CN114863341B - Online course learning supervision method and system - Google Patents
Online course learning supervision method and system Download PDFInfo
- Publication number
- CN114863341B CN114863341B CN202210535724.0A CN202210535724A CN114863341B CN 114863341 B CN114863341 B CN 114863341B CN 202210535724 A CN202210535724 A CN 202210535724A CN 114863341 B CN114863341 B CN 114863341B
- Authority
- CN
- China
- Prior art keywords
- course
- user
- returning
- feature
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000013145 classification model Methods 0.000 claims abstract description 24
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 19
- 230000002068 genetic effect Effects 0.000 claims abstract description 18
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 239000013598 vector Substances 0.000 claims description 54
- 230000006399 behavior Effects 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 14
- 238000012417 linear regression Methods 0.000 claims description 10
- 238000003860 storage Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000012549 training Methods 0.000 description 30
- 238000000605 extraction Methods 0.000 description 20
- 238000010586 diagram Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 8
- 230000000694 effects Effects 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 230000007246 mechanism Effects 0.000 description 6
- 238000007637 random forest analysis Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000005065 mining Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001351 cycling effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/20—Education
- G06Q50/205—Education administration or guidance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/817—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level by voting
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Educational Technology (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Bioinformatics & Computational Biology (AREA)
- Medical Informatics (AREA)
- Educational Administration (AREA)
- Biomedical Technology (AREA)
- Physiology (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Genetics & Genomics (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of artificial intelligence, and provides an online course learning supervision method and system, wherein the method comprises the following steps: collecting the prediction data of the course returning of the user to be supervised; preprocessing the course-returning prediction data and extracting features to obtain course-returning features, and inputting a weighted soft voting integrated classification model to obtain course-returning probability of a user to be supervised on a selected course; if the class returning probability exceeds the set value, sending reminding information to the user to be supervised; the weighted soft voting integrated classification model integrates a plurality of base classifiers, the weight of each base classifier is determined by a genetic algorithm, and the results of the base classifiers are weighted and summed to obtain the class-returning probability output by the model. The generalization capability of the model to different data is improved, and the accuracy of online course learning supervision is improved.
Description
Technical Field
The invention belongs to the technical field of artificial intelligence, and particularly relates to an online course learning supervision method and system.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Large-scale open online curriculum (Massive Open Online Courses, MOOC for short) is a new model of "internet + education". The number of MOOC users is continuously increased, but the high course-returning rate is always a problem which is difficult to solve in online education. The higher course returning rate of the online courses indicates that a large number of users give up learning without completing the courses in the learning process, and the learning effect of the users and the effective implementation of the online courses are seriously affected.
The existing online course learning supervision method mainly surrounds feature extraction and classification model development when online course withdrawal prediction is carried out, and the existing feature extraction method is divided into manual extraction and automatic extraction. The manual extraction refers to the digital statistical characteristics of the artificial construction data, and is used for evaluating the learning behavior state of the user, and has the defect of consuming a great deal of manpower and time; and automatically extracting and utilizing the existing models such as machine learning, deep learning, word vectors and the like to extract the class returning features. For example, convolutional Neural Network (CNN) is applied to automatically extract lesson-returning features, where CNN is composed of a feature extraction layer and a feature mapping layer, and can perform convolutional operation on original features to generate new features, but cannot capture long-distance dependency due to limitation of convolutional kernel; the Recurrent Neural Network (RNN) can capture time series in the class-withdrawal prediction data, extract the class-withdrawal time characteristics of the user and represent the time characteristics as characteristic vectors, but has the problem of gradient disappearance when learning long text. Word2vec Word vector model performs contextual semantic analysis on courses and video sequences based on Word dimensions, but ignores the Word order problem of the sequences. The Doc2vec Word vector model adds paragraph vectors on the basis of Word2vec, and makes up the deficiency in this aspect. However, doc2vec can only capture semantic relationships between courses and cannot express attribute differences between courses.
The online course-returning prediction classification model in the existing online course learning supervision method is divided into a machine learning model and a deep learning model. The machine learning model comprises logistic regression, a naive Bayesian classifier, a Support Vector Machine (SVM), a decision tree and the like, and has strong interpretability, but has poor classification effect when the class-returning feature dimension is higher. The deep learning model comprises CNN, RNN, LSTM models and the like, is suitable for processing high-dimensional features, but when the features have no structural association, the fitting effect of the model is poor, and the classification effect is influenced.
At present, the online course learning supervision method mainly has the following problems:
(1) The user course-returning data comprises user behavior data, course information data and video information data, and the existing feature extraction method is poor in effect when extracting different types of data features, so that feature extraction is insufficient.
(2) For extracting text information in courses, the existing word vector model can only capture semantic relations among courses, and cannot express attribute differences among courses.
(3) When a single classification model has no associated high-dimensional feature on a processing structure, the classification effect is poor, and the accuracy of course-leaving prediction is low.
Disclosure of Invention
In order to solve the technical problems in the background technology, the invention provides an online course learning supervision method and system, and provides a weighted soft voting integrated classification model aiming at high-dimension course returning characteristics, and the weighting coefficient of a base classifier is optimized through a genetic algorithm, so that the generalization capability of the model on different data is improved, and the accuracy of online course learning supervision is improved.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
a first aspect of the present invention provides an online course learning supervision method, comprising:
collecting the prediction data of the course returning of the user to be supervised;
preprocessing the course-returning prediction data and extracting features to obtain course-returning features, and inputting a weighted soft voting integrated classification model to obtain course-returning probability of a user to be supervised on a selected course;
If the class returning probability exceeds the set value, sending reminding information to the user to be supervised;
The weighted soft voting integrated classification model integrates a plurality of base classifiers, the weight of each base classifier is determined by a genetic algorithm, and the results of the base classifiers are weighted and summed to obtain the class-returning probability output by the model.
Further, the course returning feature comprises a user behavior statistical feature, a difference feature of each course selected by the user, a course similarity feature, a course attribute feature, a video attribute feature and a course difficulty feature.
Further, the user behavior statistical features comprise total watching time length of each course of the user, proportion of time length of each course watched by the user to the total time length and watching interval time length among different videos watched by the user.
Further, the method for acquiring the course attribute features comprises the following steps:
coding the course sequence of the user by adopting a word vector model to obtain a feature vector;
Performing attention operation on the feature vector and the feature matrix, and calculating to obtain an attention weight vector;
and multiplying the attention weight vector by the feature vector to obtain the course attribute feature of the user.
Further, the course difficulty characteristics are obtained by adopting a meta learning strategy model;
The meta learning strategy model consists of a basic learner and a meta learner two-layer architecture;
The basic learner of the meta learning strategy model counts the average course withdrawal rate of each course, builds splicing characteristics based on the average course withdrawal rate, learns the splicing characteristics, builds a nonlinear regression model, and predicts the preliminary course difficulty value;
the primary course difficulty value predicted by the primary learner is learned by the primary learner, a multiple linear regression model is established, and course difficulty characteristics are obtained by the multiple linear regression model.
Further, the genetic algorithm uses the mean square error of the sum of the predictive values and the true value of each base classifier as an fitness function, and uses the sum of the weights as a constraint condition.
A second aspect of the present invention provides an online course learning supervision system comprising:
A data acquisition module configured to: collecting the prediction data of the course returning of the user to be supervised;
A lecture prediction module configured to: preprocessing the course-returning prediction data and extracting features to obtain course-returning features, and inputting a weighted soft voting integrated classification model to obtain course-returning probability of a user to be supervised on a selected course;
a user reminder module configured to: if the class returning probability exceeds the set value, sending reminding information to the user to be supervised;
The weighted soft voting integrated classification model integrates a plurality of base classifiers, the weight of each base classifier is determined by a genetic algorithm, and the results of the base classifiers are weighted and summed to obtain the class-returning probability output by the model.
Further, the genetic algorithm uses the mean square error of the sum of the predictive values and the true value of each base classifier as an fitness function, and uses the sum of the weights as a constraint condition.
A third aspect of the present invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs steps in an online course learning supervision method as described above.
A fourth aspect of the invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in an online course learning supervision method as described above when the program is executed.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides an online course learning supervision method, which provides a multi-mode feature extraction method, converts different study objects to extract course-returning features from multiple dimensions, improves the course-returning feature dimension, and improves the overfitting resistance of a course-returning prediction model.
The invention provides an online course learning supervision method, which provides a course difficulty representation method and constructs a meta learning strategy model, wherein the model enhances the original course characteristics and improves the characteristic representation capability by calculating course difficulty values.
The invention provides an online course learning supervision method, which aims at high-dimension course-returning characteristics, provides a weighted soft voting integrated classification model (XLCR-SV), optimizes a base classifier weight coefficient through a genetic algorithm, improves generalization capability of the model to different data, and improves course-returning prediction accuracy.
The invention provides an online course learning supervision method, which aims at course sequence text data, provides an AD-sequence model, optimizes a course sequence text feature extraction method, and digs semantic features of courses and attribute features implicit in course sequences.
The invention provides an online course learning supervision method, which adopts a data mining technology and a deep learning model to replace a manual feature extraction method, realizes automatic course-returning feature extraction, more accurately learns course-returning behaviors of users, screens out users with higher course-returning rate, and finally reminds the users to watch courses through a mail sending mechanism, saves the users and reduces MOOC course-returning rate.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
FIG. 1 is a flow chart of an online course learning supervision method according to an embodiment of the invention;
FIG. 2 is a diagram showing the structure of an AD-sequence model according to a first embodiment of the present invention;
FIG. 3 is a block diagram of a course difficulty representation method according to a first embodiment of the present invention;
FIG. 4 is a diagram of a classifier training model in accordance with a first embodiment of the present invention;
fig. 5 is a system block diagram of a second embodiment of the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Example 1
The embodiment provides an online course learning supervision method, as shown in fig. 1, specifically comprising the following steps:
And step 1, collecting the course-returning prediction data of the user to be supervised, and collecting the course-returning prediction data of each user in the training set.
Each user in the training set has completed the course video, wherein the completion refers to that the user watches all videos in the course or carries out course returning after watching part of videos in the course; the user to be supervised has watched part of the video in the course and has not yet returned to the course.
The collected prediction data of course withdrawal is divided into: course sequence, video sequence, and user behavior data. Course sequence is a set of course IDs for user-selected courses, course IDs are character string types, exemplified by plurse-v 1: JXUST + JXUST2016001+2016_T2, wherein attribute characteristics such as school source, course number, course release time and course type are implied; the video sequence is a video ID set corresponding to each course, and the video ID format is a character string type; the user behavior data includes: video viewing sequence (i.e., video is arranged according to the order in which the user views the video), watching _count (number of times the user views each video), video_duration (total length of each video), local_ watching _time (actual viewing time of each video by the user), video_progress_time (playing time after the user considers the multiple speed for each video), video_start_time (earliest time point in which the user views the video), video_end_time (latest time point in which the user views the video), local_ starrt _time (actual viewing start time of each video by the user), local_end_time (actual viewing end time of each video by the user).
Step2, data preprocessing: and carrying out data cleaning and data reconstruction on the online course-returning prediction data.
To facilitate feature extraction, a normalized data format is required.
Data cleansing, i.e., standardized data formats: the line feed in the dataset is replaced with comma delimiters and brackets are normalized and the missing values are filled with 0.
The data reconstruction is to encode the course IDs and video IDs of all users LabelEncoder, convert the character string type data into a numeric type, and then establish a user-course-video chain relationship according to the relationship between the user and the course and the relationship between the course and the video.
Step 3: feature extraction: for a user to be supervised, through feature extraction, the behavior of the user watching the courses is sorted into multi-dimensional numerical features (course returning features) G user, wherein G user＝{g1,g2...,gm, m is the dimension of the numerical features, G i represents the course returning features of the ith dimension, and G user comprises user behavior statistics features, difference features of each course selected by the user, course similarity features, course attribute features, video attribute features and difficulty features of each course selected by the user. For the user in the training set, besides the lecture-taking feature G user, the lecture-taking tag y user,yuser =0 (no lecture-taking) or 1 (lecture-taking) of the user needs to be obtained, so as to obtain the lecture-taking feature and the lecture-taking tag of the user in the training set. The method comprises the following specific steps: firstly, a multi-mode feature extraction method is provided aiming at user behavior information, and class returning features are respectively constructed for three different objects of a user, a video and the relation between the user and a course; secondly, aiming at courses and video sequences, constructing a course sequence representation model AD-sequence, establishing a word vector model, extracting course text sequence characteristics, introducing an attention mechanism, and mining hidden attribute characteristics of the course sequence; finally, a course difficulty representation method is put forward, a meta learning strategy model is constructed, and feature enhancement is carried out on original features (the features refer to combinations of user behavior features, differential features, course similarity features and course and video attribute features).
Step 301: the multimode feature extraction method constructs the course returning feature: constructing user behavior statistical features aiming at user objects; constructing a video index sequence aiming at a video object, and calculating the difference characteristic of the video index sequence by taking difference between two adjacent elements of the video index sequence; aiming at the relation between the user and the courses, calculating similarity values of the user and all the courses through a cosine similarity function, wherein the similarity values are used for evaluating similarity between the courses to obtain course similarity characteristics. The multi-mode feature extraction method respectively constructs lesson-returning features from different study objects, and counts the maximum value, the minimum value, the average value, the variance and the like of the features. Specifically, the method comprises the following 3 substeps:
step 30101: constructing user behavior statistical features according to user behavior data, and constructing user behavior features (user behavior statistical features) of watching videos of users such as total watching time length of each course, proportion of time length of watching each course by users to total time length, watching interval time length among different videos watched by users and the like by original features such as difference, comparison, accumulation summation and the like;
Step 30102: the difference features (including first order difference features and second order difference features) represent the video's position features in the lesson as well as the video interval density features. First, an index position X (k), X (k) = (k 1,k2,...,kv), of all videos of a certain course k is calculated, wherein k i represents an index position of a user for an ith video in a video viewing sequence of the certain course k in a video sequence of a corresponding course, i=1, 2, …, v, v represents the number of videos of the user in the certain course, for example, in a first course video sequence, 1-1 represents a 1 st video in the first course, 1-2 represents a2 nd video in the first course, and 1-2 is a next video position of 1-1. Then, a first-order differential feature Y (k) of X (k) is calculated, Y (k) =x (k+1) -X (k) = (0, k 2-k1,k3-k2,...,kv-kv-1), and a second-order differential feature Z (k) =y (k+1) -Y (k) =x (k+2) -2X (k+1) +x (k) is calculated. X (k), Y (k) and Z (k) reflect the user's preference for the lesson, and the first order difference feature Y (k) and the second order difference feature Z (k) describe the continuity between videos (e.g., the user directly jumps to watch the 5 th video after watching the 1 st video of a lesson, the continuity is poor).
Step 30103: course similarity features refer to the similarity between courses under the same user. The course similarity feature needs to obtain vectorization representation of the user and the course selected by the user, a cosine similarity calculation method is used for calculating similarity values between the user and the course selected for evaluating relevance between courses, the similarity values of all courses are calculated to be similarity through a cosine similarity formula, a threshold value is set to be 0.05, when the similarity of the similarity values between courses is within 0.05, the two courses are judged to be similar, and the user has similar course returning behaviors for the courses, wherein the specific process is as follows:
Firstly, inputting user ID and corresponding course sequence into a Doc2vec word vector model for training, wherein the vectorized user ID and course sequence are r-dimensional vectors, and the vectorized representation D= (D 1,d2,…,dr) of a certain user and the vectorized representation of a course selected by the user Wherein D refers to a vector in which a certain user ID is represented as r-dimension; c i represents the vectorized representation (1.ltoreq.i.ltoreq.Q) of the i-th course selected by the user, Q being the number of courses selected by the user.
Secondly, calculating a relation value of a user and each course by using a cosine similarity algorithm to obtain the relation value of each course, if the relation values are similar, the courses have similarity, and setting a current course as theta (theta is more than or equal to 1 and less than or equal to Q) for formulation expression, wherein the formula is as follows:
and finally, adding the calculated similarity value into the original feature, and expanding the feature dimension.
Step 302: the course sequence represents the model AD-sequence: setting a curriculum sequence S= { S 1,s2,…,slen } with the length of len, wherein school sources, curriculum numbers, curriculum release time and curriculum type attribute features are implied, firstly extracting semantic features which are semantic features of curriculum IDs through a word vector model, and respectively representing the curriculum IDs as semantic features of 16 dimensions after the original curriculum sequence is trained and encoded through the word vector model; and then introducing an attention mechanism to extract weight vectors, and further mining course attribute characteristics.
Course sequence representation model AD-sequence is an important model part for extracting course sequence characteristics, the model structure is shown in figure 2, the model consists of a Doc2vec word vector model and an attention mechanism, doc2vec is a three-layer neural network, a user ID is taken as a paragraph vector, a course sequence or a video sequence is expressed as a plurality of words w i, and a maximization function isWhere T refers to the number of words, I refers to the I words (context) before and after w t, pro (w t-I,...,wt+I; w) is the predictive probability, and the output of the model is finally expressed as: y=b+uh (w t-k,...,wt+k; w). Where U is a parameter of the b function, W is a word vector matrix, and h represents a method of fusing context words together. Training to encode the course and video sequence corresponding to the sequence user into vector features. And then adding an attention mechanism to further extract the weight vector. The method comprises the steps of providing K users, and encoding a course sequence of the user i (i is more than or equal to 1 and less than or equal to K, and K represents the total number of users contained in a user to be supervised or training set) into a 16-dimension vector to be represented by adopting a word vector model: alpha i＝[p1,p2,…,p16 ], the combined video sequence of the video sequences of all courses selected by the user is encoded as beta i＝[q1,q2,…,q32 ]. The lesson sequence of all users is denoted α= [ α 1,α2,…,αK ], and the video sequence of all users is denoted β= [ β 1,β2,…,βK ]. In order to extract attribute features in courses and video sequences, an attention mechanism is added to highlight important positions in the sequences, weight vectors of vector features are extracted, and course attribute features and video attribute features are mined. Coding the course sequence of the user by adopting a word vector model to obtain a feature vector; performing attention operation on the feature vector and the feature matrix, and calculating to obtain an attention weight vector; and multiplying the attention weight vector by the feature vector to obtain the course attribute feature of the user. The method comprises the following specific steps:
First, a course sequence is taken as an example, and a video sequence is the same. Alpha i is denoted feature i, representing the feature vector of the current match, feature i＝αi＝[p1,p2,…,p16 ] (1.ltoreq.i.ltoreq.K).
Secondly, the current feature vector feature i performs attention operation with the feature matrix alpha to calculate an attention weight vector weight i, and the calculation method is as follows:
weighti＝∑ai·α
Finally, feature vector feature i is multiplied by weight vector weight i to obtain a weighted feature vector
Weighted feature vectorCompared with the original feature vector feature i, the method has the advantages that the method has the representing capability of courses and video sequences, can capture the relation between course or video attributes, and has better representing effect.
Step 303: providing a course difficulty representation method, and constructing a meta learning strategy model: the model consists of a base learner and a meta learner which are connected in stacking mode. The first layer is a base learner, firstly, the average course withdrawal rate of each course is counted based on the course withdrawal condition of a user, splicing features are built based on the average course withdrawal rate (the average course withdrawal rate is spliced with data features (combination features of user behavior features, difference features, course similarity features, courses and video attribute features) after feature extraction), the base learner learns the spliced features, a nonlinear regression model is built, and preliminary course difficulty values are predicted; the second layer is a primary course difficulty value predicted by the learning-based learner, a multiple linear regression model is established, a final course difficulty value is calculated by the multiple linear regression model, and course difficulty characteristics are obtained, wherein the course difficulty value represents the difficulty level of a certain course, the interval is [0,1], and the larger the value is, the higher the course difficulty is represented.
The course difficulty representing method is to connect the two-layer architecture of the base learner and the meta learner in stacking mode to represent course difficulty. Factors affecting the course withdrawal of the user include not only personal factors of the user but also factors of courses themselves, so that based on data in a training set, an average course withdrawal rate of each course is calculated in advance, the average course withdrawal probability is marked as an initial difficulty feature, the course difficulty feature is calculated through a course difficulty representation framework based on stacking-element learning strategies, and a calculation process is shown in fig. 3.
First, a course average withdrawal P is calculated. Based on the data in the training set, counting the total number of selected courses and the total number of returned courses in each course, wherein the ratio of the total number of returned courses to the total number of selected courses is the average course returning rate of the courses.
And secondly, splicing the average course-returning rate characteristic and the original characteristic (the user behavior characteristic, the difference characteristic, the course similarity characteristic and the combined characteristic of the course and the video attribute characteristic) into a training characteristic, inputting the training characteristic into a nonlinear regression model of the first layer of the integrated framework, and training to generate a preliminary course difficulty characteristic. And (5) meta learning of a prediction model. L 1、L2、L3 is three nonlinear regression models, and three base learners respectively output a lesson-withdrawal probability predicted value P 1、P2、P3. M is a linear regression model, which serves as a meta learner.
And finally, predicting the course difficulty value. The meta learner of the second layer learns P 1、P2、P3 to predict the course difficulty value diff, i.e., the course difficulty feature, in a linear regression manner.
Step 4: and training a weighted soft voting integrated classification model by adopting the class returning characteristics and the class returning labels of the users in the training set. The weighted soft voting integrated classification model integrates a plurality of base classifiers, the weight of each base classifier is determined by a genetic algorithm, and the results of the base classifiers are weighted and summed to obtain the class-returning probability output by the model.
The weighted soft voting integrated classification model XLCR-SV integrates the limit gradient lifting tree (XGBoost), lightweight gradient lifting Machine (LIGHT GRADIENT Boosting Machine, lightGBM), catBoost (Gradient Boosting + Categorical Features) and random forest model 4 base classifiers in a weighted soft voting manner. The weights of the base classifiers are determined by a genetic algorithm, and the final classification result is generated by soft voting after weighting each base classifier.
XLCR-SV classifier: as shown in fig. 4, after the model inputs the features, the steps of training, forming a base classifier, determining the weights of the base classifier, weighting integration and generating a prediction result are respectively performed, and the specific processes are as follows:
Firstly, dividing the features into 4 parts according to the characteristics of the 4 base classification models, respectively serving as training features of the base classification models, and training the training features after the training of the base classification models is equal, so as to obtain the base classifier. And inputting the constructed characteristics into each base model for training. The XGBoost, lightGBM and CatBoost models integrate the tree model in boosting form with the expression: (wherein F k(χi represents each weak learner, N is the number of weak learners, and F is the basic tree model). XGBoost is directed to objective function/> (L is a loss function representing the predicted value/>Error with the true value y i, Ω is a function for regularization to prevent overfitting, and f k(χi) represents each weak learner), has excellent capability of resisting overfitting, so that overfitting is not easy to occur when training the lesson returning feature; lightGBM the classification expression is Fn(x)＝λ0f0(x)+λ1f1(x)+…+λnfn(x)(fi(x), n is the number of weak classifiers, lambda i is the coefficient of the weak classifiers, i is more than or equal to 1 and less than or equal to n, and the characteristics are pre-ordered in advance before training the class-returning characteristics, so that the classification efficiency is improved; catBoost can convert the category type characteristics into digital characteristics, and is suitable for processing the category type characteristics in the class returning characteristics; the random forest model randomly extracts data and randomly selects features, so that the random forest is not easy to fall into overfitting when the high-dimensional class-returning features are processed. The original feature C is divided into the following 4 parts of fe 1、fe2、fe3、fe4, and the 4 base models are respectively trained: fe 1 is a class feature in the class feature and is used for training CatBoost model; fe 2 is a user behavior feature in the course-returning feature, and is used for training XGBoost model; fe 3 is a differential feature in the course-returning feature and is used for training a LightGBM model; and fe 4 is the rest of the lesson-returning features and is used for training a random forest model. Each base model is trained with different types of features, and finally trained into 4 base classifiers. The base classifier training different features ensures the diversity of the base classifier and improves the generalization capability of XLCR-SV model.
Secondly, setting the range of weight parameters to be 0-1, the iteration times of the genetic algorithm to be 1000, the number of initial populations to be 200, initializing 4 base classifier weights omega i (1 is less than or equal to i is less than or equal to 4), taking each base classifier weight as an independent variable of an fitness function in the genetic algorithm, then randomly generating P groups of initial values, taking the mean square error of a true value and a predicted value (the weighted sum of the predicted values of each base classifier) as the fitness function, taking the sum of the weights to be equal to 1 as a constraint condition, selecting a certain proportion of excellent individuals from the populations, intersecting and mutating the selected excellent individuals, thereby generating new individuals, and cycling the process until the condition is met, and selecting the optimal value from the calendar population as a final result. Finally, the base classifier weights (omega 1,ω2,ω3,ω4) are continuously optimized through the selection, crossing and mutation processes, the weights of all the base classifiers are determined, and a XLCR-SV model is constructed. The fitness calculation formula is as follows:
ω1+ω2+ω3+ω4＝1
Wherein omega 1、ω2、ω3、ω4 is the weight of XGBoost, lightGBM, catBoost and random forest based classifier respectively, And the predicted value of each base classifier and y T are the true values of the training set.
And (3) optimizing the weights through a genetic algorithm, then giving weight to each base classifier, and weighting soft voting to form XLCR-SV classifiers.
And 5, preprocessing and extracting features of the class-returning prediction data of the user to be supervised, inputting a weighted soft voting integrated classification model, and outputting a class-returning prediction result:
wherein, o=4 (the number of the base classifiers), x i is the prediction result of the base classifier, and when result is more than or equal to 0.5, the user is judged to be in class; when result < 0.5. And judging that the user does not leave the class.
And analyzing each course data of the user aiming at each user, wherein the course data comprises user behavior data, courses and video sequence data of the courses, and predicting the users with the course returning probability of more than 50% by training the course returning characteristics after extracting the characteristics. And sequencing the obtained high-risk users (the course returning rate is more than 50%) from high to low according to the magnitude of the course returning rate, counting courses with higher course returning rate, displaying the course returning probability of the high-risk users in a histogram mode, and displaying the courses with higher course returning rate.
The user with higher course returning rate is timely reminded through the E-mail, the user is prompted that courses are not completed, and the user is encouraged to continue learning.
According to the method, characteristics such as user behavior characteristics, differential characteristics, course similarity and the like are built through a multi-mode characteristic extraction method, a course difficulty representation method is put forward, course characteristics are enhanced, and the representation capability of the characteristics is improved; constructing a course sequence representation model AD-sequence, extracting course text sequence characteristics, and mining hidden attribute characteristics of the sequence course sequence; and constructing XLCR-SV model in a weighted soft voting mode, so as to improve the accuracy of course-leaving prediction of the model.
Example two
The embodiment provides an online course learning supervision system, as shown in fig. 5, which specifically includes the following modules:
A data acquisition module configured to: collecting the prediction data of the course returning of the user to be supervised;
A lecture prediction module configured to: preprocessing the course-returning prediction data and extracting features to obtain course-returning features, and inputting a weighted soft voting integrated classification model to obtain course-returning probability of a user to be supervised on a selected course;
a results visualization module configured to: and sequencing the obtained high-risk users (the course returning rate is more than 50%) from high to low according to the magnitude of the course returning rate, counting courses with higher course returning rate, displaying the course returning probability of the high-risk users in a histogram mode, and displaying the courses with higher course returning rate.
A user reminder module configured to: if the class returning probability exceeds the set value, sending reminding information to the user to be supervised;
The weighted soft voting integrated classification model integrates a plurality of base classifiers, the weight of each base classifier is determined by a genetic algorithm, and the results of the base classifiers are weighted and summed to obtain the class-returning probability output by the model.
The genetic algorithm takes the mean square error of the sum of the predictive values and the true value of each base classifier as an adaptability function and takes the sum of the weights as a constraint condition.
The course returning feature comprises a user behavior statistical feature, a difference feature of each course selected by a user, a course similarity feature, a course attribute feature, a video attribute feature and a course difficulty feature.
It should be noted that, each module in the embodiment corresponds to each step in the first embodiment one to one, and the implementation process is the same, which is not described here.
Example III
The present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of an online course learning supervision method as described in the above embodiment one.
Example IV
The present embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the steps in an online course learning supervision method according to the above embodiment.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disc, a Read-Only Memory (ROM), a Random access Memory (Random AccessMemory, RAM), or the like.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (7)
1. An on-line course learning supervision method, comprising:
collecting the prediction data of the course returning of the user to be supervised;
preprocessing the course-returning prediction data and extracting features to obtain course-returning features, and inputting a weighted soft voting integrated classification model to obtain course-returning probability of a user to be supervised on a selected course; the course returning feature comprises a user behavior statistical feature, a difference feature of each course selected by a user, a course similarity feature, a course attribute feature, a video attribute feature and a course difficulty feature;
The course attribute feature acquisition method comprises the following steps:
Coding the course sequence of the user by adopting a word vector model to obtain a feature vector; performing attention operation on the feature vector and the feature matrix, and calculating to obtain an attention weight vector; multiplying the attention weight vector by the feature vector to obtain course attribute features of the user;
the course difficulty characteristics are obtained by adopting a meta learning strategy model;
The meta learning strategy model consists of a basic learner and a meta learner two-layer architecture;
The basic learner of the meta learning strategy model counts the average course withdrawal rate of each course, builds splicing characteristics based on the average course withdrawal rate, learns the splicing characteristics, builds a nonlinear regression model, and predicts the preliminary course difficulty value;
The primary course difficulty value predicted by the primary learner is learned by the primary learner, a multiple linear regression model is established, and course difficulty characteristics are obtained by the multiple linear regression model;
If the class returning probability exceeds the set value, sending reminding information to the user to be supervised;
The weighted soft voting integrated classification model integrates a plurality of base classifiers, the weight of each base classifier is determined by a genetic algorithm, and the results of the base classifiers are weighted and summed to obtain the class-returning probability output by the model.
2. The online course learning supervision method of claim 1, wherein the user behavior statistics include a total user viewing time for each course, a ratio of the user viewing time for each course to the total time, and a viewing interval time between different videos viewed by the user.
3. The online course learning supervision method of claim 1, wherein the genetic algorithm uses a mean square error between a weighted sum of predicted values and a true value of each base classifier as a fitness function, and uses a constraint condition that the weighted sum is equal to 1.
4. An online course learning supervision system, comprising:
A data acquisition module configured to: collecting the prediction data of the course returning of the user to be supervised;
A lecture prediction module configured to: preprocessing the course-returning prediction data and extracting features to obtain course-returning features, and inputting a weighted soft voting integrated classification model to obtain course-returning probability of a user to be supervised on a selected course; the course returning feature comprises a user behavior statistical feature, a difference feature of each course selected by a user, a course similarity feature, a course attribute feature, a video attribute feature and a course difficulty feature; the course attribute feature acquisition method comprises the following steps: coding the course sequence of the user by adopting a word vector model to obtain a feature vector; performing attention operation on the feature vector and the feature matrix, and calculating to obtain an attention weight vector; multiplying the attention weight vector by the feature vector to obtain course attribute features of the user; the course difficulty characteristics are obtained by adopting a meta learning strategy model; the meta learning strategy model consists of a basic learner and a meta learner two-layer architecture; the basic learner of the meta learning strategy model counts the average course withdrawal rate of each course, builds splicing characteristics based on the average course withdrawal rate, learns the splicing characteristics, builds a nonlinear regression model, and predicts the preliminary course difficulty value; the primary course difficulty value predicted by the primary learner is learned by the primary learner, a multiple linear regression model is established, and course difficulty characteristics are obtained by the multiple linear regression model;
a user reminder module configured to: if the class returning probability exceeds the set value, sending reminding information to the user to be supervised;
The weighted soft voting integrated classification model integrates a plurality of base classifiers, the weight of each base classifier is determined by a genetic algorithm, and the results of the base classifiers are weighted and summed to obtain the class-returning probability output by the model.
5. The online course learning supervision system of claim 4, wherein the genetic algorithm uses a weighted sum of the predicted values of the basis classifiers and a mean square error of the true values as fitness functions, and uses a constraint that the sum of the weights is equal to 1.
6. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the steps of an online course learning supervision method as claimed in any one of claims 1 to 3.
7. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of an online course learning supervision method as claimed in any one of claims 1 to 3 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210535724.0A CN114863341B (en) | 2022-05-17 | 2022-05-17 | Online course learning supervision method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210535724.0A CN114863341B (en) | 2022-05-17 | 2022-05-17 | Online course learning supervision method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114863341A CN114863341A (en) | 2022-08-05 |
| CN114863341B true CN114863341B (en) | 2024-05-31 |
Family
ID=82637219
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210535724.0A Active CN114863341B (en) | 2022-05-17 | 2022-05-17 | Online course learning supervision method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114863341B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117493566B (en) * | 2023-11-02 | 2024-05-24 | 浙江尚云物联科技有限公司 | Data processing system and method for smart campus |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110209816A (en) * | 2019-05-24 | 2019-09-06 | 中国科学院自动化研究所 | Event recognition and classification method, system, device based on confrontation learning by imitation |
| CN110414554A (en) * | 2019-06-18 | 2019-11-05 | 浙江大学 | One kind being based on the improved Stacking integrated study fish identification method of multi-model |
| CN110580341A (en) * | 2019-09-19 | 2019-12-17 | 山东科技大学 | False comment detection method and system based on semi-supervised learning model |
| CN112926397A (en) * | 2021-01-28 | 2021-06-08 | 中国石油大学(华东) | SAR image sea ice type classification method based on two-round voting strategy integrated learning |
| CN113469730A (en) * | 2021-06-08 | 2021-10-01 | 北京化工大学 | Customer repurchase prediction method and device based on RF-LightGBM fusion model under non-contract scene |
-
2022
- 2022-05-17 CN CN202210535724.0A patent/CN114863341B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110209816A (en) * | 2019-05-24 | 2019-09-06 | 中国科学院自动化研究所 | Event recognition and classification method, system, device based on confrontation learning by imitation |
| CN110414554A (en) * | 2019-06-18 | 2019-11-05 | 浙江大学 | One kind being based on the improved Stacking integrated study fish identification method of multi-model |
| CN110580341A (en) * | 2019-09-19 | 2019-12-17 | 山东科技大学 | False comment detection method and system based on semi-supervised learning model |
| CN112926397A (en) * | 2021-01-28 | 2021-06-08 | 中国石油大学(华东) | SAR image sea ice type classification method based on two-round voting strategy integrated learning |
| CN113469730A (en) * | 2021-06-08 | 2021-10-01 | 北京化工大学 | Customer repurchase prediction method and device based on RF-LightGBM fusion model under non-contract scene |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114863341A (en) | 2022-08-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111460249B (en) | Personalized learning resource recommendation method based on learner preference modeling | |
| CN112199608B (en) | Social media rumor detection method based on network information propagation graph modeling | |
| CN112687374B (en) | Psychological crisis early warning method based on text and image information joint calculation | |
| CN111400494B (en) | Emotion analysis method based on GCN-Attention | |
| CN110704510A (en) | User portrait combined question recommendation method and system | |
| CN111597340A (en) | Text classification method and device and readable storage medium | |
| CN113254675B (en) | Knowledge graph construction method based on self-adaptive few-sample relation extraction | |
| CN113344053A (en) | Knowledge tracking method based on examination question heteromorphic image representation and learner embedding | |
| CN115510814B (en) | Chapter-level complex problem generation method based on dual planning | |
| CN116484024A (en) | Multi-level knowledge base construction method based on knowledge graph | |
| CN114385801A (en) | Knowledge tracking method and system based on hierarchical refinement LSTM network | |
| CN116975776A (en) | Multi-mode data fusion method and device based on tensor and mutual information | |
| CN116402352A (en) | Enterprise risk prediction method and device, electronic equipment and medium | |
| CN114863341B (en) | Online course learning supervision method and system | |
| CN113591971A (en) | User individual behavior prediction method based on DPI time series word embedded vector | |
| CN115687610A (en) | Text intention classification model training method, recognition device, electronic equipment and storage medium | |
| CN112416358A (en) | Intelligent contract code defect detection method based on structured word embedded network | |
| CN116579347A (en) | Comment text emotion analysis method, system, equipment and medium based on dynamic semantic feature fusion | |
| CN116245110A (en) | Multi-dimensional information fusion user standing detection method based on graph attention network | |
| CN113179276B (en) | Intelligent intrusion detection method and system based on explicit and implicit feature learning | |
| CN116644759B (en) | Method and system for extracting aspect category and semantic polarity in sentence | |
| CN113378581A (en) | Knowledge tracking method and system based on multivariate concept attention model | |
| CN117473041A (en) | Programming knowledge tracking method based on cognitive strategy | |
| CN117151222A (en) | Domain knowledge guided emergency case entity attribute and relation extraction method thereof, electronic equipment and storage medium | |
| CN115795015A (en) | Comprehensive knowledge tracking method for enhancing test question difficulty |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |