associate the data of the new user identity with the data in the original user identity after the user changes the device or modifies the device parameters and the like through the deep learning manner, for synchronization or retrieval, which helps to solve the technical problem that the new user identity cannot be associated with the data under the original user identity after the user changes the device or modifies the device parameters and the like in the id-Mapping of the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 Flowchart 100 of the method for identity continuation of the present application;

[0060] Figure 2 Flowchart 200 of the method for identity continuation of the present application;

[0061] Figure 3 Flowchart 300 of the method for identity continuation of the present application;

[0062] Figure 4 Flowchart 400 of the method for identity continuation of the present application;

[0063] Figure 5 Schematic diagram of the DNN neural network architecture of the present application;

[0064] Figure 6 Schematic diagram of the device architecture for identity continuation of the present application. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical scheme and advantages of the present application clearer, further detailed description will be made to the present application with reference to the accompanying drawings and embodiments.

[0066] The id-Mapping in the prior art can solve the user tracking problem in the background art, but due to user equipment parameters, APP product iteration and other reasons, the id-Mapping method can fail, and the user tracking problem cannot be solved. The following will explain the failure reason of id-Mapping through a specific case. The reason why id-Mapping can establish an information graph between different devices is that there is common data between different devices. For example, a user has two devices, a tablet computer A and a mobile phone A, and the historical information of the tablet computer A and the mobile phone A both have a common binding mobile phone number.

[0067] Mobile phone number, so they should be the same user, so id-Mapping can aggregate the imei and binding mobile phone number and mac address of the tablet computer A and the mobile phone A into an array of six parameters, and according to the fact that both devices have an array composed of these six parameters, the association between the tablet computer A and the mobile phone A is established, and the data synchronization or mutual access of the tablet computer A and the mobile phone A is realized. However, the id-Mapping method has a loophole, that is, when the imei or the binding mobile phone number, or the mac changes, the array of the tablet computer A and the mobile phone A will no longer be the same, the array will be invalid, and the tablet computer A and the mobile phone A cannot be associated with each other, resulting in user tracking failure.

[0068] The following is a specific example to illustrate the problem in the above example:

[0069] For example, the ID information of two logs is:

[0070] Device 1: <mac1, mac2> <imei1> <tel1>;

[0071] Device 2: <mac1> <imei2><tel1,tel2>;

[0072] mac1 and mac2, imei1 and imei2, tel1 and tel2 are two different MAC addresses and two different IMEI serial numbers, as well as two different tel1 and tel2 phone numbers.

[0073] Since mac1 appears in both logs, they likely originated from the same user. Several rounds of merging were then performed to finally obtain the result.<mac1,mac2><imei1,imei2><tel1,tel2> .

[0074] Map output:

[0075] Send to Device 1:<mac1,mac2><imei1,imei2><tel1,tel2> ;

[0076] Send to device 2:<mac1,mac2><imei1,imei2><tel1,tel2> ;

[0077] Device 1 and Device 2 can establish a connection based on the map (i.e., the array). If the phone number of one of the devices is changed to tel3, then the map will become invalid, and the connection between Device 1 and Device 2 will also be lost.

[0078] Therefore, how to enable the same user to establish associations between all devices even after changing the parameters in the array after creating different device identifiers has become an urgent problem for the R&D personnel to solve.

[0079] Figure 1 This is a schematic diagram of the process 100 for the life-extending method of the present invention. (See diagram 100.) Figure 1 As shown, in one embodiment, this application provides an identification extension method, the method comprising:

[0080] S101, Obtain a fixed number of paired training samples, wherein the paired training samples are two training samples whose similarity has been determined, and the training samples include training sample identifiers and training sample feature data corresponding to the training sample identifiers.

[0081] This step provides a specific implementation method for obtaining a fixed number of paired training samples. Each paired training sample in this step contains two training samples. Because similarity is a relative concept, at least two training samples are required to obtain the similarity score. The training sample feature data can be understood as device attribute data or user behavior data, such as the user's gender, age, geographical location, etc.

[0082] In addition, in one of the pair of training samples, the similarity of the two training samples is known to be determined, and the similarity algorithm is more common in the prior art, which will not be described here. As long as the similarity between the two training samples can be expressed, it is necessary to point out that the similarity can indicate that the two samples point to the same user, or they belong to different users. As long as the similarity of the two can be determined, for example, when the similarity is greater than 80%, it means that the two training samples belong to the same user. Conversely, if it is less than or equal to 80%, it means that the two training samples do not belong to the same user. Both of these cases are acceptable, and how to handle them will be further described later.

[0083] S102, obtaining a similarity training model by deep learning training according to the pair of training samples, wherein the similarity training model is a model for calculating similarity.

[0084] In this step, a specific step of training the similarity training model according to the pair of training samples is provided. The similarity training model can be used to calculate the similarity, and the training method can be implemented by using a DNN neural network, for example, the similarity of the two training samples in the pair of training samples is known to be 76%. Then, the objective function about the training sample feature data is constructed by the neural network, and the corresponding similarity of 76% is obtained by the calculation of the objective function. Then it is indicated that the constructed objective function may be correct, and the objective function can be used as the similarity training model for subsequent processing. More specific implementation will be further described later, which will not be described here.

[0085] S103, receiving a to-be-continued sample, wherein the to-be-continued sample includes a to-be-continued identifier and a continued feature data.

[0086] In this step, a specific implementation of receiving a to-be-continued sample is provided. The to-be-continued sample is equivalent to the user identifier and the corresponding feature data of the mobile phone A, that is, the to-be-continued identifier and the continued feature data. It should be pointed out that the continued feature data can also be understood as device attribute data or user behavior data, such as user's gender, age, geographical location, etc. Only the continued feature data is associated in the to-be-continued sample.

[0087] S104, obtaining a plurality of historical samples of different sources in the historical data according to the to-be-continued sample, wherein the historical sample includes a historical sample identifier and a historical sample feature data, and the historical sample is a sample corresponding to the historical sample feature data and the continued feature data.

[0088] In this step, a specific embodiment of receiving the sample to be continued is provided. In the above example of the tablet A and the mobile phone A, the historical sample corresponds to the sample from the tablet A. Of course, there can be a tablet B, a smart watch B, a mobile phone B, and other devices. These devices can be understood as different sources. The user identification and feature data in these devices are the historical sample identification and historical sample feature data in this step. The historical sample feature data includes attribute data such as user gender and age, and behavior data such as user use of a mobile phone number and a device.

[0089] In this step, it is pointed out that the continuation feature data and the historical sample feature data should be corresponding data. This can simplify the processing flow. The reason is that the similarity needs to be calculated by comparing the same type of feature data. It is obviously unreasonable to calculate the similarity between the user age of the tablet and the mobile phone number of the mobile phone A. Therefore, if the continuation feature data and the historical sample feature data do not correspond to each other, such historical samples will not be obtained. Even if the continuation feature data and the historical sample feature data correspond to each other, there can still be many historical samples, such as the tablet B, the smart watch B, and the mobile phone B, which meet the corresponding requirements. Therefore, the historical samples corresponding to these devices can be obtained.

[0090] S105, calculating a target historical sample from the sample to be continued and the plurality of historical samples by using the similarity training model, wherein the target historical sample is the sample with the highest similarity to the sample to be continued in the historical samples.

[0091] In this step, a specific step of calculating a target historical sample from the sample to be continued and the plurality of historical samples by using the similarity training model is provided. The target historical sample is the sample with the highest similarity to the sample to be continued in the historical samples. This step can be understood as substituting the continuation feature data and the historical sample feature data into the target function. Then, the plurality of similarities between different historical samples and the sample to be continued can be calculated. The historical sample with the highest similarity is selected as the target historical sample. For example, the target historical sample is the corresponding sample of the tablet A. This indicates that the tablet A and the mobile phone A are probably the same user.

[0092] S106, mapping the sample to be continued to the target historical sample in the historical data to realize identity continuation.

[0093] In this step, a specific step of mapping the to-be-continued identity to the historical data is provided, and finally the to-be-continued identity in the mobile phone A is associated with the tablet computer A in the historical data, the data island in the tablet computer A is eliminated, and the continuation of the to-be-continued identity in the mobile phone in the historical data is completed.

[0094] A method for identifying a continuation of life is provided in the embodiment. First, a rated number of pairs of training samples are obtained from a data source for training of a similarity training model. Since the similarity is a relative concept, the pair of training samples are two training samples with a determined similarity therebetween. The rated number means a reasonable number of pairs of training samples that can be used to train the similarity training model in a subsequent step. It should be noted that the similarity does not mean that the two training samples must be similar samples. They can also be two dissimilar training samples, as long as the similarity between the two training samples can be determined. The similarity training model is obtained by deep learning training based on the pairs of training samples. The similarity training model is a model for calculating similarity. The specific training method can be implemented by those skilled in the art, and the training method will be further explained below. Therefore, it will not be described here. When the similarity model is obtained, the continuation of life of an identifier can be realized. A sample to be continued is received, wherein the sample to be continued includes an identifier to be continued and continuation characteristic data. The identifier to be continued and the continuation characteristic data in the sample to be continued can be understood as the identifier of the mobile phone A and the data characteristics of the gender, age, and geographical location of the user of the mobile phone A. Then, a plurality of historical samples of different sources are obtained based on the sample to be continued, wherein the historical sample is a sample corresponding to the continuation characteristic data in historical data, and the historical sample includes a historical sample identifier and historical sample characteristic data. The historical data can be understood as the data of the tablet computer A. Of course, the historical data can also include the data of the tablet computer B, the smart watch B, and the mobile phone B, and the like. The plurality of historical samples of different sources (i.e., different devices) should include samples corresponding to the historical sample characteristic data and the continuation characteristic data. For example, the identifier of the tablet computer A, i.e., the historical sample identifier of the tablet computer A, and the characteristic data of the tablet computer A, i.e., the historical sample characteristic data. Of course, the tablet computer B, the smart watch B, and the mobile phone B, and the like have corresponding historical sample identifiers and historical sample characteristic data. It should be noted that the sample corresponding to the historical sample characteristic data and the continuation characteristic data, i.e., the continuation characteristic data of the mobile phone A, can be many, such as the gender and age of the user, and the mobile phone number. Therefore, the historical sample characteristic data should also involve the gender and age of the user of the tablet computer A, and the mobile phone number.Then, a target historical sample is calculated by the similarity training model according to the to-be-continued sample and the plurality of historical samples, wherein the target historical sample is a sample with the highest similarity to the to-be-continued sample in the historical samples. As in the previous example, the historical samples of different sources can include the samples of the tablet A and the tablet B, and the samples of the smart watch B and the mobile phone B. Then, the target historical sample with the highest similarity to the to-be-continued sample in the devices is calculated by the similarity training model, for example, the tablet A and the mobile phone A have the highest similarity. Finally, the to-be-continued sample is mapped to the target historical sample in the historical data to realize the identification continuation. The mobile phone A and the tablet A can be associated with data, and the data island of the tablet A is eliminated. In the above process, since the similarity is relied on instead of a specific array, that is, the map above, even if the device is replaced or the parameters in the device are modified, the final data association will not be affected. This helps to solve the technical problem that the id-Mapping in the prior art cannot continue to associate data between devices after the user replaces the device or modifies the parameters of the device.

[0095] Figure 2 The flowchart 200 of the identification continuation method of the present application is shown in FIG. 1. Figure 2 As shown in FIG. 1, in an embodiment, the similarity training model is obtained by deep learning training according to the pair of training samples, wherein the similarity training model is a model for calculating similarity, and the similarity training model includes:

[0096] S201, the training sample feature data of the two training samples in the pair of training samples are respectively encoded into vectors of a first fixed dimension.

[0097] In this step, a specific step of encoding the training sample feature data into vectors of a first fixed dimension is provided. For example, the category type feature in the training sample feature data is converted into a low-dimensional dense vector through an embedding layer (i.e., a concatenation layer), and then the low-dimensional dense vector is respectively concatenated with the corresponding numerical type feature. The low-dimensional dense vectors of the two training samples are respectively converted into vectors of the first fixed dimension through the embedding layer.

[0098] First, explain the low-dimensional dense vector, such as a vector (1.0, 0.0, 3.0), which is a one-dimensional array, which can have two expressions, because there is only one dimension, so it can be understood as a low-dimensional array, and has two expressions: dense and sparse, wherein the dense type is represented as (1.0, 0.0, 3.0), which is no different from the general array, and the sparse type is represented as (3, [0, 2], [1.0, 3.0]), the first number represents the number of elements, the second number represents the element index, and the third number represents the element value. This step uses dense expression.

[0099] The fixed dimension vector is a vector with a fixed dimension, such as user gender, male represented by 0 and female represented by 1, and mobile phone number as a specific numerical value, AAAAAA. For example, first convert the user gender into a one-dimensional low-dimensional array (0 or 1), that is, the low-dimensional dense vector, and then concatenate it with the specific numerical value of the mobile phone number. For example, the first training sample, user 1 is male, and the mobile phone number is AAAAAA, and the second training sample, user 2 is male, and the mobile phone number is BBBAAAAA. The first fixed dimension vector of the first training sample is (0, AAAAAA), and the first fixed dimension vector of the second training sample is (0, BBBAAAAA).

[0100] S202, the two first fixed dimension vectors of the same pair of training samples are concatenated according to the determined similarity through deep learning to obtain a similarity neural network.

[0101] The specific implementation of the concatenation of the two first fixed dimension vectors in this step. The similarity neural network using DNN in this step can be understood as a function concatenated by the first fixed dimension vector, and the similarity of the pair of training samples can be calculated through this function. Because the similarity of the pair of training samples is a known quantity, the pair of training samples will calculate a new similarity through the similarity neural network, and continuously adjust the function of the similarity neural network to make the determined similarity consistent with the similarity calculated through the similarity neural network. At this time, the similarity neural network training is completed.

[0102] S203, the similarity neural networks of different pairs of training samples are concatenated to obtain the similarity training model.

[0103] In this step, an embodiment of training the similarity training model is provided. Because the training sample feature data of different pairs of training samples can be different, the similarity neural network trained by the pairs of training samples is different for the training sample feature data, and therefore the similarity training model obtained by full network splicing of different similarity neural networks of the pairs of training samples can calculate the similarity of different pairs of training samples that have been trained.

[0104] In this embodiment, an embodiment of training the similarity training model is provided. First, the training sample feature data of two samples in the pairs of training samples is respectively encoded into a first fixed dimension vector. Then, the two first fixed dimension vectors are spliced to obtain the similarity neural network. Finally, the similarity neural network is full network spliced to obtain the similarity model.

[0105] Figure 3 The flow 300 of the method for identifying the continuation method of the application is shown. As shown in the embodiment, the similarity training model obtained by full network splicing of different similarity neural networks of the pairs of training samples includes: Figure 3

[0106] S301, the similarity neural network of different pairs of training samples is full network spliced to obtain a to-be-detected similarity training model.

[0107] In this step, an embodiment of obtaining the to-be-detected similarity training model by splicing the similarity neural network is provided. The to-be-detected similarity training model can be understood as the objective function, but the objective function is a function that may still have inaccuracy, because the above splicing process may reduce the accuracy of calculating the similarity.

[0108] S302, the pairs of training samples are substituted into the to-be-detected similarity training model to calculate a detected similarity.

[0109] In this step, an embodiment of substituting the pairs of training samples into the to-be-detected similarity training model is provided, and then the detected similarity is calculated.

[0110] S303, according to the similarity of the pairs of training samples that have been determined and the detected similarity, a difference value is calculated as an error value.

[0111] In this step, an embodiment of subtracting the similarity that has been determined and the detected similarity to obtain the error value is provided.

[0112] ​S304, judging whether the error value is less than a predetermined error threshold value, and if the error value is less than the predetermined error threshold value, taking the to-be-detected similarity training model as the similarity training model.

[0113] In this step, the sizes of the error value and the predetermined error threshold value are judged to determine whether the to-be-detected similarity training model can be taken as the similarity training model.

[0114] In this embodiment, a specific implementation of judging whether the to-be-detected similarity training model can be taken as the similarity training model is provided. First, the similarity neural networks of different pairs of training samples are spliced to obtain a to-be-detected similarity training model. The pairs of training samples are substituted into the to-be-detected similarity training model to calculate a detection similarity. According to the similarity of the pairs of training samples which has been determined and the detection similarity, a difference value is calculated as an error value. It is judged whether the error value is less than a predetermined error threshold value. If the error value is less than the predetermined error threshold value, the to-be-detected similarity training model is taken as the similarity training model.

[0115] In an embodiment, the judging whether the error value is less than a predetermined error threshold value further includes:

[0116] If the error value is greater than or equal to the predetermined error threshold value, the parameters in the to-be-detected similarity training model are adjusted to make the error value less than the predetermined error threshold value, and then the to-be-detected similarity training model is taken as the similarity training model.

[0117] In this embodiment, a specific implementation of the error value being less than the predetermined error threshold value is provided. If the error value is greater than or equal to the predetermined error threshold value, the parameters in the to-be-detected similarity training model are adjusted to make the error value less than the predetermined error threshold value, and then the to-be-detected similarity training model is taken as the similarity training model. The parameters should be understood as the parameters in the objective function.

[0118] Figure 4 The flow 400 of the method for identifying a continuation method of the present application is shown in FIG. 4. As shown in FIG. 4, in an embodiment, the target historical sample is calculated by the similarity training model according to the to-be-continued sample and the plurality of historical samples, wherein the target historical sample is the sample with the highest similarity to the to-be-continued identifier among the historical samples, and the method includes the following steps. Figure 4

[0119] S401, encoding the continuation feature data and the historical sample feature data into a second fixed-dimension vector.

[0120] ​In this step, a specific implementation of encoding the continuation feature data and the historical sample feature data into the second fixed dimension vector is provided. Since the objective function of the similarity training model is substituted into the fixed dimension vector by special data, the fixed dimension vector should also be substituted in the subsequent calculation, so the continuation feature data and the historical sample feature data should be encoded into the second fixed dimension vector in this step.

[0121] The second fixed dimension vector has the same data structure as the first fixed dimension vector, except that the first fixed dimension vector is derived from two pairs of training samples, while the second fixed dimension vector is derived from the continuation feature data and the historical sample feature data.

[0122] S402, obtaining a target historical sample by the second fixed dimension vector through the similarity training model, wherein the target historical sample is the sample with the highest similarity to the to-be-continued identifier in the historical samples.

[0123] In this step, a specific implementation of calculating the target historical sample by the second fixed dimension vector is provided, wherein the target historical sample is the sample with the highest similarity to the to-be-continued identifier in the historical samples. The second fixed dimension vector is substituted into the objective function to obtain the similarity, and the sample with the highest similarity value is selected.

[0124] In this embodiment, a specific implementation of how to calculate the target historical sample by the second fixed dimension vector is provided. The continuation feature data and the historical sample feature data are encoded into the second fixed dimension vector. The second fixed dimension vector is substituted into the similarity training model to obtain a target historical sample, wherein the target historical sample is the sample with the highest similarity to the to-be-continued identifier in the historical samples. Finally, the target historical sample is calculated.

[0125] In an embodiment, after the step of obtaining a rated number of paired training samples, wherein the paired training samples are two training samples with determined similarity, and the training samples include training sample identifiers and training sample feature data corresponding to the training sample identifiers, the method further includes:

[0126] Cleaning the training sample feature data in the paired training samples to obtain data type and category type training sample feature data.

[0127] In this embodiment, a specific implementation of cleaning and screening the training sample data before training the paired training samples is provided.

[0128] In an embodiment, the method comprises obtaining a plurality of historical samples from different sources in the historical sample acquisition history of the to-be-continued sample, wherein the historical sample comprises historical sample identification and historical sample feature data, and the historical sample is a sample step before the historical sample feature data corresponds to the continuation feature data.

[0129] determining whether the historical sample feature data of the historical sample in the data source corresponds to the continuation feature data, and if the historical sample feature data corresponds to the continuation feature data, performing the subsequent step.

[0130] In the embodiment, a specific implementation of determining whether the historical sample feature data corresponds to the continuation feature data is provided. When the historical sample feature data corresponds to the continuation feature data, the subsequent step is performed. The essence of the embodiment is to select the historical sample that can calculate the similarity through the similarity training model, otherwise, if the historical sample feature data does not correspond to the continuation feature data, the similarity cannot be calculated between them, and therefore, such historical sample does not have the value of performing the subsequent step except increasing energy consumption.

[0131] Basic principle and working process:

[0132] The historical sample is composed of different device sources such as the tablet computer A and the tablet computer B, the smart watch B, and the mobile phone B, and the application is further explained and described by taking the newly added mobile phone A of a user as an example. The tablet computer A has user identification B, gender and age of the user, and mobile phone number B, and the mobile phone number B is XXX.

[0133] The user establishes user identification A in the mobile phone A, and the feature data mapped or associated with the user identification A, and in this example, the gender and age of the user and the mobile phone feature data, for example, the gender is male, the age is 20, and the mobile phone A is XXX.

[0134] In an embodiment, the application provides a method for identifying a continuation, which comprises:

[0135] obtaining a plurality of paired training samples, wherein the paired training sample is two training samples with a determined similarity, and the training sample comprises training sample identification and training sample feature data corresponding to the training sample identification.

[0136] In this step, a rated number of paired training samples are obtained, which should be understood as the number of correct models that can be trained, and the number of paired training samples can be set by the developer. It should be pointed out that the paired training samples should at least have two training samples whose similarity has been determined, and each training sample includes the training sample identifier and the training sample feature data. The training sample feature data should be selected as the feature data corresponding to the user identifier A, because only in this way can the model trained be applicable to the continuation of the user identifier A. For example, if the training sample feature data is not the attribute of the user or the attribute of the device, such as two paintings, although the similarity of the two paintings can be calculated, but the probability of applying the training sample to the similarity calculation of the user identifier A after training is low, and therefore it does not meet the requirements of this step. In this step, the training sample feature data should be selected as the feature data that the user identifier A may be associated with. For example, the gender and age of the user, the mobile phone number, etc.

[0137] In addition, the similarity is only a numerical value, which can represent that the two training samples of the paired training sample are similar samples, or are not similar samples. Both of the above cases can be trained, because the purpose of the paired training sample is to train a model, even if the two training samples in the paired training sample are not similar, their reverse is actually similar, so both can be trained.

[0138] The training sample feature data in the paired training sample is cleaned to obtain data type and category type training sample feature data.

[0139] In this step, the training sample feature data is cleaned to obtain data type and category type. Why is it cleaned? Because the data type and the category type data can greatly reduce the calculation amount. For example, the gender of the user can be understood as the category type, because the gender is generally male or female, and 0 or 1 can be used to represent different categories. The data type, such as the age of the user, is a specific numerical value. It is not difficult to find that, for example, to determine the similarity of the category type data, only the gender of the user needs to be determined, and the calculation amount of the data type is also small, such as the mobile phone number. Then, whether the mobile phone numbers in the two training samples are the same can be determined. Therefore, whether the data type or the category type data has the characteristics of small calculation amount when calculating the similarity. However, if the similarity of two paintings is calculated, the algorithm and calculation amount are very large, and at the same time, it is not easy to form a fixed dimension vector afterwards. Therefore, the paired training sample is cleaned in this step.

[0140] Generally, the training sample feature data can include user gender, age, and other attribute data and user behavior data using mobile phone numbers, devices, etc., which are cleaned and converted into numerical and categorical features. Gender and other data belong to the category type, and the frequency of user mobile phone use, days, etc. belong to the numerical type.

[0141] The training sample feature data of the two samples in the pair of training samples is respectively encoded into a first fixed dimension vector.

[0142] In this step, the categorical training sample feature data in the two training samples of the pair of training samples is first converted into a low-dimensional dense vector by embedding layer, then spliced with the corresponding numerical type, and finally converted into a first fixed dimension vector by embedding layer.

[0143] The two first fixed dimension vectors of the same pair of training samples are spliced according to the determined similarity by deep learning to obtain a similarity neural network.

[0144] In this step, the two first fixed dimension vectors of the same pair of training samples are spliced to obtain the similarity neural network, and the similarity neural network is only a model spliced from a pair of training samples.

[0145] The similarity neural networks of different pairs of training samples are spliced to obtain a to-be-detected similarity training model.

[0146] In this step, it is important to splice the single similarity neural network of the rated number of pairs of training samples to obtain a total to-be-detected similarity training model, but this step is only a simple DNN neural network splicing, and cannot obtain a required model, so at this time the to-be-detected similarity training model is obtained.

[0147] Figure 5 The DNN neural network architecture diagram of the present application is shown in Figure 5 The formation of the above to-be-detected similarity training model is illustrated in the form of a DNN data model.

[0148] The vector obtained by the vector generation module is first connected to a full connection module, wherein the activation function involved in each full connection layer is a tanh function, which limits the output of each layer to (-1, 1). The activation function can not be limited to this function, and functions such as sigmoid can also be used to ensure that the output of each layer in the full connection module is normalized to a certain value range. In addition, a cross network module is connected in parallel with the full connection module, which aims to automatically construct limited high-order cross features in a controllable and efficient manner. The output of each layer is xl+1=x0xlT wl+b+xl=f(xl,wl,bl)+xl. Where x0 is the input of the first layer, xl is the output of the Lth layer, xl+1 is the output of the L+1th layer, wl and b are the parameters of the Lth layer. Wherein, the number of neurons in each layer is the same, that is, the input and output dimensions of each layer are the same. Secondly, inspired by the residual network, the function f of each layer fits the residual of xl+1-xl, which can handle the problem of gradient disappearance, so that the network can be deeper. Finally, the outputs of the full connection module and the cross network module are spliced and put into a full connection layer, and the activation function involved in the full connection layer is a sigmoid function, which limits the final output to (0, 1), indicating the probability of similarity between two samples.

[0149] When the model is trained using the model, the objective function of the trained model is:

[0150]

[0151] Where y i represents the true similarity of two IDs, p i represents the ID similarity output by the similarity discrimination model, W l represents the network parameters. The regularization term uses an L2 regularization term.

[0152] The above is an explanation of the full-splicing mathematical model in the prior art, which is not a limitation of the present application, but only to better explain the present application.

[0153] The pair of training samples is substituted into the to-be-detected similarity training model to calculate the detection similarity.

[0154] In this step, a specific step of detecting the to-be-detected similarity training model is provided, and the pair of training samples is substituted into the to-be-detected similarity training model. Specifically, the two training samples in the pair of training samples are vectorized, and then substituted into the objective function.

[0155]

[0156] The difference value calculated by the similarity and the detection similarity in this step, i.e. the error value.

[0157] The difference value calculated by the similarity and the detection similarity in this step, i.e. the error value.

[0158] It is judged whether the error value is less than a predetermined error threshold value, if the error value is less than the predetermined error threshold value, the to-be-detected similarity training model is taken as the similarity training model, if the error value is less than the predetermined error threshold value, the parameters in the to-be-detected similarity training model are adjusted to make the error value less than the predetermined error threshold value, and then the to-be-detected similarity training model is taken as the similarity training model.

[0159] This step is to confirm whether the to-be-detected similarity training model can be taken as the similarity training model, if it does not meet the predetermined error threshold value, the parameters in the to-be-detected similarity training model are adjusted, and the parameters can be understood as the parameters of the objective function. Thus, the similarity training model is obtained.

[0160] A to-be-continued sample is received, wherein the to-be-continued sample includes a to-be-continued identifier and continued characteristic data.

[0161] In this step, the user identifier A of the mobile phone A and the mapped or associated characteristic data, i.e. the to-be-continued identifier and continued characteristic data, are received.

[0162] According to the to-be-continued sample, a plurality of historical samples of different sources in historical data are obtained, wherein the historical sample includes a historical sample identifier and historical sample characteristic data, and the historical sample is a sample corresponding to the historical sample characteristic data and the continued characteristic data.

[0163] In this step, the historical sample composed of the tablet computer A and the tablet computer B, the smart watch B, and the mobile phone B and other devices of different sources in the example is obtained, and the historical sample characteristic data is the characteristic data in the above devices. It should be pointed out that if the historical sample characteristic data does not correspond to the continued characteristic data, it can be directly shielded, because the similarity between the two cannot be calculated, and the similarity is a relative concept, and the historical sample characteristic data and the continued characteristic data must correspond to be calculated.

[0164] According to the to-be-continued sample and the plurality of historical samples, a target historical sample is calculated by the similarity training model, wherein the target historical sample is a sample with the highest similarity to the to-be-continued identifier in the historical sample.

[0165] In this step, an embodiment of calculating the target historical sample is provided. The target historical sample is the sample with the highest similarity to the to-be-continued identity among the historical samples. For example, the tablet A in the example corresponds to the historical sample with the highest similarity. It is illustrated that the user of the mobile phone A and the tablet A is probably the same user.

[0166] The to-be-continued identity is mapped to the target historical sample in the historical data to realize identity continuation.

[0167] In this step, an embodiment of mapping the to-be-continued identity to the target historical sample in the historical data to realize identity continuation is provided. After that, the to-be-continued identity, that is, the user identity A of the mobile phone A, can access the data in the target historical sample, that is, the data of the tablet A, and the mobile phone A can perform data synchronization or data retrieval of the data of the tablet A. Therefore, the user identity of the mobile phone A and the tablet A is associated, and identity continuation is completed.

[0168] Figure 6 The figure is a schematic diagram of the architecture of the identity continuation device of the present application. As shown in the figure, in an embodiment, the present application provides an identity continuation device, which comprises: Figure 6

[0169] The obtaining module 101 is configured to obtain a rated number of paired training samples, wherein the paired training samples are two samples with a determined similarity, and the training samples comprise training sample identities and training sample feature data.

[0170] The training module 102 is configured to train a similarity training model by deep learning according to the paired training samples, wherein the similarity training model is a model for calculating the similarity between samples.

[0171] The receiving module 103 is configured to receive a to-be-continued sample, wherein the to-be-continued sample comprises a to-be-continued identity and continuation feature data.

[0172] The obtaining module 101 is further configured to obtain a plurality of historical samples of different sources according to the to-be-continued sample, wherein the historical samples are samples corresponding to the continuation feature data in historical data, and the historical samples comprise historical sample identities and historical sample feature data.

[0173] The computing module 104 is configured to calculate a target historical sample by the similarity training model according to the to-be-continued sample and the plurality of historical samples, wherein the target historical sample is the sample with the highest similarity to the to-be-continued identity among the historical samples.

[0174] ​The mapping module 105 is configured to map the to-be-continued identity to the target historical sample in the historical data to realize identity continuation.

[0175] In an embodiment, the device further comprises:

[0176] The encoding module 106 is configured to encode the training sample feature data of two samples in the pair of training samples into first fixed-dimension vectors respectively.

[0177] The splicing module 107 is configured to splice the two first fixed-dimension vectors of the same pair of training samples according to the determined similarity through deep learning to obtain a similarity neural network.

[0178] The splicing module 107 is further configured to splice the similarity neural networks of different pairs of training samples to obtain the similarity training model.

[0179] In an embodiment, the splicing module 107 is further configured to splice the similarity neural networks of different pairs of training samples to obtain a to-be-detected similarity training model.

[0180] The device further comprises:

[0181] The substitution module 108 is configured to substitute the pair of training samples into the to-be-detected similarity training model to calculate a detected similarity.

[0182] The calculation module 104 is further configured to calculate a difference value as an error value according to the similarity of the pair of training samples that has been determined and the detected similarity.

[0183] The judgment module 109 is configured to judge whether the error value is less than a predetermined error threshold value, and if the error value is less than the predetermined error threshold value, the to-be-detected similarity training model is taken as the similarity training model.

[0184] In an embodiment, the judgment module 109 is further configured to, if the error value is greater than or equal to the predetermined error threshold value, adjust parameters in the to-be-detected similarity training model to make the error value less than the predetermined error threshold value, and then take the to-be-detected similarity training model as the similarity training model.

[0185] In an embodiment, the encoding module 106 is configured to encode the continuation feature data and the historical sample feature data into second fixed-dimension vectors.

[0186] The calculation module 104 is configured to obtain a target historical sample through the similarity training model from the second fixed-dimension vectors, wherein the target historical sample is a sample in the historical samples that has the highest similarity with the to-be-continued identity.

[0187] In an embodiment, the apparatus further comprises:

[0188] a cleaning module 110, configured to clean the training sample feature data in the pair of training samples to obtain the data-type and category-type training sample feature data.

[0189] In an embodiment, the judging module 109 is configured to judge whether the historical sample feature data of a historical sample in a data source corresponds to the continuation feature data, and if the historical sample feature data corresponds to the continuation feature data, perform the subsequent step.

[0190] In an embodiment, the present application further provides an electronic device, and the apparatus comprises a processor and a memory.

[0191] The memory stores an application executable by the processor, so as to enable the processor to perform the steps of the identification continuation method.

[0192] In an embodiment, the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the steps of the identification continuation method.

[0193] In actual application, the computer readable medium can be included in the device / apparatus / system described in the above embodiments, or can exist independently without being assembled into the device / apparatus / system. The computer readable storage medium carries one or more programs, and when the one or more programs are executed, the image data processing method of the data is implemented.

[0194] According to the embodiments disclosed in the present application, the computer readable storage medium can be a non-volatile computer readable storage medium, for example, can include but is not limited to: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any appropriate combination thereof, but is not used to limit the scope of protection of the present application. In the embodiments disclosed in the present application, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, apparatus or device.

[0195] The above only describes the preferred embodiments of the present application and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of protection of the present application.

[0196] The preferred embodiments of the present application are merely illustrated, but not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application. < / mac1> < / imei1>

Claims

1. An identification method for prolonging life, characterized by, The method comprises: obtaining a rated number of paired training samples, the paired training samples being two training samples with a similarity determined, and the training samples comprising a training sample identifier and training sample feature data corresponding to the training sample identifier, the similarity indicating that the two samples are directed to the same user, the training sample identifier being a user identifier, and the training sample feature data being device attribute data or user behavior data; training a similarity training model from the paired training samples through deep learning, wherein the similarity training model is a model for calculating similarity; receiving a sample to be continued, the sample to be continued comprising a sample to be continued identifier and a continued feature data, the sample to be continued identifier being a user identifier, and the continued feature data being device attribute data or user behavior data; obtaining a plurality of historical samples of different sources in historical data according to the sample to be continued, the historical samples comprising a historical sample identifier and historical sample feature data, the historical sample identifier being a user identifier, and the historical sample being a sample corresponding to the continued feature data and the historical sample feature data; calculating a target historical sample from the sample to be continued and the plurality of historical samples through the similarity training model, the target historical sample being a sample with the highest similarity to the sample to be continued identifier among the historical samples; mapping the sample to be continued identifier to the target historical sample in the historical data to realize identification continuation of the sample to be continued identifier; Before the step of obtaining a plurality of historical samples of different sources in historical data according to the sample to be continued, the method comprises: judging whether the historical sample feature data of the historical sample in the data source corresponds to the continued feature data, i.e., judging whether the historical sample feature data of the historical sample in the data source is the same type of feature data as the continued feature data, and if the historical sample feature data corresponds to the continued feature data, performing the subsequent steps; The similarity training model is obtained from the paired training samples through deep learning, comprising: encoding the training sample feature data of two samples in the paired training samples into first fixed dimension vectors respectively; splicing the two first fixed dimension vectors of the same paired training sample through deep learning according to the determined similarity to obtain a similarity neural network; and obtaining the similarity training model by full network splicing of the similarity neural networks of different paired training samples.

2. The method of claim 1, wherein, The similarity training model obtained by full network splicing of the similarity neural networks of different paired training samples comprises: obtaining a to-be-detected similarity training model by full network splicing of the similarity neural networks of different paired training samples; calculating a detection similarity by substituting the paired training samples into the to-be-detected similarity training model; calculating a difference value as an error value according to the similarity determined for the paired training samples and the detection similarity; judging whether the error value is less than a predetermined error threshold, if the error value is less than the predetermined error threshold, the to-be-detected similarity training model is taken as the similarity training model.

3. The method of claim 2, wherein, The method further comprises: if the error value is greater than or equal to the predetermined error threshold, adjusting the parameters in the to-be-detected similarity training model to make the error value less than the predetermined error threshold, and then taking the to-be-detected similarity training model as the similarity training model.

4. The method of claim 1, wherein, The target historical sample is calculated from the to-be-continued sample and the plurality of historical samples by using the similarity training model, wherein the target historical sample is the sample with the highest similarity to the to-be-continued sample in the historical samples. The to-be-continued sample feature data and the historical sample feature data are encoded into a second fixed-dimension vector. The target historical sample is calculated from the to-be-continued sample and the plurality of historical samples by using the similarity training model, wherein the target historical sample is the sample with the highest similarity to the to-be-continued sample in the historical samples.

5. The method of claim 1, wherein, The method further comprises: The training sample feature data in the paired training samples is cleaned to obtain data-type and category-type training sample feature data.

6. An identification device for a life-extending apparatus, characterized by The identification device comprises: The identification device comprises: The training module is configured to train a similarity training model from the paired training samples by using deep learning, wherein the similarity training model is a model for calculating similarity. The receiving module is configured to receive a to-be-continued sample, wherein the to-be-continued sample comprises a to-be-continued sample identifier and to-be-continued sample feature data, the to-be-continued sample identifier is a user identifier, and the to-be-continued sample feature data is device attribute data or user behavior data. The obtaining module is configured to obtain a plurality of historical samples of different sources from historical data according to the to-be-continued sample, wherein the historical samples comprise historical sample identifiers and historical sample feature data, the historical sample identifiers are user identifiers, and the historical samples are samples corresponding to the to-be-continued sample feature data. The computing module is configured to calculate a target historical sample from the to-be-continued sample and the plurality of historical samples by using the similarity training model, wherein the target historical sample is the sample with the highest similarity to the to-be-continued sample in the historical samples. The mapping module is configured to map the to-be-continued sample identifier to the target historical sample in the historical data to realize identification and continuation of the to-be-continued sample identifier. The judging module is configured to judge whether the historical sample feature data of the historical sample in the data source corresponds to the continuation feature data, and if the historical sample feature data corresponds to the continuation feature data, that is, whether the historical sample feature data of the historical sample in the data source is the same type of feature data as the continuation feature data, the calculating module calculates the target historical sample according to the to-be-continued sample and the plurality of historical samples by using the similarity training model. The similarity training model is obtained by deep learning according to the pair training samples, including: encoding the training sample feature data of two samples in the pair training sample into a first fixed dimension vector respectively; obtaining a similarity neural network by deep learning according to the two first fixed dimension vectors of the same pair training sample according to the determined similarity; and obtaining the similarity training model by full network splicing of the similarity neural networks of different pair training samples.

7. An electronic device, comprising: The electronic device comprises a processor and a memory; The memory stores an application program executable by the processor, so as to enable the processor to execute the steps of the method for identifying continuation according to any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the method for identifying continuation according to any one of claims 1 to 5.

Images