Student companion teaching research system based on agent modeling and simulation

Abstract

The invention discloses a student companion teaching research system based on computer simulation, and relates to modeling, simulation experiments and data analysis of student companion teaching behaviors. The system is composed of a student companion teaching behavior strategy simulation module, a social simulation module and a data collection and analysis module, can simulate various student interactive behaviors, and performs experimental analysis on the influence of emotion, experience, neighbors and other factors on the behaviors. A core experiment framework of the system is formed by five behavior selection strategies (based on emotions, win-win people and input people exchange, the highest accumulated reward, most of historical behaviors of opponents and most of neighbors) and three network structures (full connection, grids and small world networks), the complex dynamic state of student companion teaching can be efficiently simulated, and factors influencing the interactive behaviors of the students are revealed. The system adopts an agent modeling technology, and each student is regarded as an independent agent and can autonomously select behaviors and update emotional states in interaction. Experimental results show that a behavior selection strategy and a network structure have significant influence on student interaction behaviors, for example, an emotion-based strategy can better encourage companion teaching in a full-connection network, and existence of a fixed behavior agent also affects a behavior mode of a whole group. The invention provides an efficient and flexible research platform for exploring the relationship between the student companion teaching behaviors and various factors, and provides a reference basis for the school to optimize the companion teaching environment.

Description

[0001] This research was supported by the Hangzhou Municipal Major Science and Technology Innovation Project "Research and Application of AI-native Relational Data Intelligence System" (2022AIZD0056). Technical Field

[0002] This invention relates to a student peer teaching research system, the main contents of which are student peer teaching behavior modeling, simulation experiments and result analysis based on computer simulation. Background Technology

[0003] Peer teaching refers to an educational process in which students teach and learn from each other without direct teacher intervention. In peer teaching, a student may exhibit certain basic behaviors, such as teaching other students or not teaching. These behavioral changes can be influenced by factors such as the student's personality and the relationships between students. Research into the influence of these factors can provide valuable insights for improving school teaching effectiveness, such as helping teachers create an environment that encourages peer teaching.

[0004] In existing peer teaching technologies, student behavior data is collected in real-world teaching environments, such as face-to-face interactions or online forum discussions. This data can provide valuable insights into understanding peer teaching. However, conducting more in-depth research on peer teaching issues based on these technologies presents several challenges. First, the number of potential influencing factors in peer teaching can be enormous, making it extremely difficult to exhaustively study all of them. Second, due to the complexity of human behavior patterns, our understanding of student behavior is currently incomplete, making accurate quantification difficult. Third, conducting experiments in real-world teaching environments consumes significant human and other resources, resulting in slow data collection and analysis, and consequently, inefficient research. This invention differs from existing technologies by using computer simulation technology to model and experiment with various peer teaching processes, providing a more efficient method for peer teaching research. While computer simulation technology has been used in teaching-related research, no technology related to this invention is currently being used in the field of peer teaching research. Summary of the Invention

[0005] (a) Technical problem to be solved: This invention addresses the lack of computer simulation research technology in the current field of peer teaching research, and involves modeling, simulation experiments and data analysis of student peer teaching behavior patterns.

[0006] (II) Technical Solution: The implementation block diagram of this system is as follows: Figure 1As shown, the system includes a student peer teaching behavior strategy simulation module, a student peer social simulation module, and a student peer teaching data collection and analysis module. The student peer teaching behavior strategy simulation module provides five different behavior selection strategies, which may be influenced by emotions, experience, neighbors, and attributes. The student peer social simulation module provides three classic network structures to model different interpersonal relationships in real life, such as small-world networks, where most nodes are not adjacent and most arbitrary nodes can be reached within a few steps, often used to simulate friendships between students. The student peer teaching data collection and analysis module includes experimental setup and procedures. The student peer teaching behavior strategy simulation module and the student peer social simulation module are invoked within the student peer teaching data collection and analysis module.

[0007] This patent proposes a student peer teaching behavior framework based on agent modeling technology, such as... Figure 2 As shown in the proposed framework, students' peer-learning behaviors and the factors that may influence these behaviors are modeled. A student is represented by an agent, which can be viewed as an autonomously operating software entity. A group of students is represented by a multi-agent system, and the peer-learning environment is modeled as the interactions occurring between the agents. Figure 2 The left side illustrates the interaction between two agents. In this interaction, one agent acts as the guide, and the other as the learner. Both agents must decide whether to teach and learn based on their behavioral selection strategies. These strategies access internal information about the agents, including their past experiences, emotions, relationships with others, and other attributes. For example, in an emotion-based behavioral selection strategy, emotions are categorized as: happiness, fear, anger, and stress. When an agent is happy or fearful, it is more likely to engage in peer teaching, simulating the behavior of students in real-world teaching environments; when an agent is angry, it is more likely to refuse peer teaching. After the interaction, the agents update their internal information based on the outcome. For instance, when both agents choose to engage in peer teaching, their happiness increases. Conversely, when neither agent engages in peer teaching, their stress increases, simulating the emotional changes experienced by students in real-world teaching environments. Figure 2 The right side shows the environment in which all agents interact, with multiple peer teaching interactions occurring in parallel among the agents.

[0008] The relevant technologies of this invention are introduced as follows: (1) Behavioral selection strategies: In the real world, students' behavioral selection strategies vary greatly due to their different personalities. This invention studies the following five typical strategies.

[0009] Strategy 1: Emotion-Based Strategy. In this strategy, the agent selects its behavior based on its own emotions and models the changes in these emotions. Specifically, for each emotion, the agent uses a value to represent the degree of emotion accumulated through interactions with other agents. When a tutor teaches and a learner learns, it indicates that they are engaging in peer teaching, which will bring some positive emotions to the agent; when the tutor teaches but the learner does not learn, due to the conflict of desires among the agents, both agents will experience some negative emotions; when neither party engages in peer teaching, they will also experience some negative emotions. Figure 6 The above-mentioned immediate emotional feedback is displayed. Figure 7 It shows the cumulative quantification of emotions.

[0010] Strategy 2: Winners Keep, Losers Switch: In this strategy, winning can be seen as a result of the agent's consistent desire, i.e., feeling joy or pressure. Otherwise, the agent can be considered a loser. If an agent wins in a previous interaction, it retains that behavior; otherwise, it switches to choosing a different behavior.

[0011] Strategy 3: Maximum Cumulative Reward: In this strategy, the agent selects the behavior with the highest cumulative reward. This strategy is a natural way to quantify how students choose behaviors. For example, humans tend to choose behaviors that bring them many rewards. In this invention, this patent proposes a simplified and interpretable way to quantify the effect. Interacting agents receive a +2 reward when they both engage in peer teaching. Due to their shared desire, the interacting agents receive a +1 reward when they do not engage in peer teaching. This represents a situation where a student enjoys peer teaching.

[0012] Strategy 4: Most of the Opponent's Historical Behaviors: In this strategy, the agent's actions are recorded. The agent selects the actions that the opponent has performed most frequently in the past few interactions.

[0013] Strategy 5: Majority of Neighbors: In this strategy, the actions of neighbors are recorded. The agent selects the action that occurs most frequently among the neighbors, or randomly selects an action if they occur the same number of times.

[0014] (2) Experimental Setup: The number of agents was set to 30, representing a classroom for one student. In one iteration, agents were randomly divided into pairs, with each pair engaging in peer-to-peer teaching interactions simultaneously. One trial consisted of 1000 iterations. The behaviors and emotions of the interacting agents (when using an emotion-based behavior selection strategy) were recorded, resulting in a dynamic change in the ratio of joint behavior to emotion. Unless otherwise specified, all agents used the same behavior selection strategy and fully connected network in each experiment. This patent conducted 1000 trials for each parameter setting. Convergence was considered achieved when the behaviors of all agents began to remain constant.

[0015] This invention relates to three experimental directions and nine experimental setups, specifically including: Experimental Direction 1: Studying and analyzing the effects of five behavioral choice strategies; Experimental Direction 2: Studying and analyzing the effects of various attributes within the five behavioral choice strategies, this type of experiment studies the five behavioral choice strategies by changing the specific attributes of each strategy; Experimental Direction 3: Studying the effects of agents with fixed behaviors. This type of experiment aims to study the effects of fixed agents, which can represent students with specific personalities. For example, some students may always prefer to teach or learn in peer teaching.

[0016] Setup 1.1: This study investigates and analyzes the impact of five behavioral selection strategies. Since fully connected networks are better suited to simulating student interaction patterns in schools, they were used as the default network architecture for the experiment. The agent selected five behavioral selection strategies—based on emotion, winner-takes-lose exchange, maximum cumulative reward, majority of opponents' historical behavior, and majority of neighbors—for five trials. The impact of these behavioral selection strategies on peer teaching was then compared and analyzed based on the convergence results.

[0017] Setting 1.2: This study investigates and analyzes the impact of different network structures on behavioral selection strategies. Three types of network architectures were studied: fully connected networks, grid networks, and small-world networks. Under each network architecture, the agent selected five behavioral selection strategies based on emotion, winner-takes-lose trade, maximum cumulative reward, majority of opponents' historical behavior, and majority of neighbors for five trials. The convergence results obtained from the three different network architectures were compared and the impact of network structure on peer teaching was analyzed.

[0018] Setting 2.1: In emotion-based strategies, this study investigates the impact of the probability of an agent changing its behavior when experiencing stress on peer teaching. The experiment sets the probability p of stress leading to behavioral change, with p taking values ​​of 0, 0.2, 0.4, 0.6, 0.8, and 1, and conducts six trials for each value. The likelihood of an agent changing its behavior when experiencing stress is also set to p. The convergence results obtained for the six different p values ​​are compared and analyzed to assess the impact of stress on peer teaching.

[0019] Setting 2.2: In the winner-takes-all strategy, the probability of switching to another behavior when the agent's behavior is considered a loss was changed. Experiments investigated two probabilities of switching to another behavior when the agent's behavior was considered a loss: 100% and 50%. The convergence results obtained based on the two different switching probabilities were compared and the impact of the switching probabilities on peer teaching was analyzed.

[0020] Setting 2.3: In the strategy with the highest cumulative reward, different levels of reward settings are used. The reward value received by students can represent their psychological state, such as the pleasure they feel about the interaction outcome, which may influence their behavior when using this choice strategy. The experiment studied three different levels of reward differential: 0, 0.5, and 1. For example, when the reward differential is set to 1, agents who all engage in peer teaching receive a +2 reward, while agents who do not engage in peer teaching receive only a +1 reward. Based on the three different levels of reward differential, the convergence results are compared and the impact of reward differential on peer teaching is analyzed.

[0021] Setting 2.4: In the majority of strategies based on the opponent's historical behavior, the number of opponent historical behaviors the agent can remember is varied. Experiments investigated five different memory capacities: 1, 10, 20, 50, and infinite. For example, with a memory capacity of 10, the agent can remember the opponent's most recent 10 behaviors and then select the most frequent behavior from these 10. The convergence results obtained for the five different memory capacities were compared and analyzed to assess the impact of memory capacity on peer teaching.

[0022] Setting 2.5: In the majority-based neighbor strategy, the selection of the majority neighbor behavior is divided into two types: probabilistic selection and 100% selection. When the experiment is set to 100% selection, since each agent's behavior and emotions are influenced by the surrounding neighbors in the fully connected network, less frequently selected behaviors will not be selected. For example, if all tutors happen to have chosen to teach, then under the influence of 100% selection, other learners will also perform teaching behaviors when acting as tutors, and the minority non-teaching behaviors will not be selected by any agent. When the experiment is set to probabilistic selection, the probability of an agent choosing to perform peer teaching is determined by the proportion of surrounding neighbors performing teaching in the network. This probability is determined by dividing the total number of surrounding neighbors performing teaching by the total number of surrounding neighbors. The convergence results obtained based on the two different selection probabilities are compared and the impact of neighbor behavior on peer teaching is analyzed.

[0023] Setting 3.1: Investigating the effects of agents with fixed behaviors. In this experiment, this patent sets up two scenarios. The first scenario studies agents that always engage in peer teaching; the number of these fixed agents is set to {1, 3, 5}. The second scenario studies a pair of agents exhibiting opposite behaviors. In each pair of agents, one always engages in peer teaching, and the other refuses to engage in peer teaching; the number of agent pairs is set to {1, 3, 5}.

[0024] Setting 3.2: Investigating the impact of fixed agent positions in the network. In this experiment, all agents used a grid network by default. The number of agents always performing peer teaching was set to {1, 3, 5, 7, 9}. Two sets of experiments were conducted. In the first set, agents with fixed behaviors were randomly distributed, while in the second set, agents with fixed behaviors were placed in a specific region of the grid network.

[0025] (3) The experimental procedure is as follows: Step 1: Set the number of agents to 30, representing a classroom for one student. In one iteration, agents are randomly divided into two groups, and each group interacts with peer teaching simultaneously. 1000 iterations constitute one experiment. Select the behavior selection strategy and network structure according to different experimental settings. Step 2: Randomly select a group of agents A from the agents given in Step 1, randomly select one agent A_u as the tutor, and another agent A_l as the student, and remove A_u and A_l from the given agents. An agent is represented as a tuple.<b,e,E,Q,N> Where b represents the agent's behavior, e represents the agent's emotion, E represents the agent's possible emotion, Q represents the agent's experience of recording some historical information, and N represents the agent's neighbors; Step 3: Repeat step 2 until there are no more agents available among the given agents; Step 4: Each group of agents simultaneously engages in peer teaching interaction, with the agent selecting its own behavior based on a behavior selection strategy. The behavior selection strategy accesses some internal information of the agent, such as the agent's experience in past interactions, emotions, relationships with others, and other attributes. Peer teaching can be represented as a tuple.<u,Bu,l,Bl,F> Where u represents the tutor, Bu represents the available behaviors of u, l represents the learner, Bl represents the available behaviors of l, and F represents the direct impact of all possible joint behaviors of u and l. For peer teaching, this invention proposes Bu = {teach, don't teach} and Bl = {learn, don't learn}, which will be the basic behaviors of peer teaching. For the agent's emotions, this patent considers E = {happiness, fear, anger, stress}, which will be the basic emotions that students may experience in peer teaching; Step 5: After the interaction, the agent may update its internal information based on the result of the interaction, which completes one iteration; Step 6: Repeat steps 2, 3, 4, and 5 a total of 1000 times until all iterations are completed. The behaviors and emotions of both parties in each group of interacting agents will be recorded, thus forming a dynamic change in the ratio of joint behaviors and emotions.

[0026] (III) Beneficial Effects: This invention proposes a computer simulation-based student peer learning research system. It uses agent-based modeling and multi-agent systems to model student behavior patterns, thereby reducing the human and financial costs associated with experiments. It provides three classic network structures and five behavioral selection strategies, and the system exhibits good scalability and flexibility. If further research is needed or more factors are required, only a few parameters need to be modified, without having to design the research plan and experiments from scratch. The method provided by this invention can successfully study the relationship between some problems and student peer learning. In the future, it can also provide a faster and more convenient platform for studying the relationship between student peer learning and other factors. Experimental results reveal some clues that may help teachers encourage students to engage in peer learning: behavioral selection strategies greatly influence students' behavioral choices, and different types of strategies may have different effects. For example, when agents use a winner-take-all strategy, peer learning cannot be observed without considering other information, such as experience, emotions, or peer behavior. On the other hand, peer learning can be observed when this information is considered. Therefore, when encouraging peer learning, teachers can consider students' internal information, such as experience and emotions. The attributes of behavioral selection strategies also influence student behavioral dynamics. For example, when an agent uses the highest cumulative reward, setting more rewards for peer teaching encourages students to engage in peer teaching. When an agent considers both experience and emotion as influencing factors, setting a moderate memory capacity is more effective in encouraging peer teaching than setting a very small or very large one. When an agent uses the majority of strategies from the opponent's historical behavior, selecting behaviors used by most peers reduces the likelihood of failure to engage in peer teaching. Therefore, this can be helpful for teachers, for example, by rewarding students who have engaged in peer teaching and encouraging students to engage in peer teaching with peers who have recently been willing to do so. Network structure affects student behavioral dynamics. Fully connected networks encourage peer teaching better than grid and small-world networks. Therefore, teachers can encourage students to interact with more students, not just a few close friends. Agents with fixed behaviors can influence student behavioral dynamics. Having a sufficient number of agents consistently engaging in peer teaching can encourage other students to do the same. Conversely, having students who consistently refuse peer teaching can significantly hinder peer teaching. Furthermore, the position of a fixed agent within the classroom also influences student behavior dynamics, with a greater impact at random positions than at specific ones. Therefore, teachers can encourage peer learning by avoiding consistently discouraging peer interaction. Additionally, teachers can randomly select students to encourage peer learning, which is more helpful than simply encouraging specific students (such as those sitting close together in the classroom). Attached Figure Description

[0027] Figure 1 This is a general implementation framework diagram of the present invention; Figure 2 A framework diagram for modeling a student intelligent agent and the peer teaching environment; Figure 3 The emotional direct result of the agent's joint behavior during a peer-learning interaction; Figure 4 To quantify the impact of the agent's joint behavior on each emotion; Figure 5 This refers to a multi-agent concurrent interaction process. Figure 6 The convergence results are for five behavior selection strategies based on a fully connected network structure. Figure 7 The convergence results are for five behavior selection strategies based on a grid network structure. Figure 8 The convergence results are for five behavior selection strategies based on the small-world network structure. Detailed Implementation

[0028] Taking setup 1.2 in Experiment 1 as an example, this study investigates the impact of behavior selection strategies on peer teaching under different network structures. Three types of network architectures are studied: fully connected networks, grid networks, and small-world networks. Under each network architecture, agents selected five behavior selection strategies based on emotion, winner-takes-lose trade, maximum cumulative reward, majority of opponents' historical behavior, and majority of neighbors for five trials. The convergence results obtained from the three different network architectures are compared and the impact of network structure on peer teaching is analyzed. In each trial, all agents used the same behavior selection strategy and network structure.

[0029] Step 1: Set the number of agents to 30, representing a classroom for one student. In one iteration, agents are randomly divided into pairs, with each pair interacting simultaneously in peer-to-peer teaching. One trial consists of 1000 iterations. Table 1 describes the concurrent interaction process in the multi-agent system. In this experiment, all agents used an emotion-based behavior selection strategy and a fully connected network.

[0030] Step 2: Randomly select a group of agents A from the agents given in Step 1. Randomly select one agent A_u as the mentor and another agent A_l as the student. Remove A_u and A_l from the given agents.

[0031] Step 3: Repeat step 2 until there are no more agents available among the given agents;

[0032] Step 4: Each agent engages in peer-to-peer teaching interactions simultaneously. The agent accesses its own emotions, obtaining the value of each emotion; the emotion with the highest value becomes the agent's current emotion. The agent determines its behavior during the interaction based on its current emotion: when the agent is stressed or fearful, it will choose to engage in peer teaching; when the agent is angry, it will refuse to engage in peer teaching; and when the agent is happy, it will maintain the behavior selected in the previous interaction.

[0033] Step 5: After the interaction, the agent updates its internal information based on the outcome of the interaction. The impact of different behaviors on emotions is as follows: Figure 6 As shown, for the convenience of experiments, this invention quantifies emotions in detail. The quantitative impact of different behaviors on emotions is as follows: Figure 7 As shown, when both agents engage in peer teaching, their happiness increases by 1; when neither engages in peer teaching, their stress increases by 1; and when the agents exhibit opposite behaviors, the one engaging in peer teaching gains anger by 1, while the one not engaging in peer teaching gains fear by 1.

[0034] Step 6: Repeat steps 2, 3, 4 and 5 a total of 1000 times until all iterations are completed. The behavior and emotions of both parties in each group of interacting agents will be recorded, thus forming a graph showing the ratio of behavior to emotion changes.

[0035] Step 7: Then, select the four behavior selection strategies in turn: winner keeps, loser swaps, highest cumulative reward, majority of opponents' historical behavior, and majority of neighbors, and repeat step 6 to obtain the proportion changes of agent behavior and emotion under different behavior selection strategies in the fully connected network setting.

[0036] Step 8: Statistically analyze the proportion of experimental results obtained from 1000 trials for each behavior selection strategy to obtain the number of times all agents converge to these four joint behaviors (teaching, learning), (teaching, not learning), (not teaching, learning), (not teaching, not learning) and four emotions. The convergence results of the five behavior selection strategies based on the fully connected network structure are shown in Table 2.

[0037] Step 9: Repeat steps 1, 6, 7, and 8 sequentially using both the grid network and the small-world network. The convergence results for the five behavior selection strategies based on the grid network structure are shown in Table 3; the convergence results for the five behavior selection strategies based on the small-world network structure are shown in Table 4. Based on the convergence results under the three different network structures, compare and analyze the impact of network structure on peer teaching.

Claims

1. A student peer teaching research system, which mainly involves computer simulation-based student peer teaching behavior modeling, simulation experiments and result analysis.

2. The student peer teaching behavior strategy simulation module according to claim 1 provides five different behavior selection strategies, which may be influenced by emotions, experience, neighbors, and attributes.

3. The student peer social simulation module according to claim 1 provides three classic network structures to model different interpersonal relationships in real life. For example, in a small-world network, most nodes are not adjacent, and most arbitrary nodes can be reached within a few steps. It is often used to simulate friendships between students.

4. The student peer teaching data collection and analysis module according to claim 1 includes experimental setup and experimental procedures. The student peer teaching data collection and analysis module will call the student peer teaching behavior strategy simulation module and the student peer social simulation module.

5. A student peer learning behavior framework based on agent modeling technology according to claim 1. In the proposed framework, student peer learning behaviors and factors that may influence student behavior are modeled. A student is represented by an agent, which can be viewed as an autonomously running software entity. A group of students is represented by a multi-agent system, and the peer learning environment is modeled as the interactions that occur between agents.

Images