Data de-sensitization method, apparatus, device, and medium

Abstract

The present disclosure provides a data de-sensitization method, which can be applied to the field of artificial intelligence technology and the field of big data technology. The method comprises: performing encoding processing on initial operation data to obtain an operation encoded data sequence, wherein the initial operation data comprises operation time series data; performing feature extraction on the operation encoded data sequence to obtain a first feature encoding information set, wherein the first feature encoding information in the first feature encoding information set comprises first feature query encoding information; and inputting the first feature encoding information set and the operation encoded data sequence into a target decoder, so that the target decoder processes the first feature query encoding information and the operation encoded data sequence based on an attention mechanism to obtain de-sensitized operation data associated with the initial operation data. The present disclosure also provides a data de-sensitization device, an electronic device, a computer readable storage medium and a computer program product.

Description

Technical Field

[0001] This disclosure relates to the fields of artificial intelligence technology and big data technology, and more specifically, to a data desensitization method, apparatus, electronic device, computer-readable storage medium, and computer program product. Background Technology

[0002] With the increasing demand for data security, data anonymization technology is widely used in fields such as healthcare, finance, and telecommunications to prevent data leakage and ensure data security. Data anonymization technology, also known as data de-identification, transforms or modifies original sensitive data according to preset anonymization strategies or rules. The resulting anonymized data removes sensitive information from the original data, thereby preventing leakage of sensitive information and ensuring the security of the original data.

[0003] In realizing the present invention, the inventors discovered at least the following problems in the related technology. After desensitizing the original data using related data desensitization methods, the desensitized data may lose important feature information from the original data while removing sensitive information. Summary of the Invention

[0004] In view of this, the present disclosure provides a data anonymization method, apparatus, electronic device, computer-readable storage medium, and computer program product.

[0005] One aspect of this disclosure provides a data anonymization method, comprising:

[0006] The initial operational data is encoded to obtain an operational coded data sequence, wherein the initial operational data includes operational time-series data;

[0007] Feature extraction is performed on the aforementioned operational coding data sequence to obtain a first feature coding information set, wherein the first feature coding information in the first feature coding information set includes first feature query coding information; and

[0008] The first feature encoding information set and the operation encoding data sequence are input into the target decoder so that the target decoder processes the first feature query encoding information and the operation encoding data sequence based on the attention mechanism to obtain de-identified operation data associated with the initial operation data.

[0009] According to embodiments of this disclosure, the target decoder includes a second position embedding layer, a first attention layer of the decoder, and a second attention layer of the decoder connected in sequence.

[0010] The first feature encoding information set and the operational encoding data sequence are input into the target decoder to obtain de-identified operational data associated with the initial operational data, including:

[0011] The operational code data in the above operational code data sequence is input into the second position embedding layer, and the second position code information is output.

[0012] The second position encoding information is processed using the first attention layer of the decoder to obtain the second feature encoding information associated with the above-mentioned operational encoding data;

[0013] Construct a second feature encoding information set based on the aforementioned second feature encoding information; and

[0014] The second attention layer of the decoder is used to process the second feature encoding information set and the first feature query encoding information set to obtain de-identified operational data associated with the initial operational data.

[0015] According to embodiments of this disclosure, the second feature encoding information set includes second feature encoding information, the second feature encoding information includes second feature query encoding information and second feature key encoding information associated with the second feature query encoding information, and the second attention layer of the decoder includes a second attention sublayer;

[0016] The second attention layer of the decoder processes the second feature encoding information set and the first feature encoding information set to obtain de-identified operational data associated with the initial operational data, including:

[0017] The second feature query encoding information and the second feature key encoding information are hashed respectively to obtain the second feature query hash value and the second feature key hash value.

[0018] Based on the similarity distance between the second feature query hash value and the second feature key hash value, the target second feature encoding information associated with the first feature query encoding information is determined.

[0019] The second attention sublayer, based on the local attention mechanism formed by the first feature query encoding information and the target second feature encoding information, processes the first feature query encoding information and the target second feature encoding information to obtain attention feature information;

[0020] The target prediction category is obtained by classifying and predicting the above attention feature information;

[0021] Based on the target prediction categories mentioned above, de-identified operational data associated with the initial operational data is determined.

[0022] According to embodiments of this disclosure, the target second feature encoding information associated with the first feature query encoding information is determined based on the similarity distance between the second feature query hash value and the second feature key hash value, including:

[0023] Calculate the similarity distance between the second feature query hash value and the second feature key hash value to determine candidate second feature key encoding information that is similar to the second feature query encoding information.

[0024] The aforementioned candidate second feature key encoding information is stored in the same hash bucket;

[0025] Use a hash table to determine the target hash bucket that is similar to the encoding information of the first feature query mentioned above;

[0026] Based on the preset attention rule, the target second feature encoding information associated with the first query feature encoding information is determined from the target hash bucket.

[0027] According to embodiments of this disclosure, feature extraction is performed on the above-mentioned operational coding data sequence to obtain a first feature coding information set, including:

[0028] The operational code data in the above operational code data sequence is input into the first position embedding layer, and the first position code information is output.

[0029] The encoder attention layer processes the first position encoding information to obtain first feature encoding information associated with the operational encoding data; and

[0030] Based on the aforementioned first feature encoding information, the aforementioned first feature encoding information set is constructed.

[0031] According to embodiments of this disclosure, the target decoder includes a second position embedding layer and a decoder first attention layer;

[0032] Feature extraction is performed on the above operational coding data sequence to obtain the first feature coding information set, which includes:

[0033] The operational code data in the above operational code data sequence is input into the second position embedding layer, and the first position code information is output.

[0034] The first position encoding information is processed using the first attention layer of the decoder to obtain the first feature encoding information associated with the above-mentioned operational encoding data.

[0035] According to embodiments of this disclosure, the initial operating data described above also includes descriptive operating data;

[0036] The initial operational data is encoded to obtain an operational coded data sequence, including:

[0037] The above operational descriptive data is input into the first coding model, and the first operational coding data is output.

[0038] The aforementioned operational time-series data is input into the second coding model, and the second operational coding data is output; and

[0039] According to the preset rules, the above-mentioned operation code data sequence is constructed based on the above-mentioned first operation code data and the above-mentioned second operation code data.

[0040] According to embodiments of this disclosure, constructing the operation code data sequence based on the first operation code data and the second operation code data according to preset rules includes:

[0041] Based on the target time attribute information, target first operation code data is determined from the first operation code data and target second operation code data is determined from the second operation code data, wherein the target first operation code data and the target second operation code data have the same target time attribute information.

[0042] Based on the aforementioned first target operational coding data and the aforementioned second target operational coding data, operational coding data is constructed; and

[0043] When the above-mentioned operation code data includes multiple data, the above-mentioned operation code data is spliced ​​together according to the preset splicing rules to obtain the above-mentioned operation code data sequence.

[0044] According to embodiments of this disclosure, the above-described data anonymization method further includes:

[0045] Data analysis was performed on the aforementioned anonymized operational data to obtain the data analysis results for the aforementioned initial operational data.

[0046] Another aspect of this disclosure provides a data anonymization apparatus, comprising:

[0047] The encoding module is used to encode the initial operational data to obtain an operational encoded data sequence, wherein the initial operational data includes operational time-series data.

[0048] The feature extraction module is used to extract features from the aforementioned operational coding data sequence to obtain a first feature coding information set, wherein the first feature coding information in the first feature coding information set includes first feature query coding information; and

[0049] The prediction module is used to input the first feature encoding information set and the operation encoding data sequence to the target decoder, so that the target decoder processes the first feature query encoding information and the operation encoding data sequence based on the attention mechanism to obtain de-identified operation data associated with the initial operation data.

[0050] Another aspect of this disclosure provides an electronic device comprising:

[0051] One or more processors;

[0052] Memory, used to store one or more programs.

[0053] When the one or more programs are executed by the one or more processors, the one or more processors implement the data desensitization method described above.

[0054] Another aspect of this disclosure provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the data desensitization method described above.

[0055] Another aspect of this disclosure provides a computer program product including computer-executable instructions that, when executed, implement the data desensitization method described above.

[0056] According to embodiments of this disclosure, by extracting features from the operational coding data sequence, the first feature coding information in the obtained first feature coding information set retains the temporal and order-of-magnitude features of the operational temporal data. The first feature coding information set and the operational coding data sequence are input into a target decoder. The first feature query coding information and the operational coding data sequence are processed according to the attention mechanism of the target decoder, thereby obtaining a prediction result associated with the initial operational data. This prediction result, as desensitized operational data, can hide the privacy information that needs to be desensitized in the initial operational data, while retaining the temporal and order-of-magnitude features of the operational temporal data. This makes the desensitized operational data have a similar distribution trend to the initial operational data, thus at least partially overcoming the technical problem in related technologies where desensitized data may lose important feature information from the initial data. Furthermore, relevant data analysis can be performed based on the desensitized operational data, making the data analysis results highly similar to the data analysis results obtained from analyzing the initial operational data. This achieves the goal of predicting the changing trend of the initial operational data using desensitized operational data while ensuring that privacy information in the initial operational data is not leaked, thereby improving the security and timeliness of data analysis. Attached Figure Description

[0057] The above and other objects, features and advantages of this disclosure will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0058] Figure 1 This illustration schematically shows an exemplary system architecture to which data desensitization methods and apparatus can be applied according to embodiments of the present disclosure;

[0059] Figure 2 A flowchart illustrating a data desensitization method according to an embodiment of the present disclosure is shown schematically.

[0060] Figure 3 This schematically illustrates a flowchart of encoding initial operational data to obtain an operational coded data sequence according to an embodiment of the present disclosure;

[0061] Figure 4 This illustration schematically depicts an application scenario diagram of extracting features from an operational coding data sequence to obtain a first feature coding information set according to an embodiment of the present disclosure.

[0062] Figure 5 This schematically illustrates a flowchart of inputting a first feature encoding information set and an operational encoding data sequence to a target decoder to obtain de-identified operational data associated with initial operational data, according to an embodiment of the present disclosure.

[0063] Figure 6 This illustration schematically depicts an application scenario in which a first feature encoding information set and an operational encoding data sequence are input into a target decoder to obtain de-identified operational data associated with the initial operational data, according to an embodiment of this disclosure.

[0064] Figure 7 This schematically illustrates a flowchart of inputting a first feature encoding information set and an operational encoding data sequence to a target decoder to obtain de-identified operational data associated with initial operational data, according to an embodiment of the present disclosure.

[0065] Figure 8 This diagram illustrates an application scenario of the data desensitization method according to embodiments of the present disclosure.

[0066] Figure 9 A block diagram of a data desensitization apparatus according to embodiments of the present disclosure is illustrated schematically; and

[0067] Figure 10 A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Detailed Implementation

[0068] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0069] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0070] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0071] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0072] With the increasing demand for data security, data anonymization technology is widely used in fields such as healthcare, finance, and telecommunications to prevent data leaks and ensure data security. Data anonymization methods typically involve replacing the original data, such as using asterisks (*) to replace parts of the original data, or replacing it with similar data, like replacing the postal code "20000" with "82222", thus hiding the true information in the original data. Alternatively, symmetric encryption algorithms or other encryption methods can be used to encrypt the original data, effectively hiding the true information in the resulting ciphertext.

[0073] However, the inventors discovered that data anonymization methods in related technologies may cause the anonymized data to lose important characteristic information from the original data. When the initial data contains time-series characteristics, such as company annual report data or order transaction data, the anonymized data obtained through these methods forcibly alters the format or structure of the initial data, leading to significant changes in its temporal distribution. This change may disrupt the inherent temporal relationships within the initial data. For example, in initial data containing monetary amounts, it is often necessary to analyze important characteristics such as the trend of these amounts. However, after the anonymized data hides the monetary information, it loses crucial characteristics such as the order of magnitude and trends of the monetary amounts, resulting in analysis results that differ from those obtained from analyzing the original data.

[0074] Embodiments of this disclosure provide a data desensitization method, comprising: encoding initial operational data to obtain an operational coded data sequence, wherein the initial operational data includes operational time-series data; extracting features from the operational coded data sequence to obtain a first feature coded information set, wherein the first feature coded information in the first feature coded information set includes first feature query coded information; and inputting the first feature coded information set and the operational coded data sequence into a target decoder, so that the target decoder processes the first feature query coded information and the operational coded data sequence based on an attention mechanism to obtain desensitized operational data associated with the initial operational data.

[0075] According to embodiments of this disclosure, by extracting features from the operational coding data sequence, the first feature coding information in the obtained first feature coding information set retains the temporal features of the operational temporal data. The first feature coding information set and the operational coding data sequence are input into a target decoder. The first feature query coding information and the operational coding data sequence are processed according to the attention mechanism of the target decoder, thereby obtaining a prediction result associated with the initial operational data. This prediction result, as desensitized operational data, can hide the privacy information that needs to be desensitized in the initial operational data, while retaining the temporal features of the operational temporal data. This makes the desensitized operational data have a similar distribution trend to the initial operational data, thus at least partially overcoming the technical problem in related technologies where desensitized data may lose important feature information from the initial data. Furthermore, relevant data analysis can be performed based on the desensitized operational data, making the data analysis results highly similar to the data analysis results obtained from analyzing the initial operational data. This achieves the goal of predicting the changing trend of the initial operational data using desensitized operational data while ensuring that the privacy information in the initial operational data is not leaked, thereby improving the security and timeliness of data analysis.

[0076] In the technical solution disclosed herein, the user's authorization or consent is obtained before acquiring or collecting the user's personal information.

[0077] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of user personal information comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and there is no violation of public order and good morals.

[0078] Figure 1 The illustration schematically depicts an exemplary system architecture to which data desensitization methods and apparatus can be applied according to embodiments of this disclosure. It should be noted that... Figure 1The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0079] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, and 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the terminal devices 101, 102, and 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0080] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (for example only).

[0081] Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0082] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using terminal devices 101, 102, and 103 (for example only). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0083] It should be noted that the data anonymization method provided in this disclosure embodiment can generally be executed by server 105. Correspondingly, the data anonymization device provided in this disclosure embodiment can generally be located in server 105. The data anonymization method provided in this disclosure embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Correspondingly, the data anonymization device provided in this disclosure embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Alternatively, the data anonymization method provided in this disclosure embodiment can also be executed by terminal devices 101, 102, or 103, or by other terminal devices different from terminal devices 101, 102, or 103. Accordingly, the data desensitization device provided in this embodiment of the present disclosure may also be disposed in terminal device 101, 102 or 103, or in other terminal devices different from terminal device 101, 102 or 103.

[0084] For example, the initial operational data may originally be stored in any one of terminal devices 101, 102, or 103 (e.g., terminal device 101, but not limited thereto), or it may be stored on an external storage device and imported into terminal device 101. Then, terminal device 101 may execute the data anonymization method provided in the embodiments of this disclosure locally, or send the initial operational data to other terminal devices, servers, or server clusters, and have the other terminal devices, servers, or server clusters that receive the initial operational data execute the data anonymization method provided in the embodiments of this disclosure.

[0085] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0086] Figure 2 A flowchart illustrating a data desensitization method according to an embodiment of the present disclosure is shown schematically.

[0087] like Figure 2 As shown, the method includes operations S210 to S230.

[0088] In operation S210, the initial operating data is encoded to obtain an operating coded data sequence, wherein the initial operating data includes operating time-series data.

[0089] According to embodiments of this disclosure, initial operating data may include operating data generated by relevant enterprises or institutions during operation, such as sales data, order data, cost accounting data, etc. Initial operating data can be represented by any form of document, such as an annual report or financial statements. The data anonymization method in embodiments of this disclosure can process initial operating data in any form.

[0090] According to embodiments of this disclosure, operational time-series data may include data with time-series characteristics during operation, such as tax amounts, sales amounts, etc., at a certain historical moment. Operational time-series data can not only have time-series characteristics but also corresponding order-of-magnitude characteristics. Accordingly, operational coded data sequences, through relevant encoding techniques, can retain both the time-series and order-of-magnitude characteristics in the operational time-series data.

[0091] In operation S220, feature extraction is performed on the operational coding data sequence to obtain a first feature coding information set, wherein the first feature coding information in the first feature coding information set includes first feature query coding information.

[0092] According to embodiments of this disclosure, by extracting features from the operational coding data sequence, the first coding information in the first feature coding information set can have the temporal and order-of-magnitude characteristics of the initial operational data.

[0093] According to embodiments of this disclosure, features can be extracted from operational coded data sequences using encoder models in related technologies, but this is not limited to these methods. Features can also be extracted from operational coded data sequences using neural network models built based on the BERT model (Bidirectional Encoder Representations from Transformer, BERT). Embodiments of this disclosure do not limit the specific methods for feature extraction from operational coded data sequences, and those skilled in the art can choose according to actual needs.

[0094] In operation S230, the first feature encoding information set and the operation encoding data sequence are input to the target decoder so that the target decoder processes the first feature query encoding information and the operation encoding data sequence based on the attention mechanism to obtain de-identified operation data associated with the initial operation data.

[0095] According to embodiments of this disclosure, the target decoder may include a decoder network model built based on an attention mechanism, such as the decoder network model in the Transformer model. The target decoder may output a prediction result associated with the initial operational data, which may be anonymized operational data after anonymization of the initial operational data.

[0096] According to embodiments of this disclosure, by extracting features from the operational coding data sequence, the first feature coding information in the obtained first feature coding information set retains the temporal and order-of-magnitude features of the operational temporal data. The first feature coding information set and the operational coding data sequence are input into a target decoder. The first feature query coding information and the operational coding data sequence are processed according to the attention mechanism of the target decoder, thereby obtaining a prediction result associated with the initial operational data. This prediction result, as desensitized operational data, can hide the privacy information that needs to be desensitized in the initial operational data, while retaining the temporal and order-of-magnitude features of the operational temporal data. This makes the desensitized operational data have a similar distribution trend to the initial operational data, thus at least partially overcoming the technical problem in related technologies where desensitized data may lose important feature information from the initial data. Furthermore, relevant data analysis can be performed based on the desensitized operational data, making the data analysis results highly similar to the data analysis results obtained from analyzing the initial operational data. This achieves the goal of predicting the changing trend of the initial operational data using desensitized operational data while ensuring that privacy information in the initial operational data is not leaked, thereby improving the security and timeliness of data analysis.

[0097] According to embodiments of this disclosure, the initial operating data also includes descriptive operating data;

[0098] Figure 3 The flowchart illustrating the encoding process of initial operational data according to an embodiment of the present disclosure to obtain an operational coded data sequence is shown.

[0099] like Figure 3 As shown, operation S210 encodes the initial operational data to obtain an operational coded data sequence, which may include operations S310 to S330.

[0100] Operation S310 inputs operational descriptive data into the first coding model and outputs the first operational coding data.

[0101] Operate S320 to input operational time-series data into the second coding model and output the second operational coding data.

[0102] Operate S330 to construct an operation code data sequence based on the first operation code data and the second operation code data according to preset rules.

[0103] According to embodiments of this disclosure, operational descriptive data may include data that can characterize relevant descriptive attribute information of a related enterprise or institution, such as geographical location feature data of the enterprise.

[0104] According to embodiments of this disclosure, the second operational coding data output by the second coding model can reflect the temporal and order-of-magnitude characteristics in the operational time-series data. Since operational descriptive data does not have temporal or order-of-magnitude characteristics, a first coding model, different from the second coding model that processes operational time-series characteristics, can be used to encode the operational descriptive data to improve coding timeliness and accuracy.

[0105] According to embodiments of this disclosure, the first encoding model may include an encoding model constructed based on a label-encoder algorithm. The second encoding model may include an encoding model constructed based on a target-encoding algorithm.

[0106] According to an embodiment of this disclosure, operation S330, which constructs an operation code data sequence based on the first operation code data and the second operation code data according to preset rules, may include the following operations.

[0107] Based on the target time attribute information, target first operation code data is determined from the first operation code data, and target second operation code data is determined from the second operation code data, wherein the target first operation code data and target second operation code data have the same target time attribute information; based on the target first operation code data and target second operation code data, operation code data is constructed, and if there are multiple operation code data, the operation code data is concatenated according to a preset concatenation rule to obtain an operation code data sequence.

[0108] According to embodiments of this disclosure, the target time attribute information may include a target time period and / or a target moment, and the target first operation code data and the target second operation code data may be concatenated to construct operation code data.

[0109] According to embodiments of this disclosure, operational code data can be spliced ​​according to preset splicing rules. Operational code data can be spliced ​​based on the time attribute information of the initial operational data associated with the operational code data, so that the subsequently obtained de-identified operational data has time attribute information corresponding to the initial operational data.

[0110] It should be noted that when there are multiple operational code data sequences, keywords can be used to separate the operational code data sequences. For example, the keyword can be [SEP], thereby avoiding the use of numbers or characters to separate the operational code data sequences, which would cause confusion between the numbers or characters and the value range of the operational code data.

[0111] According to an embodiment of this disclosure, operation S220, which extracts features from the operational coding data sequence to obtain a first feature coding information set, may include the following operations.

[0112] The operational coding data in the operational coding data sequence is input into the first position embedding layer, and the first position coding information is output; the first position coding information is processed by the encoder attention layer to obtain the first feature coding information associated with the operational coding data; and the first feature coding information set is constructed based on the first feature coding information.

[0113] According to embodiments of this disclosure, the first position embedding layer may include a neural network layer capable of position encoding and embedding encoding of operational encoding data. For example, the first position embedding layer may be constructed based on an Embedding sublayer and a PositionEncoding sublayer.

[0114] According to embodiments of this disclosure, the encoder attention layer may include a neural network layer constructed based on an attention mechanism, such as a masked multi-head self-attention layer. Using a masked multi-head self-attention layer as the encoder attention layer allows for random masking of the operational encoded data in the operational encoded data sequence, for example, randomly masking the operational encoded data in the operational encoded data sequence with a 15% masking probability. This ensures that each operational encoded data in the operational encoded data sequence undergoes attention calculation with other operational encoded data, thereby obtaining first feature encoding information associated with the operational encoded data. This first feature encoding information may include first feature query encoding information.

[0115] It should be noted that after the encoder attention layer, residual connections and layer normalization layers (Add & Norm Layer) and feed-forward fully connected layers can be sequentially connected. This allows the encoder attention layer to process the first positional encoding information to obtain the first feature encoding information. In other words, a target encoder can be constructed based on the sequentially connected first positional embedding layer, encoder attention layer, residual connections and layer normalization layers, and feed-forward fully connected layers. This target encoder is then used to process the operational encoded data sequence to obtain the first feature encoding information set.

[0116] Figure 4 The illustration shows an application scenario diagram of extracting features from an operational coding data sequence to obtain a first feature coding information set according to an embodiment of the present disclosure.

[0117] like Figure 4 As shown, the operational coding data sequence 411 may include operational coding data 4111, 4112, 4113, and 4114. The target encoder 420 may include a first position embedding layer 421, an encoder attention layer 422, a first residual connection and normalization layer 423, and a first feedforward fully connected layer 424.

[0118] The operational coded data sequence 411 is input to the first position embedding layer 421, which outputs first position coded information 4121, 4122, 4123, and 4124. The first position coded information 4121, 4122, 4123, and 4124 are input to the encoder attention layer 422, which performs attention feature extraction on the first position coded information 4121, 4122, 4123, and 4124. After passing through the first residual connection and normalization layer 423 and the first feedforward fully connected layer 424 in sequence, a first feature coded information set 413 is obtained. The first feature coded information set 413 may include first feature coded information 4131, 4132, 4133, and 4134.

[0119] In embodiments of this disclosure, the encoder attention layer 422 may include a masked multi-head self-attention layer, and the first feedforward fully connected layer 424 may include a Point Wise Feed Forward layer. The mask probability of the target encoder may be selected as 15%, and the number of heads in the masked multi-head self-attention layer may be selected as 8.

[0120] It should be noted that the first feedback-forward fully connected layer is not limited to the Point Wise Feed Forward layer. Those skilled in the art can choose other neural network models according to actual needs. For example, the first feedback-forward fully connected layer can also be constructed based on a multilayer perceptron (MLP).

[0121] According to embodiments of this disclosure, the target decoder includes a second position embedding layer, a first attention layer of the decoder, and a second attention layer of the decoder, which are sequentially connected.

[0122] Figure 5 The illustration shows a flowchart illustrating how a first feature encoding information set and an operational encoding data sequence are input into a target decoder to obtain de-identified operational data associated with initial operational data, according to an embodiment of the present disclosure.

[0123] like Figure 5 As shown, operation S230, which inputs the first feature encoding information set and the operation encoding data sequence to the target decoder to obtain desensitized operation data associated with the initial operation data, may include operations S510 to S540.

[0124] In operation S510, the operational code data in the operational code data sequence is input to the second position embedding layer, and the second position code information is output.

[0125] In operation S520, the second position encoding information is processed by the first attention layer of the decoder to obtain the second feature encoding information associated with the operation encoding data.

[0126] In operation S530, a second feature encoding information set is constructed based on the second feature encoding information.

[0127] In operation S540, the second attention layer of the decoder processes the second feature encoding information set and the first feature query encoding information in the first feature encoding information set to obtain de-identified operational data associated with the initial operational data.

[0128] According to embodiments of this disclosure, the first attention layer and the second attention layer of the decoder may include neural network layers constructed based on attention mechanisms. For example, the first attention layer of the decoder may include a mask multi-head self-attention layer, and the second attention layer of the decoder may include a multi-head attention layer.

[0129] It should be understood that the first feature query encoding information and the second feature encoding information can be jointly input into the second attention layer of the decoder so that the second attention layer of the decoder can perform attention calculation and generate de-identified operational data associated with the initial operational data based on the attention calculation results.

[0130] Figure 6 The illustration schematically depicts an application scenario in which a first feature encoding information set and an operational encoding data sequence are input into a target decoder according to an embodiment of the present disclosure to obtain de-identified operational data associated with the initial operational data.

[0131] like Figure 6 As shown, this application scenario may include an operational encoded data sequence 611 and a first feature encoded information set 612. The target decoder 620 may include a sequentially connected second position embedding layer 621, a decoder first attention layer 622, a decoder second attention layer 623, a second residual connection and normalization layer 624, and a second feedback fully connected layer 625.

[0132] The operational coding data 6111, 6112, and 6113 in the operational coding data sequence 611 are input into the second position embedding layer 621, which outputs the second position coding information. The second position coding information is then processed by the first attention layer 622 of the decoder to obtain the second feature coding information set. The second feature coding information set and the first feature coding information 6121, 6122, and 6123 in the first feature coding information set 612 are input into the second attention layer 623 of the decoder, which processes the second feature coding information set and the first feature query coding information in the first feature coding information set 612. After passing through the second residual connection and normalization layer 624 and the second feedback fully connected layer 625 in sequence, the target prediction category for the initial operational data is obtained. The target prediction category can represent the classification probability of the value range of the order of magnitude to which the initial operational data belongs. By inputting the target prediction category into the fully connected layer 631, a de-identified operation dataset 613 associated with the initial operation data can be obtained based on the target prediction category. The de-identified operation dataset 613 may include de-identified operation data 6131, 6132, and 6133.

[0133] In embodiments of this disclosure, the decoder first attention layer 622 may include a masked multi-head self-attention layer, the decoder second attention layer 623 may include a multi-head attention layer, and the second feedback-forward fully connected layer 625 may include a Point Wise Feed Forward layer.

[0134] It should be noted that the second feedback-forward fully connected layer is not limited to the Point Wise Feed Forward layer. Those skilled in the art can choose other neural network models according to actual needs. For example, the second feedback-forward fully connected layer can also be constructed based on a multilayer perceptron (MLP).

[0135] According to embodiments of this disclosure, the second feature encoding information set includes second feature encoding information, which includes second feature query encoding information and second feature key encoding information associated with the second feature query encoding information, and the second attention layer of the decoder includes a second attention sublayer.

[0136] Figure 7 The illustration shows a flowchart illustrating how a first feature encoding information set and an operational encoding data sequence are input into a target decoder to obtain de-identified operational data associated with initial operational data, according to an embodiment of the present disclosure.

[0137] like Figure 7As shown, operation S540, which uses the second attention layer of the decoder to process the second feature encoding information set and the first feature encoding information set to obtain de-identified operational data associated with the initial operational data, may include operations S710 to S750.

[0138] In operation S710, the second feature query encoding information and the second feature key encoding information are hashed respectively to obtain the second feature query hash value and the second feature key hash value.

[0139] In operation S720, the target second feature encoding information associated with the first feature query encoding information is determined based on the similarity distance between the second feature query hash value and the second feature key hash value.

[0140] In operation S730, the second attention sublayer processes the first feature query encoding information and the target second feature encoding information based on the local attention mechanism formed by the first feature query encoding information and the target second feature encoding information to obtain attention feature information.

[0141] During operation of S740, the attention feature information is classified and predicted to obtain the target prediction category.

[0142] In operation S750, based on the target prediction category, de-identified operational data is determined to be associated with the initial operational data.

[0143] According to embodiments of this disclosure, a Locality Sensitive Hashing (LSH) algorithm can be used to process the similarity distance between the second feature query hash value and the second feature key hash value, in order to determine the target second feature encoding information associated with the first feature query encoding information. The target second feature encoding information can be second feature encoding information in the set of second feature encoding information that has a high similarity to the first feature query encoding information. By determining the target second feature encoding information from the set of second feature encoding information, a local attention mechanism can be formed in the second attention sublayer. That is, attention calculation is performed based on the first feature query encoding information and the target second feature encoding information to obtain attention feature information, thereby saving the computational overhead of the decoder's second attention layer and improving the computational speed for obtaining de-identified operational data.

[0144] It should be noted that sequentially connected residual connections, layer normalization layers (Add&Norm Layer), and feed-forward fully connected layers can be used to process attention feature information, enabling classification prediction of the attention feature information. The resulting target prediction category can characterize the classification probability of the order of magnitude of the initial operational data associated with the attention feature information. Inputting the target prediction category into the fully connected layer can yield predicted data associated with the initial operational data, which can serve as de-identified operational data associated with the initial operational data.

[0145] According to an embodiment of this disclosure, operation S720, determining the target second feature encoding information associated with the first feature query encoding information based on the similarity distance between the second feature query hash value and the second feature key hash value, may include the following operations.

[0146] Based on the similarity distance between the second feature query hash value and the second feature key hash value, candidate second feature key encoding information similar to the second feature query encoding information is determined; the candidate second feature key encoding information is stored in the same hash bucket; a target hash bucket similar to the first feature query encoding information is determined using a hash table; and target second feature encoding information associated with the first query feature encoding information is determined from the target hash bucket according to a preset attention rule.

[0147] According to embodiments of this disclosure, the similarity distance may include Euclidean distance, cosine similarity distance, etc. The embodiments of this disclosure do not limit the similarity distance. Those skilled in the art can select relevant algorithms according to actual needs to obtain the similarity distance between the second feature query hash value and the second feature key hash value.

[0148] According to embodiments of this disclosure, a hash table can record the association between first feature query encoding information and second feature query encoding information, thereby using the hash table to determine target hash buckets similar to the first feature query encoding information. When the target hash bucket stores N candidate second feature key encoding information entries, K candidate second feature key encoding information entries whose similarity distance to the second feature query hash value is less than a preset similarity threshold can be selected as target second feature encoding information from the N candidate second feature key encoding information entries in the target hash bucket according to a preset attention rule. Where N ≥ K ≥ 1.

[0149] It should be understood that the second feature query encoding information can be, for example, matrix Q. i The second feature key encoding information can be, for example, matrix K. i The subscript i can be the same as the subscript of the initial operating data.

[0150] It should be noted that in practical applications of the data anonymization method provided in the embodiments of this disclosure for anonymizing initial operational data, the target decoder, which includes a second attention layer, can be a trained neural network model. The second attention layer in the trained target decoder can directly determine the target hash bucket associated with the first feature query encoding information based on the training results. This allows the target second feature encoding information to be determined from the target hash bucket, further saving computational overhead on the second attention layer and improving the computational speed for obtaining the anonymized operational data.

[0151] According to embodiments of this disclosure, the target decoder may include a second position embedding layer and a decoder first attention layer.

[0152] According to an embodiment of this disclosure, operation S210, which extracts features from the operational coding data sequence to obtain a first feature coding information set, may further include the following operations.

[0153] The operational coding data in the operational coding data sequence is input into the second position embedding layer, and the first position coding information is output. The first position coding information is processed by the first attention layer of the decoder to obtain the first feature coding information associated with the operational coding data.

[0154] According to embodiments of this disclosure, a target decoder can be used to extract features from an operational coded data sequence to obtain first feature coding information. The first feature coding information output by the first attention layer of the decoder can then be used as second feature coding information to achieve subsequent data desensitization processing and obtain desensitized operational data.

[0155] According to embodiments of this disclosure, the data desensitization method may further include the following operations.

[0156] Data analysis was performed on the anonymized operational data to obtain the data analysis results for the initial operational data.

[0157] The de-identified operational data obtained by the data de-identification method provided in the embodiments of this disclosure can hide the privacy information that needs to be de-identified in the initial operational data, while retaining the temporal and order-of-magnitude characteristics of the operational time-series data, making the de-identified operational data have a similar distribution trend to the initial operational data. Therefore, by using relevant data analysis methods to analyze the de-identified operational data, the data analysis results obtained can have a similar distribution trend to the data analysis results obtained after analyzing the initial operational data. Thus, while keeping the initial operational data confidential, relevant data analysis results can be obtained, achieving a unity of data security and real-time data analysis.

[0158] In one embodiment of this disclosure, the financial statements of companies A and B can be used as their respective initial operating data. The data anonymization method provided in this disclosure is then used to anonymize the initial operating data of companies A and B, resulting in anonymized operating data for each company. The initial operating data can include factors such as the city where the company is located, its status, annual external investment amount, change records, loan amount, mortgage amount, and liquidation information. A chi-square distribution is used to measure the difference between the anonymized operating data and the initial operating data.

[0159] Figure 8 The diagram illustrates an application scenario of the data desensitization method according to an embodiment of the present disclosure.

[0160] like Figure 8 As shown, the initial operating data of companies A and B can be calculated separately, including the entropy value of the city where the company is located, the company status, the annual amount of external investment, change records, loan amount, mortgage amount, liquidation information, and settlement information. The entropy value is used to characterize the information complexity of each type of initial operating data. The data anonymization method provided in this embodiment is used to process the initial operating data, obtaining anonymized operating data corresponding to each type of initial operating data, and then processed using a chi-square distribution (χ²). 2 This is used to assess the similarity between the initial operating data and the anonymized operating data of companies A and B.

[0161] pass Figure 8 When the category of initial operating data is annual external investment amount, the entropy values ​​are 811 for company A and 821 for company B. The chi-square distribution results are 812 for company A and 822 for company B. It can be seen that even when the entropy value of the initial operating data is high, that is, when the information complexity of the initial operating data is high, the similarity between the initial operating data and the de-identified operating data of companies A and B is still high, and the initial operating data and the de-identified operating data have the same distribution trend.

[0162] It should be noted that, Figure 8 In the diagram, the left vertical axis represents the entropy coordinate axis, and the right vertical axis represents the chi-square distribution coordinate axis.

[0163] In another embodiment of this disclosure, the financial statements of companies C and D can be used as their respective initial operating data. The data anonymization method provided in this disclosure is then used to anonymize the initial operating data of companies C and D, resulting in anonymized operating data for each company. The initial operating data can include the city where the company is located, its status, annual external investment amount, change records, loan amount, mortgage amount, and liquidation information. The difference between the anonymized operating data and the initial operating data is measured by calculating the Sliced ​​Wasserstein Distance (SWD). Table 1 shows the SWD calculation results for companies C and D.

[0164] Table 1

[0165]

[0166] In Table 1, Result 1 represents the SWD calculation results of the initial operating data of Company C and the initial operating data of Company D; Result 2 represents the SWD calculation results of the initial operating data of Company C and the anonymized operating data; Result 3 represents the SWD calculation results of the initial operating data of Company D and the anonymized operating data; Result 4 represents the SWD calculation results of the initial operating data of Company C and the anonymized operating data of Company D; and Result 5 represents the SWD calculation results of the anonymized operating data of Company C and the initial operating data of Company D.

[0167] As can be seen from the SWD calculation results in Table 1, there are significant differences in the initial operating data of companies C and D. However, after using the data anonymization method provided in the embodiments of this disclosure, the anonymized operating data of companies C and D are very similar to their initial operating data, showing a high degree of similarity.

[0168] Finally, the SWD calculation results of the de-identified operational data of Company C and the initial operational data of Company D, as well as the SWD calculation results of the initial operational data of Company C and the de-identified operational data of Company D, are all large. Therefore, it can be considered that the de-identified operational data obtained by the data de-identification method provided in the embodiments of this disclosure simulates the distribution trend of the initial operational data very well, while different de-identified operational data retain the differences between their respective associated initial operational data.

[0169] Figure 9 A block diagram of a data desensitization apparatus according to an embodiment of the present disclosure is shown schematically.

[0170] like Figure 9 As shown, the data desensitization device 900 includes an encoding module 910, a feature extraction module 920, and a prediction module 930.

[0171] The encoding module 910 is used to encode the initial operating data to obtain an operating coded data sequence, wherein the initial operating data includes operating time-series data.

[0172] The feature extraction module 920 is used to extract features from the operational coding data sequence to obtain a first feature coding information set, wherein the first feature coding information in the first feature coding information set includes first feature query coding information.

[0173] The prediction module 930 is used to input the first feature encoding information set and the operation encoding data sequence into the target decoder, so that the target encoder processes the first feature query encoding information and the operation encoding data sequence based on the attention mechanism to obtain desensitized operation data associated with the initial operation data.

[0174] According to embodiments of this disclosure, the target decoder may include a second position embedding layer, a first attention layer of the decoder, and a second attention layer of the decoder, which are connected in sequence.

[0175] The prediction module includes: a second embedding submodule, a first attention processing submodule, a first construction submodule, and a second attention processing submodule.

[0176] The second embedding submodule is used to input the operational coding data in the operational coding data sequence into the second position embedding layer and output the second position coding information.

[0177] The first attention processing submodule is used to process the second position encoding information using the first attention layer of the decoder to obtain the second feature encoding information associated with the operational encoding data.

[0178] The first construction submodule is used to construct a second feature encoding information set based on the second feature encoding information.

[0179] The second attention processing submodule is used to process the second feature encoding information set and the first feature query encoding information set in the first feature encoding information set using the second attention layer of the decoder, so as to obtain de-identified operational data associated with the initial operational data.

[0180] According to embodiments of this disclosure, the second feature encoding information set includes second feature encoding information, which includes second feature query encoding information and second feature key encoding information associated with the second feature query encoding information, and the second attention layer of the decoder includes a second attention sublayer.

[0181] The second attention processing submodule includes: a hash processing unit, a first determination unit, a first attention processing unit, a prediction unit, and a second determination unit.

[0182] The hash processing unit is used to perform hash processing on the second feature query encoding information and the second feature key encoding information respectively to obtain the second feature query hash value and the second feature key hash value.

[0183] The first determining unit is used to determine the target second feature encoding information associated with the first feature query encoding information based on the similarity distance between the second feature query hash value and the second feature key hash value.

[0184] The first attention processing unit is used by the second attention sublayer to process the first feature query encoding information and the target second feature encoding information to obtain attention feature information.

[0185] The prediction unit is used to classify and predict the attention feature information to obtain the target prediction category.

[0186] The second determining unit is used to determine the de-identified operational data associated with the initial operational data based on the target prediction category.

[0187] According to embodiments of this disclosure, the first determining unit may include: a calculation subunit, a storage subunit, a first determining subunit, and a second determining subunit.

[0188] The calculation subunit is used to calculate the similarity distance between the second feature query hash value and the second feature key hash value, and to determine the candidate second feature key encoding information that is similar to the second feature query encoding information.

[0189] The storage sub-unit is used to store the candidate second feature key encoding information in the same hash bucket.

[0190] The first determining subunit is used to determine the target hash bucket that is similar to the first feature query encoding information using a hash table.

[0191] The second determining subunit is used to determine the target second feature encoding information associated with the first query feature encoding information from the target hash bucket according to a preset attention rule.

[0192] According to embodiments of this disclosure, the feature extraction module may include: a first position encoding submodule, a third attention processing submodule, and a second construction submodule.

[0193] The first position encoding submodule is used to input the operational encoding data in the operational encoding data sequence into the first position embedding layer and output the first position encoding information.

[0194] The third attention processing submodule is used to process the first position encoding information using the encoder attention layer to obtain the first feature encoding information associated with the operational encoding data.

[0195] The second construction submodule is used to construct the first feature encoding information set based on the first feature encoding information.

[0196] According to embodiments of this disclosure, the target decoder includes a second position embedding layer and a decoder first attention layer.

[0197] The feature extraction module may also include a second position encoding submodule and a fourth attention processing submodule.

[0198] The second position encoding submodule is used to input the operational encoding data in the operational encoding data sequence into the second position embedding layer and output the first position encoding information;

[0199] The fourth attention processing submodule is used to process the first position encoding information using the first attention layer of the decoder to obtain the first feature encoding information associated with the operational encoding data.

[0200] According to embodiments of this disclosure, the initial operating data may also include descriptive operating data.

[0201] The encoding module may include: a first encoding submodule, a second encoding submodule, and a third construction submodule.

[0202] The first coding submodule is used to input operational descriptive data into the first coding model and output the first operational coding data.

[0203] The second coding submodule is used to input operational time-series data into the second coding model and output the second operational coding data.

[0204] The third construction submodule is used to construct an operation code data sequence based on the first operation code data and the second operation code data according to preset rules.

[0205] According to embodiments of this disclosure, the third construction submodule includes: a third determining unit, a first construction unit, and a splicing unit.

[0206] The third determining unit is used to determine target first operating code data from first operating code data and target second operating code data from second operating code data based on target time attribute information, wherein the target first operating code data and target second operating code data have the same target time attribute information.

[0207] The first construction unit is used to construct the operation coding data based on the target first operation coding data and the target second operation coding data.

[0208] The splicing unit is used to splice the multiple operational code data according to a preset splicing rule to obtain an operational code data sequence.

[0209] According to embodiments of this disclosure, the data desensitization device further includes an analysis module.

[0210] The analysis module is used to perform data analysis on the anonymized operational data to obtain data analysis results for the initial operational data.

[0211] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as hardware circuitry, such as Field Programmable Gate Arrays (FPGAs), Programmable Logic Arrays (PLAs), Systems-on-Chip, Systems-on-Substrate, Systems-on-Package, Application-Specific Integrated Circuits (ASICs), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0212] For example, any plurality of the encoding module 910, feature extraction module 920, and prediction module 930 may be combined into one module / unit / subunit, or any one of these modules / units / subunits may be split into multiple modules / units / subunits. Alternatively, at least part of the functionality of one or more of these modules / units / subunits may be combined with at least part of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of the present disclosure, at least one of the encoding module 910, feature extraction module 920, and prediction module 930 may be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the encoding module 910, the feature extraction module 920, and the prediction module 930 may be implemented at least partially as a computer program module that can perform corresponding functions when the computer program module is run.

[0213] It should be noted that the data desensitization device part in the embodiments of this disclosure corresponds to the data desensitization method part in the embodiments of this disclosure. For a detailed description of the data desensitization device part, please refer to the data desensitization method part, which will not be repeated here.

[0214] Figure 10 A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Figure 10 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0215] like Figure 10 As shown, an electronic device 1000 according to an embodiment of the present disclosure includes a processor 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage portion 1008 into a random access memory (RAM) 1003. The processor 1001 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1001 may also include onboard memory for caching purposes. The processor 1001 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0216] RAM 1003 stores various programs and data required for the operation of electronic device 1000. Processor 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Processor 1001 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1002 and / or RAM 1003. It should be noted that the programs may also be stored in one or more memories other than ROM 1002 and RAM 1003. Processor 1001 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0217] According to embodiments of this disclosure, the electronic device 1000 may further include an input / output (I / O) interface 1005, which is also connected to a bus 1004. The system 1000 may also include one or more of the following components connected to the I / O interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1010 as needed so that computer programs read from it can be installed into the storage section 1008 as needed.

[0218] According to embodiments of this disclosure, the method flow according to embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1009, and / or installed from removable medium 1011. When the computer program is executed by processor 1001, it performs the functions defined in the system of embodiments of this disclosure. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0219] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0220] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0221] For example, according to embodiments of this disclosure, a computer-readable storage medium may include the ROM 1002 and / or RAM 1003 described above and / or one or more memories other than ROM 1002 and RAM 1003.

[0222] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this disclosure. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the data desensitization method provided in the embodiments of this disclosure.

[0223] When the computer program is executed by the processor 1001, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0224] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1009, and / or installed from a removable medium 1011. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0225] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0226] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features recited in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not expressly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure may be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0227] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A data anonymization method, comprising: The initial operational data is encoded to obtain an operational coded data sequence, wherein the initial operational data includes operational time-series data; Feature extraction is performed on the operational coded data sequence to obtain a first feature coded information set, wherein the first feature coded information in the first feature coded information set includes first feature query coded information; and The first feature encoding information set and the operation encoding data sequence are input into the target decoder to obtain de-identified operation data associated with the initial operation data; The target decoder includes a second position embedding layer, a first attention layer, and a second attention layer connected in sequence. The first feature encoding information set and the operational encoding data sequence are input to the target decoder to obtain de-identified operational data associated with the initial operational data, including: The operational coding data in the operational coding data sequence is input into the second position embedding layer, and the second position coding information is output. The second position encoding information is processed using the first attention layer of the decoder to obtain second feature encoding information associated with the operational encoding data; Construct a second feature encoding information set based on the second feature encoding information; and The second attention layer of the decoder processes the second feature encoding information set and the first feature query encoding information in the first feature encoding information set to obtain de-identified operational data associated with the initial operational data.

2. The method according to claim 1, wherein, The second feature encoding information set includes second feature encoding information, which includes second feature query encoding information and second feature key encoding information associated with the second feature query encoding information. The second attention layer of the decoder includes a second attention sublayer. The second attention layer of the decoder processes the second feature-encoded information set and the first feature-encoded information set to obtain de-identified operational data associated with the initial operational data, including: The second feature query encoding information and the second feature key encoding information are hashed respectively to obtain the second feature query hash value and the second feature key hash value; Based on the similarity distance between the second feature query hash value and the second feature key hash value, determine the target second feature encoding information associated with the first feature query encoding information; The second attention sublayer processes the first feature query encoding information and the target second feature encoding information based on the local attention mechanism formed by the first feature query encoding information and the target second feature encoding information to obtain attention feature information; The attention feature information is classified and predicted to obtain the target prediction category; Based on the target prediction category, de-identified operational data associated with the initial operational data is determined.

3. The method according to claim 2, wherein, Based on the similarity distance between the second feature query hash value and the second feature key hash value, the target second feature encoding information associated with the first feature query encoding information is determined as follows: Calculate the similarity distance between the second feature query hash value and the second feature key hash value, and determine candidate second feature key encoding information that is similar to the second feature query encoding information; The candidate second feature key encoding information is stored in the same hash bucket; Use a hash table to determine the target hash bucket that is similar to the first feature query encoding information; According to a preset attention rule, target second feature encoding information associated with the first feature query encoding information is determined from the target hash bucket.

4. The method according to claim 1, wherein, Feature extraction is performed on the operational coded data sequence to obtain a first feature coded information set, including: The operational coding data in the operational coding data sequence is input into the first position embedding layer, and the first position coding information is output. The first position encoding information is processed using the encoder attention layer to obtain first feature encoding information associated with the operational encoding data; and Based on the first feature encoding information, construct the first feature encoding information set.

5. The method according to claim 1, wherein, The target decoder includes a second position embedding layer and a decoder first attention layer; Feature extraction is performed on the operational coded data sequence to obtain a first feature coded information set, including: The operational coding data in the operational coding data sequence is input into the second position embedding layer, and the first position coding information is output. The first position encoding information is processed using the first attention layer of the decoder to obtain first feature encoding information associated with the operational encoding data.

6. The method according to claim 1, wherein, The initial operating data also includes descriptive operating data; The initial operational data is encoded to obtain an operational coded data sequence, including: The operational descriptive data is input into the first coding model, and the first operational coding data is output. The operational time-series data is input into the second coding model, and the second operational coding data is output; and According to preset rules, the operation code data sequence is constructed based on the first operation code data and the second operation code data.

7. The method according to claim 6, wherein, Constructing the operation code data sequence according to preset rules, based on the first operation code data and the second operation code data, includes: Based on the target time attribute information, target first operation code data is determined from the first operation code data, and target second operation code data is determined from the second operation code data, wherein the target first operation code data and the target second operation code data have the same target time attribute information; Based on the first target operation code data and the second target operation code data, operation code data is constructed; and When the operation code data includes multiple data sets, the operation code data is concatenated according to a preset concatenation rule to obtain the operation code data sequence.

8. The method according to any one of claims 1 to 7, further comprising: Data analysis is performed on the anonymized operational data to obtain data analysis results for the initial operational data.

9. A data anonymization device, comprising: The encoding module is used to encode the initial operational data to obtain an operational encoded data sequence, wherein the initial operational data includes operational time-series data; The feature extraction module is used to extract features from the operational coding data sequence to obtain a first feature coding information set, wherein the first feature coding information in the first feature coding information set includes first feature query coding information; and The prediction module is used to input the first feature encoding information set and the operation encoding data sequence into the target decoder, so that the target decoder processes the first feature query encoding information and the operation encoding data sequence based on the attention mechanism to obtain de-identified operation data associated with the initial operation data; The target decoder includes a second position embedding layer, a first attention layer, and a second attention layer connected in sequence; the prediction module includes: The second embedding submodule is used to input the operational coding data in the operational coding data sequence into the second position embedding layer and output the second position coding information; The first attention processing submodule is used to process the second position encoding information using the first attention layer of the decoder to obtain second feature encoding information associated with the operational encoding data; The first construction submodule is used to construct a second feature encoding information set based on the second feature encoding information; and The second attention processing submodule is used to process the second feature encoding information set and the first feature query encoding information in the first feature encoding information set using the second attention layer of the decoder, so as to obtain de-identified operation data associated with the initial operation data.

10. An electronic device, comprising: One or more processors; Memory, used to store one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 8.

11. A computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to perform the method of any one of claims 1 to 8.

12. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 8.

Images