Data comparison method and system based on encrypted feature analysis

Abstract

This invention provides a data comparison method and system based on encrypted feature analysis, relating to the field of artificial intelligence technology. In this invention, first encrypted behavioral feature information is decrypted to output corresponding first-behavioral feature information; second encrypted behavioral feature information is also decrypted to output corresponding second-behavioral feature information; representative information mining is performed on the first-behavioral feature information to extract first-behavioral representative information; representative information mining is also performed on the second-behavioral feature information to extract second-behavioral representative information; and comparative analysis is performed on the first-behavioral representative information and the second-behavioral representative information to output corresponding target comparative analysis results. Based on the above, the reliability of data comparison analysis can be improved to a certain extent.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a data comparison method and system based on encrypted feature analysis. Background Technology

[0002] Comparative analysis of different user behavior characteristic data to determine the correlations or differences between different users in behavioral dimensions has applications in many fields. Therefore, there is a need to provide a reliable solution for comparative analysis of different user behavior characteristic data. However, in existing technologies, similarity calculations or difference analyses are generally performed directly on the behavioral characteristic information representing user behavior characteristics, which leads to problems with the reliability of data comparison analysis. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a data comparison method and system based on encrypted feature analysis, so as to improve the reliability of data comparison analysis to a certain extent.

[0004] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0005] A data comparison method based on cryptographic feature analysis includes:

[0006] The system obtains first encrypted behavioral feature information corresponding to a first comparison user and second encrypted behavioral feature information corresponding to a second comparison user. It also decrypts the first encrypted behavioral feature information to output corresponding first behavioral feature information and decrypts the second encrypted behavioral feature information to output corresponding second behavioral feature information. The first behavioral feature information reflects the current behavioral information of the first comparison user and the second behavioral feature information reflects the current behavioral information of the second comparison user. The data formats of the first behavioral feature information and the second behavioral feature information are consistent, both belonging to image data or voice data.

[0007] The first behavioral feature information is processed by mining representative information to extract the first behavioral representative information corresponding to the first behavioral feature information. The first behavioral representative information is used to summarize the current behavioral information of the first comparison user.

[0008] The second behavioral feature information is processed by mining representative information to extract the second behavioral representative information corresponding to the second behavioral feature information. The second behavioral representative information is used to summarize the current behavioral information of the second comparison user.

[0009] The first behavioral representative information and the second behavioral representative information are compared and analyzed to output the corresponding target comparison analysis results. The target comparison analysis results are used to reflect the correlation or difference between the first comparison user and the second comparison user in the behavioral dimension.

[0010] In some preferred embodiments, in the above-described data comparison method based on encrypted feature analysis, the step of mining representative information from the first behavioral feature information to extract the first behavioral representative information corresponding to the first behavioral feature information includes:

[0011] Determine the historical behavioral feature information corresponding to the first behavioral feature information, wherein the historical behavioral feature information is used to reflect the historical behavioral information of the first comparison user;

[0012] The first behavioral feature information is subjected to feature mining processing to mine the current behavioral feature representation corresponding to the first behavioral feature information, and the historical behavioral feature information is subjected to feature mining processing to mine the historical behavioral feature representation corresponding to the historical behavioral feature information.

[0013] Based on the current behavior feature representation and the historical behavior feature representation, the matching relationship between each current feature representation parameter included in the current behavior feature representation and each historical feature representation parameter included in the historical behavior feature representation is analyzed;

[0014] Based on the matching relationship between each current feature representation parameter and each historical feature representation parameter, the system outputs a current semantic feature representation that reflects the current behavioral feature representation based on the historical behavioral feature representation, and outputs a historical semantic feature representation that reflects the historical behavioral feature representation based on the current behavioral feature representation.

[0015] Based on the current semantic feature representation and the historical semantic feature representation, the matching feature representation between the current behavior feature representation and the historical behavior feature representation is analyzed;

[0016] Based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation, the first behavior representative information corresponding to the first behavior feature information is analyzed.

[0017] In some preferred embodiments, in the above-described data comparison method based on encrypted feature analysis, the steps of performing feature mining processing on the first behavioral feature information to mine the current behavioral feature representation corresponding to the first behavioral feature information, and performing feature mining processing on the historical behavioral feature information to mine the historical behavioral feature representation corresponding to the historical behavioral feature information, include:

[0018] Using a first feature mining network, feature mining is performed on each behavior space mapping result corresponding to the first behavior feature information to output the current behavior feature representation corresponding to the first behavior feature information. The behavior space mapping result is formed by mapping a current user behavior in the first behavior feature information to the feature space.

[0019] Using a second feature mining network, feature mining is performed on each behavior space mapping result corresponding to the historical behavior feature information to output the historical behavior feature representation corresponding to the historical behavior feature information. The behavior space mapping result is formed by mapping a historical user behavior in the historical behavior feature information to the feature space.

[0020] In some preferred embodiments, in the above-described data comparison method based on encrypted feature analysis, the steps of outputting a current semantic feature representation that reflects the current behavioral feature representation based on the matching relationship between each current feature representation parameter and each historical feature representation parameter, and outputting a historical semantic feature representation that reflects the historical behavioral feature representation based on the current behavioral feature representation, include:

[0021] Based on the matching relationship between each current feature representation parameter and each historical feature representation parameter, the influence evaluation parameter of each historical feature representation parameter relative to each current feature representation parameter is analyzed, and the influence evaluation parameter of each current feature representation parameter relative to each historical feature representation parameter is also analyzed.

[0022] Based on each of the historical feature representation parameters and the influence evaluation parameter of each of the historical feature representation parameters relative to each of the current feature representation parameters, the corresponding current semantic feature representation is analyzed and output;

[0023] Based on each current feature representation parameter and the influence evaluation parameter of each current feature representation parameter relative to each historical feature representation parameter, the corresponding historical semantic feature representation is analyzed and output.

[0024] In some preferred embodiments, in the above-described data comparison method based on encrypted feature analysis, the step of analyzing the influence assessment parameter of each historical feature representation parameter relative to each current feature representation parameter based on the matching relationship between each current feature representation parameter and each historical feature representation parameter includes:

[0025] The first historical feature representation parameter is any historical feature representation parameter, and the first current feature representation parameter is any current feature representation parameter. The calculation process for the influence assessment parameter of the first historical feature representation parameter relative to the first current feature representation parameter includes:

[0026] An exponential operation is performed on the matching value represented by the matching relationship between the first current feature representation parameter and the first historical feature representation parameter to output a first exponential value;

[0027] For each current feature representation parameter, the matching value represented by the matching relationship between the current feature representation parameter and the first historical feature representation parameter is subjected to an exponential operation to output the second exponential value corresponding to the current feature representation parameter;

[0028] The second index value corresponding to each current feature representation parameter is superimposed to output a total second index value. Based on the first index value and the total second index value, an influence assessment parameter of the first historical feature representation parameter relative to the first current feature representation parameter is determined. The influence assessment parameter and the first index value have a positive correlation, and the influence assessment parameter and the total second index value have a negative correlation.

[0029] In some preferred embodiments, in the above-described data comparison method based on encrypted feature analysis, the step of analyzing the first behavior representative information corresponding to the first behavior feature information based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation includes:

[0030] Based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation, the target representative feature representation corresponding to the first behavior feature information and the historical behavior feature information is fused and output;

[0031] Based on the target representative feature representation, the first behavioral representative information corresponding to the first behavioral feature information is analyzed.

[0032] In some preferred embodiments, in the above-described data comparison method based on encrypted feature analysis, the step of fusing and outputting the target representative feature representation corresponding to the first behavior feature information and the historical behavior feature information based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation includes:

[0033] The current behavior feature representation and the current semantic feature representation are aggregated to output the corresponding current dimension aggregated feature representation. The historical behavior feature representation and the historical semantic feature representation are aggregated to output the corresponding historical dimension aggregated feature representation. The matching feature representation includes the current semantic feature representation and the historical semantic feature representation.

[0034] The current dimension aggregated feature representation and the historical dimension aggregated feature representation are concatenated to output the target representative feature representation corresponding to the first behavioral feature information and the historical behavioral feature information.

[0035] In some preferred embodiments, in the above-described data comparison method based on encrypted feature analysis, the steps of performing feature representation aggregation operations on the current behavioral feature representation and the current semantic feature representation to output the corresponding current dimension aggregated feature representation, and performing feature representation aggregation operations on the historical behavioral feature representation and the historical semantic feature representation to output the corresponding historical dimension aggregated feature representation, include:

[0036] The current behavioral feature representation and the current semantic feature representation are concatenated to output the corresponding current dimension concatenated feature representation;

[0037] Based on a predetermined first weighting parameter distribution, the current dimension concatenated feature representation is weighted to output the corresponding current dimension weighted feature representation. Based on a predetermined first bias parameter distribution, the current dimension weighted feature representation is biased to output the corresponding current dimension biased feature representation. Based on the current dimension biased feature representation, the corresponding current dimension aggregated feature representation is determined.

[0038] The historical behavioral feature representation and the historical semantic feature representation are concatenated to output the corresponding historical dimension concatenated feature representation;

[0039] Based on a predetermined second weighting parameter distribution, the historical dimension cascaded feature representation is weighted to output a corresponding historical dimension weighted feature representation. Based on a predetermined second bias parameter distribution, the historical dimension weighted feature representation is biased to output a corresponding historical dimension biased feature representation. Based on the historical dimension biased feature representation, the corresponding historical dimension aggregated feature representation is determined.

[0040] In some preferred embodiments, in the above-described data comparison method based on encrypted feature analysis, the step of analyzing the first behavioral representative information corresponding to the first behavioral feature information based on the target representative feature representation includes:

[0041] The target representative feature representation is subjected to a representative information prediction operation to predict the probability parameters of each behavioral representative segment; and, based on the probability parameters of each behavioral representative segment, at least one behavioral representative segment is determined to form the first behavioral representative information corresponding to the first behavioral feature information.

[0042] The step of performing a prediction operation on the target representative feature representation to predict the probability parameters of each behavioral representative segment includes:

[0043] Using a feature prediction network, the target representative feature representation is subjected to feature restoration to output the corresponding restored feature representation. The matching relationship between each feature representation included in the target representative feature representation and each restored feature representation parameter is analyzed. Based on the matching relationship between each feature representation included in the target representative feature representation and each restored feature representation parameter, the influence evaluation parameter of each feature representation included in the target representative feature representation relative to each restored feature representation parameter is analyzed. Based on the influence evaluation parameter of each feature representation included in the target representative feature representation and each restored feature representation parameter, the target representative semantic feature representation is determined. Based on the restored feature representation and the target representative semantic feature representation, the probability parameters corresponding to each behavioral representative fragment are determined.

[0044] This invention also provides a data comparison system based on encrypted feature analysis, including a processor and a memory. The memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-described data comparison method based on encrypted feature analysis.

[0045] The data comparison method and system based on encrypted feature analysis provided in this invention can first decrypt the first encrypted behavioral feature information to output corresponding first behavioral feature information, and then decrypt the second encrypted behavioral feature information to output corresponding second behavioral feature information; perform representative information mining on the first behavioral feature information to extract first behavioral representative information corresponding to the first behavioral feature information; perform representative information mining on the second behavioral feature information to extract second behavioral representative information corresponding to the second behavioral feature information; and perform information comparison analysis on the first and second behavioral representative information to output corresponding target comparison analysis results. Based on this, since representative information mining is performed on the first and second behavioral feature information respectively before information comparison analysis, the extracted behavioral representative information can be used for information comparison analysis. This can reduce the interference caused by invalid information in the original behavioral feature information during the information comparison analysis process to a certain extent, thereby improving the reliability of data comparison analysis to a certain extent, and thus improving the problem of poor reliability in existing technologies regarding data comparison analysis.

[0046] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0047] Figure 1 This is a structural block diagram of a data comparison system based on encrypted feature analysis provided in an embodiment of the present invention.

[0048] Figure 2 This is a flowchart illustrating the steps of the data comparison method based on encryption feature analysis provided in this embodiment of the invention.

[0049] Figure 3 This is a schematic diagram of the modules included in the data comparison device based on encryption feature analysis provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0052] like Figure 1 As shown, this embodiment of the invention provides a data comparison system based on encrypted feature analysis. The data comparison system based on encrypted feature analysis may include a memory and a processor, and may also include other devices such as a communication unit.

[0053] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, they can be electrically connected via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that exists in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby implementing the data comparison method based on encryption feature analysis provided in this embodiment of the invention.

[0054] It is understood that, in some feasible implementations, the memory may be, but is not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Programmable Read-Only Memory (PROM), or Erasable Read-Only Memory (Erasable Memory).

[0055] The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system-on-a-chip (SoC), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0056] It is understood that, in some feasible implementations, the data comparison system based on encrypted feature analysis can be a server with data processing capabilities.

[0057] Combination Figure 2 This invention also provides a data comparison method based on encrypted feature analysis, which can be applied to the aforementioned data comparison system based on encrypted feature analysis. The method steps defined in the process of the data comparison method based on encrypted feature analysis can be implemented by the data comparison system based on encrypted feature analysis.

[0058] The following will be about Figure 2 The specific process shown will be explained in detail.

[0059] Step S110: Obtain the first encrypted behavior feature information corresponding to the first comparison user, obtain the second encrypted behavior feature information corresponding to the second comparison user, and decrypt the first encrypted behavior feature information to output the corresponding first behavior feature information, and decrypt the second encrypted behavior feature information to output the corresponding second behavior feature information.

[0060] In this embodiment of the invention, the data comparison system based on encrypted feature analysis can obtain first encrypted behavioral feature information corresponding to a first comparison user and second encrypted behavioral feature information corresponding to a second comparison user. It also decrypts the first encrypted behavioral feature information to output corresponding first behavioral feature information and decrypts the second encrypted behavioral feature information to output corresponding second behavioral feature information. The first behavioral feature information reflects the current behavioral information of the first comparison user, and the second behavioral feature information reflects the current behavioral information of the second comparison user. The first and second behavioral feature information have the same data format, both being image data or voice data. For example, the behavior of the first and second comparison users can be captured as images to form corresponding image data, or other users can provide voice descriptions of the behavior of the first and second comparison users to form corresponding voice data.

[0061] Step S120: Perform representative information mining processing on the first behavioral feature information to mine the first behavioral representative information corresponding to the first behavioral feature information.

[0062] In this embodiment of the invention, the data comparison system based on encrypted feature analysis can perform representative information mining processing on the first behavioral feature information to mine the first behavioral representative information corresponding to the first behavioral feature information. The first behavioral representative information is used to summarize the current behavioral information of the first comparison user, that is, to provide a general description of the behavioral information.

[0063] Step S130: Perform representative information mining processing on the second behavioral feature information to mine the second behavioral representative information corresponding to the second behavioral feature information.

[0064] In this embodiment of the invention, the data comparison system based on encrypted feature analysis can perform representative information mining processing on the second behavioral feature information to mine the second behavioral representative information corresponding to the second behavioral feature information. The second behavioral representative information is used to summarize the current behavioral information of the second comparison user, that is, to provide a general description of the behavioral information. Furthermore, the representative information mining processing of the second behavioral feature information can be performed in the same way as the representative information mining processing of the first behavioral feature information.

[0065] Step S140: Perform information comparison analysis on the first row of representative information and the second row of representative information to output the corresponding target comparison analysis results.

[0066] In this embodiment of the invention, the data comparison system based on encrypted feature analysis can perform information comparison analysis on the first behavioral representative information and the second behavioral representative information to output corresponding target comparison analysis results. The target comparison analysis results are used to reflect the correlation or difference between the first and second compared users in the behavioral dimension. For example, information similarity calculations can be performed on the first and second behavioral representative information. Furthermore, since the first and second behavioral representative information have different data formats, different calculation methods can be used. Existing technologies for calculating data similarity in related forms can be referenced, without specific limitations here. Alternatively, feature space mapping can be performed on the first and second behavioral representative information respectively, allowing similarity calculations to be performed on the obtained mapped feature representations, such as cosine similarity calculations.

[0067] Based on this, since the first and second behavioral feature information are processed by mining representative information before information comparison analysis, the mined behavioral representative information can be compared and analyzed during the information comparison analysis. This can reduce the interference caused by invalid information in the original behavioral feature information during the information comparison analysis process, thereby improving the reliability of data comparison analysis to a certain extent and thus improving the problem of poor reliability of data comparison analysis in the existing technology.

[0068] It is understood that, in some feasible implementations, step S120 described above may further include the following:

[0069] Determine the historical behavioral feature information corresponding to the first behavioral feature information, wherein the historical behavioral feature information is used to reflect the historical behavioral information of the first comparison user;

[0070] The first behavioral feature information is subjected to feature mining processing to mine the current behavioral feature representation corresponding to the first behavioral feature information, and the historical behavioral feature information is subjected to feature mining processing to mine the historical behavioral feature representation corresponding to the historical behavioral feature information.

[0071] Based on the current behavior feature representation and the historical behavior feature representation, the matching relationship between each current feature representation parameter included in the current behavior feature representation and each historical feature representation parameter included in the historical behavior feature representation is analyzed; for example, the current feature representation parameter and the historical feature representation parameter can be multiplied to output the matching relationship between the current feature representation parameter and the historical feature representation parameter.

[0072] Based on the matching relationship between each current feature representation parameter and each historical feature representation parameter, the system outputs a current semantic feature representation that reflects the current behavioral feature representation based on the historical behavioral feature representation, and outputs a historical semantic feature representation that reflects the historical behavioral feature representation based on the current behavioral feature representation.

[0073] Based on the current semantic feature representation and the historical semantic feature representation, the matching feature representation between the current behavior feature representation and the historical behavior feature representation is analyzed. For example, the current semantic feature representation and the historical semantic feature representation can be directly used as the matching feature representation between the current behavior feature representation and the historical behavior feature representation. Alternatively, feature representation parameters can be filtered on the matching feature representation between the current behavior feature representation and the historical behavior feature representation, such as through convolution or filtering, to obtain the matching feature representation between the current behavior feature representation and the historical behavior feature representation.

[0074] Based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation, the first behavior representative information corresponding to the first behavior feature information is analyzed.

[0075] It is understood that, in some feasible implementations, the steps of performing feature mining processing on the first behavioral feature information to mine the current behavioral feature representation corresponding to the first behavioral feature information, and performing feature mining processing on the historical behavioral feature information to mine the historical behavioral feature representation corresponding to the historical behavioral feature information, may further include the following:

[0076] Using a first feature mining network, feature mining is performed on each behavior space mapping result corresponding to the first behavior feature information to output the current behavior feature representation corresponding to the first behavior feature information. This behavior space mapping result is formed by mapping a current user behavior in the first behavior feature information to the feature space. For example, the first behavior feature information may include multiple current user behaviors. Each current user behavior can be pre-mapped to the feature space to output the corresponding behavior space mapping result. Then, the first feature mining network can be used to perform feature mining on each behavior space mapping result corresponding to the first behavior feature information to output the current behavior feature representation corresponding to the first behavior feature information. The first feature mining network can be a neural network. In addition, a current user behavior can correspond to one or more consecutive image frames or one or more consecutive audio frames.

[0077] Using a second feature mining network, feature mining is performed on each behavior space mapping result corresponding to the historical behavior feature information to output a historical behavior feature representation corresponding to the historical behavior feature information. This behavior space mapping result is formed by mapping a historical user behavior in the historical behavior feature information to the feature space. For example, the historical behavior feature information may include multiple historical user behaviors. Each historical user behavior can be pre-mapped into the feature space to output a corresponding behavior space mapping result. Then, the second feature mining network can be used to perform feature mining on each behavior space mapping result corresponding to the historical behavior feature information to output a historical behavior feature representation corresponding to the historical behavior feature information. The second feature mining network can be a neural network. In addition, a historical user behavior may correspond to one or more consecutive image frames or one or more consecutive audio frames.

[0078] It is understood that, in some feasible implementations, the step of using a first feature mining network to perform feature mining on each behavior space mapping result corresponding to the first behavior feature information to output the current behavior feature representation corresponding to the first behavior feature information may further include the following:

[0079] The first behavioral feature information includes multiple current user behaviors, and based on the behavior time corresponding to each current user behavior, the temporal relationship between the multiple current user behaviors is determined. The multiple current user behaviors included in the first behavioral feature information can be a combination of behaviors, that is, multiple consecutive user behaviors.

[0080] The first behavioral feature information is loaded into a user behavior relationship analysis network formed through network optimization. This network is then used to perform relationship analysis on the first behavioral feature information, outputting the logical sequence of multiple current user behaviors included in the first behavioral feature information. The network optimization data of the user behavior relationship analysis network includes multiple typical behavioral feature information and the actual logical sequence of behaviors corresponding to each typical behavioral feature information. This actual logical sequence of behaviors can be formed based on the configuration of the corresponding user, or determined based on the actual temporal sequence of various behaviors existing in multiple behavior databases. For each of the multiple current user behaviors, based on the temporal sequence of behaviors, the first adjacent user behavior (e.g., the previous and next) is determined among other current user behaviors. Furthermore, based on the logical sequence of behaviors, the second adjacent user behavior is determined among other current user behaviors.

[0081] For each of the multiple current user behaviors, the mean of the behavior space mapping results corresponding to each first adjacent user behavior and each second adjacent user behavior is calculated to output the relevant behavior space mapping result corresponding to the current user behavior. The relevant behavior space mapping result is then transposed to output the corresponding transposed behavior space mapping result. The device behavior space mapping result and the behavior space mapping result corresponding to the current user behavior are then multiplied, and the multiplication result is then weighted. The weighting coefficient of the weighting process is negatively correlated with the dimension of the behavior space mapping result to output the corresponding weighted behavior space mapping result. Based on this, after the incentive mapping process is performed on the weighted behavior space mapping result, the relevant behavior space mapping result and the obtained incentive mapping process result are multiplied together. In this way, the current behavior sub-feature representation corresponding to the current user behavior can be obtained.

[0082] The current behavior sub-feature representations corresponding to each of the multiple current user behaviors can be concatenated according to the chronological order of the behaviors, and the current behavior feature representation corresponding to the first behavior feature information can be output.

[0083] It is understood that, in some feasible implementations, the steps of outputting a current semantic feature representation that reflects the current behavioral feature representation based on the matching relationship between each current feature representation parameter and each historical feature representation parameter, and outputting a historical semantic feature representation that reflects the historical behavioral feature representation based on the current behavioral feature representation, may further include the following:

[0084] Based on the matching relationship between each current feature representation parameter and each historical feature representation parameter, the influence evaluation parameter of each historical feature representation parameter relative to each current feature representation parameter is analyzed, and the influence evaluation parameter of each current feature representation parameter relative to each historical feature representation parameter is also analyzed.

[0085] Based on each of the historical feature representation parameters and the influence evaluation parameter of each of the historical feature representation parameters relative to each of the current feature representation parameters, the corresponding current semantic feature representation is analyzed and output. For example, the current semantic feature representation is formed based on the historical behavioral feature representation reflecting the current behavioral feature representation, and therefore has some information related to the current behavioral feature representation in the historical behavioral feature representation.

[0086] Based on each current feature representation parameter and the influence evaluation parameter of each current feature representation parameter relative to each historical feature representation parameter, the corresponding historical semantic feature representation is analyzed and output. For example, the historical semantic feature representation is formed based on the current behavioral feature representation reflecting the historical behavioral feature representation, and therefore has some related information in the current behavioral feature representation that is related to the historical behavioral feature representation.

[0087] It is understood that, in some feasible implementations, the step of analyzing the influence evaluation parameter of each historical feature representation parameter relative to each current feature representation parameter based on the matching relationship between each current feature representation parameter and each historical feature representation parameter may further include the following:

[0088] A first historical feature representation parameter and a first current feature representation parameter are determined. The first historical feature representation parameter can be any historical feature representation parameter, and the first current feature representation parameter can be any current feature representation parameter. Based on this, the calculation process of the influence assessment parameter of the first historical feature representation parameter relative to the first current feature representation parameter includes:

[0089] An exponential operation is performed on the matching value represented by the matching relationship between the first current feature representation parameter and the first historical feature representation parameter to output a first exponential value;

[0090] For each current feature representation parameter, the matching value represented by the matching relationship between the current feature representation parameter and the first historical feature representation parameter is subjected to an exponential operation to output the second exponential value corresponding to the current feature representation parameter;

[0091] The second index value corresponding to each current feature representation parameter is superimposed to output a total second index value. Based on the first index value and the total second index value, an influence assessment parameter is determined relative to the first current feature representation parameter. The influence assessment parameter and the first index value have a positive correlation, and the influence assessment parameter and the total second index value have a negative correlation. For example, the ratio of the first index value and the total second index value can be calculated to obtain the corresponding influence assessment parameter.

[0092] It is understood that, in some feasible implementations, the step of analyzing and outputting the corresponding current semantic feature representation based on each of the historical feature representation parameters and the influence evaluation parameter of each of the historical feature representation parameters relative to each of the current feature representation parameters may further include the following:

[0093] For each current feature representation parameter, the current feature representation parameter is determined as the current feature representation parameter to be processed. For each historical feature representation parameter, the historical feature representation parameter is multiplied by the influence assessment parameter of the historical feature representation parameter relative to the current feature representation parameter to be processed, to output the multiplication calculation parameter corresponding to the historical feature representation parameter. Furthermore, the multiplication calculation parameters corresponding to each historical feature representation parameter are summed to obtain the current semantic feature representation parameter corresponding to the current feature representation parameter to be processed. Thus, the current semantic feature representation parameter corresponding to each current feature representation parameter can be obtained. Finally, the current semantic feature representation parameters corresponding to each current feature representation parameter are combined to form the corresponding current semantic feature representation. It is understood that, in some feasible implementations, the step of analyzing and outputting the corresponding historical semantic feature representation based on each current feature representation parameter and the influence assessment parameter of each current feature representation parameter relative to each historical feature representation parameter may further include the following:

[0094] For each historical feature representation parameter, the historical feature representation parameter is determined as the historical feature representation parameter to be processed. For each current feature representation parameter, the current feature representation parameter is multiplied by the influence assessment parameter of the current feature representation parameter relative to the historical feature representation parameter to be processed, so as to output the multiplication calculation parameter corresponding to the current feature representation parameter. In addition, the multiplication calculation parameters corresponding to each current feature representation parameter are summed to obtain the historical semantic feature representation parameter corresponding to the historical feature representation parameter to be processed. In this way, the historical semantic feature representation parameter corresponding to each historical feature representation parameter can be obtained. Finally, the historical semantic feature representation parameters corresponding to each historical feature representation parameter are combined together to form the corresponding historical semantic feature representation.

[0095] It is understood that, in some feasible implementations, the step of analyzing the first behavior representative information corresponding to the first behavior feature information based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation may further include the following:

[0096] Based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation, the target representative feature representation corresponding to the first behavior feature information and the historical behavior feature information is fused and output; and based on the target representative feature representation, the first behavior representative information corresponding to the first behavior feature information is analyzed.

[0097] It is understood that, in some feasible implementations, the step of fusing and outputting the target representative feature representation corresponding to the first behavior feature information and the historical behavior feature information based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation may further include the following:

[0098] The current behavioral feature representation and the current semantic feature representation are aggregated to output the corresponding current dimension aggregated feature representation. The historical behavioral feature representation and the historical semantic feature representation are aggregated to output the corresponding historical dimension aggregated feature representation. The matching feature representation includes the current semantic feature representation and the historical semantic feature representation. As mentioned above, the current semantic feature representation and the historical semantic feature representation can be directly used as the matching feature representation.

[0099] The current dimension aggregated feature representation and the historical dimension aggregated feature representation are concatenated to output the target representative feature representation corresponding to the first behavioral feature information and the historical behavioral feature information. For example, during the concatenation operation, the current dimension aggregated feature representation can be first and the historical dimension aggregated feature representation can be second. For example, the target representative feature representation can be "current dimension aggregated feature representation - historical dimension aggregated feature representation".

[0100] It is understood that, in some feasible implementations, the steps of performing feature representation aggregation operations on the current behavioral feature representation and the current semantic feature representation to output the corresponding current dimension aggregated feature representation, and performing feature representation aggregation operations on the historical behavioral feature representation and the historical semantic feature representation to output the corresponding historical dimension aggregated feature representation, may further include the following:

[0101] The current behavioral feature representation and the current semantic feature representation are concatenated, as described above, to output the corresponding concatenated feature representation of the current dimension;

[0102] Based on a predetermined first weighting parameter distribution, the current dimension cascaded feature representation is weighted, such as by multiplication, to output the corresponding current dimension weighted feature representation. Also, based on a predetermined first bias parameter distribution, the current dimension weighted feature representation is biased, such as by superposition, to output the corresponding current dimension biased feature representation. Based on the current dimension biased feature representation, the corresponding current dimension aggregated feature representation is determined. The first weighting parameter distribution and the first bias parameter distribution can be used as network parameters of the corresponding neural network, and are continuously updated during the optimization process of the neural network.

[0103] The historical behavioral feature representation and the historical semantic feature representation are concatenated as described above to output the corresponding historical dimension concatenated feature representation.

[0104] Based on a predetermined second weighting parameter distribution, the historical dimension cascaded feature representation is weighted to output a corresponding historical dimension weighted feature representation. Also, based on a predetermined second bias parameter distribution, the historical dimension weighted feature representation is biased to output a corresponding historical dimension biased feature representation. Finally, based on the historical dimension biased feature representation, a corresponding historical dimension aggregated feature representation is determined. The second weighting parameter distribution and the second bias parameter distribution can serve as network parameters for the corresponding neural network, and are continuously updated during the optimization process of the neural network.

[0105] It is understood that, in some feasible implementations, the step of determining the corresponding current dimension aggregated feature representation based on the current dimension bias feature representation may further include the following: performing nonlinear activation mapping processing on the current dimension bias feature representation to obtain the corresponding current dimension aggregated feature representation. For example, the nonlinear activation mapping processing can be implemented by a configured nonlinear activation mapping function.

[0106] It is understood that, in some feasible implementations, the step of determining the corresponding historical dimension aggregated feature representation based on the historical dimension bias feature representation may further include the following: performing nonlinear activation mapping processing on the historical dimension bias feature representation to obtain the corresponding historical dimension aggregated feature representation. For example, the nonlinear activation mapping processing can be implemented by a configured nonlinear activation mapping function.

[0107] It is understood that, in some feasible implementations, the step of analyzing the first behavioral representative information corresponding to the first behavioral feature information based on the target representative feature representation may further include the following:

[0108] The target representative feature representation is used to perform a predictive operation on the representative information to predict the probability parameters of each behavioral representative segment. For example, each behavioral representative segment can be a data label of the sample data that the corresponding neural network relies on during the network optimization process. The sample data can be sample behavioral feature information, and the data label can be sample behavioral representative information.

[0109] Based on the probability parameters of each of the behavioral representative segments, at least one behavioral representative segment is determined to form the first behavioral representative information corresponding to the first behavioral feature information. For example, one or more behavioral representative segments with the largest corresponding probability parameters can be used as the first behavioral representative information corresponding to the first behavioral feature information.

[0110] It is understood that, in some feasible implementations, the step of performing a prediction operation on the target representative feature representation to predict the probability parameters of each behavioral representative segment may further include the following:

[0111] Using a feature prediction network, the target representative feature representation is subjected to feature restoration operation to output the corresponding restored feature representation. For example, the feature restoration operation may include upsampling.

[0112] The matching relationship between each feature representation included in the target representative feature representation and each restored feature representation parameter included in the restored feature representation is analyzed. Based on the matching relationship between each feature representation included in the target representative feature representation and each restored feature representation parameter, the influence evaluation parameter of each feature representation included in the target representative feature representation relative to each restored feature representation parameter is analyzed. Based on the influence evaluation parameter of each feature representation included in the target representative feature representation and each restored feature representation parameter, the target representative semantic feature representation is determined, as described above.

[0113] Based on the restored feature representation and the target representative semantic feature representation, the probability parameters corresponding to each behavioral representative segment are determined.

[0114] It is understood that, in some feasible implementations, the step of determining the probability parameters corresponding to each behavioral representative segment based on the restored feature representation and the target representative semantic feature representation may further include the following:

[0115] The restored feature representation and the target representative semantic feature representation are aggregated to output the corresponding target concatenated feature representation, as described above.

[0116] The target cascaded feature representation is mapped to feature representation parameters to output the probability parameters corresponding to each behavioral representative segment. For example, in the feature space, there can be a central feature representation corresponding to each behavioral representative segment. The central feature representation is determined based on the feature representation corresponding to the behavioral feature information of each sample corresponding to the behavioral representative segment. Then, the similarity between the target cascaded feature representation and the central feature representation corresponding to each behavioral representative segment can be calculated. In this way, the probability parameters corresponding to each behavioral representative segment can be obtained.

[0117] It is understood that, in some feasible implementations, step S130 described above may further include the following:

[0118] The process involves: determining the second historical behavioral feature information corresponding to the second behavioral feature information, which reflects the historical behavioral information of the second user being compared; performing feature mining on the second behavioral feature information to mine the second current behavioral feature representation corresponding to the second behavioral feature information, and performing feature mining on the second historical behavioral feature information to mine the second historical behavioral feature representation corresponding to the second historical behavioral feature information; analyzing the matching relationship between each current feature representation parameter included in the second current behavioral feature representation and each historical feature representation parameter included in the second historical behavioral feature representation based on the matching relationship between each current feature representation parameter and each historical feature representation parameter; outputting the second current semantic feature representation reflecting the second current behavioral feature representation based on the second historical behavioral feature representation, and outputting the second historical semantic feature representation reflecting the second historical behavioral feature representation based on the second current behavioral feature representation; analyzing the second matching feature representation between the second current behavioral feature representation and the second historical behavioral feature representation based on the second current semantic feature representation and the second historical semantic feature representation; and analyzing the second behavioral representative information corresponding to the second behavioral feature information based on the second current behavioral feature representation, the second historical behavioral feature representation, and the second matching feature representation.

[0119] Combination Figure 3 This invention also provides a data comparison device based on encryption feature analysis, which can be applied to the aforementioned data comparison system based on encryption feature analysis. The data comparison device based on encryption feature analysis may include:

[0120] The behavior feature information determination module is used to obtain first encrypted behavior feature information corresponding to a first comparison user and second encrypted behavior feature information corresponding to a second comparison user, and to decrypt the first encrypted behavior feature information to output the corresponding first behavior feature information, and to decrypt the second encrypted behavior feature information to output the corresponding second behavior feature information. The first behavior feature information is used to reflect the current behavior information of the first comparison user, and the second behavior feature information is used to reflect the current behavior information of the second comparison user. The data format of the first behavior feature information and the second behavior feature information is the same, both belonging to image data or voice data.

[0121] The first representative information mining module is used to perform representative information mining processing on the first behavioral feature information to mine the first behavioral representative information corresponding to the first behavioral feature information. The first behavioral representative information is used to summarize the current behavioral information of the first comparison user.

[0122] The second representative information mining module is used to perform representative information mining processing on the second behavioral feature information to mine the second behavioral representative information corresponding to the second behavioral feature information. The second behavioral representative information is used to summarize the current behavioral information of the second comparative user.

[0123] The information comparison and analysis module is used to perform information comparison and analysis on the first behavior representative information and the second behavior representative information to output the corresponding target comparison and analysis results. The target comparison and analysis results are used to reflect the correlation or difference between the first comparison user and the second comparison user in the behavioral dimension, such as the degree of correlation and the degree of difference.

[0124] In summary, the data comparison method and system based on encrypted feature analysis provided by this invention can first decrypt the first encrypted behavioral feature information to output the corresponding first behavioral feature information, and then decrypt the second encrypted behavioral feature information to output the corresponding second behavioral feature information; representative information mining processing is performed on the first behavioral feature information to extract the first behavioral representative information corresponding to the first behavioral feature information; representative information mining processing is also performed on the second behavioral feature information to extract the second behavioral representative information corresponding to the second behavioral feature information; and information comparison analysis is performed on the first and second behavioral representative information to output the corresponding target comparison analysis result. Based on this, since representative information mining processing is performed on the first and second behavioral feature information respectively before information comparison analysis, the extracted behavioral representative information can be used for information comparison analysis. This can, to a certain extent, reduce the interference caused by invalid information in the original behavioral feature information during the information comparison analysis process, thereby improving the reliability of data comparison analysis to a certain extent, and thus improving the problem of poor reliability in existing technologies regarding data comparison analysis.

[0125] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data comparison method based on encrypted feature analysis, characterized in that, include: The system obtains first encrypted behavioral feature information corresponding to a first comparison user and second encrypted behavioral feature information corresponding to a second comparison user. It then decrypts the first encrypted behavioral feature information to output corresponding first behavioral feature information and decrypts the second encrypted behavioral feature information to output corresponding second behavioral feature information. The first behavioral feature information reflects the current behavioral information of the first comparison user, and the second behavioral feature information reflects the current behavioral information of the second comparison user. The first and second behavioral feature information have the same data format, both being image data or voice data. The system also performs representative information mining on the first behavioral feature information to mine first behavioral representative information corresponding to the first behavioral feature information. This first behavioral representative information summarizes the current behavioral information of the first comparison user. The second behavioral feature information is processed by mining representative information to extract the second behavioral representative information corresponding to the second behavioral feature information. The second behavioral representative information is used to summarize the current behavioral information of the second comparison user. Information comparison analysis is performed on the first behavior representative information and the second behavior representative information to output the corresponding target comparison analysis results. The target comparison analysis results are used to reflect the correlation or difference between the first comparison user and the second comparison user in the behavioral dimension. The step of mining the representative information of the first behavioral feature information to mine the first behavioral representative information corresponding to the first behavioral feature information includes: determining the historical behavioral feature information corresponding to the first behavioral feature information, wherein the historical behavioral feature information is used to reflect the historical behavioral information of the first comparison user. The first behavioral feature information is subjected to feature mining processing to mine the current behavioral feature representation corresponding to the first behavioral feature information; and the historical behavioral feature information is subjected to feature mining processing to mine the historical behavioral feature representation corresponding to the historical behavioral feature information; based on the current behavioral feature representation and the historical behavioral feature representation, the matching relationship between each current feature representation parameter included in the current behavioral feature representation and each historical feature representation parameter included in the historical behavioral feature representation is analyzed; based on the matching relationship between each current feature representation parameter and each historical feature representation parameter, a current semantic feature representation reflecting the current behavioral feature representation based on the historical behavioral feature representation is output, and a historical semantic feature representation reflecting the historical behavioral feature representation based on the current behavioral feature representation is output; based on the current semantic feature representation and the historical semantic feature representation, a matching feature representation between the current behavioral feature representation and the historical behavioral feature representation is analyzed; based on the current behavioral feature representation, the historical behavioral feature representation, and the matching feature representation, the first behavioral representative information corresponding to the first behavioral feature information is analyzed.

2. The data comparison method based on encrypted feature analysis as described in claim 1, characterized in that, The steps of performing feature mining processing on the first behavioral feature information to mine the current behavioral feature representation corresponding to the first behavioral feature information, and performing feature mining processing on the historical behavioral feature information to mine the historical behavioral feature representation corresponding to the historical behavioral feature information, include: using a first feature mining network to perform feature mining on each behavioral space mapping result corresponding to the first behavioral feature information to output the current behavioral feature representation corresponding to the first behavioral feature information, wherein the behavioral space mapping result is formed by mapping a current user behavior in the first behavioral feature information to the feature space; and using a second feature mining network to perform feature mining on each behavioral space mapping result corresponding to the historical behavioral feature information to output the historical behavioral feature representation corresponding to the historical behavioral feature information, wherein the behavioral space mapping result is formed by mapping a historical user behavior in the historical behavioral feature information to the feature space.

3. The data comparison method based on encrypted feature analysis as described in claim 1, characterized in that, The steps of outputting a current semantic feature representation that reflects the current behavioral feature representation based on the historical behavioral feature representation, and outputting a historical semantic feature representation that reflects the historical behavioral feature representation based on the matching relationship between each current feature representation parameter and each historical feature representation parameter, include: analyzing and determining the influence assessment parameter of each historical feature representation parameter relative to each current feature representation parameter based on the matching relationship between each current feature representation parameter and each historical feature representation parameter; analyzing and outputting the corresponding current semantic feature representation based on the influence assessment parameter of each historical feature representation parameter relative to each current feature representation parameter; and analyzing and outputting the corresponding historical semantic feature representation based on the influence assessment parameter of each current feature representation parameter relative to each historical feature representation parameter.

4. The data comparison method based on encrypted feature analysis as described in claim 3, characterized in that, The step of analyzing the influence assessment parameter of each historical feature representation parameter relative to each current feature representation parameter based on the matching relationship between each current feature representation parameter and each historical feature representation parameter includes: the first historical feature representation parameter is any historical feature representation parameter, the first current feature representation parameter is any current feature representation parameter, and the calculation process of the influence assessment parameter of the first historical feature representation parameter relative to the first current feature representation parameter includes: performing an exponential operation on the matching value represented by the matching relationship between the first current feature representation parameter and the first historical feature representation parameter to output a first exponential value. For each current feature representation parameter, an exponential operation is performed on the matching value representing the matching relationship between the current feature representation parameter and the first historical feature representation parameter to output a second exponential value corresponding to the current feature representation parameter; the second exponential values ​​corresponding to each current feature representation parameter are superimposed to output a total second exponential value; and, based on the first exponential value and the total second exponential value, an influence assessment parameter relative to the first current feature representation parameter is determined, wherein the influence assessment parameter and the first exponential value have a positive correlation, and the influence assessment parameter and the total second exponential value have a negative correlation.

5. The data comparison method based on encrypted feature analysis as described in claim 1, characterized in that, The step of analyzing the first behavior representative information corresponding to the first behavior feature information based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation includes: fusing and outputting the target representative feature representation corresponding to the first behavior feature information and the historical behavior feature information based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation; and analyzing the first behavior representative information corresponding to the first behavior feature information based on the target representative feature representation.

6. The data comparison method based on encrypted feature analysis as described in claim 5, characterized in that, The step of fusing and outputting the target representative feature representation corresponding to the first behavior feature information and the historical behavior feature information based on the current behavior feature representation, the historical behavior feature representation, and the matching feature representation includes: performing a feature representation aggregation operation on the current behavior feature representation and the current semantic feature representation to output the corresponding current dimension aggregated feature representation; and performing a feature representation aggregation operation on the historical behavior feature representation and the historical semantic feature representation to output the corresponding historical dimension aggregated feature representation, wherein the matching feature representation includes the current semantic feature representation and the historical semantic feature representation; and performing a concatenation operation on the current dimension aggregated feature representation and the historical dimension aggregated feature representation to output the target representative feature representation corresponding to the first behavior feature information and the historical behavior feature information.

7. The data comparison method based on encrypted feature analysis as described in claim 6, characterized in that, The steps of performing feature representation aggregation operations on the current behavioral feature representation and the current semantic feature representation to output a corresponding current dimension aggregated feature representation, and performing feature representation aggregation operations on the historical behavioral feature representation and the historical semantic feature representation to output a corresponding historical dimension aggregated feature representation, include: performing a concatenation operation on the current behavioral feature representation and the current semantic feature representation to output a corresponding current dimension concatenated feature representation; performing weighted processing on the current dimension concatenated feature representation based on a predetermined first weighted parameter distribution to output a corresponding current dimension weighted feature representation; and performing weighted processing on the current dimension weighted feature representation based on a predetermined first bias parameter distribution. Bias processing is performed to output the corresponding current dimension bias feature representation, and the corresponding current dimension aggregate feature representation is determined based on the current dimension bias feature representation; the historical behavior feature representation and the historical semantic feature representation are concatenated to output the corresponding historical dimension concatenated feature representation; the historical dimension concatenated feature representation is weighted based on a predetermined second weighting parameter distribution to output the corresponding historical dimension weighted feature representation; and the historical dimension weighted feature representation is biased based on a predetermined second bias parameter distribution to output the corresponding historical dimension bias feature representation, and the corresponding historical dimension aggregate feature representation is determined based on the historical dimension bias feature representation.

8. The data comparison method based on encrypted feature analysis as described in claim 5, characterized in that, The step of analyzing the first behavioral representative information corresponding to the first behavioral feature information based on the target representative feature representation includes: performing a representative information prediction operation on the target representative feature representation to predict the probability parameters of each behavioral representative segment; and determining at least one behavioral representative segment based on the probability parameters of each behavioral representative segment to form the first behavioral representative information corresponding to the first behavioral feature information; wherein, the step of performing a representative information prediction operation on the target representative feature representation to predict the probability parameters of each behavioral representative segment includes: using a feature prediction network to perform a feature restoration operation on the target representative feature representation to output the corresponding restored feature representation; analyzing the target representative feature representation includes The matching relationship between each feature representation and each restored feature representation parameter included in the target representative feature representation is analyzed. Based on the matching relationship between each feature representation and each restored feature representation parameter included in the target representative feature representation, the influence evaluation parameter of each feature representation included in the target representative feature representation relative to each restored feature representation parameter is analyzed. Based on the influence evaluation parameter of each feature representation included in the target representative feature representation and each restored feature representation parameter included in the target representative feature representation, the target representative semantic feature representation is determined. Based on the restored feature representation and the target representative semantic feature representation, the probability parameter corresponding to each behavioral representative fragment is determined.

9. A data comparison system based on encrypted feature analysis, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, and the processor being used to execute the computer program to implement the method of any one of claims 1-8.

Images