CN114862636A - Financial intelligent teaching and privacy protection method - Google Patents
Financial intelligent teaching and privacy protection method Download PDFInfo
- Publication number
- CN114862636A CN114862636A CN202210471694.1A CN202210471694A CN114862636A CN 114862636 A CN114862636 A CN 114862636A CN 202210471694 A CN202210471694 A CN 202210471694A CN 114862636 A CN114862636 A CN 114862636A
- Authority
- CN
- China
- Prior art keywords
- model
- students
- teacher
- student
- teaching
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/20—Education
- G06Q50/205—Education administration or guidance
- G06Q50/2057—Career enhancement or continuing education service
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/06—Asset management; Financial planning or analysis
Abstract
The application provides a financial intelligent teaching and privacy protection method and system, comprising the following steps: establishing a block chain database to protect data privacy; the data stored in the block chain database identifies the class concentration degree of the student; judging whether the teacher class attendance performance and the teacher class attendance performance affect students or not based on the NLP technology; adjusting the reading amount of the electronic documents according to the classroom performance of students; the student can adjust the teaching plan of the teacher by mastering the condition of the electronic documents; protect the privacy of the teaching plan of the teacher and adjust the learning content of the students.
Description
Technical Field
The invention relates to the technical field of intelligent privacy protection, in particular to a financial intelligent teaching and privacy protection method.
Background
Nowadays, the phenomenon of order disorder often appears in classroom such as: the student leans down to play the mobile phone, the phenomenon is often related to poor learning condition of the student, the condition of the student can be influenced by whether the teaching condition of the teacher is good or not besides the factors of the student, and the interaction frequency of the teacher and the student is analyzed by extracting the speaking content of the teacher in teaching through a natural language processing technology, so that whether the teaching condition of the teacher is good or not is judged;
the system has the advantages that the system is combined with the learning condition of financial knowledge after a student class, the teaching condition of a teacher and the performance condition of a student class, the knowledge mastering degree of the student is judged to be influenced by the enthusiasm of independent learning of the student class or whether the teaching condition of the teacher is good, the problem that the teaching condition of a teacher is revealed to other users is solved through a differential privacy technology, and therefore the problems that the teacher is given data and revealed by a school, is removed and the like are solved.
If the students have the condition of poor knowledge mastering, whether the teacher needs to review the last content in the next lesson is also the core technical problem concerned by the system, the system realizes the individuation of course arrangement of the teacher by using the DRL technology, adjusts the time ratio of new content to old knowledge, and adjusts the learning plan of the students according to the course arrangement plan of the teacher, so that the daily learning of the students is more efficient, and the system better meets the requirements of giving lessons in the next lesson.
Disclosure of Invention
The invention provides a financial intelligent teaching and privacy protection method and a system, which mainly comprise the following steps:
establishing a block chain database to protect data privacy; the data stored in the block chain database identifies the class concentration degree of the student; judging whether the teacher class attendance performance and the teacher class attendance performance affect students or not based on the NLP technology; adjusting the reading amount of the electronic documents according to the classroom performance of students; the student can adjust the teaching plan of the teacher by mastering the condition of the electronic documents; protecting the privacy of teaching plans of teachers and adjusting learning contents of students;
further optionally, the establishing a blockchain database to protect data privacy includes:
establishing a block chain database for the performance of the students in the classroom, teaching information of the teachers and classroom learning data, such as: data such as financial courses and financial data are stored and a differential privacy encryption technology is used: the uncertainty of the result is added to the query result: by the Laplace mechanism: for adjacent data sets, giving a corresponding mapping function and adding Laplace noise into a result of user query; the student can obtain the financial information which the student wants to learn in the system database, and the privacy protection of the system can prevent other users from knowing the weak part of the study of the student. The method comprises the following steps: differential privacy uses a Laplace mechanism to realize privacy protection;
the differential privacy uses a Laplace mechanism to realize privacy protection, and further comprises:
and (3) assuming that the frame coefficient of the query result is based on the generalized Gauss-Laplace prior statistical distribution, and carrying out differential processing on the result through the distribution of Laplace by using maximum likelihood estimation and posterior mean estimation quantity.
Further optionally, the data stored to the blockchain database identifying student class concentration comprises:
according to the established database, storing the acquired image data into a block chain database, establishing a student face database and a cloud server, extracting face features of students and analyzing the concentration degree of financial classes of the students; the method comprises the following steps: using an MTCNN face recognition model; training an MTCNN model based on federated learning; judging the concentration degree of the student according to iris analysis; judging the classroom performance score of the student by using posture analysis;
the using the MTCNN face recognition model further comprises:
training the MTCNN, wherein the model mainly adopts three cascaded networks, namely P-Net, R-Net and O-Net, adopts the idea of adding a classifier into a candidate frame, and uses NMS, Soft-NMS and RRelu to construct a neural network; and the horizontal federal learning technology is used, different financial teaching institutions can learn through the horizontal federal, and the model is trained by using the data of each classroom: taking the image of the human face as input, and taking the recognition result as output; and training the model by using each data, and enabling the server to obtain a training result.
The training of the MTCNN model based on federated learning further comprises:
different financial classes are learned through horizontal federal, utilize the data of each college to train the model, and the result that the server will train: the parameters, gradient values, loss function values and the like of the model are returned to the server of each financial course institution, the method can effectively improve the training efficiency of the model, obtain a more robust deep learning model, and successfully solve the problem that the model cannot be well adapted to different application scenes.
The method for judging the concentration degree of the student according to iris analysis further comprises the following steps:
according to the MCNN model, the human face is aligned before the human face image is transmitted into the model, the 3D image of the target is selected, and the feature points of the adjacent frame images of the target are identified so as to align the space, wherein the human face alignment advanced ERT algorithm is used. And detecting and identifying the students by using the trained model, and judging whether the students keep being concentrated all the time by extracting the iris characteristics of the students and the characteristics of whether the iris characteristics deviate from a blackboard or a screen.
The use gesture analysis to judge student classroom performance score still includes:
identifying the student classroom behavior according to the established local feature identification convolutional neural network and using a DenseNet model: 3 DenseBlock, 32 convolution kernels of 11x11 and a maxpool pooling layer of 3x3 are used, 16 convolution kernels of 9x9 are used in the last four layers, a full-connection layer is a deep learning model formed by 4096 dimensions, the model extracts behavior characteristic pictures of students playing mobile phones with heads down as input, whether the students play mobile phones with heads down or not is output, and a corresponding neural network model is trained. The model identifies the frequency of the student leaning down to play the mobile phone and deducts the corresponding weight of the classroom expression score.
Further optionally, the determining whether the teacher's performance in class and the teacher's performance in class affect the student based on the NLP technology includes:
performing language segmentation by using a segmentation jieba function library by using NLP technology, and extracting whether a teacher has interactive words with students in the course of class
Establishing a BERT model: with teacher financial classroom utterances as input, pre-training with MLM and employing a deep two-way Transformer component: the input is a token sequence, the token sequence is firstly embedded to be called as a vector, then the vector is input to a neural network, the output is a vector sequence with the size of H, and an Attention mechanism is used. Extracting class interaction vocabularies by using a BERT model, comprising the following steps: student name, please answer, why, etc. The language database of the teacher speaking in class is input, the frequency of teacher interactive content is output, and the relevance of teacher interactive performance and student posture performance is analyzed through a machine learning Apriori correlation analysis algorithm.
Further optionally, the adjusting the reading amount of the electronic documents according to the classroom performance of the students comprises:
the system adjusts the reading quantity of electronic documents of student related knowledge through a traditional PCA algorithm according to the scoring result of classroom performance; the PCA algorithm calculates and analyzes different subject performances of the students by using a covariance matrix, a mode of inner product and base change of the matrix and a matrix diagonalization mathematical formula and performs dimension reduction analysis, intelligently analyzes course contents which need emphasis reinforcement of the students, and requires the students to learn contents which need reinforcement of subjects after class by combining with a well-established block chain database.
Further optionally, the adjusting the teaching plan of the teacher by the student for the situation of mastering the electronic documents includes:
the teacher's lecture plan is adjusted using the DRL technique: the system combines the learning condition and classroom performance condition of students for learning electronic documents, uses a differential privacy technology, and adopts a DRL deep learning model: establishing a DQN: 3 convolutional layers and two Dense layers, and initializing Memory D, wherein the capacity of the Memory D is N; and (3) circularly traversing 1 to M steps, wherein M is manually set, a reward function is established, and the recommended content of the teaching plan is simulated: the method simulates poor performance of students and the teaching performance of teachers during learning, and how to distribute review content and teaching duration of new knowledge points can reach higher reward value, so that an intelligent DRL model for financial courses is generated.
The system intelligently recommends the teacher's teaching plan by using the DRL model, and the recommended content reward value is higher than the time for adding old knowledge of the DRL model when the students have poor performance in learning and the teachers have poor performance, so that the recommended content is influenced.
The method comprises the following steps: setting specific parameters in the DQN;
the setting of specific parameters in the DQN further includes:
the agent, Action, award weight in DQN is set. Different learning conditions of students are set into different agents, each corresponding agent performs DQN simulation, different teaching contents of a teacher teaching are used as different actions, and each Action corresponds to different reward weights. And through continuously collecting the learning conditions of different students, continuous simulation training is carried out, so that a robust teaching plan recommendation model is generated.
A financial intelligent teaching and privacy protection method and system are characterized in that the system comprises:
and protecting the teaching plan privacy of the teacher by using a Laplace mechanism in a differential privacy algorithm. And (3) adding Laplace noise into the result of the user query by giving a corresponding mapping function, so that the query result has slight difference.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
1. improving the learning efficiency of a classroom, monitoring the habit of a student lowering his head to play a mobile phone 2, providing learning help for students with poor classroom performance 3. the learning situation of the student can intelligently adjust the teaching plan of a teacher 4. judging whether the classroom performance of the student is related to the teaching situation of the teacher 5. the privacy data of the teaching situation performance of the teacher are protected
[ description of the drawings ]
FIG. 1 is a flowchart of a financial intelligent lecture and privacy protection method according to the present invention.
[ detailed description ] embodiments
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.
FIG. 1 is a flow chart of a financial intelligent teaching and privacy protection method of the present invention. As shown in fig. 1, the method and system for financial intelligent lecture and privacy protection in this embodiment may specifically include:
establishing a block chain database to protect data privacy, including storing classroom student performance, teacher teaching information, financial courses and financial data, and adding uncertain privacy in query results by using a differential privacy encryption technology to protect classroom privacy: identifying the class attendance concentration degree of the student through the data stored in the block chain database; judging whether the teacher's lesson-taking performance and the teacher's lesson-taking performance influence students based on an NLP technology, extracting whether words interacting with the students exist in the lesson-taking process of a teacher by using the NLP technology, and judging the performance of the teacher; adjusting the reading amount of the electronic documents according to the classroom performance of students; the teaching plan of the teacher is adjusted by the condition that the student grasps the electronic documents; protecting the privacy of teaching plans of teachers and adjusting the learning contents of students; adjusting a teaching plan of a teacher by using a DRL reinforcement learning technology, and establishing a DRL deep learning model by combining the learning condition and classroom expression condition of electronic documents learned by students and the difference privacy technology; the DRL deep learning model further comprises the following steps of establishing a DQN: adopting a neural network with 3 convolutional layers and two Dense layers, initializing a Memory D to enable the capacity of the Memory D to be N; 1 to M steps are circularly traversed, wherein M is a parameter; establishing a reward function, and simulating the recommended content of the teaching plan; the simulation comprises the steps of simulating poor performance of students in learning and poor performance of teachers in teaching, and distributing review content and teaching duration of new knowledge points to enable the review content and the teaching duration to reach a higher reward value, so that an intelligent DRL model for financial courses is generated; the DRL model is used for intelligently recommending the teaching plan of the teacher, when the students have poor performance in learning and the teachers have poor performance in teaching, the time for adding the old knowledge of the DRL model accounts for a higher recommended content reward value, so that the recommended content is influenced; further comprising: setting parameters in the DQN; the setting of specific parameters in the DQN further includes: setting an agent, an Action and an award weight in the DQN; setting different learning conditions of students into different agents, carrying out DQN simulation on each corresponding agent, taking different teaching contents of teaching of a teacher as different actions, and enabling each Action to correspond to different rewarding weights; and through continuously collecting the learning conditions of different students, carrying out simulation training for many times, thereby generating a robust teaching plan recommendation model.
Step 101, establishing a block chain database to protect data privacy.
Establishing a block chain database for the performance of the students in the classroom, teaching information of the teachers and classroom learning data, such as: data such as financial courses and financial data are stored and a differential privacy encryption technology is used: the uncertainty of the result is added to the query result: by the Laplace mechanism: for adjacent data sets, giving a corresponding mapping function and adding Laplace noise into a result of user query; the student can obtain the financial information which the student wants to learn in the system database, and the privacy protection of the system can prevent other users from knowing the weak part of the study of the student: for example, a student cannot be proficient in the course of the stock K line, and needs to look up a large amount of electronic books with the contents, and the books are classified in class 1, and if the student inquires the classification of the books in a certain period of time, the result has slight differences: the 1 st class becomes the 2 nd class.
Differential privacy uses the Laplace mechanism to achieve privacy protection.
And (3) assuming that the frame coefficient of the query result is based on the generalized Gauss-Laplace prior statistical distribution, and carrying out differential processing on the result through the distribution of Laplace by using maximum likelihood estimation and posterior mean estimation quantity.
Step 102, the data stored in the blockchain database identifies the class concentration of the student.
And storing the acquired image data into a block chain database according to the established database, establishing a student face database and a cloud server, extracting the face characteristics of the students and analyzing the concentration degree of the financial classroom of the students.
MTCNN face recognition model was used.
Training the MTCNN, wherein the model mainly adopts three cascaded networks, namely P-Net, R-Net and O-Net, adopts the idea of adding a classifier into a candidate frame, and uses NMS, Soft-NMS and RRelu to construct a neural network; and the horizontal federal learning technology is used, different financial teaching institutions can learn through the horizontal federal, and the model is trained by using the data of each classroom: taking the image of the human face as input, and taking the recognition result as output; the model is trained using the respective data, and the server returns the training results, for example, parameters, gradient values, loss function values, and the like of the model to the respective servers. And performing face recognition by using the model, wherein the model judges whether the two images are similar by learning the mapping of the central points in the Euclidean space of the images and comparing the characteristic vectors corresponding to the two images so as to determine whether the two images are the same face.
The MTCNN model is trained based on federated learning.
Different financial classes are learned through horizontal federal, utilize the data of each college to train the model, and the result that the server will train: the parameters, gradient values, loss function values and the like of the model are returned to the server of each financial course institution, the method can effectively improve the training efficiency of the model, obtain a more robust deep learning model, and successfully solve the problem that the model cannot be well adapted to different application scenes, such as: because classroom size, lighting, etc. issues do not allow the model to be applied to a number of different scenarios.
The concentration degree of the student is judged according to iris analysis.
According to the MCNN model, the human face is aligned before the human face image is transmitted into the model, the 3D image of the target is selected, and the feature points of the adjacent frame images of the target are identified so as to align the space, wherein the human face alignment advanced ERT algorithm is used. The trained model is used for detecting and identifying the student, whether the student always keeps being concentrated is judged by extracting the iris characteristic of the student whether the iris characteristic deviates from a blackboard or a screen, for example, when the eyes of the student deviate from a blackboard camera, the system intelligently judges that the student does not keep being concentrated.
Student classroom performance scores are judged using posture analysis.
Identifying the student classroom behavior according to the established local feature identification convolutional neural network and using a DenseNet model: 3 DenseBlock, 32 convolution kernels of 11x11 and a maxpool pooling layer of 3x3 are used, 16 convolution kernels of 9x9 are used in the last four layers, a full-connection layer is a deep learning model formed by 4096 dimensions, the model extracts behavior characteristic pictures of students playing mobile phones with heads down as input, whether the students play mobile phones with heads down or not is output, and a corresponding neural network model is trained. The model identifies the frequency of the student leaning down to play the mobile phone and deducts the corresponding weight of the classroom expression score. For example, the queen frequently plays the mobile phone in the financial fund class, after the model judges that the queen is down to play the mobile phone, the performance score of the student financial fund class is deducted, the score is returned to the blockchain server, the student is judged not to be kept attentive, and the relevant learning content of the student after class is automatically adjusted.
And 103, judging whether the teacher class attendance performance and the teacher class attendance performance affect students or not based on the NLP technology.
The method comprises the steps of using an NLP technology to perform language segmentation by using a segmentation jieba function library, extracting words whether a teacher interacts with students in the course of class, such as the names, answers, reasons and other interactive words of the students, and analyzing and judging whether the teacher simply speaks PPT.
Building a BERT model: with teacher financial classroom utterances as input, pre-training with MLM and employing a deep two-way Transformer component: the input is a token sequence, the token sequence is firstly embedded to be called as a vector, then the vector is input to a neural network, the output is a vector sequence with the size of H, and an Attention mechanism is used. Extracting class interaction vocabularies by using a BERT model, comprising the following steps: student name, please answer, why, etc. The language database of the teacher speaking in class is input, the frequency of teacher interactive content is output, and the relevance of teacher interactive performance and student posture performance is analyzed through a machine learning Apriori correlation analysis algorithm. For example, if the number of words of the BERT model for identifying the classroom interaction attribute is small, the interaction frequency between the teacher and the students is low, and if the students often hold down to play the mobile phones, the association degree between the behavior of the mobile phones played by the students and the classroom performance of the teacher is high, and the enthusiasm of the students caused by the classroom performance of the teacher is judged.
And step 104, adjusting the reading amount of the electronic documents according to the classroom performance of the students.
The system adjusts the reading quantity of electronic documents of student related knowledge through a traditional PCA algorithm according to the scoring result of classroom performance; the PCA algorithm calculates and analyzes different subject performances of the students by using a covariance matrix, a mode of inner product and base change of the matrix and a matrix diagonalization mathematical formula and performs dimension reduction analysis, intelligently analyzes course contents which need emphasis reinforcement of the students, and requires the students to learn contents which need reinforcement of subjects after class by combining with a well-established block chain database. For example, the system analyzes that the performance score of the class of the stock K line of Mingming classmates is low and the concentration degree of learning is not high according to the PCA algorithm, and provides related electronic books of the stock K line of a library for reading after the class of Mingming classmates.
In step 105, the student adjusts the lecture plan of the teacher in order to grasp the electronic document.
The teacher's lecture plan is adjusted using the DRL technique: the system combines the learning condition and classroom expression condition of students for learning electronic documents, uses a differential privacy technology, and adopts a DRL deep learning model: establishing a DQN: 3 convolutional layers and two Dense layers, and initializing Memory D, wherein the capacity of the Memory D is N; and (3) circularly traversing 1 to M steps, wherein M is manually set, a reward function is established, and the recommended content of the teaching plan is simulated: the method simulates poor performance of students and the teaching performance of teachers during learning, and how to distribute review content and teaching duration of new knowledge points can reach higher reward value, so that an intelligent DRL model for financial courses is generated.
The system intelligently recommends the teacher's teaching plan by using the DRL model, and the recommended content reward value is higher than the time for adding old knowledge of the DRL model when the students have poor performance in learning and the teachers have poor performance, so that the recommended content is influenced. For example, according to the learning condition of the students in the previous lesson and the poor state of the lessons of the teacher, the system uses the DRL model to increase the teaching time of the review content in the next lesson, so as to adjust the time ratio of the new content to the old knowledge, for example, the time ratio of the review time of the old knowledge is higher.
Specific parameters in DQN are set.
The agent, Action, award weight in DQN is set. Different learning conditions of students are set into different agents, each corresponding agent performs DQN simulation, different teaching contents of a teacher teaching are used as different actions, and each Action corresponds to different reward weights. And through continuously collecting the learning conditions of different students, continuous simulation training is carried out, so that a robust teaching plan recommendation model is generated. For example, after the model is iterated continuously, the recommendation content of the recommendation model is more intelligent and scientific.
And step 106, protecting the privacy of the teaching plan of the teacher and adjusting the learning content of the students.
And protecting the teaching plan privacy of the teacher by using a Laplace mechanism in a differential privacy algorithm. And (3) giving a corresponding mapping function, and adding Laplace noise into a result queried by a user to enable the query result to have slight difference. For example, Laplace noise is added, the review time of a teacher teaching plan in a certain day accounts for 30%, the query result is 15% after the Laplace mechanism is added, and the problem of teaching plan leakage is avoided. The daily electronic document reading amount of students is adjusted through a teacher teaching plan, an MCTS search tree is established, a root node is a certain student, the weight is 0, a second layer of nodes are used for judging whether the students perform well in class and setting the weight, a third layer of nodes are used for judging whether review content in the teaching plan accounts for the larger proportion of the teaching plan, wherein different thresholds can be set according to actual conditions; the method comprises the steps of taking different teaching plans and student class-taking performances as input, outputting a weight value of old knowledge required to be learned by the student, adding the weight value into a PCA result in a system recommended electronic document reading quantity algorithm, and adjusting the result of the system recommended electronic document reading algorithm so as to adjust learning contents of the student, wherein for example, the teacher plans to use longer time for knowledge consolidation in the teaching plans, and the student uses more time to warm up the old knowledge in learning.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Programs for implementing the information governance of the present invention may be written in computer program code for carrying out operations of the present invention in one or more programming languages, including an object oriented programming language such as Java, python, C + +, or a combination thereof, as well as conventional procedural programming languages, such as the C language or similar programming languages.
The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).
In the embodiments provided in the present invention, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. For example, the above-described device embodiments are merely illustrative, and for example, the division of the units is only one logical functional division, and other divisions may be realized in practice. The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit may be implemented in the form of hardware, or in the form of hardware plus a software functional unit.
The integrated unit implemented in the form of a software functional unit may be stored in a computer readable storage medium. The software functional unit is stored in a storage medium and includes several instructions to enable a computer device (which may be a personal computer, a server, or a network device) or a processor (processor) to execute some steps of the methods according to the embodiments of the present invention.
And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
Claims (6)
1. A financial intelligent teaching and privacy protection method is characterized by comprising the following steps:
establishing a block chain database to protect data privacy, including storing classroom student performance, teacher teaching information, financial courses and financial data, and adding uncertain privacy in query results by using a differential privacy encryption technology to protect classroom privacy: identifying the class concentration degree of the student through the data stored in the block chain database; judging whether the teacher's lesson-taking performance and the teacher's lesson-taking performance influence students based on an NLP technology, extracting whether words interacting with the students exist in the lesson-taking process of a teacher by using the NLP technology, and judging the performance of the teacher; adjusting the reading amount of the electronic documents according to the classroom performance of students; the teaching plan of the teacher is adjusted by the condition that the student grasps the electronic documents; protecting the privacy of teaching plans of teachers and adjusting the learning contents of students; adjusting a teaching plan of a teacher by using a DRL reinforcement learning technology, and establishing a DRL deep learning model by combining the learning condition and classroom expression condition of electronic documents learned by students and the difference privacy technology; the DRL deep learning model also comprises the steps of establishing a DQN: adopting a neural network with 3 convolutional layers and two Dense layers, initializing a Memory D to enable the capacity of the Memory D to be N; 1 to M steps are circularly traversed, wherein M is a parameter; establishing a reward function, and simulating the recommended content of the teaching plan; the simulation comprises the steps of simulating poor performance of students in learning and poor performance of teachers in teaching, and distributing review content and teaching duration of new knowledge points to enable the review content and the teaching duration to reach a higher reward value, so that an intelligent DRL model for financial courses is generated; the DRL model is used for intelligently recommending the teaching plan of the teacher, when the students have poor performance in learning and the teachers have poor performance in teaching, the time for adding the old knowledge of the DRL model accounts for a higher recommended content reward value, so that the recommended content is influenced; further comprising: setting parameters in the DQN; the setting of specific parameters in the DQN further includes: setting an agent, an Action and an award weight in the DQN; setting different learning conditions of students into different agents, carrying out DQN simulation on each corresponding agent, taking different teaching contents of teaching of a teacher as different actions, and enabling each Action to correspond to different rewarding weights; and through continuously collecting the learning conditions of different students, carrying out simulation training for many times, thereby generating a robust teaching plan recommendation model.
2. The method of claim 1, wherein the establishing a blockchain database protects data privacy, comprising:
by the Laplace mechanism: for adjacent data sets, giving a corresponding mapping function and adding Laplace noise into a result of user query; the students can obtain the financial information to be learned from the system database, and the privacy protection of the system can prevent other users from knowing the weak part of the study of the students; the method comprises the following steps: differential privacy uses a Laplace mechanism to realize privacy protection;
the differential privacy uses a Laplace mechanism to realize privacy protection, and further comprises:
and (3) assuming that the frame coefficient of the query result is based on the generalized Gauss-Laplace prior statistical distribution, and carrying out differential processing on the result through the distribution of Laplace by using maximum likelihood estimation and posterior mean estimation quantity.
3. The method of claim 1, wherein identifying student class concentration through data stored to a blockchain database comprises:
according to the established database, storing the acquired image data into a block chain database, establishing a student face database and a cloud server, extracting face features of students and analyzing the concentration degree of financial classes of the students; the method comprises the following steps: using an MTCNN face recognition model; training an MTCNN model based on federated learning; judging the concentration degree of the student according to iris analysis; judging the classroom performance score of the student by using posture analysis;
the using the MTCNN face recognition model further comprises:
training the MTCNN, wherein the model mainly adopts three cascaded networks, namely P-Net, R-Net and O-Net, adopts the idea of adding a classifier into a candidate frame, and uses NMS, Soft-NMS and RRelu to construct a neural network; and the horizontal federal learning technology is used, different financial teaching institutions can learn through the horizontal federal, and the model is trained by using the data of each classroom: taking the image of the human face as input, and taking the recognition result as output; training the model by using each data, and enabling the server to obtain a training result;
the training of the MTCNN model based on federated learning further comprises:
different financial classes are learned through horizontal federal, utilize the data of each college to train the model, and the result that the server will train: parameters, gradient values, loss function values and the like of the model are returned to servers of financial course institutions, the method can effectively improve the training efficiency of the model, obtain a more robust deep learning model, and successfully solve the problem that the model cannot be well adapted to different application scenes;
the method for judging the concentration degree of the student according to iris analysis further comprises the following steps:
according to the MCNN model, before the human face image is transmitted into the model, the human face is aligned, the 3D image of the target is selected, and the feature points of the adjacent frame images of the target are identified so as to align the space, wherein an advanced ERT algorithm of human face alignment is used; detecting and identifying the students by using the trained model, and judging whether the students keep concentrating all the time by extracting the iris characteristics of the students whether the iris characteristics deviate from the blackboard or the screen;
the use gesture analysis to judge student classroom performance score still includes:
identifying the student classroom behavior according to the established local feature identification convolutional neural network and using a DenseNet model: using 3 DenseBlock, 32 convolution kernels of 11x11 and a maxpool pooling layer of 3x3, using 16 convolution kernels of 9x9 in the last four layers, wherein the full connection layer is a deep learning model formed by 4096 dimensions, extracting behavior characteristic pictures of students who are low head to play mobile phones as input by the model, outputting whether the students are low head to play mobile phones or not, and training a corresponding neural network model; the model identifies the frequency of the student leaning down to play the mobile phone and deducts the corresponding weight of the classroom expression score.
4. The method of claim 1, wherein the extracting words whether the teacher has interaction with the student during the course of the teacher using the NLP technology and judging the performance of the teacher comprises:
establishing a BERT model: with teacher financial classroom utterances as input, pre-training with MLM and employing a deep two-way Transformer component: inputting a token sequence, firstly embedding the token sequence to be called a vector, then inputting the vector to a neural network, outputting the vector sequence with the size of H, and using an Attention mechanism; extracting class interaction vocabularies by using a BERT model, comprising the following steps: student's name, please answer, why, etc. interaction words; the language database of the teacher speaking in class is input, the frequency of teacher interactive content is output, and the relevance of teacher interactive performance and student posture performance is analyzed through a machine learning Apriori correlation analysis algorithm.
5. The method of claim 1, wherein said adjusting an electronic document reading size based on a student's classroom performance comprises:
the system adjusts the reading quantity of electronic documents of student related knowledge through a traditional PCA algorithm according to the scoring result of classroom performance; the PCA algorithm calculates and analyzes different subject performances of the students by using a covariance matrix, a mode of inner product and base change of the matrix and a matrix diagonalization mathematical formula and performs dimension reduction analysis, intelligently analyzes course contents which need emphasis reinforcement of the students, and requires the students to learn contents which need reinforcement of subjects after class by combining with a well-established block chain database.
6. The method of claim 1, wherein said adjusting the teacher's lecture plan using DRL reinforcement learning techniques comprises:
protecting the teaching plan privacy of the teacher by using a Laplace mechanism in a differential privacy algorithm; and (3) adding Laplace noise into the result of the user query by giving a corresponding mapping function, so that the query result has slight difference.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210471694.1A CN114862636A (en) | 2022-05-01 | 2022-05-01 | Financial intelligent teaching and privacy protection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210471694.1A CN114862636A (en) | 2022-05-01 | 2022-05-01 | Financial intelligent teaching and privacy protection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114862636A true CN114862636A (en) | 2022-08-05 |
Family
ID=82634921
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210471694.1A Withdrawn CN114862636A (en) | 2022-05-01 | 2022-05-01 | Financial intelligent teaching and privacy protection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114862636A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115115483A (en) * | 2022-08-31 | 2022-09-27 | 广州数园网络有限公司 | Student comprehensive capacity evaluation method integrating privacy protection |
-
2022
- 2022-05-01 CN CN202210471694.1A patent/CN114862636A/en not_active Withdrawn
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115115483A (en) * | 2022-08-31 | 2022-09-27 | 广州数园网络有限公司 | Student comprehensive capacity evaluation method integrating privacy protection |
| CN115115483B (en) * | 2022-08-31 | 2022-11-25 | 广州数园网络有限公司 | Student comprehensive ability evaluation method integrating privacy protection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109801525B (en) | Teacher-student multidimensional matching method and system for network teaching | |
| JP4908495B2 (en) | Systems and methods for semantic knowledge evaluation, teaching and acquisition | |
| US20160293036A1 (en) | System and method for adaptive assessment and training | |
| KR20150122669A (en) | Matching users of a network based on profile data | |
| US20150147728A1 (en) | Self Organizing Maps (SOMS) for Organizing, Categorizing, Browsing and/or Grading Large Collections of Assignments for Massive Online Education Systems | |
| CN112257966B (en) | Model processing method and device, electronic equipment and storage medium | |
| CN114254208A (en) | Identification method of weak knowledge points and planning method and device of learning path | |
| WO2021174827A1 (en) | Text generation method and appartus, computer device and readable storage medium | |
| Al-Alwani | Mood extraction using facial features to improve learning curves of students in e-learning systems | |
| US20210192973A1 (en) | Systems and methods for generating personalized assignment assets for foreign languages | |
| CN114862636A (en) | Financial intelligent teaching and privacy protection method | |
| US20200160437A1 (en) | Systems and methods for an intelligent interrogative learning platform | |
| CN115511670A (en) | Online education method and online education platform | |
| Liu et al. | An evaluation of gpt-4v and gemini in online vqa | |
| CN115935071A (en) | Knowledge point recommendation method and device, storage medium and electronic equipment | |
| CN109871430A (en) | A kind of method, apparatus, electronic equipment and the storage medium of intelligent recognition text | |
| Chen | Design of piano intelligent teaching system based on neural network algorithm | |
| Tittenbrun | Neither Capital, Nor Class: A Critical Analysis Of Pierre Bourdieu's Theoretical Framework | |
| Van Gasse et al. | Feedback opportunities of comparative judgement: An overview of possible features and acceptance at different user levels | |
| Wang et al. | A Learning Analytics Model Based on Expression Recognition and Affective Computing: Review of Techniques and Survey of Acceptance | |
| CN111179135A (en) | Self-adaptive virtual reality teaching method and system based on learning style model | |
| Fekry et al. | The relation between individual student behaviours in video presentation and their modalities using vark and paei results | |
| Wang | Evaluation and measurement of student satisfaction with online learning under integration of teaching resources | |
| Yu et al. | A transfer learning based interpretable user experience model on small samples | |
| Yoon et al. | A non‐verbal teaching behaviour analysis for improving pointing out gestures: The case of asynchronous video lecture analysis using deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220805 |
|
| WW01 | Invention patent application withdrawn after publication |