CN116720098A - Abnormal behavior sensitive student behavior time sequence modeling and academic early warning method - Google Patents
Abnormal behavior sensitive student behavior time sequence modeling and academic early warning method Download PDFInfo
- Publication number
- CN116720098A CN116720098A CN202310300117.0A CN202310300117A CN116720098A CN 116720098 A CN116720098 A CN 116720098A CN 202310300117 A CN202310300117 A CN 202310300117A CN 116720098 A CN116720098 A CN 116720098A
- Authority
- CN
- China
- Prior art keywords
- student
- behavior
- term
- representation
- short
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 25
- 206010000117 Abnormal behaviour Diseases 0.000 title claims abstract description 24
- 230000006399 behavior Effects 0.000 claims abstract description 199
- 230000007246 mechanism Effects 0.000 claims abstract description 33
- 230000007774 longterm Effects 0.000 claims abstract description 30
- 230000006870 function Effects 0.000 claims description 13
- 230000004927 fusion Effects 0.000 claims description 13
- 230000003542 behavioural effect Effects 0.000 claims description 12
- 230000001133 acceleration Effects 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 230000004931 aggregating effect Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 3
- 230000009471 action Effects 0.000 claims description 2
- 230000004913 activation Effects 0.000 claims description 2
- 235000021152 breakfast Nutrition 0.000 claims description 2
- 238000006116 polymerization reaction Methods 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims 1
- 230000035945 sensitivity Effects 0.000 claims 1
- 238000007418 data mining Methods 0.000 abstract description 4
- 230000008447 perception Effects 0.000 abstract description 2
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000012466 permeate Substances 0.000 description 1
- 230000004622 sleep time Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Educational Technology (AREA)
- Educational Administration (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Health & Medical Sciences (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
A student behavior time sequence modeling and academic early warning method sensitive to abnormal behaviors belongs to the field of education data mining in big data mining. According to the method, student campus behavior heterogeneous information networks are constructed by utilizing student campus behavior data collected by mobile equipment, meta-path examples capable of revealing student campus behaviors are extracted to be encoded, each meta-path example representation is aggregated through a attention mechanism, student campus behavior pattern representation is learned, and the discriminability of student behavior pattern embedding based on the mobile equipment data is improved. Meanwhile, in order to improve timeliness of early perception of abnormal behaviors, a attention mechanism-based abnormal behavior sensitive gating module is provided, long-term behaviors and short-term behaviors of students are effectively fused, student campus time sequence behavior representation with semantic information is established, and accuracy of student academic level prediction is improved.
Description
Technical Field
The invention belongs to the field of education data mining in big data mining, and particularly relates to a student behavior time sequence modeling and academic early warning method sensitive to abnormal behaviors.
Background
Along with the increase of the learning difficulty and the development of the learning scale of universities, the learning pressure of students at the universities is gradually increased, and events such as open class, hanging department, and delay are frequently occurred. For student management workers, student academic problems are discovered and intervened in advance, help is provided for the students, the situations of delay, study return and the like can be avoided in time, and healthy growth of the students is ensured. At present, most schools rely on regular examination for the mastering of student academic situations, and the examination has a limitation on operation time. For example, a survey of a student's learning situation is typically conducted for several fixed times, such as mid-term or end-of-school. Student management workers can only provide assistance to students who have hung or are close to hanging by virtue of examination results, and early warning cannot be carried out on students with early risks in the academic industry.
From the perspective of educational psychology, student performance is closely related to student performance. Thus, student status changes can be monitored in time by analyzing student behaviors, and related researches have been conducted in this field by researchers. The method for monitoring the student academic status by analyzing the student behaviors makes up for a plurality of defects of the conventional examination method. Particularly, with the widespread use of mobile devices, a great deal of individual activity information, such as sleep time, exercise information, etc., is collected and accumulated by the mobile devices, and how to analyze behavior characteristics by using mobile sensing data, predict student academic achievements, provide process management basis for student management workers, and start to draw attention of more researchers.
In recent years, with the rapid development of leading edge computer science and technology such as artificial intelligence, machine learning and the like, the artificial intelligence technology provides a more accurate and efficient solution for mining and analyzing big data. Under the environment, depending on national policy support and driving, the artificial intelligence gradually permeates into various traditional industries such as education, finance, medical treatment and the like, and forms an emerging cross application field. By means of artificial intelligence technology, mass student data accumulated in universities and student behavior data collected by mobile equipment can be utilized to analyze and sense campus student behaviors, influence of student status in different time periods on behaviors is explored, learning enthusiasm of students is monitored in time, and theoretical support and practice basis are provided for campus student academic management.
Disclosure of Invention
The invention aims to provide a student behavior time sequence modeling and academic early warning method sensitive to abnormal behaviors, which utilizes student on-school behavior data collected by mobile equipment to construct a student campus behavior heterogeneous information network, extracts meta-path examples capable of revealing student campus behaviors to encode, aggregates each meta-path example representation through an attention mechanism, learns student campus behavior mode representation, and improves the discriminability of student behavior mode embedding based on mobile equipment data. Meanwhile, in order to improve timeliness of early perception of abnormal behaviors, a attention mechanism-based abnormal behavior sensitive gating module is provided, long-term behaviors and short-term behaviors of students are effectively fused, student campus time sequence behavior representation with semantic information is established, and accuracy of student academic level prediction is improved.
In order to achieve the above purpose, the present invention adopts the following technical scheme: a student behavior time sequence modeling and academic early warning method sensitive to abnormal behaviors. Firstly, collecting student behavior data by using mobile equipment, constructing a student campus short-term behavior heterogeneous information network by taking single-day behaviors as short-term behaviors, designing a meta-path capable of revealing student campus behaviors, guiding the extraction of examples on the student campus short-term behavior heterogeneous information network, and learning student campus short-term behavior mode representation with node attribute semantics by means of a meta-path example encoder and an attention mechanism; secondly, taking the whole learning period behavior of the student as a long-term behavior, and weighting important behavior patterns in the behavior time sequence sensitive to learning score by using a attention mechanism, thereby overcoming the defect that the current research does not pay attention to abnormal behaviors in the historical behavior data; finally, the gate control mechanism is utilized to perform feature fusion on short-term behavior and long-term behavior information of students, historical behavior and current behavior are cooperatively utilized, more discriminative student time sequence behavior characterization is learned, the loss of behavior features related to the academic industry in the behavior modeling process is effectively reduced, and the academic performance prediction performance of the students is improved.
A student behavior time sequence modeling and academic early warning method sensitive to abnormal behaviors comprises the following steps:
step 1, preprocessing data of student campus behavior data collected by mobile equipment, and inputting the data into a model.
And 2, constructing a student campus short-term behavior heterogeneous information network based on the single-day campus behavior data.
And 2.1, representing the attribute of each node in the heterogeneous information network by using word embedding algorithm representation as an initial representation of the node.
And 2.2, constructing a student campus short-term behavior heterogeneous information network.
And 3, performing student campus short-term behavior pattern representation learning based on the meta-path.
Step 3.1, designing a meta-path capable of revealing the track behavior of the student campus.
And 3.2, extracting an instance in the student campus short-term behavior heterogeneous information network based on the meta-path, and encoding the behavior instance by using a meta-path instance encoder.
And 3.3, aggregating the path instance representations by using an attention mechanism to obtain a student campus short-term behavior pattern representation.
And 4, constructing a long-term behavior modeling module based on an attention mechanism by taking the academic behavior as the long-term behavior of the student. The attention mechanism is utilized to extract important behavior patterns most relevant to student academia in time sequence behaviors so as to pay attention to abnormal behaviors in historical behavior data.
And 5, constructing a fusion module based on a gating mechanism. And carrying out feature fusion on short-term behavior and long-term behavior information of the students by using a gating mechanism, and learning student time sequence behavior representation which fully senses abnormal behaviors.
And 6, inputting the student time sequence behavior representation to the full connection layer to realize the prediction of student academic achievements.
Compared with the prior art, the invention has the following obvious advantages:
compared with other methods in the field proposed before, the method can accurately, differentially and discriminatively model the time sequence behaviors of students in the campus based on the mobile equipment data, design a student campus single-day short-term behavior heterogeneous information network to fully model semantic information in student behaviors, sense abnormal behaviors of the students in the early stage of learning abnormal by designing a learning period long-term behavior modeling module based on an attention mechanism, and cooperatively utilize the historical behaviors and the current behaviors by a fusion module based on a gating mechanism to enhance the discriminability of the student campus time sequence behaviors. The invention can assist university campus management workers to realize early discovery of abnormal students in the academic industry, discover and intervene the academic problems of the students in advance, provide assistance for the students, and avoid the occurrence of hanging, waiting for the students, returning to the students and the like in time.
Drawings
FIG. 1 is a diagram of the overall model architecture of the present invention;
FIG. 2 is a flow chart of the method of the present invention;
FIG. 3 is a schematic diagram of a long-term behavior modeling module;
fig. 4 is a schematic diagram of a fusion module.
Detailed Description
The invention will be described in further detail below with reference to specific embodiments and with reference to the accompanying drawings.
The general structure diagram of the invention is shown in fig. 1, and the flow of the method is shown in fig. 2. The method specifically comprises the following steps:
step 1, data preprocessing is carried out on student acceleration sensor data, sound sensor data and WI-FI data collected by mobile equipment, student samples with more missing values are removed, seven time semantics of early morning, breakfast, morning, lunch, afternoon, supper and late night are added to the time when student behaviors occur, places where the student behaviors occur are replaced by specific place names from longitude and latitude, and specific place semantics of canteens, sports grounds, teaching buildings and the like are added. And dividing student behavior data by utilizing a sliding window, wherein the size of the window is one day, taking the student behavior data divided by the day as the input of a model, and setting the single-day behavior as the short-term behavior of the student.
And 2, constructing a student campus short-term behavior heterogeneous information network based on the preprocessed student single-day campus behavior data.
And 2.1, representing the attribute of each node in the heterogeneous information network by using word embedding algorithm representation as an initial representation of the node.
The student campus short-term behavior heterogeneous information network comprises four types of nodes: student nodes, acceleration sensor data nodes, sound sensor data nodes, WI-FI data nodes. The various nodes have corresponding basic attributes, such as the attribute of the student including the sex, the specialty, the grade and the like of the student, and the WI-FI data node includes the time and the place of occurrence, so the invention utilizes the Glove algorithm in the word embedding field to characterize the attributes of the various nodes as the initial representation of the nodes.
And 2.2, constructing a student campus short-term behavior heterogeneous information network.
The student campus short-term behavior heterogeneous information network comprises four types of nodes and three types of link relations. The four types of nodes are the student nodes, acceleration sensor data, sound sensor data nodes and WI-FI data nodes mentioned in the step 2.1. The three link relations comprise a relation of student movement record connection between the student nodes and the acceleration sensor data, a relation of student surrounding sound environment record connection between the student nodes and the sound sensor data, and a relation of student internet surfing record connection between the student nodes and the WI-FI data.
And 3, performing student campus short-term behavior pattern representation learning based on the meta-path.
Step 3.1, designing a meta-path capable of revealing the track behavior of the student campus.
According to the student school activity scene, three metse:Sup>A paths for revealing student campus behaviors are defined on se:Sup>A student campus short-term behavior heterogeneous information network, wherein the metse:Sup>A paths are S-A-S, S-M-S, S-W-S respectively, S represents se:Sup>A node of se:Sup>A student type, A represents se:Sup>A node of an acceleration sensor datse:Sup>A type, M represents se:Sup>A node of se:Sup>A microphone sensor datse:Sup>A type, and W represents se:Sup>A node of se:Sup>A WI-FI datse:Sup>A type. The semantics of the metse:Sup>A-path S-A-S are that two students have the same action behaviors in the same time period, the semantics of the metse:Sup>A-path S-M-S are that two students have the same sound environment in the same time period, and the semantics of the metse:Sup>A-path S-W-S are that two students use WI-FI of the same place in the same time period.
And 3.2, extracting an instance in the student campus short-term behavior heterogeneous information network based on the meta-path, and encoding the behavior instance by using a meta-path instance encoder. .
And 3.1, extracting examples in a heterogeneous information network by using the student campus behavior meta-path designed in the step 3.1 to obtain a plurality of examples reflecting student campus behaviors, and encoding the examples by using a meta-path example encoder, wherein the encoding process is as follows:
wherein MeanenCODER is an average number encoder, P (v, u) is an example in which a starting node is v and a terminating node is u, which are extracted based on a meta-path, and x is P(v,u) To get meta-path instance representation, x v For the start node representation, x u To terminate the representation, m P(v,u) In the meta-path example, t is one of intermediate nodes, x t Represented as intermediate nodes.
Step 3.3, calculating the influence of each element path instance representation on the target student node by using an attention mechanism, calculating the attention score of each element path instance representation, and weighting and aggregating each element path instance representation to obtain the short-term behavior mode representation of the target student node
And 4, constructing a long-term behavior modeling module based on an attention mechanism. The long-term behavior is the overall behavior of the student in the first school stage. And extracting important behavior patterns most relevant to student learning in single-day behaviors by using an attention mechanism, and focusing on abnormal behaviors in historical behavior data, so that aggregation is carried out to obtain the behavior representation of the whole student learning period.
Through the academic early warning task, the invention obtains the long-term behavior mode representation of the student for perceiving the early abnormal behavior of the student through training the weight of the attention mechanism, as shown in figure 3. Namely, the attention score of short-term behaviors on different dates is calculated according to the importance of the generated behavior pattern characterization on the prediction result, and the formula is as follows:
wherein ,for the short term behavior representation of student on day k, W is a learnable parameter, ++>For the calculated attention score for each short term behavior. And normalizing the attention score by using a Softmax function, wherein the normalized result is the attention distribution on each short-term behavior input during long-term behavior mode polymerization, each numerical value corresponds to the original input, and the formula is as follows:
wherein ,normalized attention weight for each short-term behavior, n is total number of short-term behaviors of student,/-for each short-term behavior>Short term behavioral attention score for student on day j. Weighted summation of the attention weight and each short-term behavioral representation to obtain a long-term behavioral representation L of the student u The formula is as follows:
where n is the total number of short-term behaviors of student u,as short term behavioral attention weight on day j of normalized student u, +.>Short term behavioral representation for student on day j.
And 5, constructing a fusion module based on a gating mechanism. And carrying out feature fusion on short-term behavior and long-term behavior information of the students by using a gating mechanism, and learning student time sequence behavior representation which fully senses abnormal behaviors.
In the fusion module, the invention introduces a gating mechanism to balance the influence of the current behaviors and the past behaviors of the students on the behavior representation, as shown in fig. 4. In particular, the invention uses a learning gating to calculate the importance of short-term and long-term behavior to the overall behavior of the student that is ultimately used for academic prediction. The mathematical formula is as follows:
G u =σ(W S S u +W L L u )
where σ represents the activation function, the invention uses a sigmoid function. W (W) S and WL As a trainable parameter S u For student u's current short-term behavioral representation, L u Is a representation of the long-term behavior pattern of students. G u Is the gating weight generated. By utilizing the learned gating weight, the invention fuses the short-term behavior representation and the long-term behavior representation of the students to obtain the final student time sequence behavior representation, and the formula is as follows:
P u =G u S u +(1-G u )L u
and 6, inputting the student time sequence behavior representation to the full connection layer to realize the prediction of student academic achievements.
The invention divides the student academic status into four grades of excellent, good, medium and bad, and uses the full connection layer to realize the prediction of the student academic grade. In the training process, the model is optimized by using a cross entropy loss function, unbalance of student academic grade distribution is improved, the convergence rate of training is accelerated, and the mathematical formula of the loss function is as follows:
where N is the number of samples used for training, T is the number of categories, y ic As a sign function (1 if the true class of sample i is equal to c, 0 otherwise), p ic The predicted probability that sample i belongs to category c.
By minimizing the loss function, W, W in the model can be determined S 、W L Optimum value of the parameter.
And finally, in the testing stage, inputting the tested student samples into the trained model, and outputting the result of student academic level prediction through the full-connection layer.
Thus, the implementation process of the invention is described.
Claims (2)
1. A student behavior time sequence modeling and academic early warning method for abnormal behavior sensitivity is characterized in that:
the method comprises the following steps:
step 1, preprocessing data of student campus behavior data collected by mobile equipment, and inputting the data into a model;
step 2, constructing a student campus short-term behavior heterogeneous information network based on the single-day campus behavior data;
step 2.1, representing the attribute of each node in the heterogeneous information network by using word embedding algorithm representation as the initial representation of the node;
step 2.2, constructing a student campus short-term behavior heterogeneous information network;
step 3, performing student campus short-term behavior pattern representation learning based on the meta-path;
step 3.1, designing a meta-path capable of revealing the track behavior of the student campus;
step 3.2, extracting an instance in the student campus short-term behavior heterogeneous information network based on the meta-path, and encoding the behavior instance by using a meta-path instance encoder;
step 3.3, aggregating the path instance representations by using an attention mechanism to obtain a student campus short-term behavior pattern representation;
step 4, taking the learning behavior as the long-term behavior of the student, and constructing a long-term behavior modeling module based on an attention mechanism; extracting important behavior patterns most relevant to student academia in time sequence behaviors by using an attention mechanism so as to pay attention to abnormal behaviors in historical behavior data;
step 5, constructing a fusion module based on a gating mechanism; characteristic fusion is carried out on short-term behavior and long-term behavior information of students by using a gating mechanism, and student time sequence behavior representation which fully senses abnormal behaviors is learned;
and 6, inputting the student time sequence behavior representation to the full connection layer to realize the prediction of student academic achievements.
2. The method according to claim 1, comprising the steps of:
step 1, data preprocessing is carried out on student acceleration sensor data, sound sensor data and WI-FI data collected by mobile equipment, student samples with more missing values are removed, seven time semantics of early morning, breakfast, morning, lunch, afternoon, supper and late night are added to the time when student behaviors occur, places where the student behaviors occur are replaced by specific place names from longitude and latitude, and at least the specific place semantics of canteens, sports grounds and teaching buildings are added; dividing student behavior data by utilizing a sliding window, wherein the size of the window is one day, taking the student behavior data divided by the day as the input of a model, and setting the single-day behavior as the short-term behavior of the student;
step 2, constructing a student campus short-term behavior heterogeneous information network based on the preprocessed student single-day campus behavior data;
step 2.1, representing the attribute of each node in the heterogeneous information network by using word embedding algorithm representation as the initial representation of the node;
the student campus short-term behavior heterogeneous information network comprises four types of nodes: student nodes, acceleration sensor data nodes, sound sensor data nodes and WI-FI data nodes; the various nodes have corresponding basic attributes, the student nodes comprise the attributes of gender, specialty, grade and the like of students, and the WI-FI data nodes comprise the occurrence time and place of the students, so that the attributes of the various nodes are characterized by utilizing a Glove algorithm in the word embedding field and are used as initial representation of the nodes;
step 2.2, constructing a student campus short-term behavior heterogeneous information network;
the student campus short-term behavior heterogeneous information network comprises four types of nodes and three types of link relations; the four types of nodes are student nodes, acceleration sensor data, sound sensor data nodes and WI-FI data nodes mentioned in the step 2.1; the three link relations comprise a relation of student movement record connection between a student node and acceleration sensor data, a relation of student surrounding sound environment record connection between a student node and sound sensor data, and a relation of student Internet surfing record connection between a student node and WI-FI data;
step 3, performing student campus short-term behavior pattern representation learning based on the meta-path;
step 3.1, designing a meta-path capable of revealing the track behavior of the student campus;
different metse:Sup>A paths express different semantic relations, three metse:Sup>A paths for revealing student campus behaviors are defined on se:Sup>A student campus short-term behavior heterogeneous information network according to student school internal activity scenes, wherein the metse:Sup>A paths are respectively S-A-S, S-M-S, S-W-S, S represents se:Sup>A node of se:Sup>A student type, A represents se:Sup>A node of an acceleration sensor datse:Sup>A type, M represents se:Sup>A node of se:Sup>A microphone sensor datse:Sup>A type, and W represents se:Sup>A node of se:Sup>A WI-FI datse:Sup>A type; the semantics of the metse:Sup>A-path S-A-S are that two students have the same action behavior in the same time period, the semantics of the metse:Sup>A-path S-M-S are that two students have the same sound environment in the same time period, and the semantics of the metse:Sup>A-path S-W-S are that two students use WI-FI of the same place in the same time period;
step 3.2, extracting an instance in the student campus short-term behavior heterogeneous information network based on the meta-path, and encoding the behavior instance by using a meta-path instance encoder;
and 3.1, extracting examples in a heterogeneous information network by using the student campus behavior meta-path designed in the step 3.1 to obtain a plurality of examples reflecting student campus behaviors, and encoding the examples by using a meta-path example encoder, wherein the encoding process is as follows:
wherein MeanenCODER is an average number encoder, P (v, u) is an example in which a starting node is v and a terminating node is u, which are extracted based on a meta-path, and x is P(v,u) To get meta-path instance representation, x v For the start node representation, x u To terminate the representation, m P(v,u) In the meta-path example, t is one of intermediate nodes, x t Represented as intermediate nodes;
step 3.3, calculating the influence of each element path instance representation on the target student node by using an attention mechanism, calculating the attention score of each element path instance representation, and weighting and aggregating each element path instance representation to obtain the short-term behavior mode representation of the target student node
Step 4, constructing a long-term behavior modeling module based on an attention mechanism; the long-term behavior is the overall behavior of the student in the first school period; extracting important behavior patterns most relevant to student learning in single-day behaviors by using an attention mechanism, and focusing on abnormal behaviors in historical behavior data so as to aggregate to obtain behavior representation of the whole student learning period;
through academic early warning tasks, a student long-term behavior pattern representation for perceiving early abnormal behaviors of students is obtained through training weights of attention mechanisms, namely attention scores of short-term behaviors on different dates are calculated according to importance of generated behavior pattern representation on prediction results, and the formula is as follows:
wherein ,for the short term behavior representation of student on day k, W is a learnable parameter, ++>An attention score for each short term behavior calculated; and normalizing the attention score by using a Softmax function, wherein the normalized result is the attention distribution on each short-term behavior input during long-term behavior mode polymerization, each numerical value corresponds to the original input, and the formula is as follows:
wherein ,normalized attention weight for each short-term behavior, n is total number of short-term behaviors of student,/-for each short-term behavior>Short term behavioral attention score for student on day j; weighted summation of the attention weight and each short-term behavioral representation to obtain a long-term behavioral representation L of the student u The formula is as follows:
where n is the total number of short-term behaviors of student u,for short term behavioral attention weights on day j of the normalized student u,for student on day jShort term behavioral representation;
step 5, constructing a fusion module based on a gating mechanism; characteristic fusion is carried out on short-term behavior and long-term behavior information of students by using a gating mechanism, and student time sequence behavior representation which fully senses abnormal behaviors is learned;
in the fusion module, a gating mechanism is introduced to balance the influence of the current behaviors and the past behaviors of the students on behavior representation, and specifically, the importance of a learnable gating calculation short-term behaviors and long-term behaviors on the overall behaviors of the students finally used for academic prediction is used; the mathematical formula is as follows:
G u =σ(W S S u +W L L u )
wherein sigma represents an activation function, and a sigmoid function is used; w (W) S and WL As a trainable parameter S u For student u's current short-term behavioral representation, L u A long-term behavior pattern representation for the student; g u Gating weights for the generation; and fusing the short-term behavior representation and the long-term behavior representation of the students by using the learned gating weight to obtain final student time sequence behavior representation, wherein the formula is as follows:
P u =G u S u +(1-G u )L u
step 6, inputting the student time sequence behavior representation to the full connection layer to realize the prediction of student academic achievement;
dividing the student academic status into four grades of excellent, good, medium and bad, and using a full connection layer to realize the prediction of the student academic grade; in the training process, the model is optimized by using a cross entropy loss function, unbalance of student academic grade distribution is improved, the convergence rate of training is accelerated, and the mathematical formula of the loss function is as follows:
where N is the number of samples used for training, T is the number of categories, y ic Taking 1 if the real category of the sample i is equal to c, or taking 0 if the real category of the sample i is equal to c; p is p ic The prediction probability of the sample i belonging to the category c;
by minimizing the loss function, W, W in the model can be determined S 、W L Optimum value of the parameter;
and finally, in the testing stage, inputting the tested student samples into the trained model, and outputting the result of student academic level prediction through the full-connection layer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310300117.0A CN116720098A (en) | 2023-03-26 | 2023-03-26 | Abnormal behavior sensitive student behavior time sequence modeling and academic early warning method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310300117.0A CN116720098A (en) | 2023-03-26 | 2023-03-26 | Abnormal behavior sensitive student behavior time sequence modeling and academic early warning method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116720098A true CN116720098A (en) | 2023-09-08 |
Family
ID=87864865
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310300117.0A Pending CN116720098A (en) | 2023-03-26 | 2023-03-26 | Abnormal behavior sensitive student behavior time sequence modeling and academic early warning method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116720098A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117493566A (en) * | 2023-11-02 | 2024-02-02 | 浙江尚云物联科技有限公司 | Data processing system and method for smart campus |
-
2023
- 2023-03-26 CN CN202310300117.0A patent/CN116720098A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117493566A (en) * | 2023-11-02 | 2024-02-02 | 浙江尚云物联科技有限公司 | Data processing system and method for smart campus |
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 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112446591A (en) | Evaluation system for student comprehensive capacity evaluation and zero sample evaluation method | |
CN114509266B (en) | Bearing health monitoring method based on fault feature fusion | |
CN111967712B (en) | Traffic risk prediction method based on complex network theory | |
CN110197332A (en) | A kind of overall control of social public security evaluation method | |
CN103544393B (en) | Method for tracking development of language abilities of children | |
CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
CN109272160A (en) | Score on Prediction system and prediction technique | |
CN109508429A (en) | Personalized adaptive learning recommended method based on teaching platform big data analysis | |
CN116720098A (en) | Abnormal behavior sensitive student behavior time sequence modeling and academic early warning method | |
CN113408852A (en) | Meta-cognition ability evaluation model based on online learning behavior and deep neural network | |
CN110084291B (en) | Student behavior analysis method and device based on big data extreme learning | |
Al-Tameemi et al. | Predictive learning analytics in higher education: Factors, methods and challenges | |
Jiang et al. | Short-term pm2. 5 forecasting with a hybrid model based on ensemble gru neural network | |
Chu et al. | [Retracted] Data Analysis of College Students’ Mental Health Based on Clustering Analysis Algorithm | |
Liu et al. | Rapid assessment of flood loss based on neural network ensemble | |
CN114117033B (en) | Knowledge tracking method and system | |
CN115293249A (en) | Power system typical scene probability prediction method based on dynamic time sequence prediction | |
Chen | Hotel management evaluation index system based on data mining and deep neural network | |
Cheng et al. | A novel learning early-warning model based on random forest algorithm | |
CN115753102A (en) | Bearing fault diagnosis method based on multi-scale residual error sub-domain adaptation | |
CN114722721A (en) | Power transformer health state evaluation method based on deep coding and decoding convolutional network | |
Hu et al. | A security risk plan search assistant decision algorithm using deep neural network combined with two-stage similarity calculation | |
CN114386327A (en) | College professional course score prediction method based on deep learning | |
Zhu | [Retracted] Research on Education Teaching Quality Analysis Based on the Neural Network Model | |
CN113935869A (en) | Student subjective and objective factor combined score prediction method and system |
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 |