Personalized learning path recommendation method and device based on enhanced cognitive diagnosis

Abstract

The invention provides a personalized learning path recommendation method and device based on enhanced cognitive diagnosis. The method comprises the following steps: constructing a learning behavior heterogeneous graph to describe richer learning behavior characteristics of students; according to a knowledge forgetting simulation algorithm based on reinforcement learning, a training agent learns a knowledge forgetting rule of a student in a learning process according to a reward mechanism; evaluating knowledge cognition levels of the students at different moments according to the learning behavior sequence fused with the forgetting rules of the students; building a knowledge sequential graph by using a finite-state automatic machine so as to build a hyponym and a first-to-second relationship between knowledge points; a learning path recommendation method based on Monte Carlo search enhancement is adopted, and personalized learning paths are recommended for students in an unobservable environment according to the knowledge cognition level and the knowledge sequential diagram. According to the method, the cognitive level of the student can be accurately evaluated, the knowledge sequential graph can be automatically generated, and the real demand of recommending a reasonable and effective personalized learning path to the student in real time in an unobservable environment is met.

Description

Technical Field

[0001] The present application relates to a method and device for recommending personalized learning paths based on enhanced cognitive diagnosis, and belongs to the technical field of educational data mining. Background Art

[0002] In recent years, online education has developed rapidly as a supplement to traditional education. More and more online course resources have been developed, and students can learn the knowledge they are interested in more conveniently and independently. However, unlike traditional teaching, which can recommend effective learning paths for students based on learning goals, online education platforms cannot provide reasonable and effective learning path guidance for students based on their knowledge and cognitive level, resulting in students facing the dilemma of "learning wandering" when learning online education resources. "Learning wandering" is mainly reflected in students' casual browsing and learning courses in a limited way, lacking overall logical control of the course system, and the knowledge learned shows characteristics such as discreteness, repetition, and discontinuity, which leads to students' frustration and continuously increases the cost of learning and the risk of failure.

[0003] Learning path recommendation is an effective solution to the problem of "learning wandering". It analyzes students' previous learning records to assess their knowledge level and then recommends appropriate subsequent knowledge points for them to learn. Although learning path recommendation technology has been widely studied, it still has the following limitations:

[0004] First, most learning path recommendation methods convert the personalized learning path recommendation task into a sequence recommendation problem, and directly use the sequence recommendation algorithm in the field of product recommendation to recommend personalized learning paths for students. Since this recommendation algorithm cannot take into account the students' knowledge level, the recommended learning path is not suitable for the students' own learning situation.

[0005] Secondly, in order to make up for the defects of directly copying the product sequence recommendation, some learning path recommendation algorithms incorporate students' knowledge and cognitive level into the recommendation process. A common practice is to introduce forget gate neural units in the sequence neural network to randomly forget certain knowledge points to simulate the students' forgetting process. However, although randomly discarding knowledge points by introducing forget gates can meet the knowledge forgetting requirements to a certain extent, the process is arbitrary and discontinuous, and cannot truly simulate the students' knowledge forgetting rules.

[0006] Finally, existing learning path recommendation algorithms based on reinforcement learning usually use Markov decision processes to construct reinforcement learning environments, formalizing the learning path recommendation problem as a reinforcement learning problem in an observable environment. However, the knowledge learning process is obviously an unobservable problem, which severely limits the ability of learning path recommendation algorithms based on observable environments. Summary of the invention

[0007] In view of the above-mentioned deficiencies in the prior art, the present application provides a personalized learning path recommendation method and device based on enhanced cognitive diagnosis, which combines students' knowledge forgetting characteristics and learning cognitive laws, analyzes the sequential characteristics of students' knowledge cognitive level and knowledge structure, and recommends personalized learning paths for students.

[0008] In a first aspect, the present application embodiment provides a personalized learning path recommendation method based on enhanced cognitive diagnosis, including:

[0009] Construct a learning behavior heterogeneous graph to describe students' richer learning behavior characteristics;

[0010] The knowledge forgetting simulation algorithm based on reinforcement learning trains the intelligent agent to learn the knowledge forgetting rules of students in the learning process according to the reward mechanism;

[0011] Evaluate students' knowledge and cognition levels at different times based on a learning behavior sequence that incorporates students' forgetting rules;

[0012] Use finite state automata to construct a knowledge sequential graph to build the hierarchical and sequential relationships between knowledge points;

[0013] A learning path recommendation method based on Monte Carlo search enhancement is adopted to recommend personalized learning paths for students in an unobservable environment according to the knowledge cognition level and the knowledge sequence graph.

[0014] Based on the above method, optionally, the construction of a learning behavior heterogeneous graph to depict students' richer learning behavior characteristics includes:

[0015] Constructing a heterogeneous graph of learning behaviors at different times to describe students' richer learning behavior characteristics, wherein the heterogeneous graph of learning behaviors includes three types of nodes: students, exercises, and knowledge points, and the interactive behavior relationships between them;

[0016] The heterogeneous graphs of learning behaviors at different times are aggregated into a sequence set, denoted as G = (G 1 ,G 2 ,…,G t ,…,G T ), where G T Representing the heterogeneous graph of learning behaviors constructed at time t, the learning behavior sequence set G can characterize the dynamic evolution of students’ knowledge cognition from the time and space dimensions.

[0017] Based on the above method, optionally, the knowledge forgetting simulation algorithm based on reinforcement learning, which trains the intelligent agent to learn the knowledge forgetting law of students in the learning process according to the reward mechanism, includes:

[0018] Construct a knowledge forgetting simulation algorithm based on reinforcement learning, and train the intelligent agent to discard the learning behavior heterogeneous graph G according to the reward mechanismT One or more "student-knowledge point" edges in simulate the process of students forgetting knowledge. Among them, the discarded "student-knowledge point" edges indicate that the students have forgotten the knowledge point. The learning behavior sequence after knowledge forgetting is recorded as

[0019] The knowledge forgetting simulation algorithm dynamically selects the knowledge points that students forget by strengthening the learning reward mechanism, and fully considers the knowledge forgetting characteristics of students in the learning process.

[0020] Based on the above method, optionally, evaluating the students' knowledge cognition levels at different times according to the learning behavior sequence integrating the students' forgetting rules includes:

[0021] Construct a learning behavior heterogeneous graph embedding method based on double-layer attention learning, and calculate the vector representation Z of nodes and relationships in the learning behavior heterogeneous graph t ,Z t Represents the heterogeneous graph of learning behavior at time t Embedded representation of ;

[0022] Aggregate the learned learning behavior heterogeneous graph vector representation Z = (Z1, Z2, ..., Z t ,…,Z T );

[0023] The temporal recurrent neural network is used to learn the dynamic evolution patterns of heterogeneous graphs of learning behaviors in different time series to evaluate students' knowledge and cognitive level.

[0024] Based on the above method, optionally, the use of a finite state automaton to construct a knowledge sequential graph to construct the superior-subordinate and precedent-successor relationships between knowledge points includes:

[0025] Use Jieba word segmentation and NLTK word segmentation tools to automatically extract knowledge points from electronic documents, including Chinese and English textbooks, reference books, and video lectures;

[0026] The knowledge sequential graph is automatically constructed using syntactic dependency trees and finite state automata, the dependencies between knowledge points are extracted, and the superior-subordinate and precedence-successor dependencies between knowledge points are obtained.

[0027] Based on the above method, optionally, the learning path recommendation method based on Monte Carlo search enhancement is adopted to recommend personalized learning paths for students in an unobservable environment according to the knowledge cognition level and the knowledge sequence diagram, including:

[0028] The student's learning process is modeled as a reinforcement learning problem in an unobservable environment. According to the student's knowledge level at time t, the corresponding knowledge point K is matched in the knowledge sequence graph. t ;

[0029] Generate and save knowledge points K in the knowledge sequential graph t A set of successor knowledge points connected to each other is selected from the set of successor knowledge points using the Monte Carlo search strategy. t+1 Join the learning path;

[0030] Re-evaluate students’ learning of knowledge point K t+1 The knowledge level after t+1 is calculated from K t+1 Recommend the knowledge point K that students need to learn at time t+2 from the set of subsequent knowledge points t+2 ;

[0031] Monte Carlo search and knowledge cognition diagnosis are performed alternately to recommend effective and reasonable personalized learning paths for students based on their latest knowledge cognition level in an unobservable environment.

[0032] In a second aspect, the embodiment of the present application further provides a personalized learning path recommendation device based on enhanced cognitive diagnosis, which includes:

[0033] Learning behavior modeling unit, used to construct learning behavior heterogeneous graphs to characterize students' richer learning behavior characteristics;

[0034] The knowledge forgetting simulation unit is used for the knowledge forgetting simulation algorithm based on reinforcement learning, which trains the intelligent agent to learn the knowledge forgetting rules of students in the learning process according to the reward mechanism;

[0035] Knowledge cognition diagnosis unit, used to evaluate students' knowledge cognition levels at different times based on a learning behavior sequence that incorporates students' forgetting rules;

[0036] A knowledge sequential graph construction unit is used to construct a knowledge sequential graph using a finite state automaton to construct the superior-subordinate and superior-superordinate and predecessor-successor relationships between knowledge points;

[0037] A learning path recommendation unit is used to recommend personalized learning paths for students in an unobservable environment according to the knowledge cognition level and the knowledge sequence graph by adopting a learning path recommendation method based on Monte Carlo search enhancement.

[0038] In the personalized learning path recommendation method and device based on enhanced cognitive diagnosis provided by this application, a learning behavior heterogeneous graph is constructed to characterize the richer learning behavior characteristics of students; a knowledge forgetting simulation algorithm based on reinforcement learning is used to train the intelligent agent to learn the knowledge forgetting law of students in the learning process according to the reward mechanism; the students' knowledge cognition level at different times is evaluated according to the learning behavior sequence that integrates the students' forgetting law; a knowledge sequential graph is constructed using a finite state automaton to construct the superior and subordinate and precedence relationships between knowledge points; a learning path recommendation method based on Monte Carlo search enhancement is used to recommend personalized learning paths for students according to the knowledge cognition level and the knowledge sequential graph in an unobservable environment. This application can accurately evaluate the students' cognitive level and automatically generate a knowledge sequential graph, meeting the practical needs of recommending reasonable and effective personalized learning paths to students in real time in an unobservable environment.

[0039] More specifically, the technical solution provided by this application has the following beneficial effects:

[0040] 1) Breaking the drawback of traditional personalized learning path recommendation methods that formalize learning path recommendation as observable reinforcement learning, formalize learning path recommendation as a more realistic unobservable reinforcement learning problem, and designing a learning path recommendation method based on Monte Carlo search enhancement to meet the practical needs of recommending reasonable and effective personalized learning paths for students in unobservable environments.

[0041] 2) Use time-series heterogeneous graphs to model the intrinsic interactions between students, exercises, and knowledge points, and design a knowledge forgetting simulation algorithm based on reinforcement learning; solve the drawbacks of traditional forget gate neural networks that arbitrarily lose knowledge points, break through the problem that forget gate neural networks cannot consider the continuity of knowledge forgetting, and learn more realistic and continuous knowledge forgetting rules through reinforcement learning reward strategies.

[0042] 3) Design a method for automatically constructing a knowledge sequential graph to get rid of the dependence on expert knowledge and manual relationship extraction, and use finite state automata to automatically extract the context and precedence and successor dependency relationships between knowledge points from the knowledge description text.

[0043] 4) It has been proved through the use of the prototype system that the solution of this application can effectively diagnose students' knowledge cognition level, and recommend reasonable and effective personalized learning paths for them based on their knowledge cognition level and knowledge sequence diagram. Moreover, the solution described in the embodiment of this application is easy to deploy in the existing online education system, simple to operate, safe and reliable, and has significant economic and social benefits and broad market promotion and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification are used to explain the principles of the present application. In addition, these drawings and text descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application for those skilled in the art by referring to specific embodiments.

[0045] Figure 1 A flowchart of a personalized learning path recommendation method based on enhanced cognitive diagnosis provided in one embodiment of the present application;

[0046] Figure 2 A schematic diagram of seven meta-paths provided for one embodiment of the present application;

[0047] Figure 3 A schematic diagram of a learning path recommendation principle provided for an embodiment of the present application;

[0048] Figure 4 A schematic diagram of the structure of a personalized learning path recommendation device based on enhanced cognitive diagnosis provided by an embodiment of the present application;

[0049] Figure 5 A schematic diagram of the operating principle of a personalized learning path recommendation device based on enhanced cognitive diagnosis provided by one embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0051] In view of the shortcomings of the above-mentioned prior art, this application proposes a personalized learning path recommendation scheme based on enhanced cognitive diagnosis, combining the student's knowledge forgetting characteristics and learning cognitive laws, analyzing the sequential characteristics of the student's knowledge cognitive level and knowledge structure, and recommending personalized learning paths for students. The basic idea is: first, use heterogeneous time sequence diagrams to model the relationship between students, exercises, and knowledge points, and characterize the dynamic evolution pattern between students and knowledge points from the time and space dimensions; then, design a knowledge cognitive diagnosis method that integrates students' forgetting behavior, and evaluate students' knowledge cognitive level in real time according to their historical learning records and forgetting laws; secondly, use finite state automata to construct knowledge sequential graphs, and construct the upper and lower and pre-repair and subsequent relationships between knowledge points; finally, design a learning path recommendation method based on Monte Carlo search enhancement, and recommend personalized learning paths for students according to the student's cognitive level and knowledge sequential graph in an unobservable environment. Based on this, the student's cognitive level can be accurately evaluated and the knowledge sequential graph can be automatically generated, meeting the practical needs of recommending reasonable and effective personalized learning paths for students in real time in an unobservable environment. The following is a non-restrictive description of the specific implementation scheme through several examples or embodiments.

[0052] Reference Figure 1 , Figure 1 The flowchart of the personalized learning path recommendation method based on enhanced cognitive diagnosis provided by one embodiment of the present application is shown in FIG. The method of this embodiment can be executed by a computer or a server. Figure 1 As shown, the method comprises at least the following steps:

[0053] Step S101: construct a learning behavior heterogeneous graph to characterize students' richer learning behavior characteristics.

[0054] Specifically, a heterogeneous graph is a modeling method for complex interactive systems, which is used to model various types of objects and the complex associations between them. In this embodiment, it is mainly used to construct three types of nodes, namely students, exercises and knowledge points, and the interactive behavior relationships between them, namely, a learning behavior heterogeneous graph, which is used to characterize the rich learning behavior characteristics of students.

[0055] Furthermore, in some embodiments, the specific implementation process of step S101 includes:

[0056] Construct a heterogeneous graph of learning behavior at different times to describe the richer learning behavior characteristics of students. The heterogeneous graph of learning behavior includes three types of nodes: students, exercises, and knowledge points, as well as the interactive behavior relationships between them. Gather the heterogeneous graphs of learning behavior at different times into a sequence set, denoted as G = (G 1 ,G 2 ,…,G t ,…,G T ), where G TRepresents the learning behavior heterogeneous graph constructed at time t. The learning behavior sequence set G can characterize the dynamic evolution of students' knowledge cognition from the time and space dimensions.

[0057] Specifically, in the personalized learning path recommendation task, the student’s learning behavior at time t can be formalized as a learning behavior heterogeneous graph G T =(V t ,E t ), where V t represents a node set, E t Represents the edge set, each node v∈V t A specific node type belonging to the node type set T:ψ(v)∈T, which includes three types: students, exercises, and knowledge points. Each edge e∈V t A specific edge type belonging to the edge type set R:Ψ(ν)∈R. Based on the formal description of the learning behavior heterogeneous graph, learning behavior triples are iteratively extracted from the student learning records to construct the association relationship between students, exercises and knowledge points. The learning behavior triples include (student i , learning, knowledge points j ),(student i ,Correct answer,Exercises j ),(student i ,Wrong Answer,Exercises j ),(exercise i ,Investigation,Knowledge Points j The learning behavior heterogeneous graphs at different times can be aggregated into a learning behavior heterogeneous graph sequence set G = (G 1 ,G 2 ,…,G t ,…,G T ).

[0058] Step S102: Based on the knowledge forgetting simulation algorithm of reinforcement learning, the intelligent agent is trained to learn the knowledge forgetting rules of students in the learning process according to the reward mechanism.

[0059] Specifically, humans will forget some of the knowledge they have learned during the process of learning knowledge. If the knowledge forgetting characteristics of students are ignored, the knowledge cognition diagnosis is inaccurate, which will seriously affect the accuracy and rationality of personalized learning path recommendation. Existing technologies often introduce forget gates in neural networks to randomly discard students' learned knowledge points, roughly simulating students' knowledge forgetting characteristics, exposing the drawbacks of randomness and discontinuity. In order to simulate the continuous knowledge forgetting process of each student more scientifically and realistically, this application designs a knowledge forgetting simulation algorithm based on reinforcement learning.

[0060] Furthermore, in some embodiments, the specific implementation process of step S102 includes:

[0061] Construct a knowledge forgetting simulation algorithm based on reinforcement learning, and train the intelligent agent to discard the learning behavior heterogeneous graph G according to the reward mechanism T One or more "student-knowledge point" edges in simulate the process of students forgetting knowledge. Among them, the discarded "student-knowledge point" edges indicate that the students have forgotten the knowledge point. The learning behavior sequence after knowledge forgetting is recorded as The knowledge forgetting simulation algorithm dynamically selects the knowledge points that students forget by strengthening the learning reward mechanism, and fully considers the knowledge forgetting characteristics of students in the learning process.

[0062] Specifically, this embodiment designs a knowledge forgetting simulation algorithm based on reinforcement learning, and formalizes the knowledge forgetting process into a Markov decision process MDP=<S,A,P,R,r> , where S = {s i} represents the set of all learning behavior heterogeneous graph states, A={a i} represents the set of activities for selecting subsequent knowledge points, P is the state transition operation, R(s t ) indicates reaching state s t The knowledge forgetting process is modeled as a continuous state transition sequence (s0, a0, r0, …, s n ,a n ,r n ,), where s n Represents the final generated learning behavior heterogeneous graph.

[0063] Each update process can be expressed as

[0064] This embodiment is designed to guide the knowledge forgetting process based on the adversarial reward function, which is formalized as follows:

[0065]

[0066] Among them, π θ is a reinforcement learning policy network, D φ represents the discriminator network, x represents the learning behavior heterogeneous graph input, P data Represents data distribution. Given a learning behavior heterogeneous graph sequence (G 1 ,G 2 ,…,G t ,…,G T ) as input, based on the adversarial reward function and the policy network π θ ,Automated Pruning(G 1 ,G 2 ,…,G t ,…,G T) in a certain number of "students i -Knowledge Points j "The side information pruned by the strategy network indicates that student i has forgotten knowledge point j. The learning behavior sequence after knowledge forgetting is recorded as Learning behavior sequences that incorporate forgetting characteristics can more accurately assess students' knowledge and cognitive levels.

[0067] Step S103: Evaluate students' knowledge cognition levels at different times based on a learning behavior sequence that incorporates students' forgetting rules.

[0068] Specifically, after obtaining the students' learning behavior characteristics and knowledge forgetting rules in the aforementioned steps, the two can be integrated to evaluate the students' knowledge cognition levels at different times.

[0069] Furthermore, the specific implementation process of step S103 includes:

[0070] Construct a learning behavior heterogeneous graph embedding method based on double-layer attention learning, and calculate the vector representation Z of nodes and relationships in the learning behavior heterogeneous graph t ,Z t Represents the heterogeneous graph of learning behavior at time t The embedding representation of the learning behavior heterogeneous graph vector Z = (Z1, Z2, ..., Z t ,…,Z T ); Use time series recurrent neural networks to learn the dynamic evolution patterns of heterogeneous graphs of learning behaviors in different time series to evaluate students' knowledge and cognitive level.

[0071] Specifically, in the embodiment of the present application, a double-layer attention graph neural network learning algorithm is first proposed to learn the vector representation of the heterogeneous graph of learning behavior at different moments from two dimensions inside and outside the meta-path. Based on the double-layer attention graph neural network, the corresponding embedding representation Z=(Z1, Z2, ..., Z t ,…,Z T ). In the embodiment of the present application, the knowledge cognition diagnosis is transformed into a time series heterogeneous graph link prediction problem solution, and the student's historical learning behavior sequence is represented by Z = (Z1, Z2, ..., Z t ,…,Z T ) is input into the temporal recurrent neural network to learn students’ knowledge cognition patterns and evaluate their knowledge cognition levels.

[0072] More specifically, refer to Figure 2 In the embodiment of the present application, 7 metapaths are first designed to explore the relationship between students, exercises and knowledge points, and then a double-layer attention graph neural network is designed to learn the heterogeneous graph embedding representation of learning behavior at different times Z = (Z1, Z2, ..., Zt ,…,Z T ).

[0073] Among them, in the learning behavior heterogeneous graph, through the meta-path p m The two nodes h connected i and h j The node-level attention weight is recorded as Formally:

[0074]

[0075] in, Represents node h i For node h j The importance of node represents the node-level attention neural network, h i and h j Represent the feature vectors of node i and node j respectively, p m is the connection node h i and h j A meta-path instance of . The normalized node-level attention weights can be converted to:

[0076]

[0077] Among them, || represents the aggregation operation, σ is the activation function, is the meta-path p m The attention vector of node h i In the metapath p m The embedding representation below can be formalized as:

[0078]

[0079] Different meta-paths have different contributions to the heterogeneous graph of characterizing learning behaviors. In view of this, the attention weight values ​​of different meta-paths are further learned in the embodiment of the present application. Given a meta-path set P = {p1, p2, ..., p m ,…,p M}, the attention weights of different meta-paths can be formalized as:

[0080]

[0081] in, Represents the meta-path p m For the importance of embedding graph nodes, is the node-level embedding set, σ path is the meta-path level activation function. The normalized meta-path weight coefficient It can be expressed as:

[0082]

[0083] in, Indicates that in the meta-path p m The weight vector under the instance is formalized as:

[0084]

[0085] Then, the final node embedding of the learning behavior heterogeneous graph at time t is formalized as:

[0086]

[0087] Then, the learning behavior sequence after knowledge forgetting is The corresponding embedding representation is denoted as Z = (Z1, Z2, ..., Z t ,…,Z T ).

[0088] Furthermore, in the embodiment of the present application, the knowledge cognition diagnosis is transformed into a time series heterogeneous graph link prediction problem solution, and the student's historical learning behavior sequence is represented by Z = (Z1, Z2, ..., Z t ,…,Z T ) is input into the temporal recurrent neural network to learn students’ knowledge cognition patterns and evaluate their knowledge cognition levels.

[0089] Step S104: Use a finite state automaton to construct a knowledge sequential graph to construct the hierarchical and superior and precedence relationships between knowledge points.

[0090] Specifically, the knowledge sequential graph can describe the hierarchical and sequential relationships between knowledge points, which is the basic work for personalized learning path recommendation. This step proposes an automatic construction method of knowledge sequential graph based on finite state automata to automatically extract the sequential relationships between knowledge points from the knowledge description text.

[0091] Furthermore, the specific implementation method of step S104 may include:

[0092] Jieba word segmentation and NLTK (Natural Language Toolkit) word segmentation tools are used to automatically extract knowledge points from electronic documents, including Chinese and English textbooks, reference books and video lectures. Syntactic dependency trees and finite state automata are used to automatically construct knowledge sequence graphs, extract dependency relationships between knowledge points, and obtain superior-subordinate and superior-subordinate dependency relationships between knowledge points.

[0093] Specifically, in the embodiments of the present application, the finite state automaton is formalized as a five-tuple A = <Φ, S, T, α0, α ∞>, which assigns a weight to each knowledge point state transition, where each dimension represents the following meaning: Φ represents a finite knowledge point input set |Φ|=V; S represents a finite knowledge point state set |S|=K; T∈V V×K×K represents the knowledge point state transfer weight vector, T[σ,I,J] is the state of the knowledge point from s to s when σ is input i Jump to knowledge point status j The transfer weight, where T[σ]∈R K×K represents the transfer matrix of σ; α0∈R K represents the initialization state weight, where α0[i] represents the initialization time t = 0 when s i The state weight of ∞ ∈R K represents the final knowledge point state weight, α ∞ [i] is the time after the system receives all inputs i The state weight of the knowledge point.

[0094] More specifically, for a knowledge point description sequence x, the weight of the weighted finite automaton after receiving x can be calculated using the forward algorithm and the Viterbi algorithm. <u1,…,u n+1 > represents the knowledge point state index sequence of x, then the state weight of the knowledge point in path P can be formalized as:

[0095]

[0096] Among them, α0[u1] represents the weight of x in the initial state of knowledge point u1, α ∞ [u n+1 ] indicates reaching knowledge point u n+1 The final weight value of the state. In the embodiment of the present application, the forward algorithm is used to calculate the sum of the weights of the knowledge transfer path, which is formalized as:

[0097]

[0098] Among them, ∏(x) represents the transition from the initial knowledge point state s0 to the final state s after traversing the knowledge point description text x. i ∈S ∞ Then use the Viterbi algorithm to calculate the knowledge point transfer path with the largest state weight:

[0099]

[0100] The knowledge point transfer path contains the sequential relationship between knowledge points. Given electronic files such as textbooks, reference books, video handouts, and teaching guides, the finite state automaton can automatically extract the superior-subordinate and predecessor-successor relationships between knowledge points, and integrate the extracted knowledge points and sequential relationship increments into the knowledge sequential graph KG.

[0101] Step S105: adopting a learning path recommendation method based on Monte Carlo search enhancement to recommend personalized learning paths for students in an unobservable environment according to their knowledge cognition level and knowledge sequence graph.

[0102] Specifically, this embodiment converts the learning path recommendation problem into an unobservable environment reinforcement learning problem, and discloses a learning path recommendation method based on Monte Carlo search enhancement, which recommends personalized learning paths for students in an unobservable environment according to their knowledge cognition level and knowledge sequence graph.

[0103] Furthermore, the specific implementation method of step S105 includes:

[0104] The student's learning process is modeled as a reinforcement learning problem in an unobservable environment. According to the student's knowledge level at time t, the corresponding knowledge point K is matched in the knowledge sequence graph. t ; Generate knowledge points K in the knowledge sequential graph t A set of successor knowledge points connected to each other is selected from the set of successor knowledge points using the Monte Carlo search strategy. t+1 Add to the learning path; evaluate again if students have completed knowledge point K t+1 The knowledge level after t+1 is calculated from K t+1 Recommend the knowledge point K that students need to learn at time t+2 from the set of subsequent knowledge points t+2 ; Alternately perform Monte Carlo search and knowledge cognition diagnosis to recommend effective and reasonable personalized learning paths for students based on their latest knowledge cognition level in an unobservable environment.

[0105] Specifically, according to the knowledge cognition level of the student at time t obtained in the aforementioned step S103, it is matched to the corresponding knowledge point K in the knowledge sequential graph. t , and then generate the knowledge point K in the knowledge sequential graph t A set of successor knowledge points connected to each other is selected from the set of successor knowledge points using the Monte Carlo search strategy. t+1 Add to the learning path and formalize as:

[0106]

[0107] Among them, score(a) is used to evaluate the gain value of executing activity a, N(a) is the number of times action a has been executed, and Q(a) is the action value calculated by the previous N(a) simulations, which is determined by the win rate and value function. a∈A describes the activity of selecting a certain knowledge point from the set of candidate knowledge points to add to the learning path, and s∈S represents a specific knowledge point in the set of all knowledge points. In the personalized learning path recommendation task, learning path planning is a typical unobservable environment for reinforcement learning agents, so the state set s in the policy network π(a|s; θ) is unobservable. In view of this, this embodiment allows the policy network to use all the collected historical observations O 1:t =[O1,O2,…,O t ,] to make decisions, the policy network can be formalized as:

[0108] f t =π(a t |O 1:t ;θ)(13)

[0109] Among them, historical observation O 1:t =[O1,O2,…,O t ,] is obtained by training the recurrent neural network. According to the student’s cognitive level at time t, the corresponding knowledge point K in the knowledge sequence graph is matched t , and then generate the knowledge point K in the knowledge sequential graph t The score(a) of all actions is calculated according to formula (12), and the knowledge point with the largest score(a) is selected from the set of successor knowledge points as K using the Monte Carlo search strategy. t+1 Add to Learning Path.

[0110] like Figure 3 As shown in the figure, when the student has mastered the linear equation at time t, the knowledge point of the linear equation system with two variables (corresponding to activity a1) should be selected first according to the gain value of score (a) to add to the subsequent learning path K t+1 Then, the knowledge cognitive diagnosis method is used to evaluate the students’ learning of knowledge point K again. t+1 The knowledge level after that, and then according to the knowledge level at time t+1, from K t+1 Recommend the knowledge point K that students need to learn at time t+2 in the subsequent knowledge point set t+2 After that, Monte Carlo tree search and knowledge cognition diagnosis are continuously and alternately performed to recommend effective and reasonable personalized learning paths for students based on their latest cognitive status in an unobservable environment.

[0111] In summary, based on the above scheme, by constructing a heterogeneous learning behavior graph to characterize students' richer learning behavior characteristics; based on the knowledge forgetting simulation algorithm of reinforcement learning, the intelligent agent is trained to learn the knowledge forgetting law of students in the learning process according to the reward mechanism; the students' knowledge cognition level at different times is evaluated according to the learning behavior sequence that integrates the students' forgetting law; the knowledge sequence graph is constructed using a finite state automaton to construct the superior and subordinate and precedence relationships between knowledge points; the learning path recommendation method based on Monte Carlo search enhancement is adopted to recommend personalized learning paths for students according to the knowledge cognition level and the knowledge sequence graph in an unobservable environment. This application can accurately evaluate the students' cognitive level and automatically generate a knowledge sequence graph to meet the practical needs of recommending reasonable and effective personalized learning paths to students in real time in an unobservable environment.

[0112] More specifically, the technical solution provided by this application has the following beneficial effects:

[0113] 1) Breaking the drawback of traditional personalized learning path recommendation methods that formalize learning path recommendation as observable reinforcement learning, formalize learning path recommendation as a more realistic unobservable reinforcement learning problem, and designing a learning path recommendation method based on Monte Carlo search enhancement to meet the practical needs of recommending reasonable and effective personalized learning paths for students in unobservable environments.

[0114] 2) Use time-series heterogeneous graphs to model the intrinsic interactions between students, exercises, and knowledge points, and design a knowledge forgetting simulation algorithm based on reinforcement learning; solve the drawbacks of traditional forget gate neural networks that arbitrarily lose knowledge points, break through the problem that forget gate neural networks cannot consider the continuity of knowledge forgetting, and learn more realistic and continuous knowledge forgetting rules through reinforcement learning reward strategies.

[0115] 3) Design a method for automatically constructing a knowledge sequential graph to get rid of the dependence on expert knowledge and manual relationship extraction, and use finite state automata to automatically extract the context and precedence and successor dependency relationships between knowledge points from the knowledge description text.

[0116] 4) It has been proved through the use of the prototype system that the solution of this application can effectively diagnose students' knowledge cognition level, and recommend reasonable and effective personalized learning paths for them based on their knowledge cognition level and knowledge sequence diagram. Moreover, the solution described in the embodiment of this application is easy to deploy in the existing online education system, simple to operate, safe and reliable, and has significant economic and social benefits and broad market promotion and application prospects.

[0117] In addition, the present application embodiment provides a personalized learning path recommendation device based on enhanced cognitive diagnosis, referring to Figure 4 , the device comprises:

[0118] The learning behavior modeling unit 41 is used to construct a learning behavior heterogeneous graph to describe students' richer learning behavior characteristics;

[0119] A knowledge forgetting simulation unit 42 is used for a knowledge forgetting simulation algorithm based on reinforcement learning, and trains the intelligent agent to learn the knowledge forgetting law of students in the learning process according to a reward mechanism;

[0120] The knowledge cognition diagnosis unit 43 is used to evaluate the students' knowledge cognition level at different times according to the learning behavior sequence integrating the students' forgetting rules;

[0121] A knowledge sequential graph construction unit 44 is used to construct a knowledge sequential graph using a finite state automaton to construct the superior-subordinate and superior-subordinate relationships between knowledge points;

[0122] The learning path recommendation unit 45 is used to recommend personalized learning paths for students in an unobservable environment according to the knowledge cognition level and the knowledge sequence graph by adopting a learning path recommendation method based on Monte Carlo search enhancement.

[0123] Among them, regarding the specific implementation methods of each module of the above-mentioned personalized learning path recommendation device based on enhanced cognitive diagnosis, reference can be made to the corresponding contents in the aforementioned method embodiment, which will not be repeated here.

[0124] in addition, Figure 5 The operating principle of the personalized learning path recommendation device based on enhanced cognitive diagnosis is shown.

[0125] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0126] It should be noted that, in the description of the embodiments of the present application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the embodiments of the present application, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0127] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred implementation in the embodiments of the present application includes alternative implementations in which the functions may not be performed in the order shown or discussed, including performing the functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments in the embodiments of the present application belong.

[0128] It should be understood that the various parts in the embodiments of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0129] A person skilled in the art may understand that all or part of the steps in the above-mentioned embodiment method may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0130] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The above-mentioned storage medium can be a read-only memory, a disk or an optical disk, etc.

[0131] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiments of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0132] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the embodiments of the present application. Ordinary technicians in the field can change, modify, replace and modify the above embodiments within the scope of the embodiments of the present application.

Claims

1. A personalized learning path recommendation method based on enhanced cognitive diagnosis, characterized in that: include: Construct a learning behavior heterogeneous graph to describe students' richer learning behavior characteristics; The knowledge forgetting simulation algorithm based on reinforcement learning trains the intelligent agent to learn the knowledge forgetting rules of students in the learning process according to the reward mechanism; Evaluate students' knowledge and cognition levels at different times based on a learning behavior sequence that incorporates students' forgetting rules; Use finite state automata to construct a knowledge sequential graph to build the hierarchical and sequential relationships between knowledge points; A learning path recommendation method based on Monte Carlo search enhancement is adopted to recommend personalized learning paths for students in an unobservable environment according to the knowledge cognition level and the knowledge sequence graph.

2. The method according to claim 1, characterized in that The construction of the learning behavior heterogeneous graph is used to describe the students' richer learning behavior characteristics, including: Constructing a heterogeneous graph of learning behaviors at different times to describe students' richer learning behavior characteristics, wherein the heterogeneous graph of learning behaviors includes three types of nodes: students, exercises, and knowledge points, and the interactive behavior relationships between them; The heterogeneous graphs of learning behaviors at different times are aggregated into a sequence set, denoted as G = (G 1 ,G 2 ,…,G t ,…,G T ), where G T Representing the heterogeneous graph of learning behaviors constructed at time t, the learning behavior sequence set G can characterize the dynamic evolution of students’ knowledge cognition from the time and space dimensions.

3. The method according to claim 2, characterized in that The reinforcement learning-based knowledge forgetting simulation algorithm trains the intelligent agent to learn the knowledge forgetting rules of students in the learning process according to the reward mechanism, including: Construct a knowledge forgetting simulation algorithm based on reinforcement learning, and train the intelligent agent to discard the learning behavior heterogeneous graph G according to the reward mechanism T One or more "student-knowledge point" edges in simulate the process of students forgetting knowledge. The discarded "student-knowledge point" edges indicate that students have forgotten the knowledge point. The learning behavior sequence after knowledge forgetting is recorded as The knowledge forgetting simulation algorithm dynamically selects the knowledge points that students forget by strengthening the learning reward mechanism, and fully considers the knowledge forgetting characteristics of students in the learning process.

4. The method according to claim 3, characterized in that The method of evaluating students' knowledge cognition levels at different times according to the learning behavior sequence integrating students' forgetting rules includes: Construct a learning behavior heterogeneous graph embedding method based on double-layer attention learning, and calculate the vector representation Z of nodes and relationships in the learning behavior heterogeneous graph t ,Z t Represents the heterogeneous graph of learning behavior at time t Embedded representation of ; Aggregate the learned learning behavior heterogeneous graph vector representation Z = (Z1, Z2, ..., Z t ,…,Z T ); The temporal recurrent neural network is used to learn the dynamic evolution patterns of heterogeneous graphs of learning behaviors in different time series to evaluate students' knowledge and cognitive level.

5. The method according to claim 4, characterized in that The method of using a finite state automaton to construct a knowledge sequential graph to construct the upper and lower level and precedence and successor relationship between knowledge points includes: Use Jieba word segmentation and NLTK word segmentation tools to automatically extract knowledge points from electronic documents, including Chinese and English textbooks, reference books, and video lectures; The knowledge sequential graph is automatically constructed using syntactic dependency trees and finite state automata, the dependencies between knowledge points are extracted, and the superior-subordinate and precedence-successor dependencies between knowledge points are obtained.

6. The method according to claim 5, characterized in that The learning path recommendation method based on Monte Carlo search enhancement is used to recommend personalized learning paths for students in an unobservable environment according to the knowledge cognition level and the knowledge sequence graph, including: The student's learning process is modeled as a reinforcement learning problem in an unobservable environment. According to the student's knowledge level at time t, the corresponding knowledge point K is matched in the knowledge sequence graph. t ; Generate and save knowledge points K in the knowledge sequential graph t A set of successor knowledge points connected to each other is selected from the set of successor knowledge points using the Monte Carlo search strategy. tt1 Join the learning path; Re-evaluate students’ learning of knowledge point K t+1 The knowledge level after t+1 is calculated from K t+1 Recommend the knowledge point K that students need to learn at time t+2 from the set of subsequent knowledge points t+2 ; Monte Carlo search and knowledge cognition diagnosis are performed alternately to recommend effective and reasonable personalized learning paths for students based on their latest knowledge cognition level in an unobservable environment.

7. A personalized learning path recommendation device based on enhanced cognitive diagnosis, characterized in that: include: Learning behavior modeling unit, used to construct learning behavior heterogeneous graphs to characterize students' richer learning behavior characteristics; The knowledge forgetting simulation unit is used for the knowledge forgetting simulation algorithm based on reinforcement learning, which trains the intelligent agent to learn the knowledge forgetting rules of students in the learning process according to the reward mechanism; Knowledge cognition diagnosis unit, used to evaluate students' knowledge cognition levels at different times based on a learning behavior sequence that incorporates students' forgetting rules; A knowledge sequential graph construction unit is used to construct a knowledge sequential graph using a finite state automaton to construct the superior-subordinate and superior-superordinate and predecessor-successor relationships between knowledge points; A learning path recommendation unit is used to recommend personalized learning paths for students in an unobservable environment according to the knowledge cognition level and the knowledge sequence graph by adopting a learning path recommendation method based on Monte Carlo search enhancement.

Images