CN112508334A - Personalized paper combining method and system integrating cognitive characteristics and test question text information - Google Patents

Personalized paper combining method and system integrating cognitive characteristics and test question text information Download PDF

Info

Publication number
CN112508334A
CN112508334A CN202011233044.0A CN202011233044A CN112508334A CN 112508334 A CN112508334 A CN 112508334A CN 202011233044 A CN202011233044 A CN 202011233044A CN 112508334 A CN112508334 A CN 112508334A
Authority
CN
China
Prior art keywords
learner
test
test question
knowledge
text information
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.)
Granted
Application number
CN202011233044.0A
Other languages
Chinese (zh)
Other versions
CN112508334B (en
Inventor
王志锋
余新国
左明章
叶俊民
张思
闵秋莎
罗恒
夏丹
姚璜
杨洋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central China Normal University
Original Assignee
Central China Normal University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central China Normal University filed Critical Central China Normal University
Priority to CN202011233044.0A priority Critical patent/CN112508334B/en
Publication of CN112508334A publication Critical patent/CN112508334A/en
Application granted granted Critical
Publication of CN112508334B publication Critical patent/CN112508334B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Educational Administration (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Educational Technology (AREA)
  • Tourism & Hospitality (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Development Economics (AREA)
  • Mathematical Physics (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • General Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Game Theory and Decision Science (AREA)
  • Primary Health Care (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Electrically Operated Instructional Devices (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention belongs to the technical field of intelligent education, and discloses a personalized paper combining method and system integrating cognitive characteristics and test question text information, wherein a cognitive diagnosis model is used for predicting scores of learners on specific test questions based on cognitive level; then predicting the score of the learner on the specific test question based on the text information by utilizing a recurrent neural network model; then, constructing a probability matrix decomposition objective function based on the obtained predicted scores of the learner based on the cognitive level and the text information, and predicting the potential scores of the learner on the specific test questions; and finally, calculating KL divergence by utilizing the estimated knowledge mastering vector of the learner and the incremental knowledge mastering vector of the learner, and selecting test questions with increased knowledge mastering tendency and proper difficulty to form the test paper for the personalized test by combining the potential scores of the learner on the test questions. The invention can self-define the paper-making result according to the test target and the test question difficulty, thereby greatly increasing the self-learning efficiency of learners.

Description

Personalized paper combining method and system integrating cognitive characteristics and test question text information
Technical Field
The invention belongs to the technical field of intelligent education, and particularly relates to a personalized paper combining method and system integrating cognitive characteristics and test question text information.
Background
At present, with the rapid development of the internet and the arrival of the big data era, the traditional education industry gradually starts to transform to digital education, and massive education resources are shared as information on an online education platform for learners to download and learn. The test questions of each subject are used by learners as important resources in education in a large amount to consolidate the knowledge of learners in classroom, however, learners are difficult to directly screen out test questions really suitable for the learners from the large amount of test questions, and more, the learners are trained by adopting the tophai tactics. The personalized test paper organizing system can quickly master the text information of the matched test questions according to the knowledge of the learner, and provide the test questions which are suitable for the learning difficulty of the learner and aim to enhance the knowledge growth of the learner, so that the learner can be better trained, the learning efficiency of the learner is improved, and meanwhile, the targeted test paper organizing exercise is performed on different learning conditions of each learner, so that the demands of the learners on the personalized test paper organizing on the online learning platform are increasingly urgent.
The traditional paper-making mode adopts the recommendation idea of popular collaborative filtering in the E-commerce field, which analogizes commodities into test questions, analogizes users into learners and analogizes scores for the commodities into scores of the learners for the test questions, so that the test paper-making mode is applied to collaborative filtering to screen and obtain proper test questions, and a complete set of test paper is formed. However, the examination paper test in this way tends to dig the learning commonality of learners, that is, the examination questions given to the target learner are more from learners with high similarity to the examination questions, so that the simple examination questions and popular examination questions have a greater recommended probability, the way is not given for the specific learning state of the learner, so the examination paper result cannot be given for weak links mastered by knowledge points of the learner, the interpretability of the examination paper result is also lacking, and the learner prefers to form a set of examination questions suitable for the cognitive characteristics of the learner through the examination paper system so as to perform targeted examination question training on the weak knowledge points of the learner, so the way cannot be well applied to the technical field of intelligent education. The cognitive diagnosis method in cognitive psychology is also used in the examination paper of the learner in the later stage, the cognitive diagnosis model can diagnose the learning state of the learner, analyze the mastering conditions of the learner for each known point according to the score of the learner on the test paper of the labeled knowledge point, so as to obtain the correct answer probability of the learner for the given knowledge point, and then screen the test paper. Meanwhile, the traditional test paper combining method usually ignores the text information of the test questions, and the subject keywords contained in the test question text usually have larger association with the probability that the learner correctly answers the test questions, so that the analysis of the test question text is necessary to be added into the personalized test paper of the learner.
Through the above analysis, the problems and defects of the prior art are as follows:
(1) in the prior art, the traditional personalized test paper combining method neglects the influence of test question text information on the answer of a learner.
(2) In the prior art, the method for test paper composition by using cognitive diagnosis ignores the commonality among learners and the parameter estimation in the model is sensitive to a data set and may generate larger errors;
(3) in the prior art, the method for combining examination papers by using collaborative filtering cannot fully consider knowledge point mastering conditions of learners, can only combine papers according to learning commonalities of similar learners, neglects learning characteristics of the learners in the learning process, and has poor interpretability of a paper combining result.
The difficulty in solving the above problems is:
(1) how to integrate test question text information into the cognitive process of a learner, and connecting the text with the learner score to obtain a mapping relation between the text and the score;
(2) how to combine the learner cognitive characteristics represented in the cognitive diagnosis result with the learner learning commonalities represented in the collaborative filtering result, namely comprehensively considering the learning commonalities and characteristics and making a tendency group exercise according to the learning characteristics of the learner.
(3) How to measure or adjust the influence of the learner's learning characteristics, learning commonalities and test question text information on the final test paper result and improve the application capability of the test paper result under different expected conditions.
Through the above analysis, the significance of solving the above problems and defects is:
(1) in the aspect of test question text analysis, the invention uses the recurrent neural network model RNN to extract the text content of the test question and constructs the mapping relation between the learner learning state and the test question text information through a full connection layer.
(2) The learner cognitive characteristics represented in the cognitive diagnosis result are combined with the learner learning commonalities represented in the collaborative filtering result, the learning commonalities and characteristics and the test question text information are comprehensively considered, and the test questions which are difficult to test or improve knowledge mastering can be selected according to the teaching targets.
(3) The invention takes the result of the model as prior information, can carry out personalized learning and paper grouping according to the learning conditions of different learners, and provides a plurality of learning and analyzing results with the learning state of the learner, the text information of the test question and the commonality among the learners, the interpretability of the paper grouping result is strong, and the education by the factors is realized.
(4) Compared with the traditional test question grouping algorithm, the test result shows that the method has higher performance improvement compared with the traditional method, and the defects of the traditional method in the aspect of test question grouping are improved. The method utilizes richer information, gives more accurate personalized test for the learner, and improves the learning efficiency of the learner.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a personalized paper combining method and a personalized paper combining system integrating cognitive characteristics and test question text information. The invention aims to solve the problem that the traditional personalized paper combining method neglects the influence of test question text information on the answer of a learner in the prior art; the method for testing the test paper group by using cognitive diagnosis ignores the commonality among learners and may generate larger errors when the parameter estimation in the model is sensitive to the data set; the method for testing and composing the test paper by utilizing the collaborative filtering can not practice aiming at the knowledge point mastering condition of the learner, and can only perform the composing of the test paper according to the learning commonality of similar learners, neglects the learning characteristics of the learner in the learning process and has poor interpretability of the composing result.
The invention is realized in this way, a personalized paper combining method for combining the learner cognitive characteristic and the test question text information, the personalized paper combining method for combining the learner cognitive characteristic and the test question text information comprises the following steps:
estimating and calculating knowledge mastering conditions of learners by using a cognitive diagnosis model according to real answering conditions and test question knowledge point distribution of the learners, and predicting scores of the learners on specific test questions based on cognitive levels;
extracting the text content of the test questions by using a recurrent neural network model, constructing a mapping relation between the learning state of the learner and the text information of the test questions through a full connection layer, and predicting the score of the learner on the specific test questions based on the text information;
thirdly, constructing a probability matrix decomposition objective function based on the obtained predicted scores of the learner based on the cognitive level and the text information, and predicting the potential score of the learner on the specific test question;
and step four, calculating KL divergence by utilizing the estimated knowledge mastering vector of the learner and the incremental knowledge mastering vector of the learner, and selecting the test paper which enables the knowledge mastering trend of the learner to be increased and has proper difficulty to form the personalized test by combining the potential scores of the learner on the test paper.
Further, in step one, the estimating and calculating knowledge mastering conditions of the learner by using the cognitive diagnosis model according to the real answer condition of the learner and the distribution of the knowledge points of the test questions and predicting the score of the learner on the specific test questions based on the cognitive level comprises:
(1.1) collecting test question knowledge point distribution labeled by domain experts and response data of learners; calculating a Q matrix represented by test question knowledge points required to be used in the cognitive diagnosis model according to the distribution of the test question knowledge points labeled by domain experts;
(1.2) calculating the ideal response condition of the learner according to the prior knowledge point mastering mode eta of the learner:
Figure BDA0002765838660000041
wherein, piijShows the ideal answer situation of the learner i on the jth test question etaikRepresents the mastery condition of learner i on known recognition point k, qjkWhether the known jth test question is examined or not is shownA knowledge point k;
(1.3) estimating the probability s of the learner doing wrong test questions under the condition of mastering a certain test question and examining all knowledge points according to an expectation-maximization algorithm, and estimating the probability g of the learner doing wrong test questions under the condition of not mastering all corresponding knowledge points;
(1.4) calculating the probability that the learner answers correctly according to the estimated ideal answer condition of the learner, the probability s that the learner makes wrong test questions under the condition that the learner masters a certain test question and examines all known points, and the probability g that the learner does wrong test questions under the condition that the learner does not master corresponding knowledge points:
Figure BDA0002765838660000051
(1.5) obtaining the total likelihood function of the DINA model:
Figure BDA0002765838660000052
wherein L is 2K；
(1.6) calculating knowledge grasping conditions of the learner using maximum likelihood estimation based on the obtained total likelihood function:
Figure BDA0002765838660000053
(1.7) calculating the scoring condition of the learner on a new test question according to the knowledge mastering condition of the learner:
Figure BDA0002765838660000054
further, in the second step, the extracting of the text content of the test question by using the recurrent neural network model, and the construction of the mapping relationship between the learner learning state and the test question text information through the full connection layer, and the predicting of the score of the learner on the specific test question based on the text information includes:
(2.1) performing text word segmentation, stop word removal and other preprocessing on the test question text; processing the test questions by using a continuous Word bag model of a Word vector model Word2vec, predicting Word vectors of target words according to the Word vectors of a plurality of words in the context of the target words, vectorizing the pre-processed test questions, and obtaining an embedded expression of the Word level of the test questions;
(2.2) using the test question word vector as input, acquiring test question context characteristic representation by adopting a bidirectional long-and-short time memory neural network, and acquiring test question sentence level representation through mean pooling to obtain test question word embedding representation;
(2.3) constructing a full-connection deep neural network, performing feature fusion on the embedded vector of the test question words and the knowledge mastering vector of the learner to serve as input of the full-connection deep neural network, and outputting scores of specific students on the test question;
(2.4) bonding the results of the layers to each other1,y2,...,yn) Processed by an output unit:
Figure BDA0002765838660000061
obtaining a value between [0, 1], representing the probability of correct answer of the student to the test question text, comparing the value with real score data, and performing weight correction in the network;
and (2.5) training the network model by using the training set data to obtain a trained neural network model, and predicting the learner response result based on the text information by using the obtained trained neural network model.
Further, in the step (2.1), the text word segmentation of the test question text includes: based on the mixed dictionary, a method combining a bidirectional maximum matching method and statistics is adopted to perform mixed word segmentation on the test question text.
Further, in the step (2.1), the removing stop words from the test question text includes: and adding words which are irrelevant to sentences and test question text themes, do not contribute to test question labeling tasks and have low frequency in the test question text into the disabled word bank, and deleting words of mixed participles in the test question text appearing in the disabled word bank.
Further, in the step (2.2), the obtaining of the contextual feature representation of the test question by using the bidirectional long-and-short term memory neural network includes:
the BilSTM network adopts two LSTMs to obtain the context characteristics of different test questions from opposite directions, and the formula is as follows:
Figure BDA0002765838660000062
Figure BDA0002765838660000063
wherein a is1,a2,b1And b2G (-) is a hidden layer activation function, which is a weight coefficient,
Figure BDA0002765838660000064
for the forward hidden layer output at time t,
Figure BDA0002765838660000065
outputting the backward hidden layer at the time t; and fusing hidden layer outputs of two directions at each moment to construct a final output ht：
Figure BDA0002765838660000066
Wherein c is1And c2F (-) is the output activation function for the weight coefficients.
Further, in step (2.2), the obtaining of the sentence-level representation of the test question through the mean pooling process includes:
obtaining the embedded expression E of the test words by average pooling at the test sentence levelh：
Eh＝mean(h1,...,ht)
Where mean (-) is the average pooling operation, i.e., taking the average of the eigenvalues as output within the domain.
Further, in step (2.3), the fully-connected deep neural network includes:
in the fully-connected deep neural network, the calculation method of the nth node value of the mth layer is as follows:
Figure BDA0002765838660000071
wherein N is the number of the units of the upper layer,
Figure BDA0002765838660000072
represents a weight coefficient from the ith cell of the m-1 th layer to the nth cell of the m-1 th layer;
and (5) obtaining the potential mapping relation between the knowledge mastery of the learner, the test question text information and the score by adopting a relu activation function.
Further, in step three, the probability matrix decomposition objective function is constructed based on the obtained predicted scores of the learner based on the cognitive level and the text information, and predicting the potential scores of the learner on the specific test question comprises:
(3.1) the learners are marked as R by the obtained score prediction based on the cognitive level and the score prediction based on the text information1,R2And constructing a potential answer representation of the learner on the test question by using a probability matrix decomposition algorithm:
Figure BDA0002765838660000073
u and V are a learner characteristic matrix and a score characteristic matrix in probability matrix decomposition respectively, and alpha and beta are adjusting parameters of the learning condition of the learner and test question text information respectively;
(3.2) constructing a final objective function of the probability matrix decomposition:
Figure BDA0002765838660000074
(3.3) optimizing the objective function by using a gradient descent method to obtain an optimal learner result characteristic matrix U, V:
Figure BDA0002765838660000075
Figure BDA0002765838660000081
(3.4) predicting the performance of the learner by using the optimal feature matrix U, V obtained by training:
Figure BDA0002765838660000082
further, in the fourth step, the step of calculating the KL divergence by using the estimated learner knowledge and mastery vector and the learner incremental knowledge and mastery vector, and combining the potential scores of the learner on the test questions to select the test paper which enables the learner to increase the knowledge and mastery trend and has proper difficulty to form the personalized test comprises the following steps:
(4.1) obtaining all incremental knowledge base vectors of the learner based on the analysis to obtain knowledge base vectors of the learner
Figure RE-GDA0002903752560000081
Figure RE-GDA0002903752560000081
0≤d≤D；
(4.2) calculating learner knowledge learning vector eta estimated by the cognitive diagnosis modeliMastery with incremental knowledge of all learners
Figure RE-GDA0002903752560000082
KL divergence measure of (1):
Figure BDA0002765838660000083
(4.3) selecting test papers which enable learners to have an increased knowledge mastering trend and are suitable in difficulty to form a personalized test:
Figure BDA0002765838660000084
another objective of the present invention is to provide a personalized test paper combining learner cognitive characteristics and test question text information system implementing the personalized test paper combining learner cognitive characteristics and test question text information method, wherein the personalized test paper combining learner cognitive characteristics and test question text information system comprises:
the cognitive level-based score prediction module is used for estimating and calculating the knowledge mastering condition of the learner by utilizing a cognitive diagnosis model according to the real answer condition of the learner and the test question recognition point distribution and predicting the cognitive level-based score of the learner on a specific test question;
the score prediction module based on the text information is used for extracting the text content of the test questions by utilizing the recurrent neural network model, constructing a mapping relation between the learning state of the learner and the text information of the test questions through a full connection layer and predicting the score of the learner on the specific test questions based on the text information;
the question selecting strategy module is used for constructing a probability matrix decomposition objective function based on the obtained predicted scores of the learner based on the cognitive level and the text information and predicting the potential scores of the learner on the specific test questions; and calculating KL divergence by utilizing the estimated learner knowledge control vector and the learner incremental knowledge control vector, and selecting the test paper which ensures that the knowledge control trend of the learner is increased and the test paper with proper difficulty is formed by combining the potential scores of the learner on the test paper.
Another object of the present invention is to provide a computer apparatus comprising a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, causes the processor to execute the personalized group paper method of fusing learner cognitive characteristics and test question text information.
Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which when executed by a processor, causes the processor to execute the personalized test paper combining learner cognitive features and test question text information.
The invention also aims to provide an information data processing terminal which is used for realizing the personalized paper combining method integrating the cognitive characteristics of learners and test question text information.
By combining all the technical schemes, the invention has the advantages and positive effects that: the invention provides proper training test questions for the learner by integrating the cognitive characteristics of the learner and the test question text information, thereby getting rid of the problems of traditional one-thousand-person-one questions, problem sea tactics and the like, and the learner can also carry out targeted training better, thereby improving the learning efficiency. The invention has wide application prospect in the fields of personalized learning, adaptive learning and intelligent education.
The invention provides an individualized test paper combining method integrating learner cognitive characteristics and test question text information, which combines a cognitive diagnosis model, and carries out test paper combining practice for a learner based on model collaborative filtering and a test question text information extraction model. The invention analyzes the knowledge point mastering condition of the learner so as to obtain the learning state of the learner, and the used cognitive diagnosis model is a widely used cognitive diagnosis DINA model. In the aspect of test question text analysis, the invention uses the recurrent neural network model RNN to extract the text content of the test question, and constructs the mapping relation between the learner learning state and the test question text information through a full connection layer. The method takes the result of the model as prior information, and is used for training with other information in a collaborative filtering method based on probability matrix decomposition, so that the result of the group paper has the learning state of the learner, the text information of the test question and the commonality among the learners at the same time.
Compared with the traditional group paper calculation method, the personalized group paper method fusing the cognitive characteristics and the test question text information has the advantages that the performance of the method is greatly improved compared with the traditional method according to the experimental result, and the defects of the traditional method in the aspect of test question recommendation are overcome. The invention utilizes richer information to provide more accurate personalized test question recommendation for the learner, thereby improving the learning efficiency of the learner.
The invention realizes the personalized test paper combining method for combining the cognitive characteristics of the learner and the test paper text information, can combine the cognitive diagnosis, the test paper text information and the common information of the learner to recommend the test paper to the target learner, obtains a more accurate personalized test paper combining the cognitive characteristics and the test paper text information, greatly increases the self-learning efficiency of the learner, and helps the learner to make up the short knowledge. The method can be applied to the fields of intelligent education technology, education data mining and the like, and can also provide effective support for subsequent education resource recommendation and the like, help an online education platform and a digital education platform to better predict the score of the learner, thereby efficiently discovering weak links of knowledge points of the learner and taking accurate remedial measures.
Compared with other test question knowledge estimation methods, the personalized test paper combining method combining the learner cognitive characteristics and the test question text information provided by the invention has the advantages that the comparison results of the accuracy, the recall rate and the F1 value of the personalized test paper combining method combining the learner cognitive characteristics and the test question text information and other methods in the same data set are shown in table 1.
TABLE 1 comparison of the results
Figure BDA0002765838660000101
Figure BDA0002765838660000111
The experimental results show that: the personalized paper combining method for integrating the cognitive characteristics of the learner and the test question text information combines the cognitive characteristics of the learner, and the learner learns the commonalities and the test question text information. The accuracy of the volume result is obviously better than that of other comparative experiments. Therefore, experiments show that the personalized paper combining method fusing the cognitive characteristics of learners and test question text information is more effective than other methods in the aspects of accuracy, recall ratio, F1 value and the like.
Meanwhile, analysis shows that the paper grouping method and the probability matrix decomposition method based on the DINA are slightly unstable, and the accuracy rate of the paper grouping method and the probability matrix decomposition method is reduced along with the increase of the number of test questions of the paper grouping. Although the conventional probability matrix decomposition is easy to implement, the potential information in the extracted data set is insufficient, which leads to the low precision of the probability matrix decomposition under general conditions, especially when a large amount of training data is faced. In a word, the personalized test paper combining method combining the cognitive characteristics of the learner and the text information of the test questions has the best experimental effect, and the best test paper which can combine the test questions with different difficulty levels and promote the cognitive growth of the learner can be constructed according to the test difficulty and the examination and check target.
The invention introduces test question text information as an important measurement index for recommending the educational test questions, so that the key information in the test question text can be utilized by the method provided by the invention.
The invention integrates test question text information, cognitive diagnosis technology and a collaborative filtering method, the learner learning condition obtained from the test question text information and the cognitive diagnosis is integrated into an objective function with collaborative filtering optimization, and the relation among the three is introduced into adjustment parameter adjustment, so as to obtain an optimal paper group model matched with the current data set.
In conclusion, the personalized test paper combining method for the learner cognitive characteristics and the test paper text information provided by the invention realizes more accurate test paper recommendation and personalized test paper combination for the learner, combines the cognitive diagnosis, the test paper text information and the learner learning commonality to recommend the test paper for the target learner, can customize the test paper combination result according to the test target and the test paper difficulty, greatly increases the self-learning efficiency of the learner, and more quickly helps the learner to make up the short board on the knowledge of the learner in a classroom. The method can be applied to the fields of education resource recommendation and evaluation, education data mining and the like, so that effective support is provided for follow-up education resource recommendation and the like, an online education platform is assisted, a digital education platform can better predict the score of a learner and provide an individualized test paper scheme, and therefore weak links of knowledge points of the learner can be efficiently diagnosed and accurate remedial measures can be taken.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings used in the embodiments of the present application will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic diagram of a personalized test paper combining learner cognitive characteristics and test question text information according to an embodiment of the present invention.
Fig. 2 is a flowchart of a personalized test paper combining learner cognitive characteristics and test question text information according to an embodiment of the present invention.
FIG. 3 is a schematic structural diagram of a personalized test paper system integrating the cognitive characteristics of learners and test question text information according to an embodiment of the present invention;
in the figure: 1. a score prediction module based on cognitive level; 2. a score prediction module based on the text information; 3. and a topic selection strategy module.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Aiming at the problems in the prior art, the invention provides a personalized paper combining method for learner cognitive characteristics and test question text information, and the invention is described in detail below with reference to the accompanying drawings.
The symbols involved in the present invention are as follows:
Figure BDA0002765838660000121
Figure BDA0002765838660000131
Figure BDA0002765838660000141
as shown in fig. 1-2, the personalized test paper combining method for learner cognitive characteristics and test question text information according to the embodiment of the present invention includes the following steps:
s101, estimating and calculating knowledge mastering conditions of the learner by using a cognitive diagnosis model according to the real answering conditions and test question knowledge point distribution of the learner, and predicting the score of the learner on a specific test question based on the cognitive level;
s102, extracting the text content of the test question by using a recurrent neural network model, constructing a mapping relation between the learning state of the learner and the text information of the test question through a full connection layer, and predicting the score of the learner on the specific test question based on the text information;
s103, constructing a probability matrix decomposition objective function based on the obtained learner cognitive level and text information-based prediction score, and predicting the potential score of the learner on a specific test question;
s104, calculating KL divergence by utilizing the estimated learner knowledge mastering vector and the learner incremental knowledge mastering vector, and selecting the test paper which enables the knowledge mastering trend of the learner to be increased and enables the test paper with proper difficulty to be formed into the personalized test by combining the potential scores of the learner on the test paper.
In step S101, the estimating and calculating knowledge mastering conditions of the learner according to the real answer condition of the learner and the distribution of knowledge points of the test questions by using the cognitive diagnosis model and predicting the score of the learner on the specific test question based on the cognitive level according to the embodiment of the present invention includes:
(1.1) collecting test question knowledge point distribution labeled by domain experts and response data of learners; calculating a Q matrix represented by test question knowledge points required to be used in the cognitive diagnosis model according to the distribution of the test question knowledge points labeled by domain experts;
(1.2) calculating the ideal response condition of the learner according to the prior knowledge point mastering mode eta of the learner:
Figure BDA0002765838660000151
wherein, piijShows the ideal answer situation of the learner i on the jth test question etaikRepresents the mastery condition of learner i on known recognition point k, qjkWhether the known jth test question examines a knowledge point k is represented;
(1.3) estimating the probability s of the learner doing wrong test questions under the condition of mastering a certain test question and examining all knowledge points according to an expectation-maximization algorithm, and estimating the probability g of the learner doing wrong test questions under the condition of not mastering all corresponding knowledge points;
(1.4) calculating the probability that the learner answers correctly according to the estimated ideal answer condition of the learner, the probability s that the learner makes wrong test questions under the condition that the learner masters a certain test question and examines all known points, and the probability g that the learner does wrong test questions under the condition that the learner does not master corresponding knowledge points:
Figure BDA0002765838660000152
(1.5) obtaining the total likelihood function of the DINA model:
Figure BDA0002765838660000153
wherein L is 2K；
(1.6) calculating knowledge grasping conditions of the learner using maximum likelihood estimation based on the obtained total likelihood function:
Figure BDA0002765838660000154
(1.7) calculating the scoring condition of the learner on a new test question according to the knowledge mastering condition of the learner:
Figure BDA0002765838660000155
in step S102, the extracting text content of the test question by using the recurrent neural network model, and constructing a mapping relationship between the learner learning state and the test question text information through the full connection layer according to the embodiment of the present invention, and predicting the score of the learner on the specific test question based on the text information includes:
(2.1) performing text word segmentation, stop word removal and other preprocessing on the test question text; processing the test questions by using a continuous Word bag model of a Word vector model Word2vec, predicting Word vectors of target words according to the Word vectors of a plurality of words in the context of the target words, vectorizing the pre-processed test questions, and obtaining an embedded expression of the Word level of the test questions;
(2.2) using the test question word vector as input, acquiring test question context characteristic representation by adopting a bidirectional long-and-short time memory neural network, and acquiring test question sentence level representation through mean pooling to obtain test question word embedding representation;
(2.3) constructing a full-connection deep neural network, performing feature fusion on the embedded vector of the test question words and the knowledge mastering vector of the learner to serve as input of the full-connection deep neural network, and outputting scores of specific students on the test question;
(2.4) bonding the results of the layers to each other1,y2,...,yn) Processed by an output unit:
Figure BDA0002765838660000161
obtaining a value between [0, 1], representing the probability of correct answer of the student to the test question text, comparing the value with real score data, and performing weight correction in the network;
and (2.5) training the network model by using the training set data to obtain a trained neural network model, and predicting the learner response result based on the text information by using the obtained trained neural network model.
In step (2.1), the text word segmentation for the test question text provided by the embodiment of the present invention includes: based on a mixed dictionary, a method combining a bidirectional maximum matching method and statistics is adopted to perform mixed word segmentation on the test question text.
In step (2.1), the removing stop words from the test question text provided by the embodiment of the present invention includes: and adding words which are irrelevant to sentences and test question text themes, do not contribute to test question labeling tasks and have low frequency in the test question text into the disabled word bank, and deleting words of mixed participles in the test question text appearing in the disabled word bank.
In step (2.2), the obtaining of the context feature representation of the test question by using the bidirectional long-short time memory neural network provided by the embodiment of the invention comprises:
the BilSTM network adopts two LSTMs to obtain the context characteristics of different test questions from opposite directions, and the formula is as follows:
Figure BDA0002765838660000162
Figure BDA0002765838660000171
wherein a is1,a2,b1And b2G (-) is a hidden layer activation function, which is a weight coefficient,
Figure BDA0002765838660000172
for the forward hidden layer output at time t,
Figure BDA0002765838660000173
outputting the backward hidden layer at the time t; and fusing hidden layer outputs of two directions at each moment to construct a final output ht：
Figure BDA0002765838660000174
Wherein c is1And c2F (-) is the output activation function for the weight coefficients.
In step (2.2), the obtaining of the sentence-level representation of the test question through the mean pooling process provided by the embodiment of the present invention includes:
obtaining the embedded expression E of the test words by average pooling at the test sentence levelh：
Eh＝mean(h1,...,ht)
Where mean (-) is the average pooling operation, i.e., taking the average of the eigenvalues as output within the domain.
In step (2.3), the fully-connected deep neural network provided by the embodiment of the present invention includes:
in the fully-connected deep neural network, the calculation method of the nth node value of the mth layer is as follows:
Figure BDA0002765838660000175
wherein N is the number of the units of the upper layer,
Figure BDA0002765838660000176
represents a weight coefficient from the ith cell of the m-1 th layer to the nth cell of the m-1 th layer;
and (5) obtaining the potential mapping relation between the knowledge mastery of the learner, the test question text information and the score by adopting a relu activation function.
In step S103, the method for constructing a probability matrix decomposition objective function based on the obtained predicted score of the learner based on the cognitive level and the text information according to the embodiment of the present invention includes:
(3.1) the learners are marked as R by the obtained score prediction based on the cognitive level and the score prediction based on the text information1,R2And constructing a potential answer representation of the learner on the test question by using a probability matrix decomposition algorithm:
Figure BDA0002765838660000177
u and V are a learner characteristic matrix and a score characteristic matrix in probability matrix decomposition respectively, and alpha and beta are adjusting parameters of the learning condition of the learner and test question text information respectively;
(3.2) constructing a final objective function of the probability matrix decomposition:
Figure BDA0002765838660000181
(3.3) optimizing the objective function by using a gradient descent method to obtain an optimal learner result characteristic matrix U, V:
Figure BDA0002765838660000182
Figure BDA0002765838660000183
(3.4) predicting the performance of the learner by using the optimal feature matrix U, V obtained by training:
Figure BDA0002765838660000184
in step S104, the step of calculating KL divergence by using the estimated learner knowledge mastering vector and the learner incremental knowledge mastering vector according to the embodiment of the present invention, and combining the potential scores of the learner on the test questions to select the test paper which enables the learner to have an increased knowledge mastering trend and is formed by the test questions with proper difficulty to perform the personalized test includes:
(4.1) obtaining all incremental knowledge base vectors of the learner based on the analysis to obtain knowledge base vectors of the learner
Figure RE-GDA0002903752560000181
Figure RE-GDA0002903752560000181
0≤d≤D；
(4.2) calculating learner knowledge learning vector eta estimated by the cognitive diagnosis modeliMastery with incremental knowledge of all learners
Figure RE-GDA0002903752560000182
KL divergence measure of (1):
Figure BDA0002765838660000185
(4.3) selecting test papers which enable learners to have an increased knowledge mastering trend and are suitable in difficulty to form a personalized test:
Figure BDA0002765838660000186
as shown in fig. 3, the personalized test paper system integrating the learner's cognitive characteristics and the test question text information provided in the embodiment of the present invention includes:
the cognitive level-based score prediction module 1 is used for estimating and calculating the knowledge mastering condition of the learner by utilizing a cognitive diagnosis model according to the real answering condition of the learner and the test question recognition point distribution, and predicting the cognitive level-based score of the learner on a specific test question;
the score prediction module 2 based on the text information is used for extracting the text content of the test questions by utilizing a recurrent neural network model, constructing a mapping relation between the learning state of the learner and the text information of the test questions through a full connection layer, and predicting the score of the learner on the specific test questions based on the text information;
the question selecting strategy module 3 is used for constructing a probability matrix decomposition objective function based on the obtained predicted scores of the learner based on the cognitive level and the text information and predicting the potential scores of the learner on the specific test questions; and calculating KL divergence by utilizing the estimated learner knowledge control vector and the learner incremental knowledge control vector, and selecting the test paper which ensures that the knowledge control trend of the learner is increased and the test paper with proper difficulty is formed by combining the potential scores of the learner on the test paper.
The technical effects of the present invention will be further described with reference to specific embodiments.
Example 1:
the invention discloses a personalized paper combining method integrating the cognitive characteristics of learners and test question text information, which comprises the following steps:
step one, mining learning states such as knowledge mastering conditions of learners by using a cognitive diagnosis model according to real answering conditions and test question knowledge point distribution of the learners.
And step two, extracting the text content of the test question by using a recurrent neural network model, and constructing a mapping relation between the learning state of the learner and the text information of the test question through a full connection layer.
And step three, constructing a probability matrix decomposition objective function by integrating the learning state of the learner, the text information of the test questions and the cognitive commonality of the learner, and excavating the potential scores of the learner on the specific test questions.
And step four, calculating KL divergence by utilizing the estimated knowledge mastering vector of the learner and the incremental knowledge mastering vector of the learner, and selecting test questions with increased knowledge mastering tendency and proper difficulty to form the test paper for the personalized test by combining the potential scores of the learner on the test questions.
Further, the first step comprises:
step a): collecting test question knowledge point distribution marked by domain experts and answer data of learners;
step b): calculating a test question knowledge point representation Q matrix required to be used in the cognitive diagnosis model according to the distribution of test question knowledge points labeled by domain experts;
step c): according to the prior knowledge point mastering mode eta of the learner, calculating the ideal response condition of the learner:
Figure BDA0002765838660000201
πijexpressing the ideal answer of the learner i on the jth test questionCase, ηikRepresenting the mastery of the knowledge point k by the learner i, qjkWhether the known jth test question examines a knowledge point k is represented;
step d): estimating the probability s of a learner doing wrong test questions under the condition of mastering a certain test question and examining all knowledge points according to an expectation maximization algorithm, and estimating the probability g of the learner doing wrong test questions under the condition of not mastering all corresponding knowledge points;
step e): calculating the probability that the learner answers correctly according to the estimated ideal answering condition of the learner, the probability s that the learner answers wrong questions under the condition that the learner examines all knowledge points while mastering a certain test question, and the probability g that the learner answers the test questions under the condition that the learner does not master all the corresponding knowledge points:
Figure BDA0002765838660000202
step f): the overall likelihood function of the DINA model is thus obtained:
Figure BDA0002765838660000203
wherein L is 2KDue to the inclusion of an implicit variable ηlAnd the maximum likelihood estimation cannot be directly carried out, so that the expectation maximization method is adopted to solve the following steps:
step g): step E, utilizing the s obtained in the previous roundjAnd gjEstimating the calculation matrix P (R | eta) [ P (R) ]i|ηl)]I×LAnd using P (R | α) to calculate a matrix P (η | R) ═ P (η | R)l|Ri)]L×IThe value of (c).
Step h): and M: respectively order
Figure BDA0002765838660000204
And
Figure BDA0002765838660000205
the following can be obtained:
Figure BDA0002765838660000206
wherein, therein
Figure BDA0002765838660000207
Representing the expectation of the number of learners who lack at least one required knowledge point of the j-th question in the learners who belong to the first knowledge point grasping mode,
Figure BDA0002765838660000208
to represent
Figure BDA0002765838660000209
The number of people answering the correct jth question expects,
Figure BDA00027658386600002010
and
Figure BDA00027658386600002011
means of
Figure BDA00027658386600002012
And
Figure BDA00027658386600002013
similarly, the difference lies in
Figure BDA00027658386600002014
And
Figure BDA00027658386600002015
is an expectation in the case where the learner mastered all the required knowledge points for the jth question. So that it can be calculated from the estimate obtained in step E
Figure BDA0002765838660000211
And
Figure BDA0002765838660000212
and thus a new value of s is obtainedjAnd gjAnd (6) estimating.
Step i): calculating knowledge mastery of the learner using maximum likelihood estimation using the total likelihood function:
Figure BDA0002765838660000213
step j): calculating the scoring condition of the learner on a new test question according to the knowledge mastering condition of the learner:
Figure BDA0002765838660000214
further, the second step comprises:
step 1): preprocessing a test question text, wherein the preprocessing mainly comprises text word segmentation and stop word removal;
step 2): and performing text word segmentation on the test question. Based on a mixed dictionary, performing mixed word segmentation on the test text by adopting a method of combining a bidirectional maximum matching method with statistics;
step 3): and stopping words according to the mixed word segmentation result. And words which are irrelevant to the subjects of the sentences and the test questions and do not contribute to the test question labeling task are removed, and words with low frequency also do not contribute to the test question labeling task, so that the words with low frequency are also treated as stop words. According to the two rules, a stop word bank is established, words appearing in the stop word bank are deleted, and words with low frequency are deleted;
step 4): processing the test questions by using a Continuous Bag Of Words (CBOW) model Of a Word vector model Word2vec, vectorizing the pre-processed input test questions, and obtaining an embedded expression Of the Word level Of the test questions. The CBOW model predicts word vectors of the target words according to the word vectors of a plurality of words in the context of the target words, and vectorizes the test questions;
step 5): the test word vector is used as input, a Bidirectional Long Short-Term Memory neural network (BilSTM) is firstly adopted to obtain a test context feature representation, then a mean pooling operation is introduced to obtain a test sentence level representation, and thus a test word embedding representation is obtained.
Step 6): the BilSTM network adopts two LSTMs to acquire the context characteristics of different test questions from opposite directions, and the calculation is defined as:
Figure BDA0002765838660000215
Figure BDA0002765838660000221
wherein a is1,a2,b1And b2G (-) is a hidden layer activation function, which is a weight coefficient,
Figure BDA0002765838660000222
for the forward hidden layer output at time t,
Figure BDA0002765838660000223
for the backward hidden layer output at the time t, finally fusing the hidden layer outputs in two directions at each time to construct a final output ht：
Figure BDA0002765838660000224
Wherein c is1And c2F (-) is the output activation function, which is the weight coefficient;
step 7): obtaining the embedded expression E of the test words by average pooling at the test sentence levelh：
Eh＝mean(h1,...,ht)
Mean (-) is average pooling operation, namely average of characteristic values is taken as output in the field, representative information in the whole window information can be obtained, and feature dimensions of test question texts and the number of model network parameters are reduced.
Step 8): and constructing a fully-connected deep neural network, performing characteristic fusion on the test question word embedded vector and the learner knowledge mastering vector to serve as input of the network, and outputting the score of a specific student on the test question.
Step 9): in the fully-connected deep neural network, the calculation method of the nth node value of the mth layer is as follows:
Figure BDA0002765838660000225
wherein N is the number of the units of the upper layer,
Figure BDA0002765838660000226
represents a weight coefficient from the ith cell of the m-1 th layer to the nth cell of the m-1 th layer;
obtaining a potential mapping relation between knowledge mastering of the learner and test question text information and scores by adopting a relu activation function;
step 10): will fully connect the results of the layers (y)1,y2,...,yn) Processed by an output unit:
Figure BDA0002765838660000227
obtaining a value between [0, 1], representing the probability of correct answer of the student to the test question text, and comparing the value with the real score data, thereby realizing weight correction in the network;
step 11): after the training of the training set data, a trained neural network model can be obtained, so that the response performance of learners based on text information can be predicted.
Further, the third step comprises:
step A): respectively recording the learner predicted by the cognitive level and the learner predicted by the text information obtained by the second step and the third step as R1,R2And constructing a potential answer representation of the learner on the test question by using a probability matrix decomposition algorithm:
Figure BDA0002765838660000231
u and V are a learner characteristic matrix and a score characteristic matrix in probability matrix decomposition respectively, and alpha and beta are adjusting parameters of the learning condition of the learner and test question text information respectively;
step B): and deducing a final objective function of probability matrix decomposition according to a result calculation formula integrating the learning condition of the learner, the test question text information and the commonality of the learner:
Figure BDA0002765838660000232
step C): optimizing an objective function by using a gradient descent method, and firstly, respectively solving partial derivatives of two characteristic matrixes U and V according to the objective function:
Figure BDA0002765838660000233
Figure BDA0002765838660000234
respectively setting the partial derivatives as 0 to obtain the recursive formula iterative calculation of the method until the result is converged or the maximum iterative times are reached, and finally obtaining the optimal learner result characteristic matrix U, V:
Figure BDA0002765838660000235
Figure BDA0002765838660000236
step D): and finally, obtaining a result of performance prediction of the learner by using the optimal feature matrix U and V obtained by training:
Figure BDA0002765838660000237
step E): and adjusting hyper-parameters in the experiment and adjusting parameters alpha and beta of the learner learning condition and test question text information to obtain the parameters most suitable for the data set, thereby obtaining a final training model.
Further, the fourth step comprises:
step I): recording the learners' knowledge vector obtained by analysis as etaiFrom ηiObtaining all incremental knowledge mastering vectors eta of learnersi (d)，0≤d≤D；
Step II): calculating learner knowledge mastering vector eta estimated by cognitive diagnosis modeliWith all learners' incremental knowledgei (d)KL divergence measure of (1):
Figure BDA0002765838660000241
step III): therefore, the test paper for the personalized test is formed by selecting the test questions with the appropriate difficulty, which increase the knowledge mastering trend of the learner, so that a recommendation result with the learning condition of the learner, the text information of the test questions and the commonality of the learner is obtained.
Figure BDA0002765838660000242
Compared with other test question knowledge estimation methods, the personalized test paper combining method fusing the cognitive characteristics of learners and test question text information compares the Precision @ K, the Recall rate Recall @ K and the F1 value F1@ K, and the calculation method comprises the following steps:
Figure BDA0002765838660000243
Figure BDA0002765838660000244
Figure BDA0002765838660000245
wherein l (i) represents the customized learning test questions formulated for the ith learner, m (i) represents the test questions matched with the learner in the question bank, and l (i) andm (i) represents the intersection of the two. The Precision @ K represents the probability of the correct recommendation in the recommendation result, the Recall ratio Recall @ K is also called Recall ratio and represents the degree that the recommendation result matches the correct recommendation in the question bank, and the Precision ratio and the Recall ratio are in a certain amount of contradiction, namely the Recall ratio is low when the Precision ratio is high. In order to conveniently display the experimental results, the traditional cognitive diagnosis method is recorded as DINA, and the traditional collaborative filtering method is recorded as PMF.
The comparison results of the accuracy, recall rate and F1 value of the personalized test paper combining method with the learner cognitive characteristics and the test question text information and other methods in the same data set are shown in Table 1.
TABLE 1 comparison of the results
Figure BDA0002765838660000251
The experimental results show that: the personalized paper combining method for integrating the cognitive characteristics of the learner and the test question text information combines the cognitive characteristics of the learner, and the learner learns the commonalities and the test question text information. The accuracy of the volume result is obviously better than that of other comparative experiments. Therefore, experiments show that the personalized paper combining method fusing the cognitive characteristics of learners and test question text information is more effective than other methods in the aspects of accuracy, recall ratio, F1 value and the like.
Meanwhile, analysis shows that the paper grouping model and the probability matrix decomposition method based on the DINA are slightly unstable, and the accuracy rate of the paper grouping model and the probability matrix decomposition method is reduced along with the increase of the number of paper grouping test questions. Although the conventional probability matrix decomposition is easy to implement, the potential information in the extracted data set is insufficient, which leads to the low precision of the probability matrix decomposition under general conditions, especially when a large amount of training data is faced. In a word, the personalized test paper combining method combining the cognitive characteristics of the learner and the text information of the test questions has the best experimental effect, and the best test paper which can combine the test questions with different difficulty levels and promote the cognitive growth of the learner can be constructed according to the test difficulty and the examination and check target.
The invention introduces test question text information as an important measurement index of the personalized group paper, so that key information in the test question text can be utilized by the method provided by the invention.
The invention integrates test question text information, cognitive diagnosis technology and a collaborative filtering method, the learner learning condition obtained from the test question text information and the cognitive diagnosis is integrated into an objective function with collaborative filtering optimization, and the relation among the three is introduced into adjustment parameter adjustment, so as to obtain an optimal paper group model matched with the current data set.
In conclusion, the personalized test paper combining method for combining the cognitive characteristics of the learner and the test question text information provided by the invention realizes a more accurate test paper combining method, the method combines three aspects of information of cognitive diagnosis, test question text information and learning commonality of the learner to make a test paper combining strategy for the target learner, the test paper combining result can be defined according to the test target and the test question difficulty, the self-learning efficiency of the learner is greatly improved, and the learner is helped to make up short boards on the knowledge of the learner in a classroom more quickly. The method can be applied to the fields of education resource recommendation and evaluation, education data mining and the like, so that effective support is provided for follow-up education resource recommendation and the like, an online education platform is assisted, a digital education platform can better predict the score of a learner, and an individualized test paper combination scheme is provided, so that the diagnosis of weak links of knowledge points of the learner is efficiently carried out, and accurate remedial measures are taken.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any modification, equivalent replacement, and improvement made by those skilled in the art within the technical scope of the present invention disclosed in the present invention should be covered within the scope of the present invention.

Claims (10)

1. A personalized paper-making system integrating the cognitive characteristics of learners and test question text information is characterized in that an information data processing terminal is applied, and the personalized paper-making system integrating the cognitive characteristics of learners and the test question text information comprises:
the cognitive level-based score prediction module is used for estimating and calculating the knowledge mastering condition of the learner by utilizing a cognitive diagnosis model according to the real answering condition and the test question knowledge point distribution of the learner and predicting the cognitive level-based score of the learner on a specific test question;
the score prediction module based on the text information is used for extracting the text content of the test questions by utilizing the recurrent neural network model, constructing the mapping relation between the learning state of the learner and the text information of the test questions through a full connection layer and predicting the score of the learner on the specific test questions based on the text information;
the topic selection strategy module is used for constructing a probability matrix decomposition objective function based on the obtained predicted scores of the learner based on the cognitive level and the text information and predicting the potential scores of the learner on the specific test questions; and calculating KL divergence by utilizing the estimated learner knowledge control vector and the learner incremental knowledge control vector, and selecting test questions with increased learner knowledge control trend and proper difficulty to form the test paper for the personalized test by combining the potential scores of the learners on the test questions.
2. A personalized paper combining method for combining the cognitive characteristics of learners and test question text information is characterized by being applied to an information data processing terminal and comprising the following steps:
estimating and calculating knowledge mastering conditions of the learner by using a cognitive diagnosis model according to the real answering conditions and the test question knowledge point distribution of the learner, and predicting the score of the learner on a specific test question based on the cognitive level;
extracting the text content of the test questions by using a recurrent neural network model, constructing a mapping relation between the learning state of the learner and the text information of the test questions through a full connection layer, and predicting the score of the learner on the specific test questions based on the text information;
constructing a probability matrix decomposition objective function based on the obtained predicted scores of the learner based on the cognitive level and the text information, and predicting the potential scores of the learner on the specific test questions;
and calculating KL divergence by utilizing the estimated learner knowledge control vector and the learner incremental knowledge control vector, and selecting test questions with increased learner knowledge control trend and proper difficulty to form the test paper for the personalized test by combining the potential scores of the learners on the test questions.
3. The method as claimed in claim 2, wherein the step of estimating and calculating knowledge mastery of the learner using the cognitive diagnosis model according to the real answer situation of the learner and the distribution of the knowledge points of the test questions and predicting the cognitive level-based score of the learner on the specific test questions comprises:
(1.1) collecting test question knowledge point distribution labeled by domain experts and response data of learners; calculating a test question knowledge point representation Q matrix required to be used in the cognitive diagnosis model according to the distribution of test question knowledge points labeled by domain experts;
(1.2) calculating the ideal response condition of the learner according to the prior knowledge point mastering mode eta of the learner:
Figure FDA0002765838650000021
wherein, piijShows the ideal answer situation of the learner i on the jth test question etaikRepresenting the mastery of the knowledge point k by the learner i, qjkWhether the known jth test question examines a knowledge point k is represented;
(1.3) estimating the probability s of the learner doing wrong test questions under the condition of mastering a certain test question and examining all knowledge points according to an expectation-maximization algorithm, and estimating the probability g of the learner doing wrong test questions under the condition of not mastering all corresponding knowledge points;
(1.4) calculating the probability that the learner answers correctly according to the estimated ideal answer condition of the learner, the probability s that the learner makes a wrong test question under the condition that the learner masters a certain test question and examines all knowledge points, and the probability g that the learner does a wrong test question under the condition that the learner does not master corresponding knowledge points:
Figure FDA0002765838650000022
(1.5) obtaining the total likelihood function of the DINA model:
Figure FDA0002765838650000023
wherein L is 2K；
(1.6) calculating knowledge mastery of the learner using maximum likelihood estimation based on the obtained total likelihood function:
Figure FDA0002765838650000031
(1.7) calculating the scoring condition of the learner on a new test question according to the knowledge mastering condition of the learner:
Figure FDA0002765838650000032
4. the method as claimed in claim 2, wherein the step of extracting the text contents of the test questions by using the recurrent neural network model and constructing the mapping relationship between the learning state of the learner and the text information of the test questions through the full link layer comprises the steps of:
(2.1) performing text word segmentation, stop word removal and other preprocessing on the test question text; processing the test questions by using a continuous Word bag model of a Word vector model Word2vec, predicting Word vectors of target words according to the Word vectors of a plurality of words in the context of the target words, vectorizing the pre-processed test questions, and obtaining Word-level embedded representation of the test questions;
(2.2) using the test word vector as input, adopting a bidirectional long-and-short time memory neural network to obtain the context characteristic representation of the test question, obtaining sentence-level representation of the test question through mean pooling treatment, and obtaining the embedded representation of the test word;
(2.3) constructing a full-connection deep neural network, performing feature fusion on the test question word embedded vector and the learner knowledge mastering vector to serve as input of the full-connection deep neural network, and outputting scores of specific students on the test questions;
(2.4) bonding the results of the layers to each other1,y2,...,yn) Processed by an output unit:
Figure FDA0002765838650000033
obtaining a value between [0, 1], representing the probability of correct answer of the student to the test question text, comparing the value with real score data, and performing weight correction in the network;
and (2.5) training the network model by using the training set data to obtain a trained neural network model, and predicting the learner response result based on the text information by using the obtained trained neural network model.
5. The personalized test paper method combining learner's cognitive characteristics and test question text information as claimed in claim 4, wherein in the step (2.1), the text-segmenting the test question text comprises: based on a mixed dictionary, performing mixed word segmentation on the test question text by adopting a method of combining a bidirectional maximum matching method with statistics;
in the step (2.1), the removing of stop words from the test question text includes: adding words which are irrelevant to sentences and test question text themes, do not contribute to test question labeling tasks and are low in frequency in the test question text into a disabled word bank, and deleting words of mixed participles in the test question text appearing in the disabled word bank;
in the step (2.2), the obtaining of the context feature representation of the test question by using the bidirectional long-and-short term memory neural network comprises:
the BilSTM network adopts two LSTMs to obtain the context characteristics of different test questions from opposite directions, and the formula is as follows:
Figure FDA0002765838650000041
Figure FDA0002765838650000042
wherein a is1,a2,b1And b2G (-) is a hidden layer activation function, which is a weight coefficient,
Figure FDA0002765838650000043
for the forward hidden layer output at time t,
Figure FDA0002765838650000044
outputting the backward hidden layer at the time t; and fusing hidden layer outputs of two directions at each moment to construct a final output ht：
Figure FDA0002765838650000045
Wherein c is1And c2F (-) is the output activation function for the weight coefficients.
6. The personalized test paper combining learner's cognitive characteristics and test question text information as claimed in claim 4, wherein in step (2.2), said obtaining the sentence-level representation of the test question through the mean pooling process comprises:
in the test question sentence level representation layer, test question word embedding is obtained through average poolingIn represents Eh：
Eh＝mean(h1,...,ht)；
Wherein mean (-) is an average pooling operation, i.e. taking the average of the eigenvalues as output in the domain;
in step (2.3), the fully-connected deep neural network includes:
in the fully-connected deep neural network, the calculation method of the nth node value of the mth layer is as follows:
Figure FDA0002765838650000046
wherein N is the number of the units of the upper layer,
Figure FDA0002765838650000051
represents a weight coefficient from the ith cell of the m-1 th layer to the nth cell of the m-1 th layer;
and (5) obtaining a potential mapping relation between knowledge mastering of the learner and test question text information and scores by adopting a relu activation function.
7. The method as claimed in claim 2, wherein the constructing the probability matrix decomposition objective function based on the obtained learner's cognitive characteristics and text information based prediction score based on cognitive level and text information comprises:
(3.1) recording the obtained learner as R according to the cognitive level score prediction and the text information score prediction1,R2And constructing a potential answer representation of the learner on the test question by using a probability matrix decomposition algorithm:
Figure FDA0002765838650000052
u and V are a learner characteristic matrix and a score characteristic matrix in probability matrix decomposition respectively, and alpha and beta are adjusting parameters of the learning condition of the learner and test question text information respectively;
(3.2) constructing a final objective function of the probability matrix decomposition:
Figure FDA0002765838650000053
(3.3) optimizing the objective function by using a gradient descent method to obtain an optimal learner result characteristic matrix U, V:
Figure FDA0002765838650000054
Figure FDA0002765838650000055
(3.4) predicting the performance of the learner by using the optimal feature matrix U, V obtained by training:
Figure FDA0002765838650000056
the method for calculating KL divergence by utilizing the estimated learner knowledge mastering vector and the learner incremental knowledge mastering vector and combining the potential scores of the learner on the test questions to select the test paper which enables the knowledge mastering trend of the learner to be increased and enables the test questions with proper difficulty to form the personalized test comprises the following steps:
(4.1) obtaining all incremental knowledge base vectors of the learner based on the analysis to obtain knowledge base vectors of the learner
Figure DEST_PATH_FDA0002903752550000046
0≤d≤D；
(4.2) calculating learner knowledge learning vector eta estimated by the cognitive diagnosis modeliWith all learners' incremental knowledge
Figure DEST_PATH_FDA0002903752550000051
KL divergence measure of (1):
Figure FDA0002765838650000061
(4.3) selecting test papers which enable learners to have an increased knowledge mastering trend and are suitable in difficulty to form a personalized test:
Figure FDA0002765838650000062
8. a computer device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the personalized group paper method of fusing learner cognitive characteristics and test question text information of claims 1-7.
9. A computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the personalized test paper composition method fusing learner's cognitive characteristics and test question text information as set forth in claims 1 to 7.
10. An information data processing terminal, characterized in that, the information data processing terminal is used to realize the personalized paper-making method of integrating learner cognitive characteristics and test question text information as claimed in claims 1-7.
CN202011233044.0A 2020-11-06 2020-11-06 Personalized paper grouping method and system integrating cognition characteristics and test question text information Active CN112508334B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011233044.0A CN112508334B (en) 2020-11-06 2020-11-06 Personalized paper grouping method and system integrating cognition characteristics and test question text information

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011233044.0A CN112508334B (en) 2020-11-06 2020-11-06 Personalized paper grouping method and system integrating cognition characteristics and test question text information

Publications (2)

Publication Number Publication Date
CN112508334A true CN112508334A (en) 2021-03-16
CN112508334B CN112508334B (en) 2023-09-01

Family

ID=74955388

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011233044.0A Active CN112508334B (en) 2020-11-06 2020-11-06 Personalized paper grouping method and system integrating cognition characteristics and test question text information

Country Status (1)

Country Link
CN (1) CN112508334B (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113127734A (en) * 2021-03-30 2021-07-16 西安理工大学 Personalized test paper pushing method based on genetic algorithm
CN113221007A (en) * 2021-05-21 2021-08-06 合肥工业大学 Method for recommending answering behavior
CN113377942A (en) * 2021-07-12 2021-09-10 北京乐学帮网络技术有限公司 Test paper generation method and device, computer equipment and storage medium
CN113569870A (en) * 2021-07-31 2021-10-29 西北工业大学 Cross-modal problem Q matrix automatic construction method based on heterogeneous graph neural network
CN113743083A (en) * 2021-09-06 2021-12-03 东北师范大学 Test question difficulty prediction method and system based on deep semantic representation
CN114238613A (en) * 2022-02-22 2022-03-25 北京一起航帆科技有限公司 Method and device for determining mastery degree of knowledge points and electronic equipment
CN114418443A (en) * 2022-01-28 2022-04-29 华南师范大学 Test paper quality detection method, system, device and storage medium
CN116431800A (en) * 2023-03-03 2023-07-14 广州秒可科技有限公司 Examination interface generation method, device and readable storage medium
CN117763361A (en) * 2024-02-22 2024-03-26 泰山学院 Student score prediction method and system based on artificial intelligence
CN117952797A (en) * 2024-03-21 2024-04-30 深圳市华师兄弟教育科技有限公司 Internet education supervision system and method based on artificial intelligence

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2896037A1 (en) * 2014-07-03 2016-01-03 Mentorum Solutions Inc. Adaptive e-learning system and method
CN108711125A (en) * 2018-07-24 2018-10-26 电子科技大学 A kind of blocks of knowledge centrad and difficulty quantization method
CN110264091A (en) * 2019-06-24 2019-09-20 中国科学技术大学 Student's cognitive diagnosis method
US20190318650A1 (en) * 2018-04-11 2019-10-17 Electronics And Telecommunications Research Institute Method and apparatus for learner diagnosis using reliability of cognitive diagnostic model
CN110502636A (en) * 2019-08-27 2019-11-26 华中师范大学 A kind of joint modeling and method for digging and system towards subjective and objective examination question
CN110516116A (en) * 2019-08-27 2019-11-29 华中师范大学 A kind of the learner's human-subject test method for digging and system of multistep layering
CN111241243A (en) * 2020-01-13 2020-06-05 华中师范大学 Knowledge measurement-oriented test question, knowledge and capability tensor construction and labeling method
CN111445153A (en) * 2020-03-31 2020-07-24 华中师范大学 Objective test question attribute mode estimation and correction method and system for education measurement

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2896037A1 (en) * 2014-07-03 2016-01-03 Mentorum Solutions Inc. Adaptive e-learning system and method
US20190318650A1 (en) * 2018-04-11 2019-10-17 Electronics And Telecommunications Research Institute Method and apparatus for learner diagnosis using reliability of cognitive diagnostic model
CN108711125A (en) * 2018-07-24 2018-10-26 电子科技大学 A kind of blocks of knowledge centrad and difficulty quantization method
CN110264091A (en) * 2019-06-24 2019-09-20 中国科学技术大学 Student's cognitive diagnosis method
CN110502636A (en) * 2019-08-27 2019-11-26 华中师范大学 A kind of joint modeling and method for digging and system towards subjective and objective examination question
CN110516116A (en) * 2019-08-27 2019-11-29 华中师范大学 A kind of the learner's human-subject test method for digging and system of multistep layering
CN111241243A (en) * 2020-01-13 2020-06-05 华中师范大学 Knowledge measurement-oriented test question, knowledge and capability tensor construction and labeling method
CN111445153A (en) * 2020-03-31 2020-07-24 华中师范大学 Objective test question attribute mode estimation and correction method and system for education measurement

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
朱天宇;黄振亚;陈恩红;刘淇;吴润泽;吴乐;苏喻;陈志刚;胡国平;: "基于认知诊断的个性化试题推荐方法", 计算机学报, no. 01, pages 176 - 191 *
李全;刘兴红;许新华;林松;: "基于联合概率矩阵分解的个性化试题推荐方法", 计算机应用, no. 03, pages 639 - 643 *
齐斌；邹红霞；王宇；李冀兴: "基于协同过滤和认知诊断的试题推荐方法", 《计算机软件及计算机应用》, pages 235 - 240 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113127734A (en) * 2021-03-30 2021-07-16 西安理工大学 Personalized test paper pushing method based on genetic algorithm
CN113127734B (en) * 2021-03-30 2022-11-25 江苏乐易智慧科技有限公司 Personalized test paper pushing method based on genetic algorithm
CN113221007B (en) * 2021-05-21 2022-09-23 合肥工业大学 Method for recommending answering behavior
CN113221007A (en) * 2021-05-21 2021-08-06 合肥工业大学 Method for recommending answering behavior
CN113377942A (en) * 2021-07-12 2021-09-10 北京乐学帮网络技术有限公司 Test paper generation method and device, computer equipment and storage medium
CN113569870A (en) * 2021-07-31 2021-10-29 西北工业大学 Cross-modal problem Q matrix automatic construction method based on heterogeneous graph neural network
CN113743083A (en) * 2021-09-06 2021-12-03 东北师范大学 Test question difficulty prediction method and system based on deep semantic representation
CN113743083B (en) * 2021-09-06 2024-03-12 东北师范大学 Test question difficulty prediction method and system based on deep semantic characterization
CN114418443A (en) * 2022-01-28 2022-04-29 华南师范大学 Test paper quality detection method, system, device and storage medium
CN114238613A (en) * 2022-02-22 2022-03-25 北京一起航帆科技有限公司 Method and device for determining mastery degree of knowledge points and electronic equipment
CN116431800A (en) * 2023-03-03 2023-07-14 广州秒可科技有限公司 Examination interface generation method, device and readable storage medium
CN116431800B (en) * 2023-03-03 2023-11-03 广州秒可科技有限公司 Examination interface generation method, device and readable storage medium
CN117763361A (en) * 2024-02-22 2024-03-26 泰山学院 Student score prediction method and system based on artificial intelligence
CN117763361B (en) * 2024-02-22 2024-04-30 泰山学院 Student score prediction method and system based on artificial intelligence
CN117952797A (en) * 2024-03-21 2024-04-30 深圳市华师兄弟教育科技有限公司 Internet education supervision system and method based on artificial intelligence
CN117952797B (en) * 2024-03-21 2024-06-14 深圳市华师兄弟教育科技有限公司 Internet education supervision system and method based on artificial intelligence

Also Published As

Publication number Publication date
CN112508334B (en) 2023-09-01

Similar Documents

Publication Publication Date Title
CN112508334B (en) Personalized paper grouping method and system integrating cognition characteristics and test question text information
CN111241243B (en) Test question, knowledge and capability tensor construction and labeling method oriented to knowledge measurement
CN107590127B (en) Automatic marking method and system for question bank knowledge points
CN110941723A (en) Method, system and storage medium for constructing knowledge graph
CN112257966B (en) Model processing method and device, electronic equipment and storage medium
CN117150151B (en) Wrong question analysis and test question recommendation system and method based on large language model
CN113283488B (en) Learning behavior-based cognitive diagnosis method and system
CN107544956A (en) A kind of text wants point detecting method and system
CN111326040A (en) Intelligent test and intelligent tutoring system and method for Chinese reading understanding
CN109582974A (en) A kind of student enrollment's credit estimation method and device based on deep learning
CN114201684A (en) Knowledge graph-based adaptive learning resource recommendation method and system
CN112667776B (en) Intelligent teaching evaluation and analysis method
CN114254127A (en) Student ability portrayal method and learning resource recommendation method and device
CN111460101A (en) Knowledge point type identification method and device and processor
CN116258056A (en) Multi-modal knowledge level assessment and learning performance prediction method, system and medium
Wang et al. Utilizing artificial intelligence to support analyzing self-regulated learning: A preliminary mixed-methods evaluation from a human-centered perspective
Gaheen et al. Automated students arabic essay scoring using trained neural network by e-jaya optimization to support personalized system of instruction
CN115455186A (en) Learning situation analysis method based on multiple models
CN117473041A (en) Programming knowledge tracking method based on cognitive strategy
CN114780723A (en) Portrait generation method, system and medium based on guide network text classification
Arifin et al. Automatic essay scoring for Indonesian short answers using siamese Manhattan long short-term memory
CN110334204A (en) A kind of exercise similarity calculation recommended method based on user record
CN112785039B (en) Prediction method and related device for answer score rate of test questions
CN114358579A (en) Evaluation method, evaluation device, electronic device, and computer-readable storage medium
CN113919983A (en) Test question portrait method, device, electronic equipment and storage medium

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
GR01 Patent grant
GR01 Patent grant