A method for dynamically updating teaching agent instructions based on Bayesian network

Abstract

The embodiment of the present invention discloses a method for dynamically updating teaching agent instructions based on a Bayesian network, comprising: obtaining a teaching business model, the teaching business model is used to characterize the relationship between the learner's situation and the teaching agent's instructions; based on the teaching business model, constructing a Bayesian network; inputting the learner's real-time learning situation into the Bayesian network for reasoning to obtain a teaching prompt word template and a teaching instruction label; according to the teaching prompt word template and the teaching instruction label, dynamically updating the prompt word and ability plug-in of the teaching agent to constrain the agent to finally output the teaching guidance content. This embodiment can dynamically update the prompt word and ability plug-in of the teaching agent.

Description

Technical Field

[0001] The embodiments of the present invention relate to the fields of artificial intelligence and educational technology, and in particular to a method for dynamically updating teaching agent instructions based on a Bayesian network. Background Art

[0002] In the development of artificial intelligence and educational technology, personalized teaching has become an important development trend in the field of online education. To achieve personalized teaching, the intelligent system needs to accurately evaluate the learner's cognitive ability and knowledge mastery, and adjust the teaching strategy for different learners based on the evaluation results. However, the existing intelligent strategies of online teaching platforms are based on pre-agreed rules, lacking a mechanism for dynamic adjustment based on process data, and a deep understanding and dynamic response to students' personalized needs.

[0003] Specifically, in the existing teaching agent technology stack, the teaching agents based on large models mainly focus on solving problems in specific vertical professional business scenarios, and are usually designed as a single responsibility with the goal of solving a specific teaching scenario. For example, these agents may be specifically used for teaching case guidance, subject knowledge questions and answers, test question explanations, etc. Their prompt words and integration capabilities are often static, preset, and non-personalized, and lack a dynamic update mechanism to adapt to the personalized needs of different learners and different learning stages. This limitation greatly limits the effect of personalized teaching. In the prior art, patent CN110297844A provides an artificial intelligence teaching system and data processing method based on a Bayesian model, and patent CN118821949A provides a multi-agent dynamic task allocation method, system and medium based on fuzzy reasoning, neither of which can solve the above technical problems. Summary of the invention

[0004] The embodiment of the present invention provides a method for dynamically updating teaching agent instructions based on a Bayesian network to solve the above technical problems.

[0005] In a first aspect, an embodiment of the present invention provides a method for dynamically updating teaching agent instructions based on a Bayesian network, comprising:

[0006] Acquire a teaching business model, wherein the teaching business model is used to characterize the relationship between the learner's situation and the teaching agent's instructions;

[0007] Based on the teaching business model, construct a Bayesian network;

[0008] Inputting the learner's real-time learning situation into the Bayesian network for reasoning to obtain a teaching prompt word template and a teaching instruction label;

[0009] According to the teaching prompt word template and the teaching instruction label, the prompt words and the ability plug-in of the teaching intelligent body are dynamically updated to constrain the intelligent body to finally output the teaching guidance content.

[0010] In a second aspect, an embodiment of the present invention provides an electronic device, the electronic device comprising:

[0011] one or more processors;

[0012] a memory for storing one or more programs,

[0013] When the one or more programs are executed by the one or more processors, the one or more processors implement the method of dynamically updating teaching agent instructions based on a Bayesian network as described in any embodiment.

[0014] In summary, the embodiment of the present invention provides a method for dynamically updating teaching agent instructions based on a Bayesian network, using a Bayesian network algorithm to select the most suitable teaching strategy, and dynamically adjusting and updating the prompt words and ability plug-ins in the teaching agent to provide learners with adaptive guidance. The method can achieve the following beneficial effects:

[0015] 1. The invention integrates machine learning and the large model agent framework to achieve effective connection between the two. Through this integration, the invention fully utilizes the advantages of adaptive technology in teaching data processing and analysis, and the capabilities of large model agents in generative artificial intelligence and personalized guidance. It solves the limitations of existing teaching agents in personalized guidance, and enables teaching agents to more accurately understand learners' cognitive states and learning needs, thereby providing more accurate and personalized teaching guidance.

[0016] 2. A dynamic update mechanism for teaching agents based on the Bayesian network algorithm is proposed. This mechanism can analyze learners' cognitive status, learning progress and target requirements in real time, and dynamically adjust the prompt words and plug-in capabilities in the teaching agent by combining multi-dimensional data such as knowledge graphs, cognitive measurements and cognitive goals. This dynamic update mechanism enables the teaching agent to flexibly adapt to the personalized needs of different learners under the premise of single responsibility design. Through this mechanism, the teaching agent can provide personalized teaching guidance based on learners' real-time feedback and cognitive status, just like a real teacher, optimizing the learners' online learning experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of a method for dynamically updating teaching agent instructions based on a Bayesian network provided by an embodiment of the present invention;

[0019] Figure 2 It is a schematic diagram of a Bayesian network-based dynamic update teaching agent solution model provided by an embodiment of the present invention;

[0020] Figure 3 It is a schematic diagram of a method for updating instructions based on Bayesian network reasoning and teaching agents provided by an embodiment of the present invention;

[0021] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0023] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0024] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0025] Figure 1 is a flow chart of a method for dynamically updating teaching agent instructions based on a Bayesian network provided by an embodiment of the present invention. The method is executed by an electronic device, such as Figure 1 As shown, the specific steps include:

[0026] S110, obtaining a teaching business model, wherein the teaching business model is used to characterize the relationship between the learner's situation and the teaching agent's instructions.

[0027] The teaching agent refers to an agent designed specifically for the teaching field based on an application framework of a large language model, specifically a software program or system that serves online teaching or personalized learning platforms and can dynamically adjust teaching content, methods and strategies according to learners' cognitive states, learning progress and target requirements. This embodiment aims to use the Bayesian network model to dynamically update the instructions of the teaching agent based on the learners' real-time learning status. Therefore, this step first obtains the teaching business model to guide the construction of the Bayesian network.

[0028] Figure 2 An exemplary teaching business model is shown. To illustrate the model, the key concepts in the model are explained first. Among them, cognitive objectives refer to the systematic summary of knowledge mastery by level based on teaching tasks. The main function is to clearly show learners their mastery of each knowledge point; optionally, cognitive objectives can be subdivided into six levels from low to high and from simple to complex: knowledge, comprehension, application, analysis, synthesis and evaluation.

[0029] Cognitive measurement refers to the process of quantitatively evaluating individual cognitive abilities and knowledge mastery. In this invention, it refers to the use of online teaching technology to evaluate learners' achievement of learning goals in a specific course, specifically covering knowledge mastery and cognitive ability assessment.

[0030] Teaching strategies refer to the teaching procedures and implementation measures carefully planned by intelligent algorithms based on knowledge graphs, cognitive goals, and cognitive measurement results in order to achieve teaching goals and adapt to students' cognitive needs during online teaching. Specific forms include case analysis, project practice, directional exercises, resource recommendations, and self-evaluation, etc. Its purpose is to provide learners with personalized adaptive guidance and improve the efficiency and quality of online teaching.

[0031] Based on the above key concepts, Figure 2 The teaching business model in is described as follows: learners learn in the intelligent teaching platform according to the established learning objectives and learning resources provided by the system. These learning resources and learning objectives have established corresponding relationships with the corresponding course knowledge points in the learning platform. After a period of learning, the system will conduct comprehensive and staged cognitive measurements based on the learner's behavioral data and evaluation data. Usually, the cycle of this measurement is usually based on the chapters of course learning. After generating cognitive measurement data, the system will select the most suitable teaching strategy for the current learner based on the measurement results. The teaching agent will dynamically update the prompt words and ability plug-ins according to the teaching strategy selected by the system. In this way, when the teaching agent outputs guidance information and strategies to learners, it can be done in a way that is personalized for the learners.

[0032] S120: constructing a Bayesian network based on the teaching business model.

[0033] A Bayesian network is a graphical network based on probabilistic reasoning, consisting of nodes (representing random variables) and directed edges (representing dependencies between variables). It describes the probability distribution between variables through a conditional probability table and is used to deal with uncertainty and incompleteness problems. Based on the above teaching business model, this embodiment constructs a Bayesian network suitable for teaching agent instruction recommendation through the following steps:

[0034] Step 1: Node modeling. Optionally, construct knowledge point nodes, knowledge point relationship nodes, cognitive goal nodes, learning behavior nodes, teaching task nodes, cognitive measurement nodes, and teaching strategy nodes. Specifically, the definition of each node is as follows:

[0035] Knowledge Point (KP): represents an independent knowledge point in the course. Each knowledge point k i There is a state S ki , indicating the learner’s mastery of the knowledge point.

[0036] Knowledge Point Relationship (KPR): represents the connection relationship between knowledge points, such as pre-relationship, parallel relationship, etc. These relationships are represented by conditional probability tables (CPTs), such as P(S kj |S ki ) indicates that at knowledge point k i When the state is known, knowledge point k j The probability distribution of the state.

[0037] Cognitive Dimension (CD): represents the learner's cognitive dimension, including knowledge, comprehension, application, analysis, synthesis and evaluation. i There is a difficulty level d j , forming a combination of cognitive goals and difficulty levels P(c i ,d j ).

[0038] Learning Behavior Data (LBD): includes learning behavior records (such as learning time, number of learning times, etc.), homework data (such as homework completion status, accuracy rate, etc.) and assessment data (such as test scores, assessment results, etc.).

[0039] Teaching Task Node (TT): represents a specific teaching task generated based on the knowledge graph and cognitive objectives. Each task t i There is a state S ti , indicating the completion of the task. Among them, the knowledge graph is a knowledge base that systematically organizes the knowledge points in the course and their relationships in a structured way. It not only displays the content of each knowledge point, but also depicts the logical relationship, hierarchical structure and interdependence between them, providing a comprehensive knowledge framework for online teaching and a basis for dynamically adjusting teaching strategies.

[0040] Cognitive Measurement (CM): represents the result of quantitative evaluation of learners’ cognitive state. This evaluation is based on knowledge points, learning behavior data, homework data, assessment data, and teaching task completion to form cognitive measurement state S. CM .

[0041] Teaching Strategy Node (TS): represents specific teaching strategies, such as case analysis, project practice, and directional practice. i There is a state S si , indicating the applicability of the strategy.

[0042] Step 2: Relationship modeling. Optionally, the relationship between nodes is represented by constructing directed edges. Specifically, directed edges represent the dependency between nodes. For example, from knowledge point node k i To the knowledge point relationship node KPR (k i , k j ) indicates that k i It is k j Pre-knowledge points; from the teaching task node TT(t i ) to the cognitive measurement node CM represents TT(t i ) affects the status of CM. In a specific implementation, the following directed edges can be constructed:

[0043] Knowledge point node (KP) and cognitive target node (CD): directed edge KP(k i ) → CD(c i ,d j ), represents knowledge point k i The degree of mastery affects the cognitive dimension c i At difficulty level d j This type of directed edge is called the first directed edge.

[0044] Cognitive target node (CD) and cognitive measurement node (CM): directed edge CD(c i ,d j ) → CM, indicating cognitive dimension c i At difficulty level d j The performance on the CM affects the result of cognitive measurement. This type of directed edge is called the second directed edge.

[0045] Learning behavior node (LBD) to cognitive measurement node (CM): Directed edge LBD → CM, indicating that learning behavior data (such as learning time, homework completion, etc.) affects the results of cognitive measurement CM. This type of directed edge is called the third directed edge.

[0046] From cognitive measurement node (CM) to teaching strategy node (TS): the directed edge CM → TS indicates that the result of cognitive measurement affects the selection of teaching strategy TS. This type of directed edge is called the fourth directed edge.

[0047] Step 3: Probability distribution modeling. Optionally, the probability distribution between node states can be described by a conditional probability table (CPT). For example, for a knowledge point relationship node KPR(k i ,k j ), its CPT can be expressed as P(S kj |S ki ), that is, at knowledge point k iWhen the state is known, knowledge point k j In a specific implementation, based on the above four types of directed edges, the following conditional probability tables can be constructed:

[0048] KP(k i ) → CD(c i ,d j )’s CPT: The key table structure is shown in Table 1, where each row represents the i Under different mastery levels, cognitive dimension c i At difficulty level d j The probability distribution on .

[0049] Table 1

[0050] <![CDATA[S ki (Knowledge point k i degree of mastery)]]> <![CDATA[P(c i ,d j |S ki )]]> Not mastered <![CDATA[P(c i ,d j |Not mastered)]]> Partial Mastery <![CDATA[P(c i ,d j |Partially mastered)]]> Complete control <![CDATA[P(c i ,d j |Fully mastered)]]>

[0051] CD(c i ,d j ) → CPT of CM: The key table structure is shown in Table 2. Each row represents the probability distribution of cognitive measurement CM under a specific combination of cognitive dimension and difficulty level.

[0052] Table 2

[0053] <![CDATA[(c i ,d j ) (combination of cognitive dimension and difficulty level)]]> <![CDATA[P(CM|(c i ,d j ))]]> <![CDATA[(K1,d1)]]> <![CDATA[P(CM|(K1,d1))]]> <![CDATA[(K1,d2)]]> <![CDATA[P(CM|(K1,d2))]]> ...... ...... <![CDATA[(E6,d n )]]> <![CDATA[P(CM|(E6,d n ))]]>

[0054] CPT of LBD → CM: The key table structure is shown in Table 3, where P1_low, P1_medium, P1_high, etc. represent the probability that the cognitive measurement is “low”, “medium”, or “high” under a specific combination of learning behavior data.

[0055] Table 3

[0056] Study time Homework completion Learning frequency P(CM=Low|LBD combination) P(CM=Medium|LBD combination) P(CM=High|LBD combination) short Difference Low Frequency <![CDATA[P1_Low]]> <![CDATA[P1_Chinese]]> <![CDATA[P1_High]]> short middle IF <![CDATA[P2_Low]]> <![CDATA[P2_Chinese]]> <![CDATA[P2_High]]> ...... ...... ...... ...... ...... ...... long good high frequency <![CDATA[P n _Low]]> <![CDATA[P n _zhong]]> <![CDATA[P n _High]]>

[0057] CPT of CM → TS: The key table structure is shown in Table 4. Each row represents the selection probability of the teaching strategy TS under a specific cognitive measurement state SCM.

[0058] Table 4

[0059] SCM (State of Cognitive Measurement) P(TS|SCM) Low P(TS|Low) middle P(TS|Medium) high P(TS|High)

[0060] S130, inputting the learner's real-time learning situation into the Bayesian network for reasoning to obtain a teaching prompt word template and a teaching instruction label.

[0061] After the Bayesian network is constructed, this step performs fuzzy reasoning based on the network, and obtains a teaching prompt word template and teaching instruction label suitable for the current learning situation of the current learner according to the learner's real-time learning situation. In a specific implementation, the process may include the following steps:

[0062] Step 1: Initialize the node state. Optionally, initialize the state of each node in the Bayesian network based on the input data (course knowledge graph, cognitive goals, learning behavior data, etc.). For knowledge point nodes and teaching task nodes, the initial state can be set based on historical data or teaching expert scores. For learning behavior data nodes, data can be directly obtained from the learning system.

[0063] Step 2: Update the state of cognitive measurement nodes. Optionally, based on the learner's learning behavior data, homework and assessment data, and the completion of teaching tasks, the state of cognitive measurement nodes is updated through the reasoning mechanism of the Bayesian network. Specifically, according to the conditional probability table P(S CM |S k1 ,S k2, ...,S kn ,S t1 ,S t2 ,...,S tm ), calculate the cognitive measurement state S CM The probability distribution of . Among them, P(S CM |S k1 ,S k2, ...,S kn ,S t1 ,S t2 ,...,S tm ) is a joint probability, S ki represents the state of knowledge point ki, S ti It represents the state of teaching task ti, n and m represent the number of knowledge points and teaching tasks respectively.

[0064] Step 3: Calculate the conditional probability of teaching strategies. Optionally, the conditional probability of different teaching strategies can be calculated by combining the state of knowledge point nodes, cognitive dimensions and difficulty levels, as well as cognitive measurement data, through the reasoning mechanism of the Bayesian network. Specifically, according to the conditional probability table P(S si |S k1 ,S k2 ,...,S kn ,C1,C2,...,C6,D1,D2,D3,S CM ), calculate each teaching strategy s i Status S si The probability distribution of x and D y denote the cognitive dimension and difficulty level indexed as x and y, respectively.

[0065] Step 4: Select the most appropriate teaching strategy. Optionally, based on the conditional probabilities of different teaching strategies, select the teaching strategy with the highest probability as the most appropriate teaching strategy. Specifically, calculate the applicability score Score(s) of each teaching strategy. i )=P(S si =1|S k1 ,S k2 ,...,S kn ,C1,C2,...,C6,D1,D2,D3,SCM), and select the teaching strategy with the highest score ∗ Among them, P(S si =1|...) represents the probability of teaching strategy si being selected under a given series of conditions; "S si =1" indicates that teaching strategy si is selected or considered applicable, whereas "S si =0” means that teaching strategy si was not selected or was not considered applicable.

[0066] Step 5: Get the teaching prompt word template and instruction label. ∗ , obtain the corresponding teaching prompt word template from the teaching prompt word template library. Exemplarily, a predefined teaching prompt word template library can be used to select a suitable teaching prompt word template according to the corresponding relationship between the type of teaching strategy and the difficulty level and the prompt word template.

[0067] At the same time, the instruction tag is used to maintain the type, behavior Action and corresponding parameters of the specific API call in the teaching agent. These instruction tags will guide the teaching agent on how to execute the selected teaching strategy and generate other teaching-related operations. For example, calling related knowledge point cases, calling related teaching activities, recommending suitable teaching resources and other behaviors. Exemplarily, the appropriate instruction tag can be selected according to the correspondence between the teaching prompt word template and the instruction tag.

[0068] S140. Dynamically update the prompt words and capability plug-in of the teaching agent according to the teaching prompt word template and the teaching instruction label, so as to constrain the agent to finally output the teaching guidance content.

[0069] After the reasoning of S130, the final output of the Bayesian network is: the most appropriate teaching strategy s ∗ As well as corresponding teaching prompt templates and instruction labels.

[0070] In a specific implementation, the teaching agent can dynamically adjust the parameters and content in the prompt word template based on the learner's learning situation and needs, generate personalized prompt words, and constrain the agent to finally output the teaching guidance content. At the same time, the teaching agent can dynamically adjust the activation and configuration of the agent's ability plug-in according to the teaching strategy suggestions output by the Bayesian network and the learner's real-time learning needs. For example, when a learner encounters difficulties in a certain knowledge point, the corresponding auxiliary learning plug-in is activated, such as animation demonstration, interactive exercises, etc. According to the learner's feedback and learning effect, the activation strategy and configuration parameters of the plug-in are continuously optimized.

[0071] Furthermore, the above-mentioned method for dynamically updating the teaching agent instructions based on the Bayesian network can be applied to different learning stages of the same learner, and the prompt words and ability plug-ins of the teaching agent are dynamically updated according to the learning conditions of the learners at different stages. That is, the real-time learning conditions of the learners at different learning stages are input into the Bayesian network for reasoning, and the teaching prompt word templates and teaching instruction labels suitable for different learning stages are obtained; and according to different teaching prompt word templates and teaching instruction labels, the prompt words and ability plug-ins of the teaching agent are dynamically updated to constrain the agent to finally output the teaching guidance content at different learning stages.

[0072] At the same time, the above-mentioned method for dynamically updating the teaching agent instructions based on the Bayesian network can be applied to different learners, and the prompt words and ability plug-ins of the teaching agent can be dynamically updated according to the learning conditions of different learners. That is, the conditions of different learners are input into the Bayesian network for reasoning, and the teaching prompt word templates and teaching instruction labels suitable for different learners are obtained; and according to different teaching prompt word templates and teaching instruction labels, the prompt words and ability plug-ins of the teaching agent are dynamically updated to constrain the agent to finally output teaching guidance content for different learners.

[0073] Therefore, the method of this embodiment can provide differentiated prompt words and ability plug-ins for different learning times or different learners, so that the teaching intelligent body can better output teaching guidance content.

[0074] Furthermore, in practical applications, we can continuously collect learners’ learning data to train and optimize the Bayesian network model. Specifically, we can use historical data to update the conditional probability table to improve the accuracy and reliability of the model. At the same time, we can also use machine learning algorithms to automatically adjust model parameters to adapt to new teaching environments and learners’ needs.

[0075] Furthermore, in another specific implementation, when constructing a Bayesian network, in addition to constructing knowledge point nodes, knowledge point relationship nodes, cognitive goal nodes, learning behavior nodes, teaching task nodes, cognitive measurement nodes, and teaching strategy nodes, a prompt word template node can also be constructed, and the prompt word template can be used as a member of the Bayesian network reasoning.

[0076] Correspondingly, when constructing directed edges, in addition to constructing the above four types of directed edges, a fifth type of directed edge can be constructed between the teaching strategy node and the prompt word template node, called the fifth directed edge; and a conditional probability table is constructed based on the fifth directed edge. The conditional probability table is used to describe the probability distribution of the prompt word template under a specific teaching strategy.

[0077] Correspondingly, in the Bayesian network reasoning, after executing step 4, the prompt word template with the highest probability under the most appropriate current teaching strategy can be determined according to the conditional probability table based on each fifth directed edge.

[0078] Optionally, each teaching strategy may correspond to a plurality of available prompt word templates, and the probability distribution of each available prompt word template may be represented by the mean and standard deviation of the Gaussian distribution, that is, for a specific teaching strategy, each fifth directed edge connected thereto corresponds to a set of means and standard deviations, and the sum of the means of all prompt word templates connected by the same teaching strategy is 1. Accordingly, when determining the prompt word template with the highest probability under the currently most suitable teaching strategy according to the conditional probability table based on each fifth directed edge, the probability of each prompt word template may be randomly generated according to the mean and standard deviation satisfied by the probability of each prompt word template from the currently most suitable teaching strategy; then, a set with the probability and the closest to 1 is selected as the final probability combination, and the prompt word template with the highest probability is selected from the final probability combination.

[0079] For example, if the most suitable teaching strategy currently has 10 prompt word template nodes connected, and each prompt word template node corresponds to the mean and standard deviation of a Gaussian distribution, then a value under the Gaussian distribution can be generated according to the mean and standard deviation of the Gaussian distribution corresponding to a prompt word template node, as the probability of the prompt word template node; the same operation is performed on the remaining 9 prompt word template nodes, and 9 probabilities can be formed respectively, and these 10 probabilities together constitute a group of probabilities. Repeating this operation N times can generate N groups of such probabilities, from which a group of 10 probabilities and the probability closest to 1 are selected as the final probability combination. The prompt word template with the highest probability is selected from the final probability combination as the prompt word template to be used.

[0080] Compared with other nodes of the Bayesian network, the probability distribution tables between other nodes need to be more accurately counted based on learning data, while the prompt word template node can be flexibly set according to teaching experience, and even because the prompt word is closer to the human-computer interaction link, there are higher requirements for vividness and flexibility. Therefore, this embodiment represents the probability of each prompt word template under a specific teaching strategy as a random distribution, which can not only improve the fuzzy reasoning ability of the network, but also improve the diversity of the prompt word template and enhance the students' experience in human-computer interaction.

[0081] Furthermore, in the process of using the Bayesian network, the learning data of the learners can be continuously collected to optimize the probability distribution table from the teaching strategy node to the prompt word template node. Optionally, the facial expressions, body movements and other information of the learners can be collected through the camera. If the collected information can identify that the learner is in a happy and positive state, it means that the current prompt word template has received positive feedback from the learner, and the mean corresponding to the prompt word template selected in this state can be increased and the standard deviation can be decreased; if the collected information can identify that the learner is in a confused and negative state, it means that the current prompt word template has received negative feedback from the learner, and the mean corresponding to the prompt word template selected in this state can be decreased and the standard deviation can be increased.

[0082] At the same time, in this process, if the similarity between the two prompt word templates is greater than the first threshold, and the corresponding mean difference is greater than the second threshold, it indicates that during the use of the two prompt word templates, the dynamic slots are filled with key information containing relatively different content. Then, the content filled in the dynamic slots of the two prompt word templates during use can be analyzed to separate the new slot fields. For example, under the teaching strategy of case analysis, one of the available prompt word templates is:

[0083] Please call up matching cases based on the students' [learning objectives] and [completion of recent learning tasks], and guide the students to conduct case analysis.

[0084] Another available prompt template is:

[0085] Please generate cases suitable for students based on their [cognitive goals] and [latest learning situation], and guide students to analyze the case elements step by step.

[0086] The similarity between the above two prompt word templates is relatively large, and the probability of being selected should also be relatively similar. However, if it is found that the mean values ​​in the probability distribution of the two prompt word templates are quite different according to the adjustment of the learning data, it indicates that there is a large difference in the probability of their selection. At this time, the content filled in the dynamic slots of the two prompt word templates during use can be extracted for analysis. For example, if it is finally found that the content filled in the [learning goal] of the first prompt word template includes [cognitive dimension] and [expected difficulty level for knowledge point A], and the content filled in the [cognitive goal] of the second prompt word template includes [cognitive dimension] and [expected difficulty level for knowledge point B], then it can be analyzed that "which key knowledge point is expected to have a high difficulty level" is an important reason for the large difference in the probability of the two prompt words being selected. At this time, the dynamic slot [target knowledge point] can be decomposed from the above-mentioned filled content and added to the two prompt word templates respectively to obtain the following two new prompt word templates:

[0087] Please call matching cases and guide students to conduct case analysis based on their [learning objectives], the expected difficulty level of [target knowledge points], and [completion of recent learning tasks].

[0088] Please generate cases suitable for students based on their [cognitive goals], the expected difficulty level of [target knowledge points], and [latest learning situation], and guide students to analyze the case elements step by step.

[0089] Through the above method, it is possible to continuously expand prompt word templates that are more in line with actual usage situations, and improve the prompt word template's ability to recognize and adapt to key teaching information.

[0090] It should be noted that the user data involved in this application (including facial expression data, body movement data, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0091] In summary, this embodiment provides a method for dynamically updating teaching agent instructions based on a Bayesian network. Based on relevant data such as knowledge graphs, cognitive measurements, cognitive goals, and learning behaviors, the Bayesian network algorithm is used to select the most suitable teaching strategy, and dynamically adjust and update the prompt words and ability plug-ins in the teaching agent to provide learners with adaptive guidance. This method can achieve the following beneficial effects:

[0092] 1. The invention integrates machine learning and the large model agent framework to achieve effective connection between the two. Through this integration, the invention fully utilizes the advantages of adaptive technology in teaching data processing and analysis, and the capabilities of large model agents in generative artificial intelligence and personalized guidance. It solves the limitations of existing teaching agents in personalized guidance, and enables teaching agents to more accurately understand learners' cognitive states and learning needs, thereby providing more accurate and personalized teaching guidance.

[0093] 2. A dynamic update mechanism for teaching agents based on the Bayesian network algorithm is proposed. This mechanism can analyze learners' cognitive status, learning progress and target requirements in real time, and dynamically adjust the prompt words and plug-in capabilities in the teaching agent by combining multi-dimensional data such as knowledge graphs, cognitive measurements and cognitive goals. This dynamic update mechanism enables the teaching agent to flexibly adapt to the personalized needs of different learners under the premise of single responsibility design. Through this mechanism, the teaching agent can provide personalized teaching guidance based on learners' real-time feedback and cognitive status, just like a real teacher, optimizing the learners' online learning experience.

[0094] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 4 As shown, the device includes a processor 60, a memory 61, an input device 62 and an output device 63; the number of processors 60 in the device can be one or more. Figure 4 A processor 60 is taken as an example; the processor 60, the memory 61, the input device 62 and the output device 63 in the device can be connected by a bus or other means. Figure 4 The example of connecting through bus is taken in the following.

[0095] The memory 61 is a computer-readable storage medium that can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the method for dynamically updating the teaching agent instructions based on the Bayesian network in the embodiment of the present invention. The processor 60 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 61, that is, realizing the above-mentioned method for dynamically updating the teaching agent instructions based on the Bayesian network.

[0096] The memory 61 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 61 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 61 may further include a memory remotely arranged relative to the processor 60, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0097] The input device 62 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 63 may include a display device such as a display screen.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamically updating teaching agent instructions based on a Bayesian network, characterized in that: include: Acquire a teaching business model, wherein the teaching business model is used to characterize the relationship between the learner's situation and the teaching agent's instructions; Based on the teaching business model, a Bayesian network is constructed, wherein the Bayesian network includes knowledge point nodes, knowledge point relationship nodes, cognitive target nodes, learning behavior nodes, teaching task nodes, cognitive measurement nodes, teaching strategy nodes and prompt word template nodes; Input the learner's real-time learning situation into the Bayesian network for reasoning to obtain a teaching prompt word template and a teaching instruction label; specifically, initialize the state of each node in the Bayesian network according to the course knowledge map, cognitive goals, and learning behavior data; dynamically update the state of the cognitive measurement node through the Bayesian network reasoning mechanism based on the learner's learning behavior data, homework and assessment data, and the completion of the teaching task; calculate the conditional probabilities of different teaching strategies through the Bayesian network reasoning mechanism in combination with the state of the knowledge point node, cognitive difficulty and difficulty level, and cognitive measurement data; select the teaching strategy with the highest probability as the most appropriate teaching strategy at present according to the conditional probabilities of different teaching strategies; determine the corresponding teaching prompt word template and instruction label according to the most appropriate teaching strategy at present; According to the teaching prompt word template and the teaching instruction label, dynamically update the prompt words and ability plug-in of the teaching agent to constrain the agent to finally output the teaching guidance content; The probability distribution of the prompt word template is represented by the mean and standard deviation of the Gaussian distribution, and the edges between each teaching strategy node and the prompt word template node correspond to a group of means and standard deviations respectively, and the sum of the means of the prompt word templates connected by the same teaching strategy is 1; accordingly, the method of determining the corresponding teaching prompt word template and instruction label according to the most suitable teaching strategy at present includes: randomly generating multiple groups of probabilities of each prompt word template according to the mean and standard deviation satisfied by the probability of the most suitable teaching strategy to each prompt word template, and selecting a group with the probability sum closest to 1 as the final probability combination; and selecting the prompt word template with the highest probability from the final probability combination; Correspondingly, after dynamically updating the prompt words and ability plug-ins of the teaching intelligent body according to the teaching prompt word template and the teaching instruction label, it also includes: collecting learners' learning data and updating the probability distribution table of the Bayesian network, wherein, if a prompt word template obtains positive feedback from the learner after use, the mean corresponding to the prompt word template is increased and the standard deviation is decreased; if a prompt word template obtains negative feedback from the learner after use, the corresponding mean is decreased and the standard deviation is increased; when the similarity between two prompt word templates is greater than a first threshold and the corresponding mean difference is greater than a second threshold, a new dynamic slot field is decomposed from the content filled in the dynamic slots of the two prompt word templates.

2. The method according to claim 1, characterized in that The cognitive target node includes a cognitive dimension and a difficulty level on the cognitive dimension; Accordingly, the Bayesian network is constructed based on the teaching business model, including: A first directed edge is constructed between the knowledge point node and the cognitive target node. The first directed edge is used to represent that the mastery of the knowledge point affects the performance of the cognitive dimension in terms of difficulty level. A second directed edge is constructed between the cognitive target node and the cognitive measurement node, and the second directed edge is used to represent that the performance of the cognitive dimension in terms of difficulty level affects the result of cognitive measurement; A third directed edge is constructed between the learning behavior node and the cognitive measurement node, and the third directed edge is used to represent the effect of the learning behavior data on the cognitive measurement result; A fourth directed edge is constructed between the cognitive measurement node and the teaching strategy node. The fourth directed edge is used to represent that the result of cognitive measurement affects the selection of teaching strategy. A fifth directed edge is constructed between the teaching strategy node and the prompt word template node; A conditional probability table between node states is constructed based on each directed edge, wherein the conditional probability table based on each first directed edge is used to describe the probability distribution of knowledge points in cognitive dimensions and difficulty levels at different mastery levels; each conditional probability table based on the second directed edge is used to describe the probability distribution of cognitive measurement under a specific combination of cognitive dimensions and difficulty levels; each conditional probability table based on the third directed edge is used to describe the probability distribution of cognitive measurement under a specific combination of learning behavior data; each conditional probability table based on the fourth directed edge is used to describe the probability of selecting a teaching strategy under a specific cognitive measurement state; each conditional probability table based on the fifth directed edge is used to describe the probability distribution of a prompt word template under a specific teaching strategy.

3. The method according to claim 1, characterized in that The state of the cognitive measurement node is dynamically updated through the Bayesian network reasoning mechanism based on the learner's learning behavior data, homework and assessment data, and the completion of the teaching task, including: According to the conditional probability table in the Bayesian network, the cognitive measurement state S is calculated. CM The probability distribution P(S CM |S k1 ,S k2, ...,S kn ,S t1 ,S t2 ,...,S tm ), where S ki represents the state of knowledge point ki, S ti It represents the state of teaching task ti, n and m represent the number of knowledge points and teaching tasks respectively.

4. The method according to claim 1, characterized in that The conditional probabilities of different teaching strategies are calculated by combining the states of knowledge point nodes, cognitive difficulty and difficulty level, and cognitive measurement data through the reasoning mechanism of the Bayesian network, including: According to the conditional probability table in the Bayesian network, calculate each teaching strategy s i Status S si The probability distribution P(S si |S k1 ,S k2 ,...,S kn ,C1,C2,...,C a ,D1,D2,D b ,S CM ), where S ki represents the state of knowledge point ki, C x and D y denotes the cognitive dimension and difficulty level indexed by x and y, respectively, a and b denote the number of cognitive dimensions and difficulty levels, respectively, S CM Represents cognitive measurement status.

5. The method according to claim 1, characterized in that The method of selecting the teaching strategy with the highest probability as the most appropriate teaching strategy at present according to the conditional probabilities of different teaching strategies includes: Calculate the applicability score of each teaching strategy: Score(s i )=P(S si =1|S k1 ,S k2 ,...,S kn ,C1,C2,...,C6,D1,D2,D3,SCM) Select the teaching strategies with the highest scores * , as the most appropriate teaching strategy at present.

6. The method according to claim 1, characterized in that The step of inputting the learner's real-time learning situation into the Bayesian network for reasoning to obtain a teaching prompt word template and a teaching instruction label comprises: inputting the situations of different learners into the Bayesian network for reasoning to obtain a teaching prompt word template and a teaching instruction label suitable for different learners; Correspondingly, the prompt words and ability plug-in of the teaching agent are dynamically updated according to the teaching prompt word template and the teaching instruction label to constrain the agent to finally output the teaching guidance content, including: according to different teaching prompt word templates and teaching instruction labels, the prompt words and ability plug-in of the teaching agent are dynamically updated to constrain the agent to finally output the teaching guidance content for different learners.

7. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method for dynamically updating teaching agent instructions based on a Bayesian network as described in any one of claims 1-6.

Images