US-12619909-B2 - Method for comprehensive performance evaluation of students based on deep learning network

Abstract

The present disclosure provides a student performance evaluation method and system based on artificial intelligence (AI) identification data, and relates to the field of intelligent education. A lightweight network model suitable for student performance evaluation takes the AI identification data as an input and evaluation results as an output. A training data generation algorithm is provided, and multidimensional AI identification data and labels are uniformly processed into training data suitable for the network model through the above algorithm, which can solve the problems that dimensions between any AI identification data and various labels are not uniform, and original data cannot meet training of a multidimensional and cross-time prediction model. A simulated data generation algorithm and a simulated label generation algorithm are provided, and simulated training data is generated using these algorithms in conjunction with the training data generation algorithm.

Inventors

• Yongduan Song
• Feng Yang
• Rui Li
• HONGYU XIA
• Qin Chen
• Shichun WANG
• Liangjie LI
• Haoyuan Zhong

Assignees

• CHONGQING UNIVERSITY
• STAR INSTITUTE OF INTELLIGENT SYSTEMS
• DB (CHONGQING) INTELLIGENT TECHNOLOGY RESEARCH INSTITUTE CO., LTD
• UNIVERSITY OF ELECTRONIC SCIENCE AND TECHNOLOGY OF CHINA

Dates

Publication Date

20260505

Application Date

20220225

Priority Date

20210909

Claims (5)

1. 1 . A method for comprehensive performance evaluation of students based on a deep learning network model, which is implemented in a system comprising a processor a memory, wherein the memory stores instructions for the processor to perform the following steps: step S 1 , designing the deep learning network model, wherein the deep learning network model is stored in the memory and capable of performing comprehensive performance evaluation of in any time span; step S 2 , generating simulated artificial intelligence (AI) identification data containing the students using a simulated data generation algorithm; step S 3 , generating simulated labels by any amount of simulated AI identification data generated in step S 2 using a simulated label generation algorithm, wherein the simulated labels are used to generate simulated training data, and the simulated training data is used for pre-training of the deep learning network model; step S 4 , processing the simulated AI identification data through a training data generation algorithm using the simulated labels in step S 3 in conjunction with all the simulated AI identification data corresponding to the labels to generate the simulated training data, wherein the simulated training data is stored in a database in the memory; step S 5 , pre-training the deep learning network model designed in step S 1 using the simulated training data generated in step S 4 ; step S 6 , determining real labels of the students, wherein the real labels represent performance evaluation of the students, and the real labels are stored in the database in the memory, and are capable of being modified; step S 7 , obtaining all real labels in the database, and then obtaining all real AI identification data corresponding to each real label in the database; and processing the obtained real AI identification data through the training data generation algorithm to generate final real training data, wherein the final real training data is stored in the database; step S 8 , obtaining the real training data of the database, and training the deep learning network model in step S 1 using the obtained real training data, wherein different real training data is used for the deep learning network model at different stages; when prediction accuracy of the deep learning network model is less than or equal to 70%, the training data generated by the training data generation algorithm is used; and when the prediction accuracy of the deep learning network model is more than 70%, training data generated by a training data generation algorithm with a discard policy is used; and step S 9 , performing comprehensive performance evaluation of the students in any time and space span using the trained deep learning network model; wherein the simulated label generation algorithm in step S 3 comprises the following specific steps: step S 3 . 1 , randomly selecting a course, randomly selecting a student which selects the course, and randomly generating a time period, wherein this time period is within a time period corresponding to the selected course; step S 3 . 2 , according to limitation of the course, the student, and the time period in step S 3 . 1 , obtaining, a set of simulated AI identification data matching the course, the student, and the time period: T={[a 11 ,a 12 , . . . ,a 1i , . . . ,a 1n ],[a 21 ,a 22 , . . . ,a 2i , . . . ,a 2n ], . . . ,[a m1 ,a m2 , . . . ,a mi , . . . ,a mn ]}, wherein a mi represents an i-th simulated AI identification data in a randomly generated m-th time period under the premise of the limitation of the course and the student, step S 3 . 3 , allocating a positive state weight and a negative state weight to each dimension feature, wherein each dimension feature defines a weight allocation function g i (x), and each weight allocation function further has a weight W i ; step S 3 . 4 , calculating label values using the set T of simulated AI identification data in step S 3 . 2 : label = ∑ i = 1 m ∑ j = 1 n T [ i ] [ j ] ⁢ g j ( x ) ⁢ W j m × ( ∑ i = 1 n W i ) × 100 ; and step S 3 . 5 , repeating steps S 3 . 1 , S 3 . 2 , and S 3 . 4 to generate the simulated labels until a number of simulated labels reaches an expected value.
2. 2 . The method for comprehensive performance evaluation of students based on the deep learning network model according to claim 1 , wherein the simulated data generation algorithm in step S 2 comprises the following specific steps: step S 2 . 1 , randomly generating information of n students, marking the student information with different student identities (IDs), wherein the student IDs are recorded as: [ s id 1 ,s id 2 , . . . ,s id i , . . . ,s id n ]; and sid i is a student ID of an i-th student generated by simulation, the student information further comprises gender, college, and name, these basic student information is permanently stored in the database, and these generated students simulate a process of selecting courses, attending classes and generating the AI identification data; step S 2 . 2 , using the student IDs and a random generation algorithm in step S 2 . 1 to give each student a unique property, wherein a set of unique properties of the n students is recorded as [property 1 , property 2 , . . . , property i , . . . , property n ], and property i is a unique property of the i-th student generated by simulation; the unique property of each student represents uniqueness of the simulated student, indicating a learning state of the simulated student, all unique properties obey normal distribution as a whole, and the set of unique properties obtained at the current stage does not have complete randomness and does not obey normal distribution for the time being; step S 2 . 3 , scrambling the set of unique properties in step S 2 . 2 using a random scrambling algorithm to obtain a scrambled set of unique properties [Lproperty 1 , Lproperty 2 , . . . , Lproperty i , . . . , Lproperty n ], wherein the scrambled unique properties are random; step S 2 . 4 , mapping values in the scrambled set of unique properties by a function, such that the scrambled set of unique properties after mapping has a normal distribution characteristic, wherein each Lproperty i in the scrambled set of unique properties is mapped by the function: f ⁡ ( Lproperty i ) = { 0.5 - 0.5 G ⁡ ( Lproperty i - 0.5 ) G ⁡ ( 0.5 ) - G ⁡ ( 0 ) , Lproperty i ≤ 0.5 0.5 + 0.5 G ⁡ ( Lproperty i - 0.5 ) G ⁡ ( 0.5 ) - G ⁡ ( 0 ) , Lproperty i > 0.5 , and G ⁡ ( x ) = 1 2 ⁢ e - π ⁡ ( Lproperty i - 0.5 ) 2 4 , and the scrambled set of unique properties after mapping is recorded as a final set of unique properties: [finalProperty 1 ,finalProperty 2 , . . . ,finalProperty i , . . . ,finalProperty n ]; step S 2 . 5 , defining Le courses, wherein each course has choice; choices, Le≥i>0, and each course has teacher, class time, and course week information; and each student generated in step S 2 . 1 randomly selects courses based on these courses; and step S 2 . 6 , sequentially generating the simulated AI identification data from a first course of a first week according to a time sequence, wherein in step S 2 . 3 scrambling the set of unique properties in step S 2 . 2 using the random scrambling algorithm comprises the following steps: step S 2 . 3 . 1 , setting a number of scrambling Y, a random seed X1, a random seed X2, a modulus M1, a modulus M2, a multiplication amount A1, a multiplication amount A2, and a loop mark I=1; step S 2 . 3 . 2 , rounding down n×(X1/M1) to obtain P 1 , and rounding down n×(X2/M2) to obtain Q 1 , wherein X1=(A1×X1+CD), and X2=(A2×X2+C2); and taking the remainder of X1 using M1 and assigning a remainder result to X1, and taking the remainder of X2 using M2 and assigning a remainder result to X2, wherein I=I+1; step S 2 . 3 . 3 , when I is less than Y, executing S 2 . 3 . 2 until I is greater than or equal to Y to finally obtain a set of scrambling instructions [P 1 , Q 1 , P 2 , Q 2 , . . . , P i , Q i , . . . , P Y , Q Y ]; and step S 2 . 3 . 4 , taking out P i , Q i , 1<<I≤Y from the beginning to the end in pairs each time from the set of scrambling instructions in step S 2 . 3 . 3 , and then exchanging two properties with subscripts of P i , Q i in the set of unique . . . properties in step S 2 . 2 to obtain [Lproperty 1 , Lproperty 2 , . . . Lproperty i , . . . Lproperty n ] after all exchanges are completed.
3. 3 . The method for comprehensive performance evaluation of students based on the deep learning network model according to claim 1 , wherein the training data generation algorithm in steps S 4 , S 7 , and S 8 comprises the following specific steps: step S 4 . 1 , obtaining all the AI identification data corresponding to the labels in step S 4 : Data=[ a 11 ,a 12 , . . . ,a 1i , . . . ,a 1n ],[a 21 ,a 22 , . . . ,a 2i , . . . ,a 2n ], . . . ,[a m1 ,a m2 , . . . ,a mi , . . . ,a mn ]}; step S 4 . 2 , determining a standard training data length to be S; when m in Data is equal to S, directly matching the data Data with the label as one of the final training data; when m in Data is less than S, calculating A=S/m, rounding down A to obtain A1, and taking a remainder of S to m to obtain B; copying A1−1 copies of random B elements in Data, copying A1+1 copies of remaining m-B elements, and matching the copied Data with the label as one of the final training data; and when m in Data is greater than S, calculating C=m/S, rounding down C to obtain B1, and taking the remainder of m to S to obtain D, summing and averaging the first D connected elements with a length of B1−1 in Data to synthesize an element, summing and averaging m-B connected elements with a length of B in the remaining elements to synthesize an element, and matching Data after synthesis with the label as one of the final training data.
4. 4 . The method for comprehensive performance evaluation of students based on the deep learning network model according to claim 2 , wherein in step S 2 . 2 , a process of using the random generation algorithm to give each student the unique property comprises the following specific steps: step S 2 . 2 . 1 , setting a prime number K greater than n, and dividing each student ID in a student ID set in step S 2 . 1 by K to obtain a set of temper numbers: [sid 1 /K, sid 2 /K, . . . , sid i /K, . . . sid n /K]=[temperNumber 1 , temperNumber 2 , . . . , temperNumber i , . . . , temperNumber n ], wherein temperNumber i is a temper number of the i-th student generated by simulation; step S 2 . 2 . 2 , setting a truncation start bit K and a truncation stop bit J, wherein J>K; and intercepting K-th to J-th bits after a decimal point of each temper number in the set of temper in numbers step S 2 . 2 . 1 obtain a set of truncation numbers to [truncation 1 , truncation 2 , . . . , truncation i , . . . , truncation n ]; and step S 2 . 2 . 3 , dividing each number in the set of truncation numbers in step S 2 . 2 . 2 by 10 J+1-k to obtain the unique property: [truncation 1 /10 J+1-K ,truncation 2 /10 J+1-K , . . . ,truncation i /10 J+1-K , . . . ,truncation n /10 J+1-K ]=[property 1 ,property 2 , . . . ,property i , . . . ,property n ].
5. 5 . The method for comprehensive performance evaluation of students based on the deep learning network model according to claim 2 , wherein step S 2 . 6 comprises the following specific steps: step S 2 . 6 . 1 , randomly generating a number B obeying (0,1) normal distribution; step S 2 . 6 . 2 , obtaining the unique property of the student finalProperty i corresponding to the simulated AI identification data being generated; and if B+finalProperty i , executing step S 2 . 6 . 3 between an interval [0,1], otherwise executing step S 2 . 6 . 1 ; and step S 2 . 6 . 3 , customizing an overall positive state rate of the currently generated simulated AI identification data as V, outputting a positive state if B+finalProperty i is greater than a in the following equation, and outputting a negative state if B+finalProperty i is less than or equal to a in the following equation, wherein the equation in step S 2 . 6 . 3 is ∫ 0 1 ∫ 0 a 1 2 ⁢ π ⁢ e - ( x - y ) 2 2 ⁢ dxdy ∫ 0 1 ∫ 0 1 1 2 ⁢ π ⁢ e - ( x - y ) 2 2 ⁢ dxdy = V , and since a value on a left side of the above equation increases with a, only B+finalProperty i is replaced with a in the equation during specific determination, and whether the value on the left side of the equation is greater than or less than V is determined after replacement, the positive state is output if the value on the left side of the equation is greater than V, and the negative state is output if the value on the left side of the equation is less than V.

Description

CROSS REFERENCE TO RELATED APPLICATION This patent application claims the benefit and priority of Chinese Patent Application No. 202111054782.3, filed on Sep. 9, 2021, the disclosure of which is incorporated by reference herein in its entirety as part of the present application. TECHNICAL FIELD The present disclosure relates to the field of intelligent education, and in particular to a student performance evaluation method and system based on artificial intelligence (AI) identification data. BACKGROUND ART With vigorous development and maturity of the artificial intelligence technology, it has been widely used in various scenarios. In classroom scenarios, the artificial intelligence technology is often used to identify emotions, gestures, and sight lines of students. At the current stage, the artificial intelligence technology in the classroom environment is usually single-technical identification, that is, each plays its own identification function, and is also single time-space identification, that is, only a single student at the current stage is identified. Such single-technical and single time-space identification cannot fully utilize the benefits brought by such technology to the field of intelligent education, and there are problems such as waste of resources and redundancy of software and hardware resources. In order to give full play to the role of various single-dimensional identification technologies to reduce resource waste and redundancy, the present disclosure provides a unified and generalized method to perform multidimensional processing on AI identification data obtained by the single-technical and single time-space dimensions, and then perform comprehensive evaluation and development prediction of the students in multi time-space dimensions. SUMMARY The present disclosure aims to solve the problem in the prior art, and particularly provides a student performance evaluation method and system based on AI identification data. To achieve the above objective of the present disclosure, according to a first aspect of the present disclosure, the present disclosure provides a student performance evaluation method based on AI identification data, including: step S1, designing a deep learning network model, where the deep learning network model is capable of performing comprehensive performance evaluation of students in any time span by using processed AI identification data;step S2, generating simulated AI identification data containing multiple students using a simulated data generation algorithm, wherethe simulated AI identification data generated in this step meets multidimensional randomness and multiple dimensions meet normal distribution, and a trend of an overall data value is controllable;step S3, generating simulated labels by any amount of simulated AI identification data generated in step S2 using a simulated label generation algorithm, wherethe simulated labels are used to generate simulated training data, and the simulated training data is used for pre-training of the deep learning network model;step S4, processing the simulated AI identification data through a training data generation algorithm using the simulated labels in step S3 in conjunction with all the simulated AI identification data corresponding to the labels to generate the simulated training data, where the simulated training data is stored in a database;step S5, pre-training the deep learning network model designed in step S1 using the simulated training data generated in step S4;step S6, determining real labels of the students, where the real labels represent performance evaluation of the students, and the real labels are stored in the database, and are capable of being modified;step S7, obtaining all real labels in the database, and then obtaining all real AI identification data corresponding to each real label in the database; andprocessing the obtained real AI identification data through the training data generation algorithm to generate final real training data, where the real training data may be stored in the database;step S8, obtaining the real training data of the database, and training the deep learning network model in step S1 using the obtained real training data, where different real training data is used for the deep learning network model at different stages;when prediction accuracy of the deep learning network model is less than or equal to 70%, the training data generated by the training data generation algorithm is used; andwhen the prediction accuracy of the deep learning network model is more than 70%, training data generated by a training data generation algorithm with a discard policy is used; andstep S9, after the training of the deep learning network model is mature, performing comprehensive performance evaluation of the students in any time and space span using the trained deep learning network model. Further, the simulated data generation algorithm in step S2 may include the following specific ste