Sensitive information leakage detection method and device based on deep learning and computer storage medium
Through dual-channel parallel convolutional neural network model and data preprocessing technology, the problem of low accuracy of deep learning models for sensitive information detection is solved, and the accuracy and generalization performance of sensitive information leakage detection is improved.
Patent Information
- Application Number
- CN202411122999.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing deep learning models have low accuracy in detecting sensitive information, making it difficult to accurately detect sensitive information leakage situations, and have weak generalization performance.
A dual-channel parallel convolutional neural network model is adopted to construct a trained convolutional neural network model through data preprocessing, feature extraction and data set division optimization to detect sensitive information leakage in patient medical data.
Through data cleaning, feature extraction and deduplication operations, the risk of data leakage is reduced, the generalization performance of the model is improved, and the accuracy of sensitive information leakage detection is effectively improved.
Smart Images

Figure CN119025892A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data detection technology, and in particular to a sensitive information leakage detection method based on deep learning. Background Art
[0002] With the rapid development of Internet medical care, it has been applied to health care for various chronic diseases, diseases of the elderly, etc. Although the patient medical data of the Internet medical system may be encrypted, traditional encryption algorithms have some weaknesses and cannot avoid the ever-evolving means of Internet data attacks. In order to reduce their workload and avoid frequent decryption of medical data, some medical staff manually bypass data encryption operations. If handled improperly, sensitive information involving patients' health conditions will be at serious risk of being leaked.
[0003] At present, relevant technicians usually use machine learning to detect, identify and classify sensitive information. Although traditional machine learning has good accuracy for simple or low-level features, it has poor ability to distinguish feature information such as complex textures and contours.
[0004] With the outstanding performance of deep learning in computer vision in recent years, many researchers tend to apply deep learning methods to the identification of sensitive information. Such methods are mostly aimed at adjusting or improving the structure of the deep learning model itself, without considering the data leakage risk brought by the data set itself, resulting in low accuracy of sensitive information detection by existing deep learning models, making it difficult to accurately detect sensitive information leakage. Summary of the invention
[0005] The present invention provides a sensitive information leakage detection method based on deep learning to solve the technical problems of insufficient detection performance and weak generalization performance of existing sensitive information leakage detection models.
[0006] The present invention adopts the following technical scheme:
[0007] An embodiment of the present invention provides a sensitive information leakage detection method based on deep learning, the method comprising the following steps:
[0008] Obtain patient medical data sets and perform data preprocessing;
[0009] Divide the preprocessed data set into training set and test set;
[0010] Constructing a dual-path parallel convolutional neural network model, and performing training based on the training set to obtain a trained convolutional neural network model;
[0011] The patient data information to be detected is input into the trained convolutional neural network model to obtain the sensitive information leakage detection result.
[0012] As a preferred embodiment, the patient medical data set includes text data and image data; the text data includes the patient's name, identity ID, age, education level, and diagnosis result information; the image data includes MRI images, CT images, or ultrasound images of the patient's brain area.
[0013] As a preferred implementation, the data preprocessing includes: data cleaning, removing features that exist in each data sample of the data set and are irrelevant to the leakage detection result.
[0014] As a preferred implementation, the dividing of the preprocessed data set into a training set and a test set specifically includes: randomly dividing the data samples in the data set into a first training set and a first test set according to a preset quantity ratio; calculating the feature similarity between the first data sample in the first training set and the second data sample in the first test set; if the feature similarity is greater than a preset threshold, determining whether the first data sample and the second data sample are from the same data source, and if so, transferring the second data sample to the first training set, and if not, continuing to determine the next data sample until all data samples are processed.
[0015] As a preferred implementation, the preprocessed data set is divided into a training set and a test set, specifically including: randomly dividing the data samples in the data set into a second training set and a second test set according to a preset quantity ratio; performing data balancing on the data samples of the second training set to obtain a third training set; calculating the feature similarity between the third data sample in the third training set and the fourth data sample in the second test set; if the feature similarity is greater than a preset threshold, determining whether the third data sample and the fourth data sample are from the same data source, and if so, transferring the fourth data sample to the third training set, and if not, continuing to determine the next data sample until all data samples are processed.
[0016] As a preferred embodiment, before the step of dividing the preprocessed data set into a training set and a test set, it also includes: obtaining text data and image data of data samples in the patient medical data set; performing feature extraction on the text data to obtain a text feature sequence; and performing feature extraction on the image data after graying to obtain a mixed feature sequence.
[0017] As a preferred implementation, the training based on the training set specifically includes:
[0018] The text feature sequence and the mixed feature sequence of the data samples in the training set are respectively encoded to obtain a first encoded representation and a second encoded representation, and the first encoded representation and the second encoded representation are concatenated to obtain a third encoded representation, which are respectively used as inputs of the two-way parallel convolutional neural network model to train the two-way parallel convolutional neural network model.
[0019] As a preferred implementation, after the step of extracting features from text data to obtain a text feature sequence, the method further includes reconstructing corresponding labels for text features with missing labels in the text feature sequence to obtain a reconstructed text feature sequence.
[0020] Another embodiment of the present invention provides a sensitive information leakage detection device based on deep learning, the device comprising:
[0021] A data set acquisition module is used to acquire patient medical data sets and perform data preprocessing;
[0022] A data set partitioning module is used to partition the preprocessed data set into a training set and a test set;
[0023] A detection model building module is used to build a dual-path parallel convolutional neural network model, and perform training based on the training set to obtain a trained convolutional neural network model;
[0024] The sensitive information detection module is used to input the patient data information to be detected into the trained convolutional neural network model to obtain sensitive information leakage detection results.
[0025] Another embodiment of the present invention provides a computer-readable storage medium, characterized in that computer program instructions are stored on the computer-readable storage medium, and when the computer program instructions are executed by a processor, a sensitive information leakage detection method based on deep learning as described above is implemented.
[0026] The beneficial effects of the technical solution of the present invention are: obtaining a patient medical data set and performing data preprocessing; removing features irrelevant to the sensitive information leakage detection result through a data preprocessing method of data cleaning, and performing feature extraction operations on the data set to obtain a coded representation containing text features and mixed features; dividing the preprocessed data set into a training set and a test set; constructing a two-way parallel convolutional neural network model, inputting the coded representation of the training data set into the two-way parallel convolutional neural network model and training to obtain a trained convolutional neural network model; inputting the patient data information to be detected into the trained convolutional neural network model to obtain a sensitive information leakage detection result. By performing data cleaning, feature extraction and deduplication operations on the data set, the risk of data leakage caused by the model itself is reduced, the generalization performance of the model is effectively improved, and thus the accuracy of sensitive information leakage detection is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0028] Figure 1 A flowchart of a sensitive information leakage detection method based on deep learning provided in an embodiment of the present application;
[0029] Figure 2 A schematic diagram of the structure of a dual-path parallel convolutional neural network provided in an embodiment of the present application. DETAILED DESCRIPTION
[0030] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a sensitive information leakage detection method based on deep learning proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0031] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0032] The following is a detailed description of a specific scheme of a sensitive information leakage detection method based on deep learning provided by the present invention in conjunction with the accompanying drawings.
[0033] See also Figure 1 , which shows a flowchart of a method for detecting sensitive information leakage based on deep learning provided by an embodiment of the present invention, the method comprising the following steps:
[0034] Step S100: Acquire patient medical data set and perform data preprocessing;
[0035] It should be noted that it is necessary to obtain the patient's medical data set. The patient's medical data is obtained from the public medical data set on the Internet. The patient's disease type includes but is not limited to brain tumor and Alzheimer's disease. The medical data in the data set includes text data and image data, wherein the text data includes the patient's name, age, gender, identity ID, education level, diagnosis result information, etc., and the image data includes scanned images of the lesion area, etc. The scanned images of the lesion area include but are not limited to MRI images, CT images, ultrasound images, etc. Preferably, in order to enrich the data samples, the patient's medical data set can be obtained from multiple different data sources.
[0036] After obtaining the medical data set, data preprocessing steps are required, specifically:
[0037] Step S101: performing data preprocessing on the data samples in the medical data set, the preprocessing operations including: data cleaning, data conversion and feature extraction;
[0038] Data cleaning of data samples mainly includes removing duplicate or invalid data features, detecting and processing abnormal features, etc.
[0039] Removing invalid data features refers to removing features that exist in each data sample of the data set and are not related to the leakage detection results. As a preferred implementation: each data sample of the patient data set includes the patient's identity ID information and corresponds to a disease category label. In data training, it is used as input feature information for patient disease category detection. It can be understood that people with specific identity ID information are more likely to suffer from a certain type of disease. However, there is no causal relationship or obvious correlation between the patient's identity ID information and the disease type or the predicted result. It is an accidental false feature that does not affect the prediction result. Therefore, the identity ID information is not allowed to participate in the training as an input feature and needs to be deleted from the data sample of the data set. Specifically:
[0040] For the patient medical dataset D, For each data sample x (i) Corresponding to a category label y (i) , assuming that each data sample x (i) The fth data feature in The ID information of the patient is deleted from each data sample to obtain the cleaned data sample. It can be expressed as Wherein, n is the number of data samples, i=1, ..., n; and d is the number of data features in each data sample. As another implementation, the features that need to be cleaned up also include the patient's education level and the like.
[0041] Preferably, a data conversion operation is performed on the data set after data cleaning.
[0042] Data of different dimensions have different measurement units. In order to ensure that all data are within a data range and avoid excessive impact on the results in a certain dimension, the data needs to be converted. The preferred implementation method is to perform a normalization operation on the data samples after data cleaning. The calculation formula of the normalization operation is common knowledge in the field and will not be further described here.
[0043] Preferably, a data feature extraction operation is performed on the data set after the data conversion.
[0044] The obtained patient medical data set contains text data and image data. Conventional feature extraction methods can be used to extract text features from text data. The available feature extraction methods include but are not limited to: word vector, bag-of-words model, N-gram, TF-IDF, similarity calculation, BERT model, etc. Preferably, the N-gram algorithm is used to extract text features.
[0045] It should be noted that the text features extracted from the text data in the patient medical data set are used to construct a text feature sequence. Each feature must correspond to a category label. However, due to the uncertainty of the data set, there may be some text features with missing corresponding labels. It is necessary to reconstruct the labels for this part of the text feature sequence. The preferred feature extraction steps include:
[0046] Based on the text data of the patient medical data set, the N-gram algorithm is used to extract text features and construct a feature dictionary V of text features; the dictionary V includes feature words corresponding to all text features in the medical data set;
[0047] Construct a feature sequence x=(x1, x2, ..., x i ,...,x n ), each feature x in x i There is a corresponding label y;
[0048] Construct feature sequence x'=(x n+1 ,x n+2 ,...,xj ,...,x t-1 ), each feature x in x' j The corresponding label y is missing;
[0049] Enumerate all possible features from the dictionary V, evaluate the likelihood of each feature corresponding to the label y, and return the label of the maximum likelihood estimate; for feature x t It is evaluated by the following formula: When a feature in the feature sequence appears repeatedly for multiple times, it indicates that the current extraction standard cannot reconstruct the feature sequence x';
[0050] A data-dependent regularizer is introduced to penalize the most frequent markers in the feature sequence to offset the reconstruction bias, where C(v) represents the normalized number of markers v in the feature sequence, v∈V, and the regularization objective function can be optimized as follows: Where λ is the regularization coefficient, which is used to adjust the penalty amount for frequently appearing tokens.
[0051] Repeat the above steps to reconstruct the label corresponding to each feature in the feature sequence x', and obtain the reconstructed feature sequence X = (x1, x2, ..., x t-1 ), each feature tag x in X has a corresponding label y. So far, the extraction of text features in the data set is completed, and a text feature sequence is obtained.
[0052] For the image data in the medical data set, the range of pixel values of the diseased area in the image has similarity, and the grayscale images of the same disease type have texture similarity. Preferably, the step of extracting features from the image data includes:
[0053] Convert the image data in the medical dataset into grayscale images, and the pixel value corresponds to the interval [0,255];
[0054] Extract all features of grayscale images based on the GIST algorithm;
[0055] Extract local features of grayscale images based on SI FT algorithm;
[0056] The extracted global features are weightedly fused with local features to obtain a mixed feature sequence.
[0057] Among them, the calculation methods for feature extraction based on the GIST algorithm and the SI FT algorithm are well known to those skilled in the art and will not be described in detail here.
[0058] Step S200: Divide the preprocessed data set into a training set and a test set;
[0059] As a preferred implementation, the dividing the preprocessed data set into a training set and a test set specifically includes: randomly dividing the data samples in the data set into a first training set and a first test set according to a preset quantity ratio.
[0060] The data samples in the patient medical data set are randomly divided into a training set and a test set according to a preset ratio. The preset ratio can be set based on experience. In the present invention, the preset ratio is selected as 7:3, that is, 70% of the data samples in the data set are divided into a training set, and 30% of the data samples are divided into a test set. The specific number of data samples may not be specifically limited. As a preferred embodiment, the divided training set contains 7,600 data samples, and the test set contains 3,200 data samples.
[0061] However, dividing the training set and the test set based on the principle of random allocation may result in similar or identical data features in the two sets, which can easily lead to data leakage and reduce the generalization ability of the model.
[0062] As a preferred implementation, the feature similarity between the first data sample in the first training set and the second data sample in the first test set is calculated; if the feature similarity is greater than a preset threshold, it is determined whether the first data sample and the second data sample are from the same data source, and if so, the second data sample is transferred to the first training set; if not, the determination of the next data sample is continued until all data samples are processed.
[0063] When the number of data samples in the patient medical data set is small and the data dimension is not high, data oversampling can be used to expand the data samples of the data set. However, data oversampling needs to be generated based on the data samples in the original data set. The similarity or repetition of the newly generated data samples may be too high, which will lead to overfitting of the model; in addition, the data samples generated by oversampling may be mistakenly included in the test set, resulting in invalid evaluation of the generated data. When the number of a certain type of data in the data set is much higher than other types, resulting in data imbalance, undersampling can be used to reduce the number of data samples of the excessively high category to achieve a balance in the number of data samples of each category. However, the data samples deleted by data undersampling may be the data samples that the training model relies on most, which may lead to overfitting of the training model and cause data leakage risks.
[0064] As a preferred implementation, the data samples in the data set are randomly divided into a second training set and a second test set according to a preset quantity ratio;
[0065] Performing data balancing on the data samples of the second training set to obtain a third training set;
[0066] Calculate the feature similarity between the third data sample in the third training set and the fourth data sample in the second test set;
[0067] If the feature similarity is greater than a preset threshold, it is determined whether the third data sample and the fourth data sample are from the same data source. If so, the fourth data sample is transferred to the third training set. If not, the determination of the next data sample is continued until all data samples are processed. The data balancing includes data expansion through data oversampling and data balancing through data undersampling.
[0068] The specific implementation steps are:
[0069] Divide the patient medical data set D into the second training set D T and the second test set D E ;
[0070] For the second training set D T Generate a third training set D based on the oversampling or undersampling function TG ;
[0071] Calculate the third training set D TG Data sample X i With the second test set D E Data sample X j The feature similarity of X i ={X i1 , X i2 , ..., X im}, X j ={X j1 , X j2 , ..., X jn}, calculate the feature similarity of each feature in Xi and each feature in Xj in turn. If X i There are m*80% features in X j The similarity of the features in is greater than 90%, or X j There are n*80% features in X i If the similarity of the features in is greater than 90%, then the data sample X i With data sample X j resemblance;
[0072] Judgment data sample X i Dataset and data sample X j Whether the data sets of X come from the same public database or from the same IP address, if so, then the data sample X i With data sample X j From the same data source, the data sample X jTransfer to the third training set D TG If not, no transfer processing is performed until the third training set D is processed. TG All original data samples.
[0073] Step S300: constructing a dual-path parallel convolutional neural network model, and performing training based on the training set to obtain a trained convolutional neural network model;
[0074] As a preferred implementation, the training based on the training set specifically includes:
[0075] The text feature sequence and the mixed feature sequence of the data samples in the training set are respectively encoded to obtain a first encoded representation and a second encoded representation, and the first encoded representation and the second encoded representation are concatenated to obtain a third encoded representation, which are respectively used as inputs of the two-way parallel convolutional neural network model to train the two-way parallel convolutional neural network model.
[0076] The encoding process specifically includes respectively encoding the text feature sequence and the mixed feature sequence into vectors, and concatenating the two vectors to obtain a vector encoding representation of the patient data sample for model training.
[0077] It should be noted that the present invention constructs a patient sensitive information detection model based on a convolutional neural network structure, and the model is composed of two parallel one-dimensional residual convolutional neural networks; common convolutional neural networks are used to identify two-dimensional or three-dimensional image data. The one-dimensional residual convolutional neural network in the present invention is improved based on the VGG network, and the convolution layer and the pooling layer are replaced with one-dimensional convolution and one-dimensional pooling, and the residual network is introduced to obtain a one-dimensional residual convolutional neural network model. For details of the model structure, please refer to Figure 2 .
[0078] As a preferred embodiment, the sensitive information detection model includes a first neural network model and a second neural network model, and the first neural network model and the second neural network model are both one-dimensional residual convolutional neural networks, wherein the first neural network includes 8 convolutional layers, 4 maximum pooling layers and 3 residual modules; the second neural network includes 8 convolutional layers, 4 average pooling layers and 3 residual modules; the residual modules are connected in a jump manner, and are additively connected to the input after two convolutional layers. This connection method can avoid the gradient descent problem in network training and improve network performance. At the same time, a batch normalization layer is set after each convolutional layer to improve the training efficiency of the model. In order to improve the performance of the model, the weights of all network layers of the first neural network and the second neural network come from the pre-trained network, wherein the weights of the first neural network are set to be fixed during the training process, and the weights of the second neural network are fine-tuned during the training period, wherein the pre-trained network can be a transfer learning network. It should be noted that the weights of the fully connected layer and the Softmax layer need to be fine-tuned during the model training process.
[0079] The vector encoding representation of the obtained patient data samples is used as the input of the first neural network and the second neural network respectively. The maximum pooling feature output by the first neural network and the average pooling feature output by the second neural network are weightedly fused to obtain the fused feature. The fused feature is input into the fully connected layer, and Softmax regression is introduced as a classifier to output the probability of the corresponding sample corresponding to each category label.
[0080] The cross-over loss function is used in the forward propagation process, and the calculation formula is as follows:
[0081]
[0082] N is the total number of training samples, y n is the actual category of the sample, The model predicts the category of the sample. During the training process, the gradient is back-propagated to each parameter, and the Adam algorithm is used to optimize and adjust the network weights and learning rate. The weights are updated using the following formula: N =W O +LR*GV, where W N is the updated weight, W O is the weight before updating, LR is the learning rate, GV is the gradient vector, and the iteration termination condition can be set to reach the preset number of iterations or model convergence.
[0083] Step S400: Evaluate the convolutional neural network model based on the test set.
[0084] As a preferred implementation, the test data set is input into the trained detection model, and the four indicators of accuracy, precision, recall rate and F1 value are calculated for evaluation.
[0085] Accuracy, that is, the percentage of correct predictions in the total samples, is calculated as:
[0086]
[0087] Precision, that is, the probability of actually being a positive sample among all samples predicted to be positive, is calculated as:
[0088]
[0089] Recall rate, that is, the probability of being predicted as a positive sample among the samples that are actually positive, is calculated as:
[0090]
[0091] F1 score (F1-Score), F1 score takes into account both precision and recall, making them both reach the highest and achieve a balance. The F1 score expression is:
[0092]
[0093] In one embodiment, during the model training process, when the training rounds reach 10 times, the model has reached a converged state, and after the model converges, the accuracy of the training set reaches 100%, and the accuracy of the test set reaches 99.54%, and there is no overfitting phenomenon, indicating that the model exhibits good performance. Experimental results show that the accuracy of the model is 99.54%, the precision is 99.61%, the recall rate is 99.42%, and the F1 value is 99.49%.
[0094] As a preferred implementation, the data set is divided into a training set, a test set and a validation set. The model is validated by cross-validation based on the validation set. After the validation is passed, the test set is input into the trained model to evaluate the performance of the model, and then step S500 is executed.
[0095] Step S500: input the patient data to be detected into the trained convolutional neural network model to obtain sensitive information leakage detection results.
[0096] The patient data to be tested is input into the trained convolutional neural network model after feature extraction and normalization for disease type detection. If the corresponding disease type is output, it indicates that sensitive information of the patient's own health status has been leaked, and an early warning message is issued. If no specific disease type is output, it means that the patient data has not reached the standard for judging leakage and the patient's sensitive information has not been leaked.
[0097] At this point, this embodiment is completed.
[0098] In another embodiment, a sensitive information leakage detection device based on deep learning is also provided, the device comprising:
[0099] A data set acquisition module is used to acquire patient medical data sets and perform data preprocessing;
[0100] A data set partitioning module is used to partition the preprocessed data set into a training set and a test set;
[0101] A detection model building module is used to build a dual-path parallel convolutional neural network model, and perform training based on the training set to obtain a trained convolutional neural network model;
[0102] The sensitive information detection module is used to input the data information to be detected into the trained convolutional neural network model to obtain the sensitive information leakage detection result.
[0103] The embodiment of the present application also provides a terminal device, which includes the sensitive information leakage detection device based on deep learning. The terminal device may include a processor and a memory storing computer program instructions. Specifically, the above-mentioned processor may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiment of the present application. The memory may include a large-capacity memory for data or instructions.
[0104] By way of example and not limitation, the memory may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. In appropriate cases, the memory may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory may be inside or outside the integrated gateway disaster recovery electronic device. In a specific embodiment, the memory is a non-volatile solid-state memory.
[0105] In a particular embodiment, the memory includes a read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM) or a flash memory or a combination of two or more of these. The memory may include a read-only memory (ROM), a random access memory (RAM), a disk storage medium electronic device, an optical storage medium electronic device, a flash memory electronic device, an electrical, optical or other physical / tangible memory storage electronic device. Therefore, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory electronic device) encoded with software including computer executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of the present disclosure. The processor implements any of the sensitive information leakage detection methods based on deep learning in the above-mentioned embodiments by reading and executing the computer program instructions stored in the memory.
[0106] In one example, the terminal device may also include a communication interface and a bus. Among them, the processor, memory, and communication interface are connected and communicate with each other through a bus. The communication interface is mainly used to realize the communication between each module, device, unit and / or electronic device in the embodiment of the present application. The bus includes hardware, software or both, and the components of the online data flow billing electronic device are coupled to each other. For example, but not limited to, the bus may include an accelerated graphics port (AGP) or other graphics bus, an enhanced industrial standard architecture (EISA) bus, a front-end bus (FSB), a hypertransport (HT) interconnection, an industrial standard architecture (ISA) bus, an infinite bandwidth interconnection, a low pin count (LPC) bus, a memory bus, a micro channel architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standard association local (VLB) bus or other suitable buses or a combination of two or more of these. In appropriate cases, the bus may include one or more buses. Although the embodiment of the present application describes and shows a specific bus, the present application considers any suitable bus or interconnection.
[0107] In addition, in combination with the sensitive information leakage detection method based on deep learning in the above embodiments, the embodiments of the present application may provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the sensitive information leakage detection methods based on deep learning in the above embodiments is implemented.
[0108] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.
[0109] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory electronic devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.
[0110] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.
[0111] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.
[0112] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.
[0113] To sum up, in the embodiments of the present invention, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A sensitive information leakage detection method based on deep learning, characterized in that: The method comprises the following steps: Obtain patient medical data sets and perform data preprocessing; Divide the preprocessed data set into training set and test set; Constructing a dual-path parallel convolutional neural network model, and performing training based on the training set to obtain a trained convolutional neural network model; The patient data information to be detected is input into the trained convolutional neural network model to obtain the sensitive information leakage detection result.
2. According to claim 1, a sensitive information leakage detection method based on deep learning is characterized in that: Also includes: The patient medical data set includes text data and image data; The text data includes the patient's name, ID, age, education level, and diagnosis information; The image data includes an MRI image, a CT image or an ultrasonic image of a patient's lesion area.
3. According to claim 1, a sensitive information leakage detection method based on deep learning is characterized in that: The data preprocessing includes: Data cleaning is used to remove features that exist in each data sample of the data set and are not related to the leakage detection result.
4. According to claim 1, a sensitive information leakage detection method based on deep learning is characterized in that: The step of dividing the preprocessed data set into a training set and a test set specifically includes: The data samples in the data set are randomly divided into a first training set and a first test set according to a preset quantity ratio; Calculate the feature similarity between the first data sample in the first training set and the second data sample in the first test set; If the feature similarity is greater than a preset threshold, it is determined whether the first data sample and the second data sample come from the same data source. If so, the second data sample is transferred to the first training set. If not, the judgment of the next data sample is continued until all data samples are processed.
5. According to claim 1, a sensitive information leakage detection method based on deep learning is characterized in that: The step of dividing the preprocessed data set into a training set and a test set specifically includes: The data samples in the data set are randomly divided into a second training set and a second test set according to a preset quantity ratio; Performing data balancing on the data samples of the second training set to obtain a third training set; Calculate the feature similarity between the third data sample in the third training set and the fourth data sample in the second test set; If the feature similarity is greater than a preset threshold, determine whether the third data sample and the fourth data sample come from the same data source. If so, transfer the fourth data sample to the third training set. If not, continue to determine the next data sample until all data samples are processed.
6. According to the method for detecting sensitive information leakage based on deep learning according to claim 1, it is characterized in that: Before the step of dividing the preprocessed data set into a training set and a test set, the method further includes: Obtain text data and image data of data samples in a patient medical dataset; Performing feature extraction on the text data to obtain a text feature sequence; The image data is grayed out and features are extracted to obtain a mixed feature sequence.
7. A sensitive information leakage detection method based on deep learning according to claim 6, characterized in that: The training based on the training set specifically includes: The text feature sequence and the mixed feature sequence of the data samples in the training set are respectively encoded to obtain a first encoded representation and a second encoded representation, and the first encoded representation and the second encoded representation are concatenated to obtain a third encoded representation, which are respectively used as inputs of the two-way parallel convolutional neural network model to train the two-way parallel convolutional neural network model.
8. The method for detecting sensitive information leakage based on deep learning according to claim 1, characterized in that: After the step of extracting features from the text data to obtain a text feature sequence, the method further includes: reconstructing corresponding labels for text features with missing labels in the text feature sequence to obtain a reconstructed text feature sequence.
9. A sensitive information leakage detection device based on deep learning, characterized in that: The device includes: A data set acquisition module is used to acquire patient medical data sets and perform data preprocessing; A data set partitioning module is used to partition the preprocessed data set into a training set and a test set; A detection model building module is used to build a dual-path parallel convolutional neural network model, and perform training based on the training set to obtain a trained convolutional neural network model; The sensitive information detection module is used to input the data information to be detected into the trained convolutional neural network model to obtain the sensitive information leakage detection result.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the sensitive information leakage detection method based on deep learning as described in any one of claims 1 to 8.
