CN114429212A

CN114429212A - Intelligent learning knowledge ability tracking method, electronic device and storage medium

Info

Publication number: CN114429212A
Application number: CN202011105897.6A
Authority: CN
Inventors: 郭力军; 罗彤
Original assignee: Beijing Ronghui Jinxin Information Technology Co ltd
Current assignee: Beijing Ronghui Jinxin Information Technology Co ltd
Priority date: 2020-10-15
Filing date: 2020-10-15
Publication date: 2022-05-03

Abstract

The invention provides an intelligent learning knowledge ability tracking method, electronic equipment and a storage medium, wherein the method comprises the following steps: inputting the identification of the user to be tested, the current question, the prior relevance weight of the current question and each knowledge point into a knowledge ability tracking model, and acquiring the answer probability of the user to be tested for the current question; the knowledge ability tracking model comprises the mastery state of the user to be tested on each knowledge point. According to the invention, the prior relation between the question and each knowledge point is introduced, and the relation between the network learning question and each knowledge point is combined, so that the explicit mapping corresponding relation between the current question and each knowledge point is established, therefore, not only can the correct prediction of the answer probability of the current question be realized, but also the knowledge point mastering state matrix can be updated according to the answer result of the current question, and the mastering state of the user to be tested on each knowledge point can be obtained.

Description

Intelligent learning knowledge ability tracking method, electronic device and storage medium

Technical Field

The invention relates to the technical field of computers, in particular to an intelligent learning knowledge ability tracking method, electronic equipment and a storage medium.

Background

The knowledge ability tracking model is based on student behavior sequences for modeling and tracking the knowledge mastering state of students, and is the core and key for constructing the self-adaptive teaching system. In the self-adaptive teaching system, no matter precise subject pushing is carried out or path planning of student learning is carried out, the first step is to precisely estimate the mastering degree of the students to the knowledge points.

However, at present, deep learning models such as a recurrent neural network that directly models student behavior time-series data cannot acquire the degree of grasp of knowledge points by students. For example, when the model inputs the answer condition of a question, the model can predict the probability of answering all questions (done and not done) next, but since the model represents the current mastery condition of the student on all knowledge points by a hidden state (hidden state), the model cannot output the mastery condition of the student on a specific knowledge point. For another example, when the model inputs the answer condition of the question corresponding to the knowledge point(s), the probability of answering all the knowledge points (done and not done) in the next step can be predicted, but the knowledge point is not the mastery degree of the student, that is, the mastery condition of the student on the specific knowledge point still cannot be output.

Disclosure of Invention

To solve the problems in the prior art, embodiments of the present invention provide an intelligent learning knowledge ability tracking method, an electronic device, and a storage medium.

Specifically, the embodiment of the invention provides the following technical scheme:

in a first aspect, an embodiment of the present invention provides a method for tracking intelligent learning knowledge ability, including:

acquiring a current question corresponding to a user to be tested, and prior relevance weights of the current question and each knowledge point;

inputting the identification of the user to be tested, the current question, the prior relevance weight of the current question and each knowledge point into a knowledge ability tracking model, and acquiring the answer probability of the user to be tested for the current question; the knowledge ability tracking model comprises the mastery states of the to-be-tested user on the knowledge points, wherein the mastery states of the to-be-tested user on the knowledge points are determined according to questions made by the to-be-tested user in a historical time period, answering results of the questions made by the to-be-tested user, and prior relevance weights of the questions made by the to-be-tested user and the knowledge points;

the knowledge ability tracking model is obtained by inputting questions made by each user, answering results of the questions made by each user and prior relevance weights of the questions made by each user and each knowledge point in historical training data into the initial knowledge ability tracking model and training the initial knowledge ability tracking model based on a machine learning mode.

Further, the intelligent learning knowledge ability tracking method further includes:

and acquiring a response result of the user to be tested for the current question, and updating the mastery state of the user to be tested for each knowledge point according to the response result.

Furthermore, the knowledge ability tracking model comprises a question knowledge point mapping model;

correspondingly, inputting the user identifier to be tested, the current question, the prior relevance weight of the current question and each knowledge point into a knowledge ability tracking model, and acquiring the pairing probability of the user to be tested for the current question specifically comprises:

inputting the question feature data of the current question into a question knowledge point mapping model, and acquiring a first relevance weight of the current question and each knowledge point;

determining prior relevance weights of the current question and each knowledge point according to the information of the knowledge points marked on the current question;

determining the final correlation weight of the current question and each knowledge point according to the first correlation weight of the current question and each knowledge point and the prior correlation weight of the current question and each knowledge point, and determining the mapping corresponding relation of the current question and each knowledge point according to the final correlation weight of the current question and each knowledge point;

acquiring the mastery state of the to-be-detected user on each knowledge point from the knowledge point mastery state matrix corresponding to the to-be-detected user; the knowledge point mastering state matrix stores the mastering state of the user to be tested on each knowledge point; the knowledge point mastering state is used for representing the mastering degree of the knowledge point of the user to be tested;

determining knowledge point state characteristic data of the user to be tested on all knowledge points in the current question according to the mastery state of the user to be tested on all knowledge points and the final relevance weight of the current question and each knowledge point;

integrating the knowledge point state characteristic data and the question characteristic data to obtain a characteristic vector simultaneously containing question information and knowledge point state information;

predicting the answer probability of the user to be tested to the current question according to the characteristic vector simultaneously containing the question information and the knowledge point state information, and updating the mastery state of the user to be tested to each knowledge point in the knowledge point mastery state matrix according to the answer result of the user to be tested to the current question and the mapping corresponding relation between the current question and each knowledge point.

Further, determining a priori correlation weight between the current topic and each knowledge point according to the information of the knowledge point marked on the current topic, including:

when n knowledge points are marked on the current question, determining that the prior relevance weights corresponding to the n knowledge points are respectively 1/n (if the specific weight of each knowledge point is marked, the relevance weights are marking weights), and determining that the prior relevance weights of other knowledge points are all 0, wherein n is more than or equal to 1.

Further, determining a final relevance weight of the current topic and each knowledge point according to the first relevance weight of the current topic and each knowledge point and the prior relevance weight of the current topic and each knowledge point comprises:

and determining the final relevance weight of the current question and each knowledge point according to a first relation model, wherein the first relation model is as follows:

including, but not limited to, a weighted average, for example,

w_ti＝α*softmax(M_im_t)+(1-α)*{c_t}

wherein alpha is [0,1]]OfAdjusting parameters; w is a_tiRepresenting the final relevance weight of the current topic and a knowledge point i; softmax (M)_im_t) Representing a first relevance weight of the current topic and each knowledge point; { c_tExpressing the prior relevance weight of the current question and each knowledge point; m represents topic knowledge point mapping matrix, used for storing the mapping between the current topic and each knowledge point, and the size is n x d_mN is the number of knowledge points, d_mDimension of the subject feature data of the current subject; m_iLine i, M, representing M_tTopic feature data representing a current topic; wherein the expression of the soft max function is

Wherein i 1.

Further, the knowledge ability tracking model also comprises a forgetting attenuation network model;

accordingly, the method further comprises:

acquiring the time interval of updating the knowledge point mastering state matrix at the latest time from the current distance; wherein, the time interval is used for inputting the forgetting attenuation network model to obtain an attenuation rate;

correspondingly, acquiring the grasping state of the knowledge point of the user to be tested from the knowledge point grasping state matrix corresponding to the user to be tested, including:

and performing attenuation updating on the knowledge point grasping state in the knowledge point grasping state matrix corresponding to the user to be tested by using the attenuation rate, and acquiring the knowledge point grasping state after the attenuation updating.

Further, the forgetting attenuation network model is as follows: d_t＝sigmoid(W_dt_t+b_d) (ii) a Wherein d is_tRepresents the attenuation ratio, t_tRepresents a time interval, W_dAnd b_dDenotes a network parameter, W_dAnd b_dThe value of (a) is obtained by training; the expression of the sigmoid function is

Correspondingly, the attenuation updating of the knowledge point grasping state in the knowledge point grasping state matrix corresponding to the user to be tested by using the attenuation rate comprises the following steps:

according to the attenuation rate, a second relation model is adopted to perform attenuation updating on knowledge point grasping states in a knowledge point grasping state matrix corresponding to the user to be detected;

wherein the second relationship model is:

wherein N is_tThe knowledge point grasping state matrix is used for storing the grasping state of the user to be tested on each knowledge point; size n x d_nN is the number of knowledge points, d_nDimension, N, representing knowledge point mastery state_t-1Knowledge point grasping state matrix representing the latest update, N_tiRepresents N_tRow i of (1), N_t-1,iRepresents N_t-1The number of the ith row of (a),

representing dot product, 1 representing all 1 vectors, w_tiRepresenting the final relevance weight of the current topic and a knowledge point i; ()^TA transposed matrix representing ().

Further, determining knowledge point state feature data of the user to be tested for all knowledge points in the current question according to the mastery state of the user to be tested for each knowledge point and the final relevance weight of the current question and each knowledge point, includes:

determining knowledge point state characteristic data of the user to be tested for all knowledge points in the current question according to a third relation model; wherein the third relation model is:

wherein k is_tIndicating the user to be tested is rightKnowledge point state characteristic data of all knowledge points in the previous topic.

Further, integrating the knowledge point state feature data and the topic feature data to obtain a feature vector simultaneously containing topic information and knowledge point state information, including:

point state feature data k of knowledge_tAnd topic feature data m_tIntegrated together, obtaining a feature vector s by a fully connected neural network_t：

s_t＝tanh(W_s[k_t,m_t]+b_s)

Wherein the expression of the tanh function is:

W_fand b_fThe two network parameters are obtained through training.

Further, the knowledge ability tracking model further comprises: a capability network model and a question difficulty network model;

correspondingly, predicting the answer probability of the user to be tested for the current question according to the feature vector simultaneously containing the question information and the knowledge point state information, comprising the following steps:

acquiring the ability value of the user to be tested for the current question according to a feature vector simultaneously containing question information and knowledge point state information and an ability network model corresponding to the user to be tested;

acquiring a difficulty value of the current question based on the question feature data and the question difficulty network model;

and predicting the answer probability of the user to be tested aiming at the current question based on the capability value and the difficulty value.

Further, obtaining the ability value of the user to be tested for the current question according to a feature vector simultaneously containing question information and knowledge point state information and an ability network model corresponding to the user to be tested, the method includes:

using a capability network model theta_tj＝tanh(W_θs_t+b_θ) Acquiring the ability value of the user to be tested for the current question; wherein, theta_tjIndicates the value of the capability, W_θAnd b_θParameters representing the capability network model are obtained through training;

obtaining a difficulty value of the current topic according to the topic feature data and the topic difficulty network model, wherein the obtaining of the difficulty value of the current topic comprises the following steps:

using topic difficulty network model beta_j＝tanh(W_βm_t+b_β) Acquiring the difficulty value of the current question; wherein, beta_jDenotes the value of the degree of difficulty, W_βAnd b_βParameters representing the question difficulty network model are obtained through training;

predicting the answer probability of the user to be tested aiming at the current question based on the capability value and the difficulty value, and the method comprises the following steps: using a prediction model p_t＝sigmoid(a*θ_tj-β_j) Predicting the answer probability of the user to be tested for the current question; wherein p is_tThe answer probability is expressed, and according to the IRT theory, the value of a is 3.0.

Further, updating the grasping state of the user to be tested for each knowledge point in the knowledge point grasping state matrix according to the response result of the user to be tested for the current question and the mapping corresponding relationship between the current question and each knowledge point, including but not limited to:

according to the answer result vector v of the user to be tested for the current question_tAnd the mapping corresponding relation between the current question and each knowledge point, firstly erasing the original memory and then writing the new memory; wherein the erase vector and the write vector are e_tAnd h_t；

e_t＝sigmoid(W_en_t+b_e)

h_t＝tanh(W_hn_t+b_h)

Wherein, W_eAnd b_eThe network parameters are obtained by training in advance; w_hAnd b_hThe network parameters are obtained by training in advance;

erasing the original memory according to a fourth relation model, wherein the fourth relation model is as follows:

writing new memory according to a fifth relation model, wherein the fifth relation model is as follows:

wherein N is_t+1,iRepresenting the updated knowledge point mastery state matrix, N_t+1,iIs N_t+1Row i of (2).

Further, when the answer result is a binary 0 and 1 result; accordingly, the loss function is a binary cross entropy; when the response result is score [0,1 ]; accordingly, the loss function is a multivariate cross entropy;

in a second aspect, an embodiment of the present invention further provides a resource pushing method based on the intelligent learning knowledge ability tracking method in the first aspect, including:

and pushing test questions and/or learning resources for the user to be tested according to the mastery state of the user to be tested on each knowledge point.

In a third aspect, an embodiment of the present invention further provides an answer prediction method and a paper composition method based on knowledge ability tracking in the first aspect, where the answer prediction method and the paper composition method include:

automatically grouping the paper according to the mastery state of each knowledge point of each user to be tested in the appointed evaluation range and the knowledge points marked on the test questions of the paper to be grouped; and/or automatically grouping the paper according to the prediction scores of the users to be tested in the specified evaluation range on the test questions of the paper to be grouped; and/or mapping student groups in different areas to the same topic space according to the mastery states of the users to be tested in all the areas to the knowledge points and the predicted scores of the users to be tested in all the areas to different topics, so as to realize score comparability in different areas.

In a fourth aspect, embodiments of the present invention further provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the intelligent learning knowledge capability tracking method according to the first aspect when executing the program, and/or implements the steps of the resource pushing method according to the second aspect when executing the program, and/or implements the steps of the volume group method according to the third aspect when executing the program.

According to the above technical solution, in the intelligent learning knowledge ability tracking method, the electronic device and the storage medium provided by the embodiment of the present invention, firstly, the prior relevance weights of the current question, the current question and each knowledge point corresponding to the user to be tested are obtained, then, the prior relevance weights of the user identification, the current question and each knowledge point to be tested are input into the knowledge ability tracking model, and the answer probability of the user to be tested for the current question is obtained; the knowledge ability tracking model comprises the mastery states of the to-be-tested user on the knowledge points, wherein the mastery states of the to-be-tested user on the knowledge points are determined according to questions made by the to-be-tested user in a historical time period, answering results of the questions made by the to-be-tested user, and prior relevance weights of the questions made by the to-be-tested user and the knowledge points; the knowledge ability tracking model is obtained by inputting questions made by each user, answering results of the questions made by each user and prior relevance weights of the questions made by each user and each knowledge point in historical training data into the initial knowledge ability tracking model and training the initial knowledge ability tracking model based on a machine learning mode. According to the embodiment of the invention, the prior relation between the question and the knowledge point is introduced, and the relation between the network learning question and each knowledge point is combined, so that the explicit mapping corresponding relation between the current question and each knowledge point is established, the answer probability of the current question can be accurately predicted, the knowledge point mastering state matrix can be updated according to the answer result of the current question, and the mastering state of the user to be tested on each knowledge point can be obtained.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.

FIG. 1 is a flow chart of a method for tracking intelligent learning knowledge ability according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a knowledge capability tracking model according to an embodiment of the present invention;

fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

At present, deep learning models such as a recurrent neural network directly modeled aiming at student behavior time series data cannot acquire the mastery degree of students on knowledge points. For example, when the model inputs the answer condition of a question, the model can predict the probability of answering all questions (done and not done) next, but since the model represents the current mastery condition of the student on all knowledge points by a hidden state (hidden state), the model cannot output the mastery condition of the student on a specific knowledge point. For another example, when the model inputs the answer condition of the question corresponding to the knowledge point(s), the probability of answering all the knowledge points (done and not done) in the next step can be predicted, but the knowledge point is not the mastery degree of the student, that is, the mastery condition of the student on the specific knowledge point still cannot be output. In addition, in the time sequence, the situation of discontinuity and fluctuation change can occur, and in the prediction accuracy, the situation of model prediction accuracy reduction can be caused due to the loss of the topic dimension information.

It can be understood that the goal of the model is to be able to calculate the knowledge point grasping condition after the current input answer result besides predicting the next answer condition, and the deep cycle neural network model cannot achieve the goal. It can be understood that an important objective of the model is to obtain the mastery state of the labeled knowledge point on the topic, so that the model needs to input the prior relationship between the labeled topic and the knowledge point as a constraint condition, that is, the correlation weight between the topic and the knowledge point is composed of the labeled prior relationship and the network learned relationship. The intelligent learning knowledge ability tracking method provided by the invention will be explained and explained in detail through specific embodiments.

Fig. 1 shows a flowchart of a method for tracking intelligent learning knowledge ability provided by an embodiment of the present invention. As shown in fig. 1, the method for tracking intelligent learning knowledge ability provided by the embodiment of the present invention includes the following steps:

step 101: acquiring a current question corresponding to a user to be tested, and prior relevance weights of the current question and each knowledge point;

step 102: inputting the identification of the user to be tested, the current question and the prior relevance weight of each knowledge point into a knowledge ability tracking model, and acquiring the answer-to-pair probability of the user to be tested for the current question; the knowledge ability tracking model comprises the mastery states of the to-be-tested user on the knowledge points, wherein the mastery states of the to-be-tested user on the knowledge points are determined according to questions made by the to-be-tested user in a historical time period, answering results of the questions made by the to-be-tested user, and prior relevance weights of the questions made by the to-be-tested user and the knowledge points; the knowledge ability tracking model is obtained by inputting questions made by each user, answering results of the questions made by each user and prior relevance weights of the questions made by each user and each knowledge point in historical training data into the initial knowledge ability tracking model and training the initial knowledge ability tracking model based on a machine learning mode.

It can be understood that the intelligent learning knowledge ability tracking method further includes:

and after the current question is answered, acquiring the answering result of the user to be tested for the current question, and updating the mastery state of the user to be tested for each knowledge point according to the answering result.

In this embodiment, the priori correlation weight between the topic and each knowledge point is determined according to the knowledge point information marked on the topic. For example, assuming that n knowledge points are marked on the current topic, it can be determined that the prior relevance weights of the corresponding knowledge points are all 1/n (if the specific weight of each knowledge point is marked, the relevance weight is a marking weight), and the prior relevance weights of the remaining knowledge points are all 0.

In this embodiment, it should be noted that, an important objective of this embodiment is to acquire a grasping state of a knowledge point labeled on a question, so that a priori relations between the labeled question and the knowledge point need to be input as a strong signal, and since the a priori relations between the question and each knowledge point are introduced and simultaneously relations between the network learning question and each knowledge point are combined, an explicit mapping corresponding relation between the question and each knowledge point can be established, so that not only can accurate prediction of question answering probability be realized, but also a grasping state of the knowledge point can be updated according to answering results of the question, and thus a grasping state of a user to be tested on each knowledge point can be known.

It can be understood that, in this embodiment, first, the prior relevance weights of the current question, and each knowledge point corresponding to the user to be tested are obtained, then, the prior relevance weights of the user identifier, the current question, and each knowledge point to be tested are input into the knowledge ability tracking model, and the answer probability of the user to be tested for the current question is obtained; the knowledge ability tracking model comprises the mastery states of the to-be-tested user on the knowledge points, wherein the mastery states of the to-be-tested user on the knowledge points are determined according to questions made by the to-be-tested user in a historical time period, answering results of the questions made by the to-be-tested user, and prior relevance weights of the questions made by the to-be-tested user and the knowledge points; the knowledge ability tracking model is obtained by inputting questions made by each user, answering results of the questions made by each user and prior relevance weights of the questions made by each user and each knowledge point in historical training data into the initial knowledge ability tracking model and training the initial knowledge ability tracking model based on a machine learning mode. According to the embodiment of the invention, the prior relation between the question and the knowledge point is introduced, and the relation between the network learning question and each knowledge point is combined, so that the explicit mapping corresponding relation between the current question and each knowledge point is established, the answer probability of the current question can be accurately predicted, the knowledge point mastering state matrix can be updated according to the answer result of the current question, and the mastering state of the user to be tested on each knowledge point can be obtained.

The intelligent learning knowledge ability tracking method provided by the invention is explained and explained in more detail through a more specific scheme.

Based on the content of the above embodiment, in this embodiment, the knowledge ability tracking model includes a topic knowledge point mapping model;

correspondingly, inputting the user identifier to be tested, the current question, the prior relevance weight of the current question and each knowledge point into a knowledge ability tracking model, and acquiring the answer probability of the user to be tested for the current question, which specifically comprises:

acquiring the mastery state of the to-be-detected user on each knowledge point from the knowledge point mastery state matrix corresponding to the to-be-detected user; the knowledge point grasping state matrix stores the dynamic grasping state of the to-be-detected user for each knowledge point; the knowledge point mastering state is used for representing the mastering degree of the knowledge point of the user to be tested;

In this embodiment, the topic knowledge point mapping model is used to represent a correlation between topic feature data and the first correlation weight of each knowledge point, so that the first correlation weight of the current topic and each knowledge point can be obtained through the topic knowledge point mapping model.

In this embodiment, according to the information of the knowledge points marked on the current topic, the priori correlation weight between the current topic and each knowledge point can be determined. For example, assuming that n knowledge points are marked on the current topic, it can be determined that the prior relevance weights of the corresponding knowledge points are all 1/n (if the specific weight of each knowledge point is marked, the relevance weight is a marking weight), and the prior relevance weights of the remaining knowledge points are all 0.

In this embodiment, it should be noted that, an important objective of this embodiment is to obtain a grasp state of a labeled knowledge point on a topic, so that a priori relationship between the labeled topic and the knowledge point needs to be input as a constraint condition, that is, a correlation weight between the topic and the knowledge point is composed of the labeled priori relationship and a relationship learned through a network.

In this embodiment, it should be noted that, because the explicit mapping corresponding relationship between the current question and each knowledge point is established according to the first relevance weight of the current question and each knowledge point and the prior relevance weight of the current question and each knowledge point, the knowledge point grasping state matrix can be subsequently updated according to the answering result of the current question, so that the grasping state of the user to be tested on each knowledge point can be known in real time.

In this embodiment, it should be noted that the knowledge point states of the user to be tested for the knowledge points, which are stored in the knowledge point mastering state matrix, are updated in real time according to the answer conditions of the user to be tested, that is, the states of the user to be tested for the knowledge points, which are stored in the knowledge point mastering state matrix, are obtained by updating the states of the knowledge points in real time according to the answer conditions of the user to be tested for the questions with the marked knowledge points, based on the history of the user to be tested; and when the user to be tested does not have a historical answer record, the knowledge point mastering state matrix initially stores the average knowledge point state of all users of training model data for the state of each knowledge point.

In this embodiment, since the current topic includes at least one knowledge point, the knowledge point state feature data of the user to be tested for all knowledge points in the current topic can be finally determined according to the mastery state of the user to be tested for each knowledge point and the final correlation weight between the current topic and each knowledge point.

In this embodiment, the knowledge point state feature data and the question feature data are integrated together to obtain a feature vector simultaneously containing question information and knowledge point state information, so that the answer probability of the user to be tested for the current question can be predicted according to the feature vector simultaneously containing the question information and the knowledge point state information. In the embodiment, in the aspect of prediction accuracy, the question dimension information is further considered, so that the prediction accuracy of the answer pair probability can be improved.

In this embodiment, after answering, according to the answer result of the user to be tested for the current question and the mapping corresponding relationship between the current question and each knowledge point, the grasping state of the user to be tested for each knowledge point in the knowledge point grasping state matrix is updated, so that the grasping state of the user to be tested for each knowledge point can be known, and meanwhile, the success rate of answering next time can be accurately predicted according to the grasping state of the user to be tested for each knowledge point.

As can be seen from the above technical solutions, in the method for tracking intelligent learning knowledge ability provided by the embodiments of the present invention, first correlation weights of a current question and each knowledge point are obtained, and meanwhile, prior correlation weights of the current question and each knowledge point are determined according to knowledge point information marked on the current question, so that final correlation weights of the current question and each knowledge point can be determined according to the first correlation weights of the current question and each knowledge point and the prior correlation weights of the current question and each knowledge point, and meanwhile, a grasping state of a user to be tested on each knowledge point is obtained from a knowledge point grasping state matrix corresponding to the user to be tested; therefore, the knowledge point state feature data of the user to be tested on all knowledge points in the current question can be determined according to the mastered state of the user to be tested on all knowledge points and the final relevance weight of the current question and each knowledge point, finally, the embodiment of the invention integrates the knowledge point state feature data with the question feature data to obtain the feature vector simultaneously containing question information and knowledge point state information, so that the answer probability of the user to be tested on the current question can be predicted according to the feature vector simultaneously containing question information and knowledge point state information, and the mastered state of the user to be tested on each knowledge point in the knowledge point mastered state matrix can be updated according to the answer result of the user to be tested on the current question and the mapping corresponding relation between the current question and each knowledge point, therefore, the mastering state of the user to be tested on each knowledge point can be known conveniently, and the success rate of answering next time can be accurately predicted according to the mastering state of the user to be tested on each knowledge point.

Based on the content of the foregoing embodiment, in this embodiment, determining a priori correlation weights of the current topic and each knowledge point according to knowledge point information marked on the current topic includes:

In this embodiment, the prior relationship information of the current question and each knowledge point is introduced, so that on one hand, an explicit mapping corresponding relationship between the current question and each knowledge point can be established, and on the other hand, the final weight of the question and each knowledge point can be determined together according to the prior weight of the question and each knowledge point and the weights learned through the network, so that the determined final weight can more accurately reflect the association relationship between the question and each knowledge point, and meanwhile, the potential relationship between the knowledge points can be mined.

Based on the content of the foregoing embodiment, in this embodiment, determining a final relevance weight of the current topic and each knowledge point according to the first relevance weight of the current topic and each knowledge point and the prior relevance weight of the current topic and each knowledge point includes:

including, but not limited to, a weighted average, for example,

w_ti＝α*soft max(M_im_t)+(1-α)*{c_t}

wherein alpha is [0,1]]The adjustable parameter(s) of (2) can take, for example, a value of 0.5; w is a_tiRepresenting the final relevance weight of the current topic and a knowledge point i; softmax (M)_im_t) Representing a first relevance weight of the current topic and each knowledge point; { c_tExpressing the prior relevance weight of the current question and each knowledge point; m represents topic knowledge point mapping matrix, used for storing the mapping between the current topic and each knowledge point, and the size is n x d_mN is the number of the slot positions of the knowledge points, d_mDimension of the subject feature data of the current subject; m_iLine i, M, representing M_tTopic feature data representing a current topic; wherein the expression of the softmax function is

Wherein i 1.

In this embodiment, the final relevance weight of the current topic and each knowledge point is determined according to the first relationship model, and the relevance weight w of the topic and each knowledge point can be seen from the first relationship model_tFrom input { c_tAnd (4) carrying out weighted average on vectors after softmax is output by the mapping matrix of the topic knowledge points. In the embodiment, the final weight of the question and each knowledge point is determined according to the prior weight of the question and each knowledge point and the initial weight learned through the network, so that the determined final weight can reflect the association relationship between the question and each knowledge point more accurately.

Based on the content of the above embodiment, in this embodiment, the knowledge ability tracking model further includes a forgetting attenuation network model; correspondingly, the intelligent learning knowledge ability tracking method further comprises the following steps:

The method is based on the above embodiments, the factor of time attenuation is added, since the mastery degree of the knowledge points by students is reduced along with time forgetting, the time factor in the answer sequence is lost by a pure answer pair and a wrong sequence, the accuracy of prediction and the calculation accuracy of the knowledge state are influenced definitely, the same subject is done before one day and one month, the current influence is different, the input of a forgetting attenuation network and the time dimension is increased, the prediction accuracy of the model can be improved, the forgetting attenuation network can learn different attenuation rates aiming at each knowledge point, an Evosforgetting curve is only suitable for certain specific scenes, and the forgetting attenuation rates learned by a large amount of historical data are more reasonable when being applied to related scenes.

Based on the content of the foregoing embodiment, in this embodiment, the forgetting attenuation network model is: d_t＝sigmoid(W_dt_t+b_d) (ii) a Wherein d is_tRepresents the attenuation ratio, t_tRepresents a time interval, W_dAnd b_dDenotes a network parameter, W_dAnd b_dThe value of (a) is obtained by training; the expression of the sigmoid function is

Correspondingly, the attenuation rate is used for carrying out attenuation updating on the knowledge point grasping state in the knowledge point grasping state matrix corresponding to the user to be tested, and the attenuation updating method comprises the following steps of:

wherein the second relationship model is:

wherein N is_tThe knowledge point grasping state matrix is used for storing the grasping state of the user to be tested on each knowledge point; size n x d_nN is the number of knowledge points, d_nMastering the dimension of the state matrix for a knowledge point, N_t-1Knowledge point grasping state matrix representing the latest update, N_tiRepresents N_tRow i of (1), N_t-1,iRepresents N_t-1The number of the ith row of (a),

In this embodiment, a forgetting attenuation network d is adopted according to the time interval_t＝sigmoid(W_dt_t+b_d) Obtaining the attenuation rate, and then adopting a second relation model according to the obtained attenuation rate

And carrying out attenuation updating on the knowledge point grasping state in the knowledge point grasping state matrix corresponding to the user to be detected, so that the knowledge point grasping state matrix after attenuation updating can be obtained.

Based on the content of the foregoing embodiment, in this embodiment, determining knowledge point state feature data of the user to be tested for all knowledge points in the current topic according to the mastery state of the user to be tested for each knowledge point and the final relevance weight of the current topic and each knowledge point includes:

wherein k is_tAnd representing the knowledge point state characteristic data of the user to be tested for all knowledge points in the current question.

In this embodiment, according to the grasping state of the user to be tested on each knowledge point and the final correlation weight between the current question and each knowledge point, the knowledge point state feature data of the user to be tested on all knowledge points in the current question is determined according to the third relationship model, so that the comprehensive grasping result of the user on all knowledge points included in the current question can be accurately obtained.

Based on the content of the foregoing embodiment, in this embodiment, integrating the knowledge point state feature data and the topic feature data together to obtain a feature vector including both topic information and knowledge point state information includes:

s_t＝tanh(W_s[k_t,m_t]+b_s)

Wherein the expression of the tanh function is:

W_sand b_sThe two neural network parameters are obtained through training.

In the embodiment, the knowledge point state feature data r_tAnd topic feature data k_tIntegrated together, the feature vector f is obtained through a fully connected neural network_tAnd then, the answer accuracy rate can be predicted according to the characteristic vector simultaneously containing the question characteristic data and the knowledge point state characteristic data.

Based on the content of the foregoing embodiment, in this embodiment, the knowledge ability tracking model further includes: a capability network model and a question difficulty network model;

In this embodiment, can understand that same user is inequality to the ability value of different topics, consequently, can be according to the eigenvector that contains topic information and knowledge punctate state information simultaneously, and, with the ability network model that the user that awaits measuring corresponds acquires the user that awaits measuring is to the ability value of current topic, even if two topics contain the same knowledge point in addition, but the degree of difficulty of two topics is also not necessarily the same, for example, some topics are difficult, some topics are simple, consequently, add topic degree of difficulty network model, acquire the difficulty value of current topic, then synthesize ability value with the difficulty value can predict the probability that the topic was answered right.

In this embodiment, it can be understood that the capability network model can output the capability values of the students on the questions and the capability values of the students on the knowledge points (i.e., knowledge point grasping states), the topic difficulty network model can output the topic difficulty values, and then the probability of the topic pairing is predicted by combining the IRT theory, so that the interpretability of the model can be enhanced.

Based on the content of the foregoing embodiment, in this embodiment, obtaining the capability value of the user to be tested for the current topic according to a feature vector that includes both topic information and knowledge point state information, and a capability network model corresponding to the user to be tested includes:

based on the topic feature data and the topic difficulty network model, obtaining a difficulty value of the current topic, including:

predicting the answer probability of the user to be tested aiming at the current question based on the capability value and the difficulty value, and the method comprises the following steps: using a prediction model p_t＝sigmoid(a*θ_tj-β_j) Predicting the answer probability of the user to be tested for the current question; wherein p is_tAnd (4) expressing the answer-pair probability, and taking a value of 3 by combining the IRT theory.

Based on the content of the foregoing embodiment, in this embodiment, the grasping state of the to-be-tested user for each knowledge point in the knowledge point grasping state matrix is updated according to the answer result of the to-be-tested user for the current question and the mapping corresponding relationship between the current question and each knowledge point, which includes but is not limited to:

e_t＝sigmoid(W_en_t+b_e)

h_t＝tanh(W_hn_t+b_h)

Wherein, W_eAnd b_eThe network parameters are obtained by training in advance; w_hAnd b_hFor network parameters, advance passingTraining to obtain;

In this embodiment, a question response vector n is obtained_t(dimension d)_n) Then erasing the original memory and then writing the new memory, the erase vector and the write vector are e_t、h_t，e_t＝sigmoid(W_en_t+b_e)，h_t＝tanh(W_hn_t+b_h) Erasing the original memory:

writing new memory:

therefore, the mastery state of the user to be tested on each knowledge point in the knowledge point mastery state matrix is updated.

Based on the contents of the above embodiments, in the present embodiment, when the answer result is a binary 0 and 1 result; accordingly, the loss function is a binary cross entropy; when the answer result is score; accordingly, the loss function is a multivariate cross entropy;

in this embodiment, it should be noted that the answer results of many subjective questions are not only correct (1 and 0), but also not correct (1 and 0), and usually, the answer results may output a score (between 0 and 1), and this embodiment supports score input, and the current answer result input is not only 1 and 0, but also may include scores such as 0.1, 0.2.

It can be understood that, if a student has a considerable number of answer records, the model provided by the embodiment can output the mastery condition of all the questions of the student in the question space, that is, the prediction scores of the student on all the questions are obtained; the model can also automatically mine the knowledge concept implied by the question, and further obtain the mastery degree of all the knowledge of the student in the knowledge space. In particular, at a first moment, if a student does not have any question history, his first knowledge state is based on a non-personalized state, which is generated based on massive training data, and the following knowledge states become more and more personalized after the first knowledge state.

The embodiment implements a knowledge ability tracking model using a memory network. The network model stores all knowledge points by using one matrix (question knowledge point mapping matrix), and stores and updates the mastery degree of the students on the knowledge points by using the other matrix (knowledge point mastering state matrix). The model can track the mastering states of different knowledge points and capture the relationship among the different knowledge points, maintains a state for each knowledge point, automatically learns the correlation between the current question input and each knowledge point, firstly selects the knowledge point related to the current question when a new question is input, and then updates the state of the related knowledge point according to the wrong answer condition of the student. In order to obtain the mastered state of each knowledge point, the model maps the knowledge point state stored in the knowledge point mastered state matrix card slot and the knowledge point labeled on the question, and uses the prior relation between the labeled question and the knowledge point as a constraint condition. In a preferred embodiment, the model of this embodiment further adds a time interval input for making questions and a forgetting attenuation network, performs forgetting attenuation on the state of the knowledge point in the knowledge point mastery state matrix, predicts the correct answer probability of the student using the state of the knowledge point, and updates the state of the knowledge point in the matrix according to the right and wrong making questions. The model network structure is shown in detail in fig. 2, and is now described in detail as follows:

1. inputting a model:

q_tfor topic id, q_tE {1, 2.., Q }, wherein Q is the number of topics; a is_tIn response, a_tBelongs to {0,1}, and the score input is a plurality of scores, qa_t＝q_t+a_t*Q，qa_t∈{1,2,...,2Q}；{c_tThe input of softmax of the prior relation between the topic and the knowledge points (namely if the topic contains n knowledge points, the values of the positions of the knowledge points are all 1/n, and the others are 0); t is t_tFor the current question making time and the last question making time interval related to the same knowledge point, for example, the first question making time interval is input as 0, and the non-first question making time interval is input as log₁₀(1+ Interval hours).

2. Acquiring mapping weight of question knowledge points:

q_tby topic embedding matrix (shape is q x d)_mQ is number of questions) to obtain an embedding vector m of the questions_t(dimension d)_m) M (shape is n x d)_mN is the number of knowledge points) as a question knowledge point mapping matrix, and storing the mapping between the question and the knowledge point; n is a radical of_t(shape is n x d)_vAnd n is the number of knowledge points) is a knowledge point grasping state matrix, and the grasping state of the knowledge points is stored. Relevance weight w of topic to each knowledge point_tFrom input { c_tAre combined with the vector after the output softmax of the topic knowledge point mapping matrix, including but not limited to a weighted average manner, for example,

w_ti＝α*soft max(M_im_t)+(1-α)*{c_t}

wherein alpha is [0,1]]For example, may default to 0.5; w is a_tiIs a correlation weight vector w_tThe ith element of (1), M_iThe ith row vector of M is the vector of M,

wherein i 1.

3. Forgetting attenuation:

input t_tConnecting a forgetting attenuation full-connection neural network, and outputting the attenuation rate d of each knowledge point_t，

d_t＝sigmoid(W_dt_t+b_d)

The knowledge point states in the knowledge point mastery state matrix are subjected to attenuation memorization, including but not limited to,

wherein 1 is a vector of all 1 s,

in order to be a dot product,

4. and (3) prediction process:

first, the state matrix N is grasped from the knowledge points_tMiddle read knowledge point state k_t，

Wherein N is_tiIs N_tThe ith row vector of (1).

Then, k is put_tAnd m_tConnected together and connected with a fully connected neural network to obtain a characteristic vector s_t，

s_t＝tanh(W_s[k_t,m_t]+b_s)

Reconnecting student capability network theta_tjAnd topic difficulty network beta_j，

θ_tj＝tanh(W_θs_t+b_θ)

β_j＝tanh(W_βm_t+b_β)

Finally, according to the IRT Theory (i.e., Item Response Theory), also called topic Response Theory, it is a mathematical model for analyzing test results or questionnaire survey data.

The model expression is as follows:

wherein, P represents the probability of answering the question, a represents the degree of distinction of the question, b represents the difficulty parameter of the question, c represents the guess parameter of the question, theta represents the ability value of the student, c is 0, a is 1), and the predicted question answering probability P_t：

p_t＝sigmoid(3.0*θ_tj-β_j)

5. Updating knowledge point mastering state matrix:

qa_tsubject answering embedding matrix (shape is 2q d)_nQ is the number of questions) to obtain a question-answering vector n_t(dimension d)_n) Then erase the original memory and then write the new memory, including but not limited to, the erase vector and the write vector being e respectively_t、h_t：

e_t＝sigmoid(W_en_t+b_e)

h_t＝tanh(W_hn_t+b_h)

Erasing the original memory:

writing new memory:

specifically, the model training and use process is as follows:

1. training model

The input of the model is the corresponding relation data of the question recording sequence, the question and the knowledge point, the direct output of the model is the probability of predicting the question pair, and the ability value of the student to the question, the ability value to the knowledge point and the question difficulty value are indirectly output.

In this embodiment, for the knowledge ability tracking model to which only the topic knowledge point prior relationship is added, when the model is trained, the data format of the input model (taking a certain discipline a as an example, and other discipline formats are the same) is shown in (1) in the following 1), and for the knowledge ability tracking model to which the topic knowledge point prior relationship input and the time interval input are added at the same time, when the model is trained, the data format of the input model (taking a certain discipline a as an example, and other discipline formats are the same) is shown in (2) in the following 1):

1) data of questions

(1) Question-free time input

(student id, [ [ topic id ], question result ], ]

For example, (101893639, [ [ [2],1], [ [5],0], [ [3],1] ])

Wherein, the topic id is the converted continuous id from 0, the topic making result is pair 1, and wrong 0 is made;

(2) increasing question making time input

(student id, [ [ subject id ], question making result, question making time ], [ subject id ], ])

For example, (101893639, [ [ [2],1, '2020-08-0216: 07: 10' ], [ [5],0, '2020-08-0216: 08: 11' ], [ [3],1, '2020-08-0216: 08: 56' ])

2) the prior relationship data of the topics and the knowledge points are shown in the following table 1:

TABLE 1

Topic id	Knowledge point id
		0	0
1	1
		2	2,3
3	4,5,6
		4	7
5	8
		...	...

Wherein, the topic id is a converted continuous id from 0; knowledge point id is the converted consecutive id starting from 0.

2. Use model

According to the embodiment, the mastery degree of the students on the knowledge points is calculated according to the real-time problem making records of the students, and the problem making probability and the like are predicted. And updating the knowledge point state of the knowledge point mastering state matrix according to the question making result, and then acquiring the corresponding knowledge point state. If the knowledge point in the ith memory card slot needs to be acquired to master the state, the process is as follows:

let w_tIs [0,. ], w_i,..0]，w_iSetting the position to 1, and reading the knowledge point state k in the knowledge point grasping state matrix_t，

And a characteristic vector s_tThe topic-related information is not connected,

s_t＝tanh(W_s[k_t,0]+b_s)

and calculating the ability value of the student to the knowledge point, namely the mastery degree of the knowledge point through a student ability network.

θ_tj＝tanh(W_θs_t+b_θ)

According to the above description, the intelligent learning knowledge ability tracking method provided by the embodiment of the invention has the following advantages:

I. the model provided by the embodiment takes the question as input, automatically learns the correlation between the question and each knowledge point, and dynamically outputs the mastery state of each knowledge point by combining the prior relationship between the question and the knowledge point. However, in the time series model represented by the deep circulation neural network, if the answer condition of the subject is input by the model, the probability of answering all the subjects (done and not done) in the next step can be predicted, but the model expresses the current mastering conditions of the student on all knowledge points by a hidden state (hidden state), so that the model cannot output the mastering conditions of the student on specific knowledge points. If the answer condition of the knowledge point(s) corresponding to the question is input by the model, the answer probability of all the knowledge points (done and not done) in the next step can be predicted, but the probability is not the mastery degree of students on the knowledge points, the situation of discontinuity and fluctuation change can occur in the time sequence, and the situation of reduction of the prediction precision can occur due to the loss of the dimension information of the question in the prediction accuracy.

In addition, the main goal of the model is to be able to calculate the knowledge point grasping condition after the current input answer result, but the deep circular neural network model cannot achieve the goal, and the model of the embodiment can use the current input result to update the knowledge point grasping state matrix and then acquire the knowledge point state of the corresponding memory card slot position.

II. The model structure of the embodiment not only has a student ability network and a subject difficulty network, but also adds a forgetting attenuation network.

The student ability network can output the ability value of the student to the question and the ability value of the student to the knowledge point (namely the knowledge point mastering state), the question difficulty network can output the question difficulty value, and then the probability of the question pairing is predicted by combining the IRT theory, so that the interpretability of the model is enhanced. The mastery degree of students on knowledge points can be reduced along with time forgetting, the time factors in the answer sequence are lost by a pure answer pair wrong sequence, the prediction accuracy and the knowledge state calculation accuracy are certainly influenced, the same question is made before one day and one month, the current influence is different, the forgetting attenuation network and the time dimension input are increased, the prediction accuracy of the model can be improved, the forgetting attenuation network can learn different attenuation rates aiming at each knowledge point, the Einghao forgetting curve is only suitable for certain specific scenes, and the forgetting attenuation rates learned through a large amount of historical data can be more reasonable when being applied to related scenes.

III, model support score input of this embodiment

The answer results of many subjective questions are not only wrong (1 and 0), but also the answer results of the indefinite choice questions are not only wrong (1 and 0), usually, the answer results can output a score (between 0 and 1), the model of the embodiment supports score input, the current answer result input is not only 1 and 0, and the score input can also comprise scores such as 0.1, 0.2.

Taking the disciplines a, b, c, d, e and f shown in the following table 2 as examples, the model evaluation indexes are auc (area under curve), acc (accuracy, threshold 0.5), and the value range is [0,1], and the larger the auc and acc indexes are, the better the model evaluation indexes are.

TABLE 2

For the knowledge ability tracking model without adding the prior relation input and the time interval input of the subject knowledge points, the model indexes are shown in the following table 3:

TABLE 3

Subject of discipline	auc	acc
			a	0.8281	0.7934
b	0.8531	0.8038
			c	0.8414	0.7958
d	0.9028	0.8630
			e	0.9086	0.8682
f	0.9405	0.9118

For the knowledge ability tracking model for adding the prior relation input and the time interval input of the topic knowledge points, the model indexes are shown in the following table 4:

TABLE 4

Subject of discipline	auc	acc
			a	0.8632	0.8295
b	0.8734	0.8224
			c	0.8674	0.8164
d	0.9212	0.8823
			e	0.9283	0.8864
f	0.9589	0.9246

From the above, the performance of each discipline model is greatly improved.

Based on the same inventive concept, another embodiment of the present invention further provides a resource pushing method based on the intelligent learning knowledge ability tracking method described in the above embodiment, including:

and carrying out test question pushing and/or learning resource pushing for the user to be tested according to the mastery state of the user to be tested on each knowledge point and/or the answer prediction result of the user to be tested on each question in the question bank.

In this embodiment, it should be noted that the knowledge ability tracking model is modeled based on a student behavior sequence, tracks the knowledge mastering state of students, and is the core and key for constructing the adaptive teaching system. In the self-adaptive teaching system, no matter precise subject pushing is carried out or path planning of student learning is carried out, the first step is to precisely estimate the mastery degree of the knowledge of the students. Therefore, the embodiment can be used for personalized teaching process. Specifically, the academic behaviors of students can be predicted, the excellent student knowledge ability tracking model needs to have score prediction ability, and an important evaluation index of the model is the accuracy of the score prediction. The score prediction is defined as: a learning history of a student is obtained, along with a representation of a topic at which the model can predict the student's score. Individualized teaching system can carry out accurate prediction to student's answer condition, and this is favorable to promoting student's autonomic learning efficiency, because the system can be progressive give the moderate exercise of the propelling movement degree of difficulty of student, avoids recommending too difficult or too simple topic, extravagant student's time and energy. In addition, the embodiment can also carry out personalized diagnosis and learning on the academic level of the student. The objective of the student ability diagnosis is to model the academic ability level of students on the knowledge level and help the students to find weak knowledge points so that the students can learn independently. For example, poor subjects mastered by knowledge points of students can be automatically pushed to strengthen the practice and the strengthening of the students on the knowledge points. Meanwhile, on the basis of predicting the result according to the knowledge points, teaching and research experiences of education experts can be combined to help students to independently learn and assist teachers to conduct targeted teaching; the personalized learning is further carried out, and appropriate learning resources such as learning videos, test questions and the like are directly recommended to students according to the knowledge states of the students and the prior knowledge in the education field. The accurate learning scheme is given by using historical learning data of a single student and related groups, the learning mode of one face of thousands of people in the traditional teaching is broken, and the learning efficiency of the student can be effectively improved.

Based on the same inventive concept, another embodiment of the present invention further provides a volume generation method based on the intelligent learning knowledge ability tracking method described in the above embodiment, which includes one or more of the following methods:

automatically grouping papers according to the mastery state of each knowledge point of each user to be tested in the appointed evaluation range and the knowledge points marked on the test questions of the papers to be grouped;

automatically grouping the paper according to the prediction scores of the users to be tested in the appointed evaluation range on the test questions of the paper to be grouped;

thirdly, according to the dynamic mastery state of the users to be tested in each region on each knowledge point and the prediction scores of the users to be tested in each region aiming at different topics, the student groups in different regions are mapped to the same topic space so as to realize the score comparability in different regions.

In this embodiment, different paper groups can be performed for different application scenarios, for example, for classroom consolidation exercise, the mastery degree of each knowledge point of each user to be tested in the specified evaluation range can be obtained according to the mastery state of each user to be tested in the specified evaluation range and the knowledge points marked on the test questions of the paper to be tested, and then the appropriate questions can be selected for paper grouping according to the mastery degree of each knowledge point of each user to be tested in the specified evaluation range, so that each user to be tested can effectively consolidate incomprehensible or inexperienced knowledge points in classroom consolidation exercise. For another example, for an application scenario of an examination class, according to the prediction score of each user to be tested in the specified evaluation range for each examination question to be formed, forming a test paper in a purposeful and targeted manner according to the normal distribution requirement of the test score or other requirements.

In addition, it should be noted that the embodiment is also beneficial to scientific evaluation of capability differences among students, and the embodiment can map student groups in different areas to the same topic space according to the mastery states of users to be tested in various areas to various knowledge points and the prediction scores of the users to be tested in various areas to different topics, so as to realize score comparability in different areas, facilitate teaching and research staff to know normalized academic capability distribution of different student groups, and further make a next-step teaching plan.

In this embodiment, it should be noted that the knowledge ability tracking model is modeled based on a student behavior sequence, tracks the knowledge mastering state of students, and is the core and key for constructing the adaptive teaching system. In the self-adaptive teaching system, no matter precise subject pushing is carried out or path planning of student learning is carried out, the first step is to precisely estimate the mastery degree of the knowledge of the students.

The knowledge ability tracking model has important significance for personalized teaching:

first, the student knowledge ability tracking model can predict the academic behaviors of students. The excellent student knowledge ability tracking model needs score prediction ability, and an important evaluation index of the model is the accuracy of the score prediction. The score prediction is defined as: a learning history of a student is obtained, and-a representation of the topic in which the model can predict the student's score. The personalized teaching system can accurately predict the answer condition of the student, and is beneficial to improving the autonomous learning efficiency of the student, because the system can gradually push exercise questions with moderate difficulty to the student, the problem that the student is difficult or simple to recommend is avoided, and the time and the energy of the student are wasted; secondly, the teaching efficiency of teachers is improved, for example, in the process of paper forming by teachers, the system can tell the teachers the prediction scores of students in each test question, and the teachers can conveniently form paper more purposefully; finally, the method is beneficial to scientifically evaluating the capability difference among students, can map student groups in different schools and areas into the same subject space through score prediction, achieves score comparison, and is convenient for teaching and research personnel to know the normalized academic capability distribution of different student groups, thereby formulating the next teaching plan.

Second, the student knowledge ability tracking model can perform personalized diagnosis on the academic level of the student. The student capability diagnosis aims at modeling the academic capability level of students on the knowledge level, and can help the students to find self weak knowledge points by combining the teaching and research experience of education experts so that the students can independently learn and assist teachers to carry out targeted teaching; the personalized learning is further carried out, and appropriate learning resources such as learning videos, test questions and the like are directly recommended to students according to the knowledge states of the students and the prior knowledge in the education field. The accurate learning scheme is given by using historical learning data of a single student and related groups, the learning mode of one face of thousands of people in the traditional teaching is broken, and the learning efficiency of the student can be effectively improved.

Based on the same inventive concept, another embodiment of the present invention provides an intelligent learning knowledge ability tracking apparatus, including: an acquisition module and a prediction module; wherein:

the acquisition module is used for acquiring the current question corresponding to the user to be tested, and the prior relevance weight of the current question and each knowledge point;

the prediction module is used for inputting the identification of the user to be tested, the current question, the prior relevance weight of the current question and each knowledge point into the knowledge capability tracking model and acquiring the answer probability of the user to be tested for the current question; the knowledge ability tracking model comprises the mastery states of the to-be-tested user on the knowledge points, wherein the mastery states of the to-be-tested user on the knowledge points are determined according to questions made by the to-be-tested user in a historical time period, answering results of the questions made by the to-be-tested user, and prior relevance weights of the questions made by the to-be-tested user and the knowledge points;

Since the intelligent learning knowledge ability tracking device provided by the embodiment of the present invention can be used for performing the intelligent learning knowledge ability tracking described in the above embodiment, and the working principle and the beneficial effect are similar, detailed description is not provided here, and specific contents can be referred to the description of the above embodiment.

In this embodiment, it should be noted that each module in the apparatus according to the embodiment of the present invention may be integrated into a whole or may be separately disposed. The modules can be combined into one module, and can also be further split into a plurality of sub-modules.

Based on the same inventive concept, another embodiment of the present invention provides an electronic device, which specifically includes the following components, with reference to fig. 3: a processor 301, a memory 302, a communication interface 303, and a communication bus 304;

the processor 301, the memory 302 and the communication interface 303 complete mutual communication through the communication bus 304;

the processor 301 is configured to call a computer program in the memory 302, and the processor implements all the steps of the above-mentioned intelligent learning knowledge ability tracking method when executing the computer program, for example, the processor implements the following processes when executing the computer program: acquiring a current question corresponding to a user to be tested, and prior relevance weights of the current question and each knowledge point; inputting the identification of the user to be tested, the current question, the prior relevance weight of the current question and each knowledge point into a knowledge ability tracking model, and acquiring the answer probability of the user to be tested for the current question; the knowledge ability tracking model comprises the mastery states of the to-be-tested user on the knowledge points, wherein the mastery states of the to-be-tested user on the knowledge points are determined according to questions made by the to-be-tested user in a historical time period, answering results of the questions made by the to-be-tested user, and prior relevance weights of the questions made by the to-be-tested user and the knowledge points; the knowledge ability tracking model is obtained by inputting questions made by each user, answering results of the questions made by each user and prior relevance weights of the questions made by each user and each knowledge point in historical training data into the initial knowledge ability tracking model and training the initial knowledge ability tracking model based on a machine learning mode.

It will be appreciated that the detailed functions and extended functions that the computer program may perform may be as described with reference to the above embodiments.

In addition, the processor may further implement the steps of the resource pushing method according to the above embodiment when executing the computer program, and/or implement the steps of the volume group method according to the above embodiment when executing the program.

Based on the same inventive concept, yet another embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements all the steps of the above-mentioned intelligent learning knowledge capability tracking method, for example, the processor implements the following processes when executing the computer program: acquiring a current question corresponding to a user to be tested, and prior relevance weights of the current question and each knowledge point; inputting the identification of the user to be tested, the current question, the prior relevance weight of the current question and each knowledge point into a knowledge ability tracking model, and acquiring the answer probability of the user to be tested for the current question; the knowledge ability tracking model comprises the mastery states of the to-be-tested user on the knowledge points, wherein the mastery states of the to-be-tested user on the knowledge points are determined according to questions made by the to-be-tested user in a historical time period, answering results of the questions made by the to-be-tested user, and prior relevance weights of the questions made by the to-be-tested user and the knowledge points; the knowledge ability tracking model is obtained by inputting questions made by each user, answering results of the questions made by each user and prior relevance weights of the questions made by each user and each knowledge point in historical training data into the initial knowledge ability tracking model and training the initial knowledge ability tracking model based on a machine learning mode.

In addition, the logic instructions in the memory may be implemented in the form of software functional units and may be stored in a computer readable storage medium when sold or used as a stand-alone product. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment of the present invention. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. Based on such understanding, the technical solutions mentioned above may be embodied in the form of a software product, which may be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the intelligent learning knowledge ability tracking method according to various embodiments or some parts of embodiments.

Moreover, in the present invention, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

Furthermore, in the present disclosure, reference to the description of the terms "one embodiment," "some embodiments," "an example," "a specific example," or "some examples" or the like means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. An intelligent learning knowledge ability tracking method is characterized by comprising the following steps:

2. The intelligent learning knowledge ability tracking method according to claim 1, further comprising:

3. The intelligent learning knowledge ability tracking method according to claim 1, wherein the knowledge ability tracking model comprises a question knowledge point mapping model;

4. The intelligent learning knowledge ability tracking method according to claim 3, wherein determining a priori correlation weights of the current topic and each knowledge point according to knowledge point information marked on the current topic comprises:

5. The method for tracking intelligent learning knowledge ability according to claim 3, wherein determining the final relevance weight of the current topic and each knowledge point according to the first relevance weight of the current topic and each knowledge point and the prior relevance weight of the current topic and each knowledge point comprises:

including, but not limited to, a weighted average, for example,

w_ti＝α*softmax(M_im_t)+(1-α)*{c_t}

wherein alpha is [0,1]]The adjustable parameters of (2); w is a_tiRepresenting the final relevance weight of the current topic and a knowledge point i; softmax (M)_im_t) Representing a first relevance weight of the current topic to a knowledge point; { c_tIndicates the current titleA priori relevance weight to a knowledge point; m represents topic knowledge point mapping matrix, is used for storing the mapping of the current topic and knowledge point, and has the size of n x d_mN is the number of knowledge points, d_mDimension of the subject feature data of the current subject; m_iLine i, M, representing M_tTopic feature data representing a current topic; wherein the expression of the softmax function is

Wherein i 1.

6. The intelligent learning knowledge ability tracking method according to any one of claims 3 to 5, wherein the knowledge ability tracking model further comprises a forgetting attenuation network model;

accordingly, the method further comprises:

7. The intelligent learning knowledge capability tracking method of claim 6, wherein the forgetting attenuation network model is: d_t＝sigmoid(W_dt_t+b_d) (ii) a Wherein d is_tRepresents the attenuation ratio, t_tRepresents a time interval, W_dAnd b_dDenotes a network parameter, W_dAnd b_dThe value of (a) is obtained by training; the expression of the sigmoid function is

wherein the second relationship model is:

8. The method for tracking intelligent learning knowledge ability according to claim 7, wherein determining knowledge point state feature data of the user to be tested for all knowledge points in the current question according to the mastery state of the user to be tested for each knowledge point and the final correlation weight of the current question and each knowledge point comprises:

9. The intelligent learning knowledge ability tracking method according to claim 8, wherein integrating the knowledge point state feature data and the topic feature data together to obtain a feature vector simultaneously containing topic information and knowledge point state information comprises:

s_t＝tanh(W_s[k_t,m_t]+b_s)

Wherein the expression of the tanh function is:

W_sand b_sThe two network parameters are obtained through training.

10. The intelligent learning knowledge ability tracking method according to claim 9, wherein the knowledge ability tracking model further comprises: a capability network model and a question difficulty network model;

11. The intelligent learning knowledge ability tracking method according to claim 10, wherein obtaining the ability value of the user to be tested for the current question according to a feature vector containing both question information and knowledge point state information and an ability network model corresponding to the user to be tested comprises:

12. The method for tracking intelligent learning knowledge ability according to claim 10, wherein the grasping state of the user to be tested on each knowledge point in the knowledge point grasping state matrix is updated according to the response result of the user to be tested on the current question and the mapping correspondence between the current question and each knowledge point, including but not limited to:

e_t＝sigmoid(W_en_t+b_e)

h_t＝tanh(W_hn_t+b_h)

13. The intelligent learning knowledge ability tracking method according to claim 11, wherein when the answer result is a binary 0 and 1 result; accordingly, the loss function is a binary cross entropy; when the response result is score [0,1 ]; accordingly, the loss function is a multivariate cross entropy.

14. A resource pushing method based on the intelligent learning knowledge ability tracking method according to any one of claims 1 to 13, comprising:

15. A volume method based on the intelligent learning knowledge ability tracking method according to any one of claims 1 to 13, comprising:

automatically grouping the paper according to the mastery state of each knowledge point of each user to be tested in the appointed evaluation range and the knowledge points marked on the test questions of the paper to be grouped;

and/or the presence of a gas in the gas,

automatically grouping the paper according to the prediction scores of the users to be tested in the specified evaluation range on the test questions of the paper to be grouped;

and/or the presence of a gas in the gas,

and mapping student groups in different areas to the same topic space according to the mastery states of the users to be tested in the areas to the knowledge points and the prediction scores of the users to be tested in the areas to different topics, so as to realize score comparability in different areas.

16. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program performs the steps of the intelligent learning knowledge capability tracking method according to any one of claims 1 to 13, and/or wherein the processor when executing the program performs the steps of the resource pushing method according to claim 14, and/or wherein the processor when executing the program performs the steps of the volume grouping method according to claim 15.