A teacher-student group privacy protection method and system for group perception

Abstract

The present invention discloses a group-perception-oriented teacher-student group privacy protection method and system, the method comprising: collecting the interactive activity data related to the on-campus learning of students and teachers and performing classification preprocessing, extracting the target data that needs to be protected for privacy; performing privacy labeling on the target data that needs to be protected for privacy, and preparing training samples for privacy identification; constructing a privacy identification model through a deep learning network, optimizing the hyperparameters of the deep learning network through an intelligent optimization algorithm, and training the privacy identification model through the training samples; performing privacy object identification on the target data to be protected through the trained privacy identification model, and performing fuzzy processing on the identified privacy objects. The present invention establishes a privacy identification model based on deep learning technology and intelligent optimization algorithms, which can automatically identify and shield the privacy of the learning interaction data of teachers and students.

Description

Technical Field

[0001] The present invention belongs to the technical field of educational collaborative learning, and specifically relates to a teacher-student group privacy protection method and system oriented to group perception. Background Art

[0002] In recent years, crowd sensing technology has made rapid progress in the field of Computer Supported Collaborative Learning (CSCL), becoming one of the effective means to reduce collaboration costs, improve collaboration efficiency and reduce collaboration conflicts, and providing a new method to promote interaction between learners.

[0003] The main idea of ​​mobile group perception technology is to use the most common users carrying mobile terminal devices as perception nodes, connect users through the mobile Internet, enable users to collaborate indirectly, realize the allocation of perception tasks and the collection of perception data, and thus complete large-scale, multi-dimensional, and complex reality perception tasks. In the process of collaborative learning, by collecting relevant data of teachers and students, mutual learning, content sharing, and peer evaluation are carried out to carry out learning assessments in various teaching environments. For example, as a way of collaborative learning, peer evaluation focuses on the evaluation of learners' learning process, which helps to improve learners' critical thinking ability, enhance learning motivation, and enhance social presence. Traditional learning evaluation often only focuses on learning results and ignores the learning process, while peer evaluation can focus on students' learning process. Students are both evaluators and learners. Studies have confirmed that peer evaluation helps to improve learners' professional knowledge and abilities, increase communication and interaction between teachers and students, and improve teaching standards.

[0004] However, in the process of collaborative learning such as peer evaluation, a large amount of teacher and student data needs to be collected through group sensing technology, which inevitably involves concerns about privacy protection. If we do not pay attention to privacy protection issues, on the one hand, it will not comply with the requirements of relevant laws and regulations, and on the other hand, it will reduce the sense of participation of teachers and students and affect the effectiveness of collaborative learning. Summary of the invention

[0005] In view of this, the present invention proposes a teacher-student group privacy protection method and system oriented to group perception, which are used to solve the problem of improper privacy protection in the process of collaborative learning.

[0006] In a first aspect, the present invention provides a group-aware teacher-student group privacy protection method, the method comprising:

[0007] Collect students' and teachers' interactive activity data related to on-campus learning and perform classification preprocessing to extract target data that needs privacy protection;

[0008] Privacy-label the target data that needs to be protected, and create training samples for privacy identification;

[0009] Constructing a privacy identification model through a deep learning network, optimizing hyperparameters of the deep learning network through an improved optimization algorithm, and training the privacy identification model through the training samples;

[0010] The trained privacy identification model is used to identify the privacy objects in the target data to be protected, and the identified privacy objects are fuzzified.

[0011] On the basis of the above technical solution, preferably, the target data that needs to be privacy protected includes image data, text data, audio data, and video data.

[0012] Based on the above technical solution, preferably, the step of performing privacy labeling on target data that needs privacy protection and preparing training samples for privacy identification specifically includes:

[0013] According to the type of target data that needs to be protected, the privacy objects are selected and the privacy categories are marked;

[0014] For image data, the included privacy objects are marked with annotation boxes and the privacy categories are recorded. For text data, the fields corresponding to the included privacy objects are recorded and the privacy categories are marked.

[0015] For audio data, perform audio recognition on the audio data and translate it into text data, extract the fields corresponding to the privacy objects in the corresponding text data and mark the privacy categories;

[0016] For video data, video frame sequences are extracted at preset time intervals, privacy objects contained in the video frame sequences are annotated with annotation boxes, and privacy categories are recorded.

[0017] On the basis of the above technical solution, preferably, the optimization of the hyper parameters of the deep learning network by the improved optimization algorithm specifically includes:

[0018] Initializing a population of a transient search optimization algorithm, each search agent in the population represents a set of hyperparameter combinations of a deep learning network, wherein the hyperparameter combination includes a penalty coefficient;

[0019] Set search range boundaries and maximum number of iterations;

[0020] Calculate the thermal resistance coefficient of the search agent based on the current iteration number;

[0021] Based on the thermal resistance coefficient, updating the position of the search agent;

[0022] Evaluate the fitness value of each search agent according to the fitness function and update the optimal fitness value;

[0023] Determine whether the end condition is met. If so, output the optimal position as the global optimal solution. Otherwise, recalculate the thermal resistance coefficient, the position of the search agent, and the fitness value until the end condition is met.

[0024] On the basis of the above technical solution, preferably, the updating of the position of the search agent based on the thermal resistance coefficient specifically includes:

[0025] When updating the position of the search agent, two random numbers r1 and r2 between 0 and 1 are generated to balance the exploration and development process, and an arithmetic optimization operator is introduced to update the position. The specific position update formula is:

[0026]

[0027] Among them, Y l , Y l+1 are the exploration positions of the lth and l+1th iterations respectively, Y l b is the optimal exploration position after l iterations, [LB, UB] is the boundary of the exploration range, MOP(l) = 1-(l-1) 1 / α / L 1 / α, α is a sensitive parameter, L is the maximum number of iterations, and u is a control parameter; C l , T is the thermal resistance coefficient.

[0028] Based on the above technical solution, preferably, the fitness function is the minimum root mean square error of the deep learning network.

[0029] Based on the above technical solution, preferably, the step of identifying the privacy object of the target data to be protected by using the trained privacy identification model and performing fuzzy processing on the identified privacy object specifically includes:

[0030] Obtain the interaction data between students, between teachers, or between teachers and students from the sending client and pre-process it according to the data type;

[0031] Input the pre-processed interactive data sent to the client into the privacy identification model, and output the identified privacy object location, field and category;

[0032] Performing fuzzy processing on the location or field of the privacy object in the corresponding interaction data;

[0033] The obfuscated interaction data of the sending client is sent to the receiving client to complete the privacy protection of the interaction data.

[0034] In a second aspect, the present invention discloses a group-aware teacher-student group privacy protection system, the system comprising:

[0035] Data collection module: collects interactive activity data related to on-campus learning of students and teachers;

[0036] Data processing module: Classify and preprocess interactive activity data to extract target data that needs privacy protection; perform privacy labeling on target data that needs privacy protection and create training samples for privacy identification;

[0037] Model training module: constructing a privacy identification model through a deep learning network, optimizing the hyperparameters of the deep learning network through an improved optimization algorithm, and training the privacy identification model through the training samples;

[0038] Privacy processing module: Use the trained privacy identification model to identify the privacy objects of the target data to be protected, and perform fuzzy processing on the identified privacy objects.

[0039] In a third aspect of the present invention, an electronic device is disclosed, comprising: at least one processor, at least one memory, a communication interface and a bus;

[0040] Wherein, the processor, memory, and communication interface communicate with each other via the bus;

[0041] The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method described in the first aspect of the present invention.

[0042] According to a fourth aspect of the present invention, a computer-readable storage medium is disclosed, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to implement the method described in the first aspect of the present invention.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1) The present invention constructs a privacy identification model through a deep learning network, and optimizes the hyperparameters of the deep learning network through an improved optimization algorithm. It can perform privacy identification and processing on various interactive types of data such as image data, text data, audio data, and video data, thereby expanding the processing scope of privacy protection;

[0045] 2) The present invention integrates the arithmetic optimization operator with the transient search optimization algorithm, expands the search space of the transient search optimization algorithm with a certain random probability through the arithmetic optimization operator, and accelerates the local search with a certain probability, thereby improving the optimization ability of the exploration and development process, thereby optimizing the hyperparameters of the deep learning network and improving the recognition accuracy of the privacy recognition model;

[0046] 3) The present invention can automatically identify and shield the privacy of the learning interaction data of teachers and students, and blur the private part during external display or interactive sharing, so that the sharer has no worries in the collaborative learning process, enhances the enthusiasm of teachers and students to participate, and is conducive to improving the learning effect of collaborative learning such as peer evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] 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.

[0048] Figure 1 The present invention is a flow chart of a group-aware teacher-student group privacy protection method. DETAILED DESCRIPTION

[0049] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1 The present invention proposes a group-aware teacher-student group privacy protection method, the method comprising:

[0051] S1. Collect the interactive activity data related to on-campus learning of students and teachers, perform classification preprocessing, and extract the target data that needs to be protected for privacy;

[0052] First, collect the interactive activity data related to on-campus learning of students and teachers, including interactive data between students, between teachers, and between students and teachers through mobile terminals, computers and other clients. Classify the interactive data according to data type, such as image data, text data, audio data, and video data, and pre-process these data to filter out target data that needs to be protected, such as data containing sensitive content that is not convenient to share directly with the outside world.

[0053] S2. Privacy label the target data that needs to be protected and create training samples for privacy identification;

[0054] Specifically, according to the type of the extracted target data that needs to be protected, the privacy objects are selected and the privacy categories are marked;

[0055] For image data, the included privacy objects are marked with annotation boxes and the privacy categories are recorded. For text data, the fields corresponding to the included privacy objects are recorded and the privacy categories are marked.

[0056] For audio data, perform audio recognition on the audio data and translate it into text data, extract the fields corresponding to the privacy objects in the corresponding text data and mark the privacy categories;

[0057] For video data, video frame sequences are extracted at preset time intervals, privacy objects contained in the video frame sequences are annotated with annotation boxes, and privacy categories are recorded.

[0058] In general, image data, text data, audio data, and video data can be converted into two major types of data. Image data and video data are ultimately converted into image data, which can be processed for privacy protection through image target recognition; while text data and audio data are ultimately converted into text data, which can be processed for privacy protection through text content recognition. Based on these two types of data, two privacy recognition models can be trained to perform privacy recognition and privacy fuzziness processing respectively. One privacy recognition model processes image data and video data, and one privacy recognition model processes text data and audio data, ultimately processing various types of privacy data to the greatest extent possible.

[0059] S3. constructing a privacy identification model through a deep learning network, optimizing hyperparameters of the deep learning network through an optimization algorithm, and training the privacy identification model through the training samples;

[0060] The deep learning network of the present invention can be various neural network models such as CNN, RNN, DNN, PNN, generative adversarial network, etc. In machine learning or deep learning, hyperparameters are parameters whose values ​​are set before starting the learning process, rather than parameter data obtained through training. Generally, it is necessary to optimize the hyperparameters and select a set of optimal hyperparameters for the learning machine to improve the performance and effect of learning.

[0061] Taking the generative adversarial network WGAN for privacy identification of image data and audio data as an example, the generative adversarial network consists of a generative model and a discriminative model. Among them, the generative model G captures the distribution of sample data, and the discriminative model D is a binary classifier. The input of the generative model is some random noise z that obeys a simple distribution. If the input is image data, the output is a generated image with the same size as the training image. When the generated sample is input to the discriminative model D, D is expected to output a low probability. For the generative model G, it is necessary to try to deceive the discriminative model D so that the discriminative model outputs a high probability, thereby forming competition and confrontation. In order to achieve better results, it is necessary to find the optimal penalty term and penalty coefficient. Therefore, the hyperparameters of the generative adversarial network can be optimized through optimization algorithms.

[0062] The present invention optimizes the hyper parameters of the deep learning network through an intelligent optimization algorithm, specifically including:

[0063] S31, initializing a population of a transient search optimization algorithm, wherein each search agent in the population represents a set of hyperparameter combinations of a deep learning network, wherein the hyperparameter combination includes a penalty coefficient;

[0064] S32, setting the search range boundary [LB, UB] and the maximum number of iterations L;

[0065] S33, calculating the thermal resistance coefficient C of the search agent based on the current number of iterations l , T, the thermal resistance coefficient C l The calculation method of T is the same as that in the transient search optimization algorithm.

[0066] S34, updating the position of the search agent based on the thermal resistance coefficient;

[0067] When updating the position of the search agent, two random numbers r1 and r2 between 0 and 1 are generated to balance the exploration and development process, and an arithmetic optimization operator is introduced to update the position. The specific position update formula is:

[0068]

[0069] Among them, Y l , Y l+1 are the exploration positions of the lth and l+1th iterations respectively, Y l b is the optimal exploration position after l iterations, [LB, UB] is the boundary of the exploration range, MOP(l) = 1-(l-1) 1 / α / L 1 / α , α is a sensitive parameter, L is the maximum number of iterations, and u is a control parameter; C l , T is the thermal resistance coefficient.

[0070] S35. Evaluate the fitness value of each search agent according to the fitness function and update the optimal fitness value; the fitness function is to minimize the root mean square error of the deep learning network.

[0071] S36, determine whether the end condition is met, if so, output the optimal position as the global optimal solution, otherwise, return to step S33 to re-calculate the thermal resistance coefficient, the position of the search agent and the fitness value until the end condition is met.

[0072] The present invention integrates the arithmetic optimization operator with the transient search optimization algorithm, expands the search space of the transient search optimization algorithm with a certain random probability through the arithmetic optimization operator, and accelerates the local search with a certain probability, thereby improving the optimization ability of the exploration and development process and achieving a balance between exploration and development. In this way, the hyperparameters of the deep learning network can be optimized, and the recognition accuracy of the privacy recognition model can be improved.

[0073] S4. Use the trained privacy identification model to identify the privacy objects in the target data to be protected, and perform fuzzification on the identified privacy objects.

[0074] Obtain the interaction data between students, between teachers, or between teachers and students from the sending client and pre-process it according to the data type;

[0075] Input the pre-processed interactive data sent to the client into the privacy identification model, and output the identified privacy object location, field and category;

[0076] Performing fuzzy processing on the location or field of the privacy object in the corresponding interaction data;

[0077] The obfuscated interaction data of the sending client is sent to the receiving client to complete the privacy protection of the interaction data.

[0078] The privacy identification model established in the present invention based on deep learning technology and intelligent optimization algorithm can automatically identify and shield the privacy of the learning interaction data of teachers and students, and blur the private part during external display or interactive sharing, so that the sharer has no worries in the collaborative learning process, enhances the enthusiasm of teachers and students to participate, and is conducive to improving the learning effect of collaborative learning such as peer evaluation.

[0079] Corresponding to the above method embodiment, the present invention also proposes a teacher-student group privacy protection system oriented to group perception, the system comprising:

[0080] Data collection module: collects interactive activity data related to on-campus learning of students and teachers;

[0081] Data processing module: Classify and preprocess interactive activity data to extract target data that needs privacy protection; perform privacy labeling on target data that needs privacy protection and create training samples for privacy identification;

[0082] Model training module: constructing a privacy identification model through a deep learning network, optimizing the hyperparameters of the deep learning network through an improved optimization algorithm, and training the privacy identification model through the training samples;

[0083] Privacy processing module: Use the trained privacy identification model to identify the privacy objects of the target data to be protected, and perform fuzzy processing on the identified privacy objects.

[0084] The above system embodiments correspond to the method embodiments. Please refer to the method embodiments for a brief description of the system embodiments.

[0085] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus; wherein the processor, memory, and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the aforementioned method of the present invention.

[0086] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to implement all or part of the steps of the method described in the embodiment of the present invention. The storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk, and other media that can store program codes.

[0087] The system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be distributed to multiple network units. A person skilled in the art may select some or all of the modules according to actual needs to achieve the purpose of the solution of this embodiment without creative effort.

[0088] 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 spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A group-aware teacher-student group privacy protection method, characterized in that: The method comprises: Collect students' and teachers' interactive activity data related to on-campus learning and perform classification preprocessing to extract target data that needs privacy protection; Privacy-label the target data that needs to be protected, and create training samples for privacy identification; Constructing a privacy identification model through a deep learning network, optimizing hyperparameters of the deep learning network through an intelligent optimization algorithm, and training the privacy identification model through the training samples; The trained privacy identification model is used to identify the privacy objects in the target data to be protected, and the identified privacy objects are fuzzified. The method of optimizing the hyperparameters of the deep learning network by the intelligent optimization algorithm specifically includes: Initializing a population of a transient search optimization algorithm, each search agent in the population represents a set of hyperparameter combinations of a deep learning network, wherein the hyperparameter combination includes a penalty coefficient; Set search range boundaries and maximum number of iterations; Calculate the thermal resistance coefficient of the search agent based on the current iteration number; Based on the thermal resistance coefficient, updating the position of the search agent; Evaluate the fitness value of each search agent according to the fitness function and update the optimal fitness value; Determine whether the end condition is met. If so, output the optimal position as the global optimal solution. Otherwise, recalculate the thermal resistance coefficient, the position of the search agent, and the fitness value until the end condition is met. The updating of the position of the search agent based on the thermal resistance coefficient specifically includes: When updating the position of the search agent, two random numbers r1 and r2 between 0 and 1 are generated to balance the exploration and development process, and an arithmetic optimization operator is introduced to update the position. The specific position update formula is: Among them, Y l , Y l+1 are the exploration positions of the lth and l+1th iterations respectively, Y l b is the optimal exploration position after l iterations, [LB, UB] is the boundary of the exploration range, MOP(l) = 1-(l-1) 1 / α / L 1 / α , α is a sensitive parameter, L is the maximum number of iterations, and u is a control parameter; C l , T is the thermal resistance coefficient.

2. The teacher-student group privacy protection method for group perception according to claim 1 is characterized in that: The target data that needs to be privacy protected includes image data, text data, audio data, and video data.

3. The teacher-student group privacy protection method for group perception according to claim 2 is characterized in that: The step of labeling the target data that needs to be privacy protected and preparing training samples for privacy identification specifically includes: According to the type of target data that needs to be protected, the privacy objects are selected and the privacy categories are marked; For image data, the included privacy objects are marked with annotation boxes and the privacy categories are recorded. For text data, the fields corresponding to the included privacy objects are recorded and the privacy categories are marked. For audio data, perform audio recognition on the audio data and translate it into text data, extract the fields corresponding to the privacy objects in the corresponding text data and mark the privacy categories; For video data, video frame sequences are extracted at preset time intervals, privacy objects contained in the video frame sequences are annotated with annotation boxes, and privacy categories are recorded.

4. The teacher-student group privacy protection method for group perception according to claim 3 is characterized in that: The fitness function is to minimize the root mean square error of the deep learning network.

5. The teacher-student group privacy protection method for group perception according to claim 1 is characterized in that: The method of identifying the privacy objects of the target data to be protected by using the trained privacy identification model and performing fuzzy processing on the identified privacy objects specifically includes: Obtain the interaction data between students, between teachers, or between teachers and students from the sending client and pre-process it according to the data type; Input the pre-processed interactive data sent to the client into the privacy identification model, and output the identified privacy object location, field and category; Performing fuzzy processing on the privacy object position or field in the corresponding interaction data; The obfuscated interaction data of the sending client is sent to the receiving client to complete the privacy protection of the interaction data.

6. A group-aware teacher-student privacy protection system, which is implemented based on the group-aware teacher-student privacy protection method according to any one of claims 1 to 5, and is characterized in that: The system includes the following modules: Data collection module: collects interactive activity data related to on-campus learning of students and teachers; Data processing module: Classify and preprocess interactive activity data to extract target data that needs privacy protection; perform privacy labeling on target data that needs privacy protection and create training samples for privacy identification; Model training module: constructing a privacy identification model through a deep learning network, optimizing the hyperparameters of the deep learning network through an improved optimization algorithm, and training the privacy identification model through the training samples; Privacy processing module: Use the trained privacy identification model to identify the privacy objects of the target data to be protected, and perform fuzzy processing on the identified privacy objects.

7. An electronic device comprising: at least one processor, at least one memory, a communication interface, and a bus; Wherein, the processor, memory, and communication interface communicate with each other via the bus; The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing computer instructions, wherein the computer instructions enable the computer to implement the method according to any one of claims 1 to 5.

Images