MOOC online user behavior monitoring method based on separation of microscopic and macroscopic information

Abstract

The invention relates to an MOOC online user behavior monitoring method based on separation of microscopic and macroscopic information. A deep learning framework based on a double-tower structure is provided, macroscopic information and microscopic information are modeled respectively, time sequence information can be effectively mined from continuous activity records of learners, and two modes of combined training and separated training can be supported. According to the method, the model with the time sequence modeling mechanism is used, so that the potential time sequence mode in the online learning interaction log can be effectively captured; the performance of the proposed double-tower frame in the combined mode or the separated mode is better than that of a single-tower baseline model, the model performance is improved, the overall performance of the model can be improved through the combined mode, and the effect is particularly remarkable when microscopic and macroscopic information is processed at the same time. And the separation mode provides higher flexibility, and is suitable for scenes requiring independent training of each part of the model.

Description

Technical Field

[0001] The present invention relates to a MOOC online user behavior monitoring method based on separation of micro and macro information. Background Art

[0002] Well-known MOOC platforms such as Coursera, EdX, and Udacity provide high-quality learning resources to learners around the world every day. Despite this progress, MOOC courses still have a high dropout rate problem. The reasons why students stop learning are difficult to accurately determine or predict, and may be due to the quality of quiz questions, poor presentation of course materials, or even simply personal problems of the learner. Online user behavior monitoring can be an effective means of early intervention; at the same time, it can also be used as a teaching improvement method to determine whether online courses need to be adjusted or modified. With the growing interest of interdisciplinary research groups in applying machine learning techniques to online education problems, many techniques have been used to build learner activity monitoring models.

[0003] Existing techniques include logistic regression and its variants [1], support vector machines [2][3], and decision trees [4][5]. In addition, in recent years, models based on deep learning have also shown great potential in mining complex latent information. For example, multi-layer perceptrons (MLPs) have been used to deeply compare various evaluation indicators [6]. Some researchers [7] proposed the ConRec model, which combines the advantages of recurrent neural networks (RNNs) and convolutional neural networks (CNNs). Other researchers [8] proposed a context-aware feature interaction model that uses context smoothing techniques, combines user and course information, and achieves information fusion through attention mechanisms. However, to the best of our knowledge, existing studies have not yet attempted to distinguish between micro and macro information and model them separately for monitoring online users' learning behavior.

[0004] References:

[0005] [1]He,J.,Bailey,J.,Rubinstein,BI,Zhang,R.:Identifying at-riskstudents in mas-sive open online courses.In:Twenty-NinthAAAI ConferenceonArtificial Intelli-gence(2015)

[0006] [2]Kloft,M.,Stiehler,F.,Zheng,Z.,Pinkwart,N.:Predicting MOOC dropoutover weeks using machine learning methods.In:Proceedings of the EMNLP2014Work-shop on Analysis of Large Scale Social Interaction inMOOCs,pp.60–65(2014)

[0007] [3]Amnueypornsakul,B.,Bhat,S.,Chinprutthiwong,P.:Predicting attritionalong the way:the UIUC model.In:Proceedings of the EMNLP 2014Workshop onAnalysis of Large Scale Social Interaction inMOOCs,pp.55–59(2014)

[0008] [4]Al-Shabandar,R.,Hussain,A.,Laws,A.,Keight,R.,Lunn,J.,Radi,N.:Machine learning approaches to predict learning outcomes in Massive openonline courses.In:2017International Joint Conference onNeuralNetworks(IJCNN),pp.713–720,IEEE(2017)

[0009] [5]Al-Shabandar,R., Hussain,A., Laws,A., Keight,R., Lunn,J.: Towards the differ-entiation of initial and final retention in massive open online courses. In: Huang,D.-S., Bevilacqua,V., Premaratne,P., Gupta,P. (eds.) ICIC 2017. LNCS, vol. 10361, pp. 26–36. Springer, Cham (2017). https: / / doi.org / 10.1007 / 978-3-319-63309-1 3

[0010] [6]Imran,A.S., Dalipi,F., Kastrati,Z.: Predicting student dropout in a MOOC: an evaluation of a deep neural network model. In: Proceedings of the 2019 5th International Conference on Computing and Artificial Intelligence, pp. 190–195 (2019)

[0011] [7]Wang,W., Yu,H., Miao,C.: Deep model for dropout prediction in MOOCs. In: Proceedings of the 2nd International Conference on Crowd Science and Engineer ing, pp. 26–32 (2017)

[0012] [8]Feng,W., Tang,J., Liu,T.X.: Understanding dropouts in MOOCs. In: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 517–524 (2019). Summary of the Invention

[0013] The purpose of the present invention is to address the defects of the prior art. Therefore, a MOOC online user behavior monitoring method based on separation of micro and macro information is provided, and a deep learning framework based on a double-tower structure is proposed, which includes a CNN-based micro module, which can effectively mine time series information from learners' continuous activity records.

[0014] To achieve the above purpose, the technical solution of the present invention is: a MOOC online user behavior monitoring method based on separating micro and macro information, proposing a deep learning framework based on a double-tower structure, modeling macro information and micro information respectively, which can effectively mine time series information from learners' continuous activity records, and can support both joint training and separate training modes.

[0015] In one embodiment of the present invention, the method includes:

[0016] Preprocess the input data;

[0017] Based on the preprocessed data, a deep learning framework based on a dual-tower structure is trained to achieve online user behavior monitoring.

[0018] In one embodiment of the present invention, the input data includes data from the course introduction and the user's historical learning records.

[0019] In one embodiment of the present invention, the input data is preprocessed, including data cleaning, normalization, and time series processing of the input data, as follows:

[0020] 1) Data cleaning:

[0021] Remove missing values, duplicate data and abnormal data from input data to ensure data integrity and consistency;

[0022] 2) Normalization processing:

[0023] Standardize or normalize the input data to ensure that the data dimensions of different features are consistent to improve the convergence and prediction effect of the model;

[0024] 3) Time serialization:

[0025] Arrange the input data in chronological order to form time series data so that the model can identify the time dependencies in the learning behavior.

[0026] In one embodiment of the present invention, a deep learning framework based on a dual-tower structure includes a macro module and a micro module; a user can select a joint mode or a separate mode to train the deep learning framework based on a dual-tower structure, the joint mode is suitable for scenarios where two modules are collaboratively optimized, and the separate mode is suitable for scenarios where a single module is optimized.

[0027] In one embodiment of the present invention, the macro module is used to process the basic information of the corresponding course and the descriptive statistical information of the interaction history between all learners and the corresponding course; for a certain user u i The interaction between the course c and the macro information is represented as G i (u i ,c)∈R h , where h is the dimension used to represent macro information; in the macro module, the macro information is modeled by GBDT, and the specific process is as follows:

[0028] (1) Input sample set

[0029] (2) Initialization

[0030]

[0031] Start traversal calculation from t = 1, t = {1, 2, ..., T};

[0032] (3) Calculate the loss function L for each sample with respect to the current model The negative gradient

[0033]

[0034] (4) Take the negative gradient as the fitting object and build a decision tree. Assume that the model parameters are

[0035]

[0036] (5) The constructed decision tree divides the sample set into J subsets, each leaf node is a subset, denoted as

[0037] R tj ,j=1,2,…,J

[0038] (6) Estimate the predicted value γ of each leaf node tj

[0039]

[0040] (7) Model update

[0041]

[0042] in, is the sample data in the input data set, γ is the model prediction value, L(y i ,γ) is the loss of the current sample; are model parameters, which are randomly initialized according to Gaussian distribution; initialization When it is a tree with only one root node, the predicted value of the root node can minimize the overall loss of the loss function L on the entire sample set.

[0043] In one embodiment of the present invention, the micro module refers to a user u i The interaction details with course c are defined as L i (u i ,c); Since a course may contain multiple teaching videos, L i (u i ,c) is further expressed as L i (u i ,c)∈R n·m ,c=v1,v2,…,v n , v i represents the i-th video of course c, n is the user u i The number of videos that have been interacted with belonging to course c, m is the dimension of the feature space; denoted as f1f m The feature representation of user u i and video v j Specifically, given a course c, L i,j For a user u i With a video v j The interaction record between L i,j (u i ,v j )∈L i (u i ,c); In the micro-module, the CNN network is used to capture micro-information, including valuable time series patterns; the specific process is represented by the input data set sample matrix X = [x1, x2, ..., x n ],X∈R n×d , where x i ∈R d Represents a specific sample, which is the specific information of an online user interacting with a resource; the convolution operation is as follows:

[0044] y=f(WX+b)

[0045] Among them, W is the weight corresponding to the input signal X, b is the bias term, and f is the convolution transformation function. After a series of consecutive convolution operations, pooling operations and fully connected layers are applied to summarize all the extracted information and generate the final prediction.

[0046] In one embodiment of the present invention, the intermediate results generated by the micro module and the macro module are combined by weighted summation before the final prediction, which is expressed as follows:

[0047] P(Y=1|x)=σ(Wmicro r micro +W macro r macro +b)

[0048] Among them, r micro and r macro are the intermediate results from the micro module and the macro module respectively; W micro and W marco are two different weights; b is the bias value; σ represents the regression function, and the mathematical form is shown as follows:

[0049]

[0050] t is the input variable of the regression function.

[0051] The present invention also provides a MOOC online user behavior monitoring system based on separation of micro and macro information, comprising a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0052] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) Product advantages:

[0055] 1. Identification of time series patterns: This paper effectively captures potential time series patterns in online learning interaction logs by using a model with a time series modeling mechanism (the proposed CNN-based model). Experiments show that the model using time series modeling performs well in processing micro information, which indicates that time series features in learning behavior are of great significance for prediction.

[0056] 2. Innovative design of dual-tower framework: The dual-tower framework proposed in this invention outperforms the single-tower baseline model in both joint mode and separation mode. Using different modules to process different types of information can further improve model performance. It is difficult to find the optimal solution by using only a single model to process all types of information.

[0057] 3. Flexible training mode: The present invention supports both joint training and separate training modes. In most cases, the joint mode can improve the overall performance of the model, especially when processing micro and macro information at the same time. The separate mode provides higher flexibility and is suitable for scenarios where each part of the model needs to be trained independently.

[0058] (2) Product use:

[0059] 1. Learner behavior monitoring: This invention analyzes learners’ interaction logs and identifies time series patterns therein, helping the MOOC platform predict which learners’ learning activities are likely to be affected, so as to take intervention measures in advance.

[0060] 2. Personalized learning intervention: Provide personalized learning suggestions or reminders based on learners’ learning behavior data to help high-risk learners re-engage in learning.

[0061] 3. Course design optimization: Through the analysis of learner behavior, it helps course designers understand which links in the learning process are likely to lead to cessation of learning, and then optimize the course content and structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The figure is a flow chart of the method of the present invention.

[0063] Figure 2 This is a structural diagram of the model of the present invention.

[0064] Figure 3 For the organization and relationship between user-video interaction and user-course interaction. DETAILED DESCRIPTION

[0065] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0066] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0067] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0068] The present invention provides a MOOC online user behavior monitoring method based on separating micro and macro information, and proposes a deep learning framework based on a double-tower structure, which models macro information and micro information respectively, can effectively mine time series information from learners' continuous activity records, and can support both joint training and separate training modes.

[0069] The following is the specific implementation process of the present invention.

[0070] This embodiment provides a method for monitoring MOOC online user behavior based on separation of micro and macro information. The main process is as follows: Figure 1 As shown, the model structure is as Figure 2 As shown, the model of the present invention mainly includes two modules and a regression function. The method of the present invention is specifically implemented as follows:

[0071] (1) Input:

[0072] The original input comes from the course introduction and the user’s historical learning records, which is defined as input x. It contains a mixture of micro and macro information.

[0073] (2) Micro modules:

[0074] Micro information refers to a user u i The interaction details with course c can be defined as L i (u i ,c). Since a course may contain multiple teaching videos, the above definition can be further expressed as L i (u i ,c)∈R n·m ,c=v1,v2,…,v n ; where n is the user u i The number of videos (belonging to course c) that have been interacted with, and m is the dimension of the feature space. Labeled as f1f m The feature representation of user u i and video v j Specifically, given a course c, L i,j For a user u i With a video v j The interaction record between L i,j (u i ,v j )∈L i (u i ,c). Figure 3 It represents the organization and relationship between user-video interaction and user-course interaction. In the micro-module, we use CNN network to capture micro-information, which may contain valuable time series patterns. The specific process can be expressed as input information X = [x1, x2, ..., x n ],X∈R n×d , where x i ∈R d The convolution operation can be constructed as shown in formula (1).

[0075] y=f(WX+b)

[0076] Where W is the weight corresponding to the input signal X, b is the bias term, and f is the convolution transformation function. After a series of consecutive convolution operations, pooling operations and fully connected layers are applied to summarize all the extracted information and generate the final prediction.

[0077] (3) Macro module:

[0078] Refers to high-level, summary information about a course. This includes basic information about the course and descriptive statistics of all learners’ interaction history with the course. i The interaction between the course c and the macro information is represented as G i (u i ,c)∈R h , where h is the dimension used to represent macro information. In the macro module, we use GBDT to model macro information. The specific process is shown in Table 1:

[0079] Table 1

[0080]

[0081]

[0082] 4) Regression function:

[0083] like Figure 2 As shown, the intermediate results produced by the micro-module and the macro-module are combined by weighted summation before the final prediction. This processing step is expressed as formula (8).

[0084] P(Y=1|x)=σ(W micro r micro +W macro r macro +b)

[0085] Among them, r micro and r macro are the intermediate results from the micro and macro modules respectively; W micro and W macro are two different weights; b is the bias value; σ represents the regression function, and the mathematical form is shown in the following formula. In this framework, the micro and macro modules can be trained in two different ways: joint mode or separate mode. In the joint mode, the two components are trained together. In contrast, in the separate mode, the two components are trained independently and do not know each other.

[0086]

[0087] t is the input variable of the regression function.

[0088] (5) Output:

[0089] The predicted value is defined as y∈{0, 1} (where 1 means stop learning and 0 means continue learning).

[0090] The use process of the software product of the method of the present invention

[0091] (1) Data input:

[0092] Derived from the course introduction and the user's historical learning records.

[0093] (2) Data preprocessing:

[0094] 1. Data cleaning:

[0095] Remove missing values, duplicate data, and abnormal data to ensure data integrity and consistency.

[0096] 2. Normalization processing:

[0097] Standardize or normalize the learner's behavior data to ensure that the data dimensions of different features are consistent to improve the convergence and prediction effect of the model.

[0098] 3. Time serialization:

[0099] The learners’ interaction logs are arranged in chronological order to form time series data so that the model can identify the temporal dependencies in learning behaviors.

[0100] (3) Model selection and training:

[0101] The platform can select the micro-module and macro-module in the dual-tower framework for training as needed. The micro-module (the proposed CNN model) is used to capture the time series patterns in the learner's behavior, while the macro-module (GBDT) processes the category and demographic information.

[0102] Users can choose "joint mode" or "separate mode" to train the model. The joint mode is suitable for scenarios where two modules are optimized together, while the separate mode provides a more flexible way to optimize modules independently.

[0103] (4) Model prediction:

[0104] The trained model is used for online prediction, real-time analysis of new learners’ interaction data, and output of their learning behavior risk scores. Based on the scores, the platform can identify learners whose learning activities are affected in advance.

[0105] The present invention also provides a MOOC online user behavior monitoring system based on separation of micro and macro information, comprising a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0106] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0107] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0108] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0109] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0111] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for monitoring online user behavior of MOOC based on separation of micro and macro information, characterized in that: A deep learning framework based on a dual-tower structure is proposed to model macro information and micro information respectively. It can effectively mine time series information from the learner's continuous activity records and support both joint training and separate training modes.

2. The MOOC online user behavior monitoring method based on separation of micro and macro information according to claim 1 is characterized in that: include: Preprocess the input data; Based on the preprocessed data, a deep learning framework based on a dual-tower structure is trained to achieve online user behavior monitoring.

3. The MOOC online user behavior monitoring method based on separation of micro and macro information according to claim 2 is characterized in that: The input data includes the course introduction and the user's historical learning records.

4. The MOOC online user behavior monitoring method based on separation of micro and macro information according to claim 2 is characterized in that: Preprocess the input data, including data cleaning, normalization, and time series processing, as follows: 1) Data cleaning: Remove missing values, duplicate data and abnormal data from input data to ensure data integrity and consistency; 2) Normalization processing: Standardize or normalize the input data to ensure that the data dimensions of different features are consistent to improve the convergence and prediction effect of the model; 3) Time serialization: Arrange the input data in chronological order to form time series data so that the model can identify the time dependencies in the learning behavior.

5. The MOOC online user behavior monitoring method based on separation of micro and macro information according to claim 2 is characterized in that: The deep learning framework based on the dual-tower structure includes macro modules and micro modules. Users can choose joint mode or separation mode to train the deep learning framework based on the dual-tower structure. The joint mode is suitable for scenarios where two modules are optimized collaboratively, and the separation mode is suitable for scenarios where a single module is optimized.

6. The method for monitoring MOOC online user behavior based on separation of micro and macro information according to claim 5 is characterized in that: The macro module is used to process the basic information of the corresponding course and the descriptive statistical information of the interaction history between all learners and the corresponding course; for a user u i The interaction between the course c and the macro information is represented as G i (u i , c)∈R h , where h is the dimension used to represent macro information; in the macro module, the macro information is modeled by GBDT, and the specific process is as follows: (1) Input sample set (2) Initialization Start traversal calculation from t=1, t={1, 2, ..., T}; (3) Calculate the loss function L for each sample with respect to the current model The negative gradient (4) Take the negative gradient as the fitting object and build a decision tree. Assume that the model parameters are (5) The constructed decision tree divides the sample set into J subsets, each leaf node is a subset, denoted as R tj ,j=1,2,...,J (6) Estimate the predicted value γ of each leaf node tj (7) Model update in, is the sample data in the input data set, γ is the model prediction value, L(y i ,γ) is the loss of the current sample; are model parameters, which are randomly initialized according to Gaussian distribution; initialization When it is a tree with only one root node, the predicted value of the root node can minimize the overall loss of the loss function L on the entire sample set.

7. The method for monitoring online MOOC user behavior based on separation of micro and macro information according to claim 5, characterized in that: The micro module refers to a user u i The interaction details with course c are defined as L i (u i ,c); Since a course may contain multiple teaching videos, L i (u i , c) is further expressed as L i (u i , c)∈R n·m , c = v1, v2, ..., v n , v i represents the i-th video of course c, n is the user u i The number of videos that have been interacted with belonging to course c, m is the dimension of the feature space; denoted as f1f m The feature representation of user u i and video v j Specifically, given a course c, L i，j For a user u i With a video v j The interaction record between L i，j (u i , v j )∈L i (u i , c); In the micro-module, the CNN network is used to capture micro-information, including valuable time series patterns; the specific process is represented as the input data set sample matrix X = [x1, x2, ..., x n ],X∈R n×d , where x i ∈R d Represents a specific sample, which is the specific information of an online user interacting with a resource; the convolution operation is as follows: y=f(WX+b) Among them, W is the weight corresponding to the input signal X, b is the bias term, and f is the convolution transformation function. After a series of consecutive convolution operations, pooling operations and fully connected layers are applied to summarize all the extracted information and generate the final prediction.

8. The method for monitoring online MOOC user behavior based on separation of micro and macro information according to claim 5 is characterized in that: The intermediate results produced by the micro and macro modules are combined by weighted summation before the final prediction, expressed as follows: P(Y=1|x)=σ(W micro r micro +W macro r macro +b) Among them, r micro and r macro are the intermediate results from the micro module and the macro module respectively; W micro and W macro are two different weights; b is the bias value; σ represents the regression function, and the mathematical form is shown as follows: t is the input variable of the regression function.

9. A MOOC online user behavior monitoring system based on separation of micro and macro information, characterized in that: The method comprises a memory, a processor and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps as claimed in any one of claims 1 to 8 can be implemented.

10. A computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor, and when the processor executes the computer program instructions, the method steps according to any one of claims 1 to 8 can be implemented.

Images