MOOC personalized learning support system based on behavioral science theory

Abstract

The present invention discloses a personalized learning support system for MOOCs based on behavioral science theory, comprising three parts: a learner model system, a learner cognitive situation judgment system, and a learner learning behavior improvement system. First, a number of learner models for reference are constructed based on the learning data of MOOC learners who have completed previous periods. When a new MOOC learner joins the course, a variety of learning modes are provided for the learner, and these modes rely on the established learner models. After the learner selects a learning mode, the learner's cognitive situation of the knowledge points to be learned that week is judged based on his or her learning behavior each week. When the cognitive situation is poor and triggers the system threshold, the learner learning behavior improvement system is activated, intervenes in the learner's learning process, improves his or her learning behavior, and ultimately makes the learner's learning behavior consistent with the learning behavior of the learner model of the selected learning mode.

Description

Technical Field

[0001] The present invention belongs to the field of new generation information technology, and in particular relates to a MOOC personalized learning support system based on behavioral science theory. Background Art

[0002] Modern information technology has promoted the transformation and development of educational informatization. Massive Open Online Courses (MOOCs) are believed to be able to provide learners with high-quality course resources anytime and anywhere, enabling people to participate in higher education and lifelong education, and alleviating the contradiction between the scarcity of educational resources and people's demand for higher education.

[0003] However, high dropout rates have constrained the continued development of MOOCs. While MOOCs currently boast scale—a significant number of registered users—few learners are able to follow the curriculum and complete the entire MOOC learning process. Furthermore, traditional MOOC teaching methods are primarily lecture-based, using micro-lesson videos and slideshows lasting from a few minutes to over ten minutes. The focus is on disseminating and replicating learning content, lacking consideration for learners' diverse learning needs. MOOC platforms employ a diverse user base, each with distinct learning needs and behavioral characteristics. A monotonous and tedious learning experience is a primary reason for low MOOC course completion scores.

[0004] In summary, for MOOCs to continue to develop healthily and rapidly and truly benefit more learners, it's imperative to reduce their high dropout rates and enhance their personalized learning experience. Therefore, tailoring MOOC instruction to individual students is essential. The more personalized MOOC learning is, the less likely learners are to drop out, and the better the learning outcomes. Summary of the Invention

[0005] The present invention aims to overcome the shortcomings of the existing technology and provide a personalized learning support system for MOOCs based on behavioral science theory. The system can determine the learner's current cognitive state, quantify the learner's learning motivation, and improve the learner's learning behavior.

[0006] The present invention is achieved through the following technical solutions:

[0007] A personalized learning support system for MOOCs based on behavioral science theory, including three parts: a learner model system, a learner cognitive status judgment system, and a learner behavior improvement system;

[0008] The learner model system includes a learning demand model, a learning motivation model, a learning behavior model, and a negative feedback relationship model between behavior and motivation;

[0009] The learner cognitive situation judgment system is used to judge the learner's cognitive situation of the course studied, and the learner cognitive situation judgment system includes a performance prediction model based on learning behavior;

[0010] The system for improving learner learning behavior constructs an adaptive neuro-fuzzy inference system based on model reference as a system for improving learner learning behavior in a personalized learning support system;

[0011] First, based on the learning data of MOOC learners who have completed previous courses, several learner models are constructed for reference; when a new MOOC learner joins the course, a variety of learning modes are provided for him or her. These modes rely on the established learner model. After the learner chooses a learning mode, his or her cognitive status of the knowledge points to be learned that week is judged based on his or her learning behavior each week; when the poor cognitive status triggers the system threshold, the system for improving the learner's learning behavior is activated, intervenes in the learner's learning process, improves his or her learning behavior, and ultimately makes the learner's learning behavior consistent with the learning behavior of the learner model of the selected learning mode.

[0012] In the above technical solution, the mathematical expression of the learning demand model is:

[0013]

[0014] Where n is the total number of knowledge points that need to be learned in the course, week is the total number of weeks of course learning, demand(t) represents the temporal variation pattern of the weekly learning demand, and ε(t) represents the step function.

[0015] In the above technical solution, the learning motivation model:

[0016] Assuming that learning motivation and learning needs meet the second-order object characteristics, the learning motivation model is:

[0017]

[0018] Where K represents the magnitude of the learning motivation under the unit step response when dynamic equilibrium is reached; ξ represents the damping ratio, and ω represents the oscillation frequency.

[0019] In the above technical solution, the learning behavior model:

[0020] behavior=base×MT

[0021] Where base is the base number of learning behaviors, MT is the coefficient of learning motivation conversion into learning behaviors, and behavior represents the number of learning behaviors that occur within a period of time; MT = K(1 + re(1-K)e -K ), re is the relative behavior coefficient;

[0022] MT indirectly represents the threshold for motivation to be converted into behavior; if the learning motivation is ultimately equal to the demand, the expected learning behavior will be achieved within this learning cycle; if the learning motivation is zero, no learning behavior will occur within this learning cycle.

[0023] In the above technical solution, the negative feedback relationship model between behavior and motivation is:

[0024]

[0025] Where e is the difference between the actual number of behaviors and the base number of learned behaviors, ω0, ξ0, and K0 are the initial values ​​at the beginning of course learning.

[0026] In the above technical solution, the performance prediction model based on learning behavior uses a DNN regression model to judge the learner's course cognition; in the historical database of MOOCs, a data set of appropriate size containing as many types of learners as possible is selected, and the DNN regression model is trained based on the data set; the data set is characterized by learning behavior and labeled as performance; learning behavior includes the number of times the course is visited, the number of days the course is visited, the number of course chapters visited, and the number of times video micro-courses are played; during the training process, 70% of the data set is divided into a training set, 15% is divided into a test set, and 15% is divided into a validation set.

[0027] In the above technical solution, the method for training the adaptive neuro-fuzzy inference system is as follows:

[0028] The training process is to select different types of learners to cover as many types of learners as possible who will use the personalized learning support system in the future; use different types of learners as learning motivation models, and then use the learner model corresponding to the learning mode as the reference learner learning motivation model, and repeat the training until the control effect of the adaptive neuro-fuzzy inference system reaches the expected effect or the number of training times reaches the maximum number of training times.

[0029] The advantages and beneficial effects of the present invention are:

[0030] (1) It can improve learners’ learning behavior and enhance the personalization of the learning process. After the intervention of the personalized learning support system based on behavioral science theory, the learners’ learning behavior can be corrected according to the learning mode chosen by the learners to prevent them from deviating from the proper learning trajectory; different learning suggestions can be provided according to each learner’s own situation and learning goals, so that the learning process can be adapted to their own cognitive situation and learning habits, making the learning process more personalized.

[0031] (2) Because learners’ learning behavior follows the system’s designated learner model, the possibility of learners dropping out of the course and giving up learning is greatly reduced, which can effectively reduce the dropout rate of MOOCs. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a structural diagram of the MOOC personalized learning support system based on behavioral science theory.

[0033] Figure 2 It is a DNN model structure diagram.

[0034] Figure 3 This is the ANFIS-MR system structure diagram in the MOOC personalized learning support system based on behavioral science theory.

[0035] Figure 4 This is a comparison chart of the changes in learning motivation in the first four weeks before and after ANFIS-MR intervention.

[0036] Figure 5.1-Figure 5.6 This is a graph showing learners' learning behavior information and learning performance before and after the intervention of the MOOC personalized learning support system based on behavioral science theory.

[0037] For ordinary technicians in this field, other relevant drawings can be obtained based on the above drawings without any creative work. DETAILED DESCRIPTION

[0038] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention are further described below with reference to specific embodiments.

[0039] A personalized learning support system for MOOCs based on behavioral science theory, including three parts: a learner model system, a learner cognitive situation judgment system, and a learner learning behavior improvement system.

[0040] (1) Learner Model System

[0041] The scientific basis of the learner model system is behavioral science theory. This theory was first proposed by Professor Mayo of Harvard University in the late 1920s and early 1930s. Behavioral science, informed by theories from psychology, human science, and socioeconomics, primarily studies human behavior. The core concept of behavioral science theory is that behavior is driven by motivation, which stems from individual needs, and that all human behavior follows this principle. When a person has an unmet psychological need or unrealized goal, they become motivated to act, which in turn triggers action. The resulting action may fully or partially satisfy the need, and the current state of the need determines the motivation for the next behavioral cycle. This new motivation then triggers new behavior, and this cycle repeats. Key points of behavioral science theory in guiding personalized learning are identifying the sources of learning motivation, clarifying the relationship between motivation and behavioral transformation, and the impact of behavioral outcomes on the motivation for the next cycle.

[0042] The learner model system needs to establish a mathematical model of learning needs, learning motivation, learning behavior, and the negative feedback relationship between behavior and motivation.

[0043] 1. Learning Demand Model: Assuming the time span of a MOOC course is reasonable, with a week as the minimum time unit, and the number of knowledge points to be learned each week is roughly equal, then the learning demand should be equal each week, then:

[0044]

[0045] Where n is the total number of knowledge points that need to be learned in the course, week is the total number of weeks of course learning, demand(t) represents the temporal variation pattern of the weekly learning demand, and ε(t) represents the step function.

[0046] 2. Learning Motivation Model: Learning motivation fluctuates rather than remains constant within each learning week. At the beginning of each learning week, motivation, fueled by the need to learn, continues to rapidly increase. However, this increase is not infinite. As the week progresses, needs are satisfied to a certain extent, and motivation declines. This decline in motivation causes the gap between needs and motivation to gradually widen, effectively giving motivation another boost. Motivation then briefly rises, but learning behavior further satisfies needs, causing motivation to decline again. This cycle continues until needs and motivation reach a dynamic equilibrium, preventing motivation from changing, or until the end of the learning week, with motivation repeating this pattern of rise and fall.

[0047] Motivation for learning MOOCs rises and falls an average of twice per week, ultimately reaching a dynamic equilibrium between demand and motivation, maintaining constant motivation. If learning demand is considered the stimulus and motivation is the response, the changing relationship between demand and motivation is similar to the closed-loop step response process of a second-order object in automatic control principles. Therefore, assuming that motivation and demand satisfy the characteristics of a second-order object, their Laplace transform in the complex frequency domain is:

[0048]

[0049] Where K represents the magnitude of the learning motivation under the unit step response when dynamic equilibrium is reached; ξ represents the damping ratio, and ω represents the oscillation frequency.

[0050] 3. Learning Behavior Model: Learning motivation triggers learning behavior. Learning motivation changes continuously over time, while learning behavior changes much more slowly than learning motivation. Rather than focusing on the specific learning behavior occurring at a specific point in time, we should focus on how many learning behaviors occurred over a period of time, i.e., the time-integrated characteristic of learning behavior. For ease of calculation, the relationship between learning behavior and learning motivation is expressed as:

[0051] behavior=base×MT (3)

[0052] In the formula, base is the base number of learning behaviors, MT is the coefficient of converting learning motivation into learning behaviors, and behavior represents the number of learning behaviors occurring within a period of time.

[0053] MT=K(1+re(1-K)e -K ) (4)

[0054] re is the relative behavior coefficient.

[0055] MT indirectly represents the threshold for motivation to be transformed into behavior. If learning motivation ultimately equals demand, the expected learning behavior is achieved within this learning cycle; if learning motivation is zero, no learning behavior occurs within this learning cycle.

[0056] 4. Negative feedback relationship model between behavior and motivation:

[0057]

[0058] Where e is the difference between the actual number of behaviors and the base number of learning behaviors, and ω0, ξ0, and K0 are the initial values ​​at the beginning of the course. The results of each week's learning behavior will affect the magnitude of the next week's learning motivation in formula (2), reflecting a negative feedback relationship. The worse the learning behavior results this week, the later it will take for the learning motivation to stabilize next week, and the weaker the intensity of the learning motivation after stabilization. Furthermore, once learning behavior deviates from expectations, it is extremely difficult to return to the desired learning state through self-correction.

[0059] (2) Learner cognitive status judgment system

[0060] The personalized learning support system needs to judge the learner's cognitive status of the course studied, and based on this, provide behavioral suggestions to change the learner's learning behavior. The present invention assumes that the learner's test score is equivalent to the learner's cognitive status. The higher the score, the better the cognitive status; the lower the score, the worse the cognitive status. The judgment of cognitive status requires the establishment of a corresponding relationship between learning behavior and grades, that is, to predict what score the current learning behavior will get. The grade is a continuous distribution data, so this is a regression problem among prediction problems. Therefore, the core of the learner cognitive status judgment system is to establish a grade prediction model based on learning behavior, and its purpose is to find the following mapping relationship:

[0061] grades=f(events,activedays,videos,chapters) (6)

[0062] The mapping f should, as far as possible, accurately map all combinations of learning behaviors to the correct scores.

[0063] The present invention uses a DNN (Deep Neural Network) regression model to determine learners' course cognition. DNN is a deepening of the BP neural network, with its hidden layer increasing from one to multiple layers. The output forward propagation and error backpropagation process between each two layers is still similar to that of a simple BP neural network. The parameters of the DNN of the present invention are set to two hidden layers, and the activation function of the hidden layer is ReLU, as shown in formula (7). The activation function is the sigmoid function, and the calculation formula is shown in (8).

[0064]

[0065]

[0066] Assume that the input of the input layer is x1, x2, x3, x4, and the output of the jth neuron in the hidden layer is shown in formula (9).

[0067]

[0068] Where w ij represents the weight between input layer neuron i and hidden layer neuron j, θ j represents the threshold of hidden layer neuron j, and g is the activation function of hidden layer neurons.

[0069] The output of the output layer is shown in formula (10). jo is the weight from hidden layer neuron j to output layer neuron o, θ o is the threshold of the output layer neuron o, and f is the activation function of the output layer neuron.

[0070]

[0071] From the MOOC historical database, select a dataset of appropriate size that includes a wide variety of learners and train a DNN model based on it. In this example, the dataset features learning behaviors and labels are grades. Learning behaviors include the number of course visits, the number of days a course was visited, the number of course chapters visited, and the number of video micro-courses played. The dataset size is 41,600 data points.

[0072] During the training process, this example divides 70% of the data set into the training set, 15% into the test set, and 15% into the validation set. There are three hidden layers, with 6, 10, and 10 neurons in the hidden layers respectively. The DNN structure diagram is as follows: Figure 2 shown.

[0073] A common metric for comparing the accuracy of regression models is R 2 , the calculation formula is shown in (6). 2The value range is [0,1], R 2 The closer it is to 1, the higher the prediction accuracy of the regression model.

[0074]

[0075] In this example, the R of the DNN performance regression model 2 It is 0.8055.

[0076] (3) Improve learners’ learning behavior system

[0077] Personalized learning support systems require the ability to provide each learner with a targeted learning plan, addressing poor learning behaviors, improving academic performance, and indirectly enhancing cognitive development while enabling more effective utilization of MOOC resources. The key to personalized learning is enhancing learning motivation, which is time-varying and nonlinear, making it difficult to accurately describe its dynamics using mathematical models.

[0078] This paper introduces model reference control into an adaptive neuro-fuzzy inference engine, constructing an adaptive neuro-fuzzy inference system based on model reference (ANFIS-MR). ANFIS-MR is used as a learning behavior improvement system within a personalized learning support system. The ANFIS-M system has five layers, and the functions of each layer are explained below.

[0079] The first layer is the fuzzy layer. This layer has variable parameters and fuzzifies the input and outputs the corresponding membership degree.

[0080]

[0081] Where O 1 i is the membership degree of the corresponding node, μA i (x) is the membership function of the corresponding node. Common functions include bell-shaped, trapezoidal, and Gaussian functions. The present invention adopts the Gaussian function.

[0082] The second layer: rule layer. This layer performs fuzzy reasoning. The output of each node represents the credibility of a rule, multiplying each input signal.

[0083] O 2 i =ω i =μA i (x)·μB j (x)·μC k (x)·μD l (x) i,j,k,l=1,2,... (13)

[0084] Where O2 i represents the credibility of the corresponding rule, ω i is the incentive intensity of the i-th rule.

[0085] The third layer: Normalization layer. This layer calculates the normalized credibility of each rule.

[0086]

[0087] Where O 3 i is the normalized credibility of the corresponding rule, is the ratio of the credibility of the i-th rule to the total credibility of all rules.

[0088] The fourth layer: Defuzzification layer. This layer calculates the output of each rule.

[0089]

[0090] Where O 4 i is the output value after defuzzification of the corresponding rule, f i It is a fuzzy rule.

[0091] The fifth layer: output layer. This layer calculates the total output of the entire fuzzy system.

[0092]

[0093] Where O 5 i is the total output value of the entire system.

[0094] See attached Figure 3 , the method for training the ANFIS-MR system is as follows:

[0095] The training process is to select different types of learners to cover as many types of learners as possible who will use the personalized learning support system in the future. Figure 3 The learning motivation model in the learning mode is used as the reference learner learning motivation model, and the training is repeated until the control effect of ANFIS-MR reaches the expected effect or the number of training times reaches the maximum number of training times.

[0096] Since there are only two learner models in this example, the one with better performance is selected as the reference learner, so only one ANFIS-MR needs to be trained, and its structure is as follows: Figure 3 shown. Figure 4 Shown are the changes in learning motivation in the first 4 weeks before and after the ANFIS-MR intervention.

[0097] The present invention designs a personalized learning support system for MOOCs based on behavioral science theory. First, it constructs several learner models for reference based on the learning data of MOOC learners who have completed the course in the past. When a new MOOC learner joins the course, it provides a variety of learning modes for him / her. These modes rely on the established learner models. After the learner selects a learning mode, the cognitive status of the knowledge points that he / she should learn that week is judged based on his / her learning behavior every week. When the cognitive status is poor and the system threshold is triggered, the adaptive neural fuzzy reasoning system based on model reference (i.e., the system for improving learner learning behavior) is started, intervenes in the learner's learning process, improves his / her learning behavior, and finally makes the learner's learning behavior consistent with the learning behavior of the learner model of the selected learning mode.

[0098] In this example, 250 learners from the MOOC's historical data were selected as new learners. The learning patterns of those who consistently learned and achieved high grades were used as the learner-selected patterns. A simulation experiment was conducted to compare the changes in learning behavior of new learners before and after the intervention of a personalized MOOC learning support system based on behavioral science theory. Figure 5 shows the results of the simulation experiment.

[0099] It is not difficult to find from the simulation experiment that the MOOC personalized learning support system based on behavioral science theory can improve learners' learning behavior; it can extend learners' MOOC learning time and effectively reduce the MOOC dropout rate.

[0100] The above is an exemplary description of the present invention. It should be noted that, without departing from the core of the present invention, any simple deformation, modification or other equivalent replacement that can be made by other skilled in the art without expending creative labor falls within the scope of protection of the present invention.

Claims

1. A personalized learning support system for MOOCs based on behavioral science theory, characterized by: It includes three parts: learner model system, learner cognitive situation judgment system, and learner learning behavior improvement system; The learner model system includes a learning demand model, a learning motivation model, a learning behavior model, and a negative feedback relationship model between behavior and motivation; The learner cognitive situation judgment system is used to judge the learner's cognitive situation of the course studied, and the learner cognitive situation judgment system includes a performance prediction model based on learning behavior; The system for improving learner learning behavior constructs an adaptive neuro-fuzzy inference system based on model reference as a system for improving learner learning behavior in a personalized learning support system; First, based on the learning data of previously completed MOOC learners, several learner models are constructed for reference. When a new MOOC learner joins the course, multiple learning modes are provided for them. These modes rely on the established learner models. After the learner selects a learning mode, their cognitive status of the knowledge points they should learn that week is judged based on their learning behavior each week. When the cognitive status is poor and the system threshold is triggered, the system for improving learner learning behavior is activated, intervening in the learner's learning process to improve their learning behavior, ultimately making the learner's learning behavior consistent with the learning behavior of the learner model for the selected learning mode. The mathematical expression of the learning demand model is: Where n is the total number of knowledge points that need to be learned in the course, week is the total number of weeks of course learning, demand(t) represents the temporal variation pattern of weekly learning demand, and ε(t) represents the step function. Assuming that learning motivation and learning needs meet the second-order object characteristics, the learning motivation model is: Where K represents the magnitude of the learning motivation under the unit step response when dynamic equilibrium is reached; ξ represents the damping ratio, and ω represents the oscillation frequency.

2. The personalized learning support system for MOOCs based on behavioral science theory according to claim 1 is characterized by: The learning behavior model: behavior=base×MT Where base is the base number of learning behaviors, MT is the coefficient of learning motivation conversion into learning behaviors, and behavior represents the number of learning behaviors that occur within a period of time; MT = K(1 + re(1-K)e -K ), re is the relative behavior coefficient; The threshold for motivation to be transformed into behavior is indirectly expressed through MT; if learning motivation is ultimately equal to demand, the expected learning behavior is achieved within this learning cycle; If the learning motivation is zero, no learning behavior will occur during this learning cycle.

3. The personalized learning support system for MOOCs based on behavioral science theory according to claim 2 is characterized by: The negative feedback relationship model between behavior and motivation is: Where e is the difference between the actual number of behaviors and the base number of learned behaviors, ω0, ξ0, and K0 are the initial values ​​at the beginning of course learning.

4. The personalized learning support system for MOOCs based on behavioral science theory according to claim 1 is characterized by: The learning behavior-based performance prediction model uses a DNN regression model to determine learners' course cognition. A dataset of appropriate size containing as many types of learners as possible is selected from the MOOC historical database, and a DNN regression model is trained based on this dataset. The dataset is characterized by learning behavior and labeled by performance. Learning behavior includes the number of course visits, the number of days a course is visited, the number of course chapters visited, and the number of video micro-courses played. During the training process, 70% of the dataset is divided into a training set, 15% into a test set, and 15% into a validation set.

5. The MOOC personalized learning support system based on behavioral science theory according to claim 1 is characterized by: The method for training an adaptive neuro-fuzzy inference system is as follows: The training process is to select different types of learners to cover as many types of learners as possible who will use the personalized learning support system in the future; use different types of learners as learning motivation models, and then use the learner model corresponding to the learning mode as the reference learner learning motivation model, and repeat the training until the control effect of the adaptive neuro-fuzzy inference system reaches the expected effect or the number of training times reaches the maximum number of training times.

Images