CN111553166B - Online learner dynamic model prediction method based on scene cognition calculation - Google Patents
Online learner dynamic model prediction method based on scene cognition calculation Download PDFInfo
- Publication number
- CN111553166B CN111553166B CN202010263233.6A CN202010263233A CN111553166B CN 111553166 B CN111553166 B CN 111553166B CN 202010263233 A CN202010263233 A CN 202010263233A CN 111553166 B CN111553166 B CN 111553166B
- Authority
- CN
- China
- Prior art keywords
- dynamic
- scene
- cognition
- model
- learner
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an online learner dynamic model prediction method based on scene cognition calculation, which comprises the steps of carrying out computable representation on information carried by a dynamic scene to obtain computable representation data of the dynamic scene, rapidly extracting important features of sparse data in the dynamic scene according to the computable representation data, modeling a relation problem between a multidimensional cognition expected level and a target expected level under a dynamic situation by using a self-attention model to obtain various mappers of the dynamic scene, determining cognition expected quantitative prediction of the dynamic scene according to the important features and the various mappers, integrating a plurality of aspects of cognition expected quantitative prediction to establish an overall prediction model, applying a graph convolution neural network technology to the overall prediction model to obtain a cognition predictable model of the dynamic learner, and predicting the dynamic scene to be predicted by adopting the cognition predictable model of the dynamic learner to realize the prediction of the dynamic scene to be predicted.
Description
Technical Field
The invention relates to the technical field of cognition of intelligent supply of personalized knowledge service for online learning, in particular to an online learner dynamic model prediction method based on scene cognition calculation.
Background
The core of the personalized knowledge service oriented to the big data environment is a learner model. Learner dynamic modeling is an abstract representation of the cognitive process of a real learner, and one important trend of the learner dynamic modeling is the diversification of application scenes of the learner dynamic modeling. The data is combined with specific scene application, so that the relation between the learner and the behavior of the learner can be further explored, and key cognitive characteristics which can be promoted are identified. Based on scene cognition analysis, the measurement of the cognition meaning of the scene computable by the data derivation is realized, the proper online learning situation is selected to really establish the association with the learner, and the dynamic model prediction of the online learner with the scene cognition computable is important for the intelligent supply of personalized knowledge service.
The research and the state of the art at home and abroad are comprehensively analyzed, and two main problems exist in the aspect of the dynamic model prediction technology of online learners under the big data environment:
1) The method is oriented to a big data environment, a dynamic model of an online learner is more in local formalization of cognition under specific situations, the prediction method needs to be established under the evolution situation of a dynamic scene, and the evolution of the dynamic model of the learner can be reliably and reasonably interpreted and evaluated to simulate various learning cognitive behavior phenomena of the learner.
2) In the space of big data mapping, the learner dynamic model evolution process relates to establishing a definite and measurable corresponding relation between dynamic situations and each cognitive element in the multi-dimensional view angle, and each element reflected in the learner dynamic model evolution process is characterized by a computable technology, and the influence between multi-dimensional correlations is quantized and computable.
Therefore, it is an urgent task to study how to predict a learner's dynamic model in a way that dynamic scene cognition can be calculated to support intelligent personalized knowledge service, and describe and explain the learner's cognitive behavior and process in this process.
Disclosure of Invention
Aiming at the problems, the invention provides an online learner dynamic model prediction method based on scene cognition calculation.
In order to achieve the purpose of the invention, the invention provides an online learner dynamic model prediction method based on scene cognition calculation, which comprises the following steps:
s10, carrying out computable representation on information carried by a dynamic scene based on a big data space associated with online learning to obtain computable representation data of the dynamic scene;
s20, rapidly extracting important features of sparse data in the dynamic scene according to the computable representation data, modeling a relation problem between a multi-dimensional cognition expected level and a target expected under a dynamic situation by adopting a self-attention model so as to obtain various mappers of the dynamic scene, and determining cognition expected quantitative prediction of the dynamic scene according to the important features and the various mappers;
and S30, integrating multiple aspects of the cognition expected quantitative prediction to establish an integral prediction model, applying a graph convolution neural network technology to the integral prediction model to obtain a cognition predictable model of a dynamic learner, and predicting a dynamic scene to be predicted by adopting the cognition predictable model of the dynamic learner.
In one embodiment, the performing, based on the big data space associated with online learning, the computable representation of the information carried by the dynamic scene, to obtain computable representation data of the dynamic scene includes:
based on online learning associated big data space, adopting a network local dynamic perception strategy of a topic model, fusing a dynamic topic model and a dynamic situation extraction method of network structure analysis, and carrying out dynamic probability topic model analysis in each topic-based sub-network of a dynamic scene so as to map a feature node semantic function based on scene cognition to the topic space, and carrying out calculation and representation on information mapped to the topic space to obtain the computable representation data of the dynamic scene.
As one embodiment, the dynamic scenario extraction method integrating the dynamic topic model and the network structure analysis includes:
each situation included in the dynamic scene can be characterized through the dynamic theme distribution of the situation, and the situation is marked by the theme with the highest probability, so that the preliminary extraction of the situation is realized;
and capturing a more complex neighborhood through the similarity propagation of the dynamic situation topics in the cognitive correlation network, and performing secondary extraction on the dynamic situation topics.
As one embodiment, the mapping the scene awareness based feature node semantic function to the topic space includes:
where θ is the probability of occurrence of a contextual topic, z is the contextual topic, w is an element of a multidimensional cognitive space, and the learned cognitive feature vector x (t) = (x) 1 (t),...,x N (t)) T N node local dynamics at time t are captured.
In one embodiment, the determining the cognitive expected quantitative prediction of the dynamic scene from the important features and the plurality of mappers comprises:
aiming at various mappers, executing an attention function in parallel, processing cognitive sequence data information by using the important features represented by different subspaces of different sequence positions, obtaining new features considering context information for each cognitive feature, learning feature dependency relationship in the cognitive space, and capturing an internal structure;
and selectively screening a small amount of important information from a large amount of situation information, focusing on the important characteristics, and forming the final cognition expected quantitative prediction through linear fusion.
According to the on-line learner dynamic model prediction method based on scene cognition calculation, the information carried by the dynamic scene is calculated and represented through a big data space based on-line learning association, the calculated and represented data of the dynamic scene is obtained, important features of sparse data in the dynamic scene are rapidly extracted according to the calculated and represented data, a relation problem between a multi-dimensional cognition expected level and a target expected level under a dynamic situation is modeled by adopting a self-attention model, multiple mappers of the dynamic scene are obtained, cognition expected quantitative prediction of the dynamic scene is determined according to the important features and the multiple mappers, multiple aspects of cognition expected quantitative prediction are integrated to establish an integral prediction model, a graph convolution neural network technology is applied to the integral prediction model, the cognition predictable model of the dynamic learner is obtained, and the cognition predictable model of the dynamic scene to be predicted is adopted to realize the prediction of the dynamic scene to be predicted. The modeling and interpretation of the dynamic cognitive evolutionary phenomenon which cannot be described and interpreted by the traditional learner modeling theory in the dynamic scene cognitive evolutionary process are realized, the uncertainty in the aspects of situation, cognition and the like in the dynamic evolutionary cognitive process is formally modeled, the problem that how to formalize and calculate the uncertainty in the cognitive process, situation and the like of a person in the large data space learner model evolutionary process is solved, and the method has more general purpose and practical value than the traditional learner modeling research.
Drawings
FIG. 1 is a flow diagram of an online learner dynamic model prediction method based on scene recognition calculation according to one embodiment;
FIG. 2 is a flowchart of another embodiment of a scene recognition calculation-based online learner dynamic model prediction method.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the present application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.
The invention aims at solving the problems of dynamic situation evolution and cognitive semantic association cognitive phenomenon in the dynamic model evolution process of the online learner in the big data environment, and provides a computable method for designing the dynamic model prediction of the potential learner and supporting intelligent personalized knowledge service under the condition of conforming to engineering practice.
The technical problems to be solved by the invention are as follows:
1) The local dynamics may calculate the representation.
Based on the scene recognition, psychological factors and human factors affecting the behavior of the learner can be understood. In a large-scale online learning environment, the change of the scene reflects the influence of static and dynamic attributes on the learning behavior cognitive process. The calculability of the scene cognition provides possibility for quantification of scene cognition attributes, and also provides basis for quantized data to participate in calculation processes such as modeling, judgment and the like. The method provides a recognizable, structural, compact and efficient characteristic computable representation method for various elements such as dynamic situation of the learner model evolution process, and realizes the expression and identification of the scene of dynamic evolution so as to measure the semantic situation and the implication potential cognitive factors of scene interaction.
2) Learner cognition expectation calculation
The cognitive judgment in advance of the learner forms a 'cognitive expectation', and the future learning state expectation and prediction formed in advance under the specific scene cognitive situation are defined. For a specific learning task, the learner can continuously expect and predict the effect of learning behavior, information alternation which may occur, and the future state of the learning situation, so as to form a dynamic cycle. And carrying out quantitative prediction on the situation cognition level of the learner, integrating information factors with different dimensions, realizing the calculability of the cognition expected level, and enabling the learner model to select a proper situation to really establish association with the learner.
Overall, the online learned big data space is understood as an ensemble with multidimensional heterogeneous complex network morphology formed by a large number of individual feature data following interaction and association rules, which map with the cognitive space. The novel data driving analysis technology-network embedded representation formed by integrating big data and deep learning is applied to a multidimensional cognitive associated network and is used as a basic framework and a core academic idea for solving the problem of online learning-oriented and dynamic scene cognition-based computable learner model prediction. The specific technical scheme is developed around three parts of computable representation of dynamic scene cognition, cognition expected calculation and structured learner dynamic model prediction. Referring to fig. 1, fig. 1 is a schematic flow chart of a scene cognition calculation-based on-line learner dynamic model prediction method according to an embodiment, including the following steps:
s10, based on the online learning associated big data space, calculating and representing the information carried by the dynamic scene to obtain the calculating and representing data of the dynamic scene.
S20, rapidly extracting important features of sparse data in the dynamic scene according to the computable representation data, modeling a relation problem between a multi-dimensional cognition expected level and a target expected under a dynamic situation by adopting a self-attention model, so as to obtain various mappers of the dynamic scene, and determining cognition expected quantitative prediction of the dynamic scene according to the important features and the various mappers.
And S30, integrating multiple aspects of the cognition expected quantitative prediction to establish an integral prediction model, applying a graph convolution neural network technology to the integral prediction model to obtain a cognition predictable model of a dynamic learner, and predicting a dynamic scene to be predicted by adopting the cognition predictable model of the dynamic learner.
According to the on-line learner dynamic model prediction method based on scene cognition calculation, the information carried by the dynamic scene is calculated and represented through a big data space based on-line learning association, the calculated and represented data of the dynamic scene is obtained, important features of sparse data in the dynamic scene are rapidly extracted according to the calculated and represented data, a relation problem between a multi-dimensional cognition expected level and a target expected level under a dynamic situation is modeled by adopting a self-attention model, multiple mappers of the dynamic scene are obtained, cognition expected quantitative prediction of the dynamic scene is determined according to the important features and the multiple mappers, multiple aspects of cognition expected quantitative prediction are integrated to establish an integral prediction model, a graph convolution neural network technology is applied to the integral prediction model, the cognition predictable model of the dynamic learner is obtained, and the cognition predictable model of the dynamic scene to be predicted is adopted to realize the prediction of the dynamic scene to be predicted. The modeling and interpretation of the dynamic cognitive evolutionary phenomenon which cannot be described and interpreted by the traditional learner modeling theory in the dynamic scene cognitive evolutionary process are realized, the uncertainty in the aspects of situation, cognition and the like in the dynamic evolutionary cognitive process is formally modeled, the problem that how to formalize and calculate the uncertainty in the cognitive process, situation and the like of a person in the large data space learner model evolutionary process is solved, and the method has more general purpose and practical value than the traditional learner modeling research.
In one embodiment, the performing, based on the big data space associated with online learning, the computable representation of the information carried by the dynamic scene, to obtain computable representation data of the dynamic scene includes:
based on online learning associated big data space, adopting a network local dynamic perception strategy of a topic model, fusing a dynamic topic model and a dynamic situation extraction method of network structure analysis, and carrying out dynamic probability topic model analysis in each topic-based sub-network of a dynamic scene so as to map a feature node semantic function based on scene cognition to the topic space, and carrying out calculation and representation on information mapped to the topic space to obtain the computable representation data of the dynamic scene.
Specifically, the dynamic situation extraction method integrating the dynamic theme model and the network structure analysis comprises the following steps:
each situation included in the dynamic scene can be characterized through the dynamic theme distribution of the situation, and the situation is marked by the theme with the highest probability, so that the preliminary extraction of the situation is realized;
and capturing a more complex neighborhood through the similarity propagation of the dynamic situation topics in the cognitive correlation network, and performing secondary extraction on the dynamic situation topics.
Specifically, the mapping the feature node semantic function based on scene recognition to the topic space comprises:
where θ is the probability of occurrence of a contextual topic, z is the contextual topic, w is an element of a multidimensional cognitive space, and the learned cognitive feature vector x (t) = (x) 1 (t),...,x N (t)) T N node local dynamics at time t are captured.
The embodiment can realize the computable representation of the dynamic scene cognition.
In one example, in a big data space associated with online learning, data describing personalized cognitive features of each learner and implicit cognitive mechanisms thereof have complex association relationships therebetween, and are self-organized into a complex association network. The network representation learning is a vectorization (vectorization) technology, and a dimension-reducing method is used to express the characteristics of structures (nodes, edges or subgraphs) in the network into a vector of hidden space, and the characteristics of the nodes or edges are naturally determined by the network environment (context) in which the nodes or edges are located. Complex network information is changed into a structured multidimensional feature vector representation, so that more convenient algorithm application is realized by using a deep learning method.
Therefore, a basic framework of network embedded representation is adopted to represent a multidimensional cognitive associated network, and a vector is cut into a graph structure. Migrating Word vector method Word2Vec in natural language processing to complex cognitive associated network, generating node sequence on the graph network according to a searching method, representing the node in complex cognitive associated network as a vector according to algorithm frame of Word2Vec to obtain vector expression of the node. As long as each point is assigned an n-dimensional vector, an embedded representation of the entire network is obtained. The network embedded vector representation technology based on deep learning is used for automatically learning all cognitive correlation characteristics by a machine, and reflects the ecological status of each node in the whole cognitive correlation network, so that a user can stand at a higher level to understand a complex learning and cognition process.
Based on the above representation, capturing local dynamics based on context transformations, taking into account spatiotemporal characteristics, accords with the general thinking of humans solving problems, where human cognition cannot be fully, completely, uniformly formalized, but millions of cognition under specific contexts can be localized formalized. The semantic expression of the dynamic context in the multidimensional cognitive associated network is essentially regarded as an extraction problem of the dynamic context theme. The network local dynamic perception strategy based on the topic model adopts a dynamic situation extraction method integrating dynamic topic model and network structure analysis, firstly, in each sub-network based on topic, adopts a dynamic probability topic model analysis method to map the feature node semantic functions (state, intention, situation and the like) based on scene cognition to the topic space,
where θ is the probability of occurrence of a contextual topic, z is the contextual topic, and w is an element of the multidimensional cognitive space. Learned cognitive feature vector x (t) = (x) 1 (t),...,x N (t)) T N node local dynamics at time t are captured.
Each context can be characterized through the dynamic theme distribution of the context, and the context is marked by the theme with the highest probability, so that the preliminary extraction of the context is realized. And then estimating the similarity measurement of the characteristics based on the situation according to the topic probability distribution of the situation by using a network structure analysis method in the cognitive space, and processing the cognitive correlation network propagation process related to the dynamic situation information based on the network embedded vector. And capturing a more complex neighborhood through the similarity propagation of the dynamic situation topics in the cognitive correlation network, and performing secondary extraction on the dynamic situation topics.
In the actual online intelligent education application situation facing to the big data space, the intelligent education system is required to rapidly understand and judge the relevance and dynamic change among various learning process objects in a complex scene, and the fact that the relevance can be abstracted and expressed is the premise of understanding and judging. The dynamic scene cognition in the online learning process is realized to realize the computable representation, so that the accuracy of the relevance of the system abstraction and the expression can be greatly improved. Based on the technology of computable representation, the learning process is further understood through perception and interaction, reasoning is performed by using assumptions and arguments, and learning is performed through data, so that the cognitive computing technology is applied to specific intelligent education application, and the learner is helped to create new value. Therefore, in the online learning process, based on the cognition computable representation technology, formal representation and computable characteristics are satisfied, the associated features in the learning and cognition process can be effectively abstracted, and the problem that the traditional theoretical driven cognition modeling method is incomplete in assumption is effectively solved.
In one embodiment, the determining the cognitive expected quantitative prediction of the dynamic scene from the important features and the plurality of mappers comprises:
aiming at various mappers, executing an attention function in parallel, processing cognitive sequence data information by using the important features represented by different subspaces of different sequence positions, obtaining new features considering context information for each cognitive feature, learning feature dependency relationship in the cognitive space, and capturing an internal structure;
and selectively screening a small amount of important information from a large amount of situation information, focusing on the important characteristics, and forming the final cognition expected quantitative prediction through linear fusion.
The present embodiment can realize learner cognition expectation calculation, specifically, attention mechanism (Attention) imitates the internal process of biological observation behavior, namely, a mechanism of aligning internal experience and external feeling to increase observation fineness of a partial region. Important features of sparse data can be rapidly extracted, and the method is widely used in various deep learning tasks such as natural language processing and the like. Self-Attention mechanism (Self-Attention) is an improvement of Attention mechanism that reduces reliance on external information, and is better at capturing internal dependencies of data or features.
The self-attention model (SelfMulti-HeadAttention) is introduced into learner cognition expectation calculation, so that the relation problem between the cognition expectation level of multiple dimensions and target expectation under the dynamic situation can be well modeled: the Source source= < key, value > is a vector composed of feature nodes in the multidimensional cognitive correlation network, and it is expected that the Target context-aware feature vector Target is generated through the deep learning framework. Essentially, given an element Query in Target, the Attention mechanism is to weight and sum the Value values of the context elements in Source, and Query and Key are used to calculate the weight coefficient corresponding to the Value. I.e. its essential idea can be rewritten as the following formula:
wherein L is X Source, representing the length of Source. Multi-headand (3) performing linear mapping for each dimension of the Query, the key and the value for h times by using various mappers learned by the attention, and executing the attention function in parallel. And performing cognitive sequence data information processing by using the characterization information of different subspaces of different sequence positions, obtaining new characterization which considers context information for each cognitive feature, learning the feature dependency relationship in the cognitive space, and capturing the internal structure. Attention can be understood as separately describing interaction between a learner and scene context information, selectively screening a small amount of important information from a large amount of context information, focusing on the important information, and then forming final cognition expectation quantitative prediction through linear fusion, so that the cognition expectation model is a computable implementation of a cognition expectation model of a multi-situation state learner.
In big data-oriented intelligent education application, the data is transited to wisdom, and a wisdom education system is required to know cognitive factors of a learner implicit in the data in advance, so that a better learning service is provided. The learner cognition expectation calculation presents immeasurable explosive effect in the aspects of acquiring and perceiving complex multidimensional education data, detecting and evaluating dynamic teaching process, providing personalized learning service, making accurate teaching decision and the like. The learning process is a continuous dynamic process for situational awareness understanding, and awareness expectation calculation provides basic technical support for intelligent personalized knowledge service provision oriented to the dynamic process. Learner cognition expectancy computing techniques may "enable" intelligent learning assessment, may "enable" learning service ecology. From the learner's perspective, will obtain suitable personalized learning service and good development experience under the anticipated computing technology support, from the teacher's perspective, the teacher can develop high-efficiency intelligent teaching methods and develop core literacy-based intelligent assessment.
In one embodiment, in step S30, multiple aspects of the cognitive prediction quantification prediction are integrated to create an overall prediction model, the graph convolution neural network technology is applied to the overall prediction model to obtain a dynamic learner cognitive prediction model, and the dynamic learner cognitive prediction model is adopted to predict a dynamic scene to be predicted, so that a structural prediction model can be implemented.
Specifically, the learner structured predictive model refers to an overall predictive model that is built by integrating aspects of the learner. Understanding the cognitive structure and function in the big data learning space, predicting the evolution of the cognitive relation network is a key for modeling the collective dynamic property in the learner cognitive process, so that the semantic structural attribute of the static function and the dynamic situation relation should be fully considered, and the cognitive characteristic variables are aggregated through learning and social behavior data. The situation and knowledge state changes expressed by the evolution of cognitive association are utilized to more carefully understand and characterize the association between a learner and the scene where the learner is located, and more abundant cognitive behavior information is mapped to a possibly associated learner cognitive model subspace.
The cognitive space is a subspace of a cognitive model that maps cognitive process information to possible associations using context and cognitive state changes exhibited by the evolution of cognitive associations. The cognitive process itself can be seen as a topological graph of cognitive feature interactions, where the context-to-cognitive feature associations in such graph-based cognitive space change as the context changes.
Applying the graph roll-up neural network technique (GCN) to structured learner model prediction is a very reasonable idea. Graph convolutional neural network (GCN) is a method that enables deep learning of graph data. GCN allows end-to-end learning of structured data, i.e. inputting a graph, which may be of arbitrary size and shape, to handle large amounts of network data. The method mainly comprises two steps: a. selecting a representative node sequence from the graph structure; b. for each node selected, a convolution neighborhood is determined. If the characteristic expression of the image layer is wanted, the expression of each node is only needed to be integrated and then is subjected to a mapping operation. However, unlike conventional graph structure data, the topology of the cognitive process is a series of time series with spatial cognitive features at each point in time and temporal characteristics between the cognitive features. Therefore, a space-time diagram convolutional network (ST-GCN) is adopted to correspondingly extract high-dimensional cognitive characteristics for space and time edges. Compared with the traditional GCN method, the method can be extended to a space-time diagram by correspondingly modifying the sampling function and the weight mapping function, and the sampling area is set to be adjacent frames. The weight mapping function sets the mapping function according to the orderly characteristics of the images of the adjacent frames, and the mapping function is used as a corresponding addressing operation, namely, the number of matrix elements in the number of dimensions, is modeled, and the cognitive semantic features are captured.
The potential structured knowledge is learned from the complex multidimensional cognitive associated network, and the characteristic information and the associated structure information of the nodes are considered at the same time to describe different situation changes in the data space, so that the influence of the situation changes is embedded into the cognitive characteristics and the associated dynamic analysis and representation, and the influence of the information from different scenes on the learning cognitive behaviors of the learner is measured. The ST-GCN method provides a new idea for solving the problem of prediction of the graph structure data. The cognitive node linear relation in the big data space is encoded by the cognitive associated network node vectors with different dimensions, different parts of the vectors correspond to different semantics, and hidden relations among the nodes are encapsulated. By adopting a graph convolution network method, joint vector representations of cognitive features with different dimensions are used as the input of ST-GCN, and a structured hidden and complex mode in a correlation network is automatically learned Xi Duowei. The prediction of the learner dynamic model is analyzed by using the method, and the cognition predictable model of the dynamic learner with semantic structure is realized by considering the cognition association relation based on the graph and the cognition semantic dynamic superposition association characteristic based on the situation.
The structured prediction of the learner dynamic model is that each factor in the learning space extends in multiple dimensions, and under a typical intelligent education application scene, the learner dynamic model is strongly correlated with multiple dimensions such as the intention, the state scene, the situation and the like of the learner besides being strongly correlated with the semantics of a knowledge layer, so that the evolution of the learner model is formed under different dimensions. As in the learning scenario of experimental knowledge and theoretical knowledge, a learner is faced with different specific situations or views in a particular time and in a particular space. In addition to the different dimensions of knowledge itself, different environmental factors such as spatial environment of learning, collaborative perception, etc. are involved. The multi-dimensional situation matching evolution can be naturally modeled and detected by embedding the influence of the situation change into the cognitive characteristics and the associated dynamic analysis and the represented structured learner model prediction, the mutual interweaving factors and the mutual relations of the factors in the scene are included, and the explanation of the mutual relations of all objects of the scene is embodied. And guiding the personalized knowledge supply and the optimization direction of the automatic recommendation service, and finally applying the personalized knowledge supply and the optimization direction to related business of intelligent education.
The embodiment has the following technical effects:
big data has become the leading field for constructing new modes of economic and social development in various countries and remolding the long-term competitiveness of the countries. The invention is oriented to big data space, the invention is oriented to the work of the learner dynamic cognition model prediction of big data space personalized knowledge service, the modeling and interpretation can not describe and interpret the dynamic cognition evolution phenomenon which is unable to be described and interpreted by the traditional learner modeling theory in the dynamic scene cognition evolution process, formally models the uncertainty in the aspects of situation, cognition and the like in the dynamic evolution cognition process, and solves the problem of how to formalize and calculate the uncertainty in the cognition process, situation and the like of people in the big data space learner model evolution process. Compared with the former learner modeling research, the method has more general purpose and practical value.
In another embodiment, the above-mentioned on-line learner dynamic model prediction method based on scene recognition calculation is further described with reference to fig. 2.
First, the learner characteristic association is expressed and explained based on the association network representation method of network embedded representation learning in the big data space. The graph structure cuts into a vector, the ecological status of each cognitive node in the whole associated network is reflected by an N-dimensional vector, the structured knowledge is extracted, and the complex learning and cognitive process is understood at a higher level.
Secondly, the cognitive expectation calculation based on the attention model generates cognitive expectation levels of different dimensions in a high-dimensional space, wherein the cognitive expectation levels influence the cognitive expectation of a learner, and the cognitive expectation levels of the different dimensions are different, so that the effect expected by the learner can be best achieved under a certain scene situation.
Finally, based on the working foundation of the first two parts, the semantic structured attribute of capturing the static function and the dynamic situation relation in the learning and cognition space is expressed by adopting a multidimensional feature joint vector, the possibility of dynamic evolution of the learner model is predicted, the dynamic learner model is found and identified under the condition of scene cognition change, and the evolution prediction of the medium scene cognition computable learner dynamic model facing to the big data environment is realized.
In the whole implementation of the embodiment, the network embedded representation and the uncertainty cognition dynamic semantic expression method are adopted to realize interaction and understanding in the dynamic learning cognition evolution process reflected by the data feature form, so that the uncertainty in the aspect of the cognition evolution of the data feature context, scene and the like reflected in the learning process is formally explained, and the problem of how to formally calculate the uncertainty of the thinking, cognition and the like of a learner in the online learner dynamic model evolution prediction process is solved.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The terms "comprising" and "having" and any variations thereof, in embodiments of the present application, are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, article, or device that comprises a list of steps or modules is not limited to the particular steps or modules listed and may optionally include additional steps or modules not listed or inherent to such process, method, article, or device.
The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.
Claims (4)
1. An online learner dynamic model prediction method based on scene cognition calculation is characterized by comprising the following steps:
s10, carrying out computable representation on information carried by a dynamic scene based on a big data space associated with online learning to obtain computable representation data of the dynamic scene;
s20, rapidly extracting important features of sparse data in the dynamic scene according to the computable representation data, modeling a relation problem between a multi-dimensional cognition expected level and a target expected under a dynamic situation by adopting a self-attention model so as to obtain various mappers of the dynamic scene, and determining cognition expected quantitative prediction of the dynamic scene according to the important features and the various mappers;
s30, integrating multiple aspects of the cognition expected quantitative prediction to establish an overall prediction model, applying a graph convolution neural network technology to the overall prediction model to obtain a dynamic learner cognition predictable model, and predicting a dynamic scene to be predicted by adopting the dynamic learner cognition predictable model;
the step of carrying out the computable representation on the information carried by the dynamic scene based on the big data space of the online learning association, and the step of obtaining the computable representation data of the dynamic scene comprises the following steps:
based on online learning associated big data space, adopting a network local dynamic perception strategy of a topic model, fusing a dynamic topic model and a dynamic situation extraction method of network structure analysis, and carrying out dynamic probability topic model analysis in each topic-based sub-network of a dynamic scene so as to map a feature node semantic function based on scene cognition to the topic space, and carrying out calculation and representation on information mapped to the topic space to obtain the computable representation data of the dynamic scene.
2. The method for predicting dynamic models of online learners based on scene recognition calculation as recited in claim 1, wherein the method for extracting dynamic context by integrating dynamic topic models and network structure analysis comprises the following steps:
each situation included in the dynamic scene can be characterized through the dynamic theme distribution of the situation, and the situation is marked by the theme with the highest probability, so that the preliminary extraction of the situation is realized;
and capturing a more complex neighborhood through the similarity propagation of the dynamic situation topics in the cognitive correlation network, and performing secondary extraction on the dynamic situation topics.
3. The method for on-line learner dynamic model prediction based on scene recognition computing according to claim 1, wherein the mapping the scene recognition based feature node semantic function to the topic space comprises:
where θ is the probability of occurrence of a contextual topic, z is the contextual topic, w is an element of a multidimensional cognitive space, and the learned cognitive feature vector x (t) = (x) 1 (t),...,x N (t)) T N node local dynamics at time t are captured.
4. A method of on-line learner dynamic model prediction based on scene recognition calculation according to any one of claims 1 to 3, wherein said determining a recognition expected quantitative prediction of the dynamic scene from the important features and the plurality of mappers comprises:
aiming at various mappers, executing an attention function in parallel, processing cognitive sequence data information by using the important features represented by different subspaces of different sequence positions, obtaining new features considering context information for each cognitive feature, learning feature dependency relationship in the cognitive space, and capturing an internal structure;
and selectively screening a small amount of important information from a large amount of situation information, focusing on the important characteristics, and forming the final cognition expected quantitative prediction through linear fusion.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010263233.6A CN111553166B (en) | 2020-04-05 | 2020-04-05 | Online learner dynamic model prediction method based on scene cognition calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010263233.6A CN111553166B (en) | 2020-04-05 | 2020-04-05 | Online learner dynamic model prediction method based on scene cognition calculation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111553166A CN111553166A (en) | 2020-08-18 |
| CN111553166B true CN111553166B (en) | 2023-05-26 |
Family
ID=72007372
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010263233.6A Active CN111553166B (en) | 2020-04-05 | 2020-04-05 | Online learner dynamic model prediction method based on scene cognition calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111553166B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112132430B (en) * | 2020-09-14 | 2022-09-27 | 国网山东省电力公司电力科学研究院 | Reliability evaluation method and system for distributed state sensor of power distribution main equipment |
| CN113283488B (en) * | 2021-05-18 | 2021-11-23 | 清华大学 | Learning behavior-based cognitive diagnosis method and system |
| CN113408852B (en) * | 2021-05-18 | 2022-04-19 | 江西师范大学 | Meta-cognition ability evaluation model based on online learning behavior and deep neural network |
| CN116955335A (en) * | 2023-07-21 | 2023-10-27 | 北京国信达数据技术有限公司 | Address data management method and system based on big data model algorithm |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287941A (en) * | 2019-07-03 | 2019-09-27 | 哈尔滨工业大学 | A kind of thorough perception and dynamic understanding method based on concept learning |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040230951A1 (en) * | 2003-04-02 | 2004-11-18 | Scandura Joseph M. | Method for Building Highly Adaptive Instruction Based on the Structure as Opposed to the Semantics of Knowledge Representations |
| CN108664614B (en) * | 2017-05-12 | 2021-11-23 | 华中师范大学 | Learner model dynamic construction method based on education big data |
| CN108510841A (en) * | 2017-07-25 | 2018-09-07 | 北京七展国际数字科技有限公司 | EPS telegraphy official's context aware learning methods |
| CN108062369B (en) * | 2017-12-08 | 2020-08-11 | 华中师范大学 | Situation-integrated polymorphic ubiquitous learning resource aggregation method and system |
| CN110598770B (en) * | 2019-08-30 | 2022-04-08 | 华中师范大学 | Multi-space fusion learning environment construction method and device |
-
2020
- 2020-04-05 CN CN202010263233.6A patent/CN111553166B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287941A (en) * | 2019-07-03 | 2019-09-27 | 哈尔滨工业大学 | A kind of thorough perception and dynamic understanding method based on concept learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111553166A (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111553166B (en) | Online learner dynamic model prediction method based on scene cognition calculation | |
| CN110168495A (en) | It can be re-used, reconfigure and be reassembled as the housebroken artificial intelligence object of artificial intelligence model | |
| Zhuge | Socio-natural thought semantic link network: A method of semantic networking in the cyber physical society | |
| CN107391505A (en) | A kind of image processing method and system | |
| CN112328801A (en) | Method for predicting group events by event knowledge graph | |
| Najar et al. | Predicting information diffusion on social networks with partial knowledge | |
| Wang et al. | Learning performance prediction via convolutional GRU and explainable neural networks in e-learning environments | |
| CN113886567A (en) | Teaching method and system based on knowledge graph | |
| CN113191154A (en) | Semantic analysis method, system and storage medium based on multi-modal graph neural network | |
| Hall et al. | Visualizing the workings of agent-based models: Diagrams as a tool for communication and knowledge acquisition | |
| Chanaa et al. | Predicting learners need for recommendation using dynamic graph-based knowledge tracing | |
| Carbone | A framework for enhancing transdisciplinary research knowledge | |
| Shen et al. | Ontology-based learning content recommendation | |
| Cerreta et al. | Complex values-based approach for multidimensional evaluation of landscape | |
| Liu | The artistic design of user interaction experience for mobile systems based on context-awareness and machine learning | |
| CN115238036A (en) | Cognitive diagnosis method and device based on graph attention network and text information | |
| Wang et al. | Human-instructed deep hierarchical generative learning for automated urban planning | |
| Pasichnyk et al. | Model of the Recommender System for the Selection of Electronic Learning Resources. | |
| CN106372147B (en) | Heterogeneous topic network construction and visualization method based on text network | |
| Karacapilidis et al. | On the exploitation of collaborative argumentation structures for inducing reasoning behavior | |
| Iwaniak et al. | Ontology driven analysis of spatio-temporal phenomena, aimed at spatial planning and environmental forecasting | |
| Liebovitch et al. | Modeling the dynamics of sustainable peace | |
| CN116702784B (en) | Entity linking method, entity linking device, computer equipment and storage medium | |
| Black et al. | Basic steps for the development of decision support systems | |
| Okoye | Process mining with semantics: Real-time application on a learning process domain |
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 |