CN113052316A

CN113052316A - Knowledge tracking method, system, equipment and storage medium based on causal reasoning

Info

Publication number: CN113052316A
Application number: CN202110223365.0A
Authority: CN
Inventors: 黄昌勤; 朱佳; 黄琼浩; 梁婉莹
Original assignee: Zhejiang Normal University CJNU
Current assignee: Zhejiang Normal University CJNU
Priority date: 2021-03-01
Filing date: 2021-03-01
Publication date: 2021-06-29
Anticipated expiration: 2041-03-01
Also published as: CN113052316B

Abstract

The invention discloses a knowledge tracking method, a knowledge tracking system, knowledge tracking equipment and a storage medium based on causal reasoning, wherein the method comprises the following steps: acquiring a random variable and a potential result, and determining a relational expression of the random variable and the potential result; acquiring an observation variable, and dividing the observation variable into a confounding variable, an adjusting variable and an irrelevant variable; analyzing by adopting a causal reasoning method, and determining the conversion output of the potential result according to the relation among the confounding variable, the adjusting variable, the random variable and the potential result; determining a first objective function according to the conversion output; acquiring sample weight and determining a second objective function; based on the second objective function and the sample weights, balancing weights are determined that can reduce the covariance between the elements to help evaluate the causal relationship between the individual variables and the effect variable.

Description

Knowledge tracking method, system, equipment and storage medium based on causal reasoning

Technical Field

The present application relates to the field of knowledge tracking, and in particular, to a knowledge tracking method, system, device, and storage medium based on causal reasoning.

Background

With the success of artificial intelligence in modeling in various fields, it is desirable to use algorithms to simulate the ability of humans to track the Knowledge state of students when they master a particular skill or concept and to predict the learning performance of the students, which motivates human research into Knowledge Tracking (KT). Knowledge tracking is a task of modeling the knowledge state of a student according to historical data, and represents the mastery level of the knowledge of the student. One notable model for solving the KT problem is a model based on a recurrent neural network, called Deep Knowledge Tracking (DKT), which, although it takes an impressive performance in the Knowledge tracking task, has a major problem, namely the transition of the prediction output. When a student performs well in a learning task associated with a skill i, the model's predictive performance for that skill may be degraded and vice versa. This is unreasonable because it is desirable for the student's knowledge status to transition gradually over time rather than alternating between mastery and non-mastery, and this wavy transition is disadvantageous in that it misleads the interpretation of the student's knowledge status.

Disclosure of Invention

The present application is directed to solving, at least to some extent, one of the technical problems in the related art. Therefore, the application provides a knowledge tracking method, a knowledge tracking system, knowledge tracking equipment and a storage medium based on causal reasoning.

In a first aspect, an embodiment of the present application provides a knowledge tracking method based on causal reasoning, including: acquiring random variables, wherein the random variables correspond to specific knowledge points; acquiring a potential result, wherein the potential result represents an answer result of a student to a specific knowledge point; determining a relation between the random variable and the potential result according to the random variable and the potential result; acquiring observation variables, wherein the observation variables comprise a confounding variable, an adjusting variable and an irrelevant variable; determining a conversion output of the potential result according to the miscellaneous variable, the adjusting variable and the relation between the random variable and the potential result; determining a first objective function of the observed variable according to the conversion output; acquiring sample weight, updating the first objective function according to the sample weight, and determining a second objective function; determining a balance weight for evaluating a causal relationship between the single variable and the potential effect based on the second objective function and the sample weights.

Optionally, the method further comprises: and performing optimization processing on the first objective function, wherein the optimization processing separates the adjusting variable from the observation variable and separates the confounding variable from the observation variable.

Optionally, the method further comprises: acquiring a global balance model; evaluating a causal relationship between a single variable and the potential outcome through the global equilibrium model.

Optionally, said evaluating causal relationships between individual variables and said potential outcomes through said global balance model comprises: encoding the mixed variable through an automatic encoder to determine a reconstruction vector; and performing data reconstruction on the reconstruction variable through a product decoder to determine the sample weight.

Optionally, the evaluating, by the global balance model, a causal relationship between a single variable and the potential outcome further comprises: the input data is mapped to a non-linear low dimensional space by a depth knowledge tracking module.

Optionally, the conversion output is specifically:

wherein Y + is the conversion output, K is the random variable, Z is the adjustment variable, X is the miscellaneous vector, i is the number of the variable.

Optionally, the first objective function is specifically:

minimize||Y⁺-h(U)||²

wherein, Y⁺For the conversion output, U is the observed variable.

In a second aspect, the present application provides a knowledge tracking system based on causal reasoning, including: the acquisition module is used for acquiring random variables, potential results, observation variables and sample weights; the data processing module is used for determining a relational expression of the random variable and the potential result according to the random variable and the potential result; and is used for determining the conversion output of the potential result according to the miscellaneous variable, the adjusting variable and the relation between the random variable and the potential result; the objective function processing module is used for determining a first objective function of the observation variable according to the conversion output; and the first objective function is updated according to the sample weight, and a second objective function is determined. And the causal reasoning module is used for determining balance weight according to the second objective function and the sample weight, and the balance weight is used for evaluating the causal relationship between the single variable and the potential result.

In a third aspect, an embodiment of the present application provides an apparatus, including: at least one processor; at least one memory for storing at least one program; when executed by the at least one processor, cause the at least one processor to implement the causal reasoning based knowledge tracking method as described in the first aspect.

In a fourth aspect, embodiments of the present application provide a computer storage medium, in which a processor-executable program is stored, the processor-executable program, when executed by the processor, being configured to implement the causal reasoning-based knowledge tracking method according to the first aspect.

The beneficial effects of the embodiment of the application are as follows: acquiring a random variable corresponding to a specific knowledge point, acquiring a potential result of a student answering the specific knowledge point, and determining a relational expression of the random variable and the potential result according to the random variable and the potential result; acquiring an observation variable, and dividing the observation variable into a confounding variable, an adjusting variable and an irrelevant variable; analyzing the relation between a random variable and a potential result by adopting a causal reasoning method, and determining the conversion output of the potential result according to the confounding variable, the adjusting variable and the relation between the random variable and the potential result; determining a first objective function of the observed variable according to the conversion output; the confounding variable can be separated from the observation variable through the first objective function, so that the purpose of recovering the causal relationship between the confounding vector and the potential result is achieved; acquiring sample weight, updating the first objective function according to the sample weight, and determining a second objective function; determining an equilibrium weight from the second objective function and the sample weight, the equilibrium weight being used to evaluate a causal relationship between a single variable and the potential outcome; learning sample weights and continuously updating the target function to make the weighted variables independent; having accurate balancing weights, the balancing weights can reduce the covariance between the elements to help evaluate the causal relationship between individual and effect variables.

Drawings

The accompanying drawings are included to provide a further understanding of the claimed subject matter and are incorporated in and constitute a part of this specification, illustrate embodiments of the subject matter and together with the description serve to explain the principles of the subject matter and not to limit the subject matter.

FIG. 1 is a diagram of the steps of a knowledge tracking method based on causal reasoning according to some embodiments of the present application;

FIG. 2 is a causal graph provided by some embodiments of the present application;

FIG. 3 is an overall framework diagram of a global balancing model provided by some embodiments of the present application;

fig. 4 is a heat map for knowledge tracking of a student in assisments 2009 dataset using a cause-and-effect reasoning knowledge tracking method as provided by some embodiments of the present application;

fig. 5 is a heatmap for knowledge tracking of a student in assistcents 2009 dataset using DKT, provided by some embodiments of the present application;

FIG. 6 provides a knowledge tracking system based on causal reasoning for some embodiments of the present application;

fig. 7 is an apparatus provided in some embodiments of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

It should be noted that although functional block divisions are provided in the system drawings and logical orders are shown in the flowcharts, in some cases, the steps shown and described may be performed in different orders than the block divisions in the systems or in the flowcharts. The terms first, second and the like in the description and in the claims, and the drawings described above, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.

In general, the knowledge tracking task can be formulated as a supervised sequence learning problem: given a student's past exercise interaction Xt ═ (X1, X2, …), the probability that the student correctly answers a new exercise in the next time step t +1, i.e. p (at +1 ═ 1| qt +1, Xt). The input xt ═ (qt, at) is a tuple indicating which exercise received the answer (qt) and whether this exercise received the correct answer (at) at time t.

One notable approach to solving the KT problem is based on a model of Recurrent Neural Networks (RNNs), known as Deep Knowledge Tracking (DKT). Preliminary studies have found that DKT outperforms traditional knowledge tracking methods without the need to acquire a large number of human features for engineering. While DKT achieves an impressive performance in the knowledge tracking task, some researchers have discovered a major problem, namely, the transition of the predicted output. When a student performs well in a learning task associated with a skill i, the model's predictive performance for that skill may be degraded, and vice versa. This is not reasonable because one would like the student's knowledge state to transition gradually over time, rather than alternating between mastery and masterless, and therefore this wavy transition is disadvantageous and misleads the interpretation of the student's knowledge state. In order to solve the above problem, some researchers propose to enlarge the original loss function of the DKT model to a regularized loss function to obtain better prediction accuracy for the next interaction. However, their solution only focuses on local dependencies and does not address global dependencies of the inputs. Global dependencies mean that the problem does not occur continuously in the form of local dependencies, but rather after a few time steps.

In order to find out and solve the root cause of the above problem in DKT, in the solution of the present application, a method of causal reasoning, which is a powerful statistical modeling tool for interpretation analysis, is first adopted. The essence of the causal reasoning method is to eliminate the confounding effect of confounding factors, thereby significantly improving the current functional effect estimation. Confounding effects are usually handled by a bias score, which is used to summarize the information needed to control confounding factors, but confounding effects treat all observed variables as confounding variables, ignoring tuning variables, which effectively reduce the variance of the estimated effect. Furthermore, most of these methods assume that whether a variable is the cause of the result is known in advance. However, in most cases, such as knowledge tracking, a priori knowledge does not define the causal structure well.

Based on the method, the system, the equipment and the storage medium for knowledge tracking based on causal reasoning are provided, firstly, the causal reasoning analysis is utilized to separate the confounding variable and the adjusting variable in all the observed variables, and a method for recovering the causal relationship is found out; in addition, the variables are reweighted by learning the sample weight values, and the reweighted variables are independent of each other, so that the causal relationship of a single variable to an effect variable is evaluated, namely the false correlation between non-causal characteristics and a prediction result is eliminated, and stable knowledge tracking is realized.

The embodiments of the present application will be further explained with reference to the drawings.

Referring to fig. 1, fig. 1 is a diagram illustrating steps of a knowledge tracking method based on causal reasoning according to some embodiments of the present application, including, but not limited to, steps S100 to S130.

Step S100, acquiring a random variable, wherein the random variable corresponds to a specific knowledge point; and obtains potential results representing the student's answer results to the particular knowledge point.

Specifically, the embodiment of the present application first performs observability learning of knowledge tracking: defining a knowledge point as a random variable K, and defining a potential result Y (K) to indicate whether a student can correctly answer a question corresponding to a specific knowledge point K ═ K, and K ∈ 0, 1; students with knowledge or so-called skills are also defined, i.e. K1, otherwise K0. Then for each data sample with index i-1, 2, … m, a knowledge point K is observed_iOne result Y_i(K_i) It is observed that the result of sample i, i.e. the relation between the random variable and the potential result, can be expressed as:

Y_i(K_i)＝K_i·Y_i(1)+(1-K_i)·Y_i(0)

it should be noted that, unless otherwise specified, a variable in the examples of the present application corresponds to a physical meaning, for example, Y represents a potential result in the above formula, the physical meaning of Y in the following formula is not changed, the meaning of Y is not described in detail, and other variables in the examples of the present application are also the same.

Step S110, observation variables are obtained, wherein the observation variables comprise confounding variables, adjusting variables and irrelevant variables.

Specifically, in the observability learning of knowledge tracing in step S100, except for the observation of the knowledge point K_iResults Y_i(K_i) It is also possible to observe a variable U of dimension n_iU is an observed variable.

In the observability learning of knowledge tracking performed in the embodiment of the present application, three assumptions are set, each of which is:

(ii) stable knowledge point values: when given an observation variable, it is assumed that the distribution of potential outcomes for one knowledge point is not affected by the particular allocation of another knowledge point.

② non-mixing: the random variables corresponding to the knowledge points are independent from the underlying outcomes when the observed variables are given.

(iii) separability: the observed variable U can be decomposed into three sets, i.e., U ═ X, Z, I, where X is a confounding variable, Z is an adjusting variable, and I is an independent variable.

Referring to fig. 2, fig. 2 is a causal graph provided by some embodiments of the present application, and the causal graph shown in fig. 2 may be used to separate confound variables, X, from tuning variables, the confound variables being associated with a random variable K (which may also be understood as knowledge point K) and possibly affecting a potential outcome Y; the adjustment variable is Z, which affects the potential result Y but is independent of the random variable K; the independent variable is I, which is independent of both the random variable K and the underlying result Y, and is therefore not shown in fig. 2.

And step S120, determining the conversion output of the potential result according to the mixed variable, the adjusting variable and the relation between the random variable and the potential result.

Specifically, the conversion output in the embodiment of the present application specifically is:

wherein Y + is the conversion output, K is the random variable, Z is the adjustment variable, X is the miscellaneous vector, i is the number of the variable. In the above formula, phi (Z) is used to linearly process Z, which helps to reduce the variance of the potential result Y with respect to the tuning variable Z, and then the tuning estimate continues to reduce the difference, which is:

E(Y+|X)＝E(Y(1)-Y(0)|X)

based on the above estimated values, a first objective function for the observed variable U can be derived:

minimize||Y⁺-h(U)||²

updating the first objective function, and determining a second objective function, specifically as follows:

wherein α and β are parameters for optimization, α is a separation tuning variable Z, β is an optimization for separation confounding variable X, γ is a coefficient constant of h (u), λ, δ, η are constants, W is a sample weight, and W (β) is specifically:

in addition, the loss function when estimating the tendency score is specifically expressed as:

additionally, ☉ refers to the Hadamard product. The above-mentioned tendency score is defined as: the probability of a student in the knowledge state (K ═ 1) is given for all observed variables U.

In addition, the first objective function is optimized, where the optimization is used to separate the tuning variables Z and β, and β is used to separate the variable X from the variable U, specifically, the coefficient vector α is optimized, and the optimization is performed by the following formula:

the analysis of steps S100 to S120 provides interpretability that once the confounding variable X and the tuning variable Z are separated, all causal relationships between the confounding variable X and the potential outcome Y can be restored, thereby achieving stable learning. In predicting the potential outcome Y, the part used to verify whether the student can correctly answer the question by the causal feature is stable, while the unstable part is mainly derived from the instability of the non-causal feature (noise feature) which is derived from the false correlation between the non-causal feature and the potential outcome Y. Thus, if the non-causal features are not related to the potential outcome Y, this spurious correlation can be eliminated. Accordingly, embodiments of the present application provide a causal regulator with a learning algorithm to eliminate such false correlations, which will be described in the following.

And step S130, determining balance weight according to the second objective function and the sample weight, wherein the balance weight is used for evaluating the causal relationship between the single variable and the potential result.

Specifically, in the embodiment of the present application, a learning algorithm is proposed, which is used for constructing balance weights for estimating causal effects, and the learning algorithm adjusts sample weights to enable continuous estimation to reach approximate balance, and the main idea is to learn the sample weights, weight the samples by using the sample weights, make the weighted variables independent of each other, so as to help evaluate the causal relationship between single variables and effect variables, and theoretically prove the existence of the sample weights.

First, refer to the following second objective function:

to minimize the second objective function, the following should be minimized:

the confound variable X is separated from the observed variable U, and then the second objective function is updated by adjusting the sample weight W, as shown in the following equation:

wherein λ is₁、λ₂、λ₃、λ₄Are all constants, W is the adjusted global sample weight, X_iIs the ith row/sample in the confounding variable X, and p is the number of covariates.

In addition, the first line in the above equation is the weighting penalty in the learning algorithm, i.e., the following equation:

setting the sample weight W to be more than or equal to 0, indicating that the weight of each sample is restricted to be a non-negative number when the norm is

The variance of the sample weights can be reduced. In addition, elastic net restraint is set

And | | | β | | | non-conducting phosphor₁Lambda 4 is less than or equal to lambda 4, and overfitting can be avoided.

In addition, the above formula includes a causal regulator, which is specifically:

wherein, X_·,-jIs the jth variable in the confounding variable X, and X_·,-j＝X\{X_·,jIndicates that all the variables remaining by removing the jth variable two are formulated for the causal regulator as X_·,jWhen the random variable corresponding to the knowledge point is set, the loss of imbalance of the covariate, which means Hadamard product, is indicated. By sample weighting using the sample weight W, the stability characteristics can be determined by the causal regulator checking whether there is a correlation between the latent outcome Y and the confounding variable X covariates.

In order to determine stable features, the embodiment of the present application proposes a global equilibrium model, in which a global sample weight may be learned, and the weight may be used to estimate the effect of each feature; while the model may control other features.

Referring to fig. 3, fig. 3 is a general block diagram of a global balancing model provided in some embodiments of the present application. To capture the non-linear structure between stable features and response variables, the global equilibrium model uses Convolutional Neural Networks (CNNs) to build the auto-encoder model, since CNNs can preserve the input neighborhood relationships and spatial locality in higher potential values. The mixed factor X is input into an automatic encoder to obtain a reconstructed vector after encoding

The auto-encoder also passes phi (-) to the global equalization section. In addition, the global balance model improves the ability to process long input sequence data by combining the depth knowledge tracking module DKT, while the depth knowledge tracking module maps the input data to a non-linear low-dimensional space, since balancing in a low-dimensional space would simplify the problem of global balance. The output data of the auto-encoder is input to a product decoder for data reconstruction and the sample weights W are determined. For each covariate j, the weight would be at X_·,jBalance comes from the dimensionality reduction Φ (X)_·,-j) The construction covariates of (1). Finally, sample weights are usedThe weights W weight the samples and learn a prediction model of KT, which is a function of a low-dimensional representation of covariates, which can be expressed in particular as follows:

through the processing of the global balancing model, the second objective function is updated again as follows:

wherein λ₅Is a constant, phi (X)^KTo represent variables

Hybrid vector X and reconstructed vector representing input

With a sample weight W between them. If there is enough data so that all realizations of the confounding variable X appear in the data, then an accurate balance weight can be derived. The balance weight can reduce the covariance between the elements, and after the balance weight is determined, the causal relationship between the single variable and the result variable can be evaluated.

The following describes, by way of an example, the causal relationship between a single variable and an effect variable that can be evaluated after a sample is weighted by a global balance model proposed in an embodiment of the present application.

Assuming that the confounding variable X is S, V, where S is a stable feature and V is a noisy feature, one equation of likelihood is assumed to exist as follows:

P(y|s)＝P(Y＝y|S＝s)＝P(Y＝y|S＝s,V＝v)

one proposal is presented below: if for all confounding variables X, there are

When the following conditions are satisfied:

there is a sample weight W, which is in fact the balance weight proposed by the present application, which satisfies:

at this time, the variables in the confounding variable X are independent of each other after being equalized by the sample weight W, as evidenced by: according to the assumption that P (Y | S) ═ P (Y ═ Y | S ═ S) and the above proposal, that is, S and V are independent of each other, it can be deduced that:

p (Y | V ═ V) ═ P (Y | S, V ═ V), that is to say

P (Y | V ═ V) ═ P (Y ═ Y | S), that is to say

P(Y＝y|V＝v)＝P(Y＝y)

It can therefore be concluded that Y is independent of V, and that the method proposed in this application makes KT stable after re-weighting, since only stable features are relevant to the outcome, in other words the causal relationship between X and Y is restored.

The following describes the result of performing an experiment on a real data set by the knowledge tracking method based on causal reasoning proposed in the embodiment of the present application.

Referring to table 1, table 1 is a case of a plurality of existing data sets provided by some embodiments of the present application. Table 1 the first two rows of datasets are from assismntents online tutoring platform and have been widely used in a variety of knowledge tracking models. The original dataset of the first line data of table 1 (ASSIST2009) was processed due to the record repetition, the processed dataset containing 328291 question-answer interactions from 4417 students from 124 skills. The second line dataset (ASSIST2015) in Table 1 contained 19917 students' responses to 100 skills, totaling 708631 question-answer interactions; it contains more interactions than the dataset ASSIST2009, but on average the number of records per skill and student is smaller, because there are more students. The third row of data set in Table 1 (Statics2011) is provided by the engineering statistics course, which is a data set of 189927 interactions from 333 students with 1223 skill labels. The fourth data set (Simulated-5) of table 1 simulates 2000 virtual students to learn the virtual concepts, five virtual concepts, each student answering 50 questions in the same order, and finally getting 100000 answers.

Data set	Number of students	Label (R)	Answer to the question
				ASSIST2009	4417	124	328291
ASSIST2015	19917	100	708631
				Statics2011	333	1223	189927
Simulated-5	2000	5	100000

TABLE 1

Performing 5-time cross validation on the training set mentioned in the table 1 to obtain a super-parameter setting, and initializing the weight of the model by using an Xavier unified initialization program, wherein the learning rate and the dropping rate are respectively set to be 0.01 and 0.5; using AUC as an evaluation index; in addition, the F1 score (another classical indicator in classification) was also used to evaluate the performance of the model.

Referring to table 2, table 2 shows the test results of the data set by the multiple knowledge tracking methods provided in some embodiments of the present application, and 4 methods are selected in the embodiments of the present application to be compared with the knowledge tracking method based on causal reasoning provided in the present application, where the 4 methods are PFA, BKT, DKT, and DKT +. Referring to table 2, in addition to the normalized-5, the knowledge tracking method based on causal reasoning proposed in the present application achieves excellent results on four data sets for both AUC and F1 evaluation indexes. For example, in terms of AUC for ASSIST2015, the proposed approach exceeds DKT + 10%; the F1 score also performed well, and the knowledge tracking method based on causal reasoning proposed in this application was significantly improved compared to other models.

Moreover, the knowledge tracing method proposed in the present application is not very excellent in performance on the fused-5 data set, one reason is that there is no long sequence in the 4 data sets used in the embodiments, and therefore the advantage of capturing the long sequence of the method of the embodiments of the present application cannot be exerted; another reason is that all data have the same sequence of problems and each problem occurs only once, so the dependence of the weighted-5 data set between data is not as strong as the other 3 data sets, which also affects the performance of the causal reasoning based knowledge tracking method in the embodiments of the present application.

TABLE 2

The above description has shown that the causal reasoning knowledge tracking method proposed by the present application has unsophisticated performance based on two evaluation criteria of AUC and F1 by comparing the performance of 4 knowledge tracking methods in the prior art with the performance of the knowledge tracking method proposed by the present application in 4 real data sets. For the problem of DKT in the knowledge tracking task, namely the transitional problem of prediction output, the cause-and-effect reasoning-based knowledge tracking method proposed by the present application also better solves the problem, and the specific result refers to fig. 4 and 5, fig. 4 is a heat map student in ASSISTments2009 data set for knowledge tracking using the cause-and-effect reasoning-based knowledge tracking method provided by some embodiments of the present application; fig. 5 is a heatmap for knowledge tracking of a student in assistcents 2009 dataset using DKT, provided by some embodiments of the present application; the horizontal axis of fig. 4 and 5 has the same meaning, and the vertical axis has the same meaning; the abscissa indicates the result of the student's answer to a certain knowledge concept, for example, (55,1) indicates the knowledge concept with answer number 55 at the current time step, and if the answer result is correct, it is marked as (55, 1); if the student answers the knowledge concept with the number of 55 at the current time step, recording as (55, 0); the horizontal axis has 35 coordinates, which indicates that the student answers 35 questions; the ordinate axes represent the numbers of knowledge concepts of the student responses, 98, 55, 45, 33, 32 respectively. In addition, in fig. 4 and 5, the degree of grasp of the student on different knowledge concepts is indicated by the light color on the right side of the picture, the darkest color represents complete grasp, the lightest color represents no grasp, and the degree of grasp is divided into 6 grades of 0.0 to 1.0.

Referring to fig. 4, after answering 35 questions, the student mastered the

knowledge concepts

32 and 55 but failed to master the

knowledge concepts

33, 45, and 99. As can be seen from fig. 4 and 5, the causal reasoning knowledge tracking method proposed by the present application has better performance in knowledge state modification than DKT. Referring to FIG. 4, when a student incorrectly answers a knowledge concept 32 repeatedly, the knowledge state of the concept 32 gradually decreases. However, the student correctly answers the concept 32 in the last three time steps, and thus the knowledge state of the student concept 32 is increased. While referring to fig. 5, i.e., prediction of DKT, it is clear that the knowledge prediction results for the student are not consistent with the method proposed in the present application, the prediction of DKT is not reasonable because the knowledge state of the student gradually changes over time but cannot alternate between mastery and not mastery, and it is shown in fig. 5 that the knowledge state of the student exhibits a wave-like transition, which misleads the interpretation of the knowledge state of the student. Therefore, in the aspect of predicting the knowledge acquisition state of the student by using the knowledge tracking method, the cause-and-effect reasoning knowledge tracking method provided by the embodiment of the application is better than the DKT method in the prior art, that is, the cause-and-effect reasoning knowledge tracking method provided by the embodiment of the application can better grasp the knowledge level of the student and provides a more useful reference for people through predictable visual analysis.

Referring to FIG. 6, FIG. 6 provides a causal reasoning based knowledge tracking system for some embodiments of the present application, the system 600 including an acquisition module 610, a data processing module 620, an objective function processing module 630, and a causal reasoning module 640; the acquisition module is used for acquiring random variables, potential results, observation variables and sample weights; the data processing module is used for determining a relation between the random variable and the potential result according to the random variable and the potential result; and is used for determining the conversion output of the potential result according to the mixed variable, the adjusting variable and the relation between the random variable and the potential result; the target function processing module is used for determining a first target function of the observation variable according to the conversion output; and is used for updating the first objective function according to the sample weight and determining the second objective function. And the causal reasoning module is used for determining balance weights according to the second objective function and the sample weights, and the balance weights are used for evaluating the causal relationship between the single variable and the potential result.

Referring to fig. 7, fig. 7 illustrates an apparatus according to some embodiments of the present application, the apparatus 700 including at least one processor 710 and at least one memory 720 for storing at least one program; fig. 7 illustrates an example of a processor and a memory.

The processor and memory may be connected by a bus or other means, such as by a bus in FIG. 7.

The memory, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs as well as non-transitory computer executable programs. Further, the memory may include high speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory located remotely from the processor, and these remote memories may be connected to the device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

Another embodiment of the present application also provides an apparatus, which may be used to perform the control method as in any of the above embodiments, e.g., to perform the above-described method steps S100 to S130 in fig. 1.

The above described embodiments of the device are merely illustrative, wherein the units illustrated as separate components may or may not be physically separate, i.e. may be located in one place, or may also be distributed over a plurality of network elements. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

The embodiment of the application also discloses a computer storage medium, wherein a program executable by a processor is stored, and the program executable by the processor is used for realizing the knowledge tracking method based on causal reasoning provided by the application when being executed by the processor.

It will be understood by those of ordinary skill in the art that all or some of the steps, systems, and methods disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those skilled in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art.

While the preferred embodiments of the present invention have been described, the present invention is not limited to the above embodiments, and those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and such equivalent modifications or substitutions are included in the scope of the present invention defined by the claims.

Claims

1. The knowledge tracking method based on causal reasoning is characterized by comprising the following steps:

acquiring a random variable, wherein the random variable corresponds to a specific knowledge point;

acquiring a potential result, wherein the potential result represents an answer result of a student to a specific knowledge point;

determining a relation between the random variable and the potential result according to the random variable and the potential result;

acquiring observation variables, wherein the observation variables comprise a confounding variable, an adjusting variable and an irrelevant variable;

determining a conversion output of the potential result according to the miscellaneous variable, the adjusting variable and the relation between the random variable and the potential result;

determining a first objective function of the observed variable according to the conversion output;

acquiring sample weight, updating the first objective function according to the sample weight, and determining a second objective function;

determining a balance weight from the second objective function and the sample weight, the balance weight being used to evaluate a causal relationship between a single variable and the potential outcome.

2. The causal inference based knowledge tracking method of claim 1, further comprising:

and performing optimization processing on the first objective function, wherein the optimization processing separates the adjusting variable from the observation variable and separates the confounding variable from the observation variable.

3. The causal inference based knowledge tracking method of claim 2, further comprising:

acquiring a global balance model;

evaluating a causal relationship between a single variable and the potential outcome through the global equilibrium model.

4. The causal inference based knowledge tracking method of claim 3, wherein said assessing causal relationships between individual variables and said potential outcomes via said global equilibrium model comprises:

encoding the mixed variable through an automatic encoder to determine a reconstruction vector;

and performing data reconstruction on the reconstruction variable through a product decoder to determine the sample weight.

5. The causal inference based knowledge tracking method of claim 4, wherein said assessing causal relationships between individual variables and said potential outcomes via said global equilibrium model further comprises:

the input data is mapped to a non-linear low dimensional space by a depth knowledge tracking module.

6. The causal reasoning-based knowledge tracking method of claim 1, wherein the conversion output is specifically:

7. The method for knowledge tracking based on causal reasoning according to claim 1, wherein said first objective function is specifically:

minimize||Y⁺-h(U)||²

wherein, Y⁺For the conversion output, U is the observed variable.

8. A knowledge tracking system based on causal reasoning, comprising:

the acquisition module is used for acquiring random variables, potential results, observation variables and sample weights;

the data processing module is used for determining a relation between the random variable and the potential result according to the random variable and the potential result; and is used for determining the conversion output of the potential result according to the miscellaneous variable, the adjusting variable and the relation between the random variable and the potential result;

the objective function processing module is used for determining a first objective function of the observation variable according to the conversion output; and the first objective function is updated according to the sample weight, and a second objective function is determined.

And the causal reasoning module is used for determining balance weight according to the second objective function and the sample weight, and the balance weight is used for evaluating the causal relationship between the single variable and the potential result.

9. An apparatus, comprising:

at least one processor;

at least one memory for storing at least one program;

when executed by the at least one processor, cause the at least one processor to implement the causal inference based knowledge tracking method of any of claims 1-7.

10. A computer storage medium having stored therein a processor-executable program, wherein the processor-executable program, when executed by the processor, is configured to implement the causal reasoning-based knowledge tracking method of any of claims 1-7.