Checkpoint generation system and method in financial education field

Abstract

The invention relates to a level generation system and method in the financial education field, which are used for solving the problems of insufficient individuation degree, inaccurate difficulty matching and single feedback mechanism in the existing financial education field. The system comprises a dynamic difficulty matching module used for adjusting a learning level of a user in a financial education learning process based on state space data of the user; the differentiated reward module is used for performing differentiated reward in the user learning process based on the risk preference data of the user; the multi-modal feedback module is used for dynamically adjusting a learning path of the user in a learning process based on the failure event data of the user; and the level path generation module is used for planning a learning path of the user based on the initial knowledge state of the user and the performance in the learning level.

Description

Technical Field

[0001] This application relates to the field of financial education technology, and in particular to a level generation system and method for financial education. Background Technology

[0002] With the rapid development of financial technology, the importance of financial education is becoming increasingly prominent. Traditional financial education models often employ uniform teaching content and pace, making it difficult to meet the personalized needs of different learners. Especially in financial education, learners exhibit significant differences in their knowledge background, learning abilities, and risk preferences. Existing systems struggle to dynamically adjust the difficulty of content based on learners' actual abilities and progress. Furthermore, feedback in existing systems is typically presented in text form, lacking multimodal feedback and failing to meet the comprehension needs of diverse learners. Therefore, existing standardized educational content is unlikely to achieve ideal learning outcomes. This application provides a level generation system and method in the field of financial education to address the problems of insufficient personalization, inaccurate difficulty matching, and a single feedback mechanism in existing financial education practices. Summary of the Invention

[0003] This application provides a level generation system and method in the field of financial education to address the problems of insufficient personalization, inaccurate difficulty matching, and simplistic feedback mechanisms in existing financial education.

[0004] Firstly, this application provides a level generation system in the field of financial education. The system includes: a dynamic difficulty matching module for adjusting the learning levels of a user in the financial education learning process based on the user's state space data; a differentiated reward module for providing differentiated rewards to the user during the learning process based on the user's risk preference data; a multimodal feedback module for dynamically adjusting the user's learning path during the learning process based on the user's failure event data; and a level path generation module for planning the user's learning path based on the user's initial knowledge state and performance in the learning levels.

[0005] In one possible design, the dynamic difficulty matching module is specifically used to: input the user's current state space data into the DQN network to obtain the predicted Q-values ​​of all possible actions in the action space; the user's state space data includes the user's learning behavior data, historical performance data, and personal characteristics in the financial education system; the action space includes adjustments for knowledge point density, interaction complexity, time pressure, and risk-reward ratio; the DQN network randomly selects an action from the action space with probability ε, selects the action with the highest predicted Q-value with probability 1-ε, and dynamically adjusts the ε value; adjusts the user's subsequent learning levels based on the selected action, and provides corresponding rewards based on the user's state space data at the next moment.

[0006] In one possible design, the DQN network is a dual DQN network; the DQN network employs experience replay technology to store and reuse historical interaction data.

[0007] In one possible design, the differentiated reward module is specifically used for: collecting users' risk preference data; the users' risk preference data includes users' initial assessment data, continuous behavior data, and historical analysis data; quantifying users' risk preference data to obtain users' risk preference scores, classifying users according to their risk preference scores to obtain users' risk preference categories; and providing differentiated rewards according to users' risk preference categories.

[0008] In one possible design, the multimodal feedback module is specifically used to: capture user failure event data; perform multi-dimensional failure cause analysis on the user failure event data to obtain analysis data; generate personalized multimodal feedback for user failure events based on the analysis data; and dynamically adjust the user's subsequent learning path according to the effect of the multimodal feedback.

[0009] In one possible design, the level path generation module is specifically used for: constructing and maintaining a knowledge graph in the fintech field; planning an initial learning path on the knowledge graph based on the user's initial knowledge state; dynamically adjusting the learning path according to the user's performance in the learning level when a specific event is triggered; and making personalized content recommendations in the user's learning path.

[0010] Secondly, embodiments of this application provide a method for generating learning levels in the field of financial education. This method includes: adjusting the learning levels of a user during the financial education learning process based on the user's state space data; the user's state space data includes the user's learning behavior data, historical performance data, and personal characteristics in the financial education system; the action space includes adjustments to knowledge point density, interaction complexity, time pressure, and risk-reward ratio; providing differentiated rewards based on the user's risk preference data during the learning process; the user's risk preference data includes the user's initial assessment data, continuous behavior data, and historical analysis data; dynamically adjusting the user's learning path during the learning process based on the user's failure event data; and planning the user's learning path based on the user's initial knowledge state and performance in the learning levels.

[0011] In one possible design, adjusting the user's learning stages in the financial education learning process based on the user's state space data includes: inputting the user's current state space data into a DQN network to obtain the predicted Q-values ​​of all possible actions in the action space; the DQN network randomly selects an action in the action space with probability ε, selects the action with the highest predicted Q-value with probability 1-ε, and dynamically adjusts the ε value; adjusting the user's subsequent learning stages based on the selected action, and providing corresponding rewards based on the user's state space data at the next moment.

[0012] In one possible design, the DQN network is a dual DQN network; the DQN network employs experience replay technology to store and reuse historical interaction data.

[0013] In one possible design, the differentiated reward based on user risk preference data during the user learning process includes: collecting user risk preference data; quantifying the user risk preference data to obtain a user risk preference score; classifying the user according to the user risk preference score to obtain a user risk preference category; and providing differentiated rewards based on the user risk preference category.

[0014] In one possible design, dynamically adjusting the user's learning path based on the user's failure event data includes: capturing the user's failure event data; performing multi-dimensional failure cause analysis on the user's failure event data to obtain analysis data; generating personalized multimodal feedback for the user's failure events based on the analysis data; and dynamically adjusting the user's subsequent learning path based on the effect of the multimodal feedback.

[0015] In one possible design, the step of planning the user's learning path based on the user's initial knowledge state and performance in learning levels includes: constructing and maintaining a knowledge graph in the fintech field; planning an initial learning path on the knowledge graph based on the user's initial knowledge state; dynamically adjusting the learning path according to the user's performance in learning levels when a specific event is triggered; and recommending personalized content within the user's learning path.

[0016] Thirdly, embodiments of this application also provide a computing device, including: Memory, used to store program instructions; A processor for invoking program instructions stored in the memory and executing the system or method as described in any possible design of the first aspect, according to the obtained program instructions.

[0017] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer-readable instructions that, when read and executed by a computer, cause the system or method described in any possible design of the first aspect to be implemented.

[0018] Fifthly, embodiments of this application also provide a computer program product, which includes computer program code that, when run on a computer, causes the computer to perform the system or method as described in any possible design of the first aspect. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic diagram of a level generation system in the field of financial education provided in this application embodiment; Figure 2 A schematic diagram of a dynamic difficulty matching module provided in an embodiment of this application; Figure 3 A schematic diagram of a differentiated reward module provided in an embodiment of this application; Figure 4 A schematic diagram of a multimodal feedback module provided in an embodiment of this application; Figure 5 A schematic diagram of a level path generation module provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the embodiments of this application, "multiple" refers to two or more. Terms such as "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or order.

[0023] Existing financial education recommendation systems primarily rely on learners' historical behavioral data, providing learning resources through collaborative filtering or content recommendation algorithms. For example, PyTorch-driven personalized learning path generation systems analyze learners' learning abilities, interests, and goals to provide customized learning content. However, these systems suffer from the following problems: Lack of dynamic adaptability: Most systems cannot dynamically adjust content based on learners' real-time performance, resulting in recommended content that does not match the learner's current ability.

[0024] Ignoring learner characteristics: Existing systems mainly focus on recommending learning content, with little consideration for learners' individual characteristics, such as risk preferences and learning styles.

[0025] The feedback mechanism is simple: the system usually provides simple correct and incorrect feedback, lacks in-depth analysis and multimodal feedback, and cannot effectively guide learners to improve.

[0026] Lack of long-term planning: Systems often focus on short-term content recommendations and lack planning and optimization for learners' long-term learning paths.

[0027] This application provides a level generation system and method for the field of financial education, aiming to address issues such as insufficient personalization, inaccurate difficulty matching, and a single feedback mechanism in existing financial education systems. The goal is to achieve the following: establish an algorithmic mechanism that can dynamically adjust level difficulty based on learner behavioral data, improving learning efficiency and experience; design a system that provides differentiated reward mechanisms based on learners' risk preferences, enhancing learning motivation and participation; construct a closed-loop system that provides multimodal feedback when learners fail, improving learning effectiveness and knowledge absorption rate; and implement an adaptive learning system that can dynamically adjust subsequent level paths based on learner performance, optimizing the personalized learning experience.

[0028] Figure 1 A schematic diagram of a level generation system in the field of financial education provided in this application embodiment, such as... Figure 1 As shown, the system includes a dynamic difficulty matching module 100, a differentiated reward module 200, a multimodal feedback module 300, and a level path generation module 400. Each module has the following functions: The dynamic difficulty matching module 100 is used to adjust the learning levels of users in the financial education learning process based on the user's state space data.

[0029] The differentiated reward module 200 is used to provide differentiated rewards based on the user's risk preference data during the user's learning process.

[0030] The multimodal feedback module 300 is used to dynamically adjust the user's learning path during the learning process based on the user's failure event data.

[0031] The level path generation module 400 is used to plan the user's learning path based on the user's initial knowledge state and performance in the learning level.

[0032] like Figure 2 As shown, when the dynamic difficulty matching module 100 adjusts the learning levels of a user in the financial education learning process based on the user's state space data, it specifically includes the following steps: Step 101: Input the user's current state space data into the DQN network to obtain the predicted Q-values ​​of all possible actions in the action space.

[0033] In this embodiment, the user's state space data includes the user's learning behavior data, historical performance data, and personal characteristics within the financial education system. The system continuously collects the user's real-time learning behavior data and accesses the user's historical performance data and personal characteristics. The learning behavior data may include trading frequency, simulated trading win rate, learning duration, test scores, etc.; historical performance data may include historical level completion status, error type distribution, etc.; personal characteristics may include risk preference scores, learning styles, etc. The system transforms the user's state space data into a vectorized state representation St=[s1,s2,...,sn], which is then input into a Deep Q-Learning (DQN) model. The DQN network includes an input layer, hidden layers, and an output layer. The input layer receives the state space data; the hidden layers consist of multiple fully connected layers for feature extraction and representation learning; and the output layer is used to adjust parameters corresponding to the action space.

[0034] The DQN network processes the vectorized state identifiers of the user's state space data and outputs the Q-value of all possible actions in the action space: Q(St,a). The action space can include adjustments for knowledge density, interaction complexity, time pressure, and risk-reward ratio. Knowledge density adjustment controls the number of knowledge points presented per unit of time; interaction complexity adjustment controls the complexity of interactive operations within a level; time pressure adjustment controls the time limit for completing a level; and risk-reward ratio adjustment controls the ratio of risk to reward within a level.

[0035] For example, the knowledge point density adjustment ΔK can be expressed as ΔK=Knew-Kcurrent, where Knew=Kcurrent×(1+αK) αK), where aK ∈ -1, 0, 1 (decreasing, maintaining, increasing) or a continuous range, and αK is a scaling factor (e.g., 0.1 represents a 10% change). The interaction complexity adjustment ΔC can be expressed as ΔC = Cnew - Ccurrent, where Cnew = Ccurrent × (1 + αK) aC), where aC is related to the number of decision points and the required input type. The time pressure adjustment ΔT can be expressed as ΔT = Tnew - Tcurrent, where Tnew = Tcurrent × (1 - αT) aT), where aT is the adjustment time limit (e.g., aT>0 indicates reduced time pressure, aT<0 indicates increased time pressure). The risk-reward ratio adjustment ΔRR can be expressed as ΔRR=RRnew-RRcurrent, where RRnew=RRcurrent×(1+αRR). aRR), aRR adjusts the inherent risk-return balance of the financial scenarios in the level.

[0036] Step 102: The DQN network randomly selects an action from the action space with probability ε, selects the action with the highest predicted Q value with probability 1-ε, and dynamically adjusts the ε value.

[0037] The DQN network randomly selects an action *at* from the action space with probability ε, and selects the action with the highest predicted Q-value *at* = argmax with probability 1-ε. a The system calculates Q(St,a) and dynamically adjusts the value of ε. The adjustment of ε is related to the system's learning progress or the user's "flow state," which is influenced by learning time and learning stability. For example, if the user remains in the "flow zone," the ε value can be decreased, allowing the DQN network to utilize known optimal actions. Conversely, if the user frequently leaves the flow zone (e.g., feeling bored or frustrated), ε can be increased to encourage exploration of new difficulty adjustments. This dynamic adjustment of ε allows users to explore more in the early stages and balances the exploration of new difficulty adjustments with the utilization of known effective adjustments in later stages.

[0038] Step 103: Adjust the user's subsequent learning levels based on the selected action and give corresponding rewards based on the user's state space data at the next moment.

[0039] The selected action (at) (e.g., "increase knowledge point density by 10%) is applied to the educational environment to generate new levels or modify the current level. Users interact with the modified level, and the system observes the new state St+1 (updated user behavior, historical performance, etc.) and calculates the corresponding reward Rt. Rewards can include completion rewards, time efficiency rewards, learning progress rewards, and challenge rewards. Completion rewards are given based on the user's level completion rate; time efficiency rewards are given based on the user's time efficiency in completing the level; learning progress rewards are given based on the user's improvement compared to previous performance; and challenge rewards are given based on the match between the level difficulty and the user's ability.

[0040] For example, the reward function can be expressed as: Rt = wcomp Rcompletion+wtime Rtimeefficiency+wprog Rprogress+wchall Rchallenge; where w is a weighting factor (wcomp + wtime + wprog + wchall = 1); the completion reward Rcompletion can be a binary reward (1 for completing a level, 0 otherwise) or a tiered reward (e.g., based on the percentage of task completion, Rcompletion = score / highest score); the time efficiency reward (Rtimeefficiency) can be expressed as Rtimeefficiency = max(0, 1 - (time used / optimal time)) or an exponentially decaying function Rtimeefficiency = exp(-β) The time taken is denoted by β, where β is the decay constant. The learning progress reward (Rprogress) is based on normalized improvement, for example, for the same knowledge point, Rprogress = (current score - previous score) / highest score, or a reduction in a specific error type. The challenge reward (Rchallenge) is 1.

[0041] In one possible implementation, the DQN network is a dual DQN network that employs an experience replay technique to store and reuse historical interaction data. Specifically, the transformation tuple (St, at, Rt, St+1) is stored in an experience replay buffer to break the correlation between consecutive samples and improve learning stability. Small batches of experience are periodically sampled randomly from the buffer for training the main DQN network. The target network is periodically updated by replicating the weights of the main DQN network (e.g., every C steps) to reduce Q-value overestimation and stabilize the Q-value target.

[0042] This application's embodiments update the weights of the main DQN network by minimizing the difference between the predicted Q-value and the target Q-value using a loss function. For example, the target Q-value Yt can be calculated using the following formula: Yt=Rt+γ Qtarget(St+1,argmax a Qmain(St+1,a)) Here, γ is a discount factor (e.g., 0.99), which determines the importance of future rewards.

[0043] The loss function is L=(Yt-Qmain(St,at)) 2The network weights are updated using an optimizer (e.g., Adam) to minimize this loss, thereby optimizing the difficulty adjustment strategy of DQN.

[0044] The aforementioned dynamic difficulty matching module can dynamically adjust the level difficulty based on the user's real-time performance, keeping the user in the "flow zone," that is, a state where the challenge matches the ability, thereby improving learning efficiency and experience.

[0045] like Figure 3 As shown, when the differentiated reward module 200 applies differentiated rewards based on the user's risk preference data during the user's learning process, it specifically includes the following steps: Step 201: Collect users' risk preference data.

[0046] In this embodiment, user risk preference data includes initial assessment data, ongoing behavioral data, and historical analysis data. Initial assessment data can be obtained from a questionnaire completed by the user. For example, when registering, the user completes a validated financial risk preference questionnaire (e.g., based on the Grable & Lytton Risk Tolerance Scale or a similar established psychometric scale). The system obtains a risk score (e.g., a value from 1 to 100) based on the questionnaire's content and uses this score as the user's initial assessment data. Ongoing behavioral data can be specific behavioral data indicating the user's risk preference continuously recorded by the system during simulated trading and other interactive learning activities, including average holding time, stop-loss frequency, asset volatility preference, and leverage usage. Average holding time refers to the average duration of holding simulated assets (shorter holding times generally indicate a higher risk appetite for rapid gains / losses); stop-loss frequency refers to the frequency with which stop-loss orders are set / triggered (higher frequency indicates more conservative risk management); asset volatility preference refers to the tendency to choose high / low volatility assets (e.g., the number of high-beta and low-beta assets in a portfolio); leverage usage refers to the frequency and extent of leverage used in simulated trading. Historical analysis data can include data from returning users, analyzing their past learning paths, simulated investment choices, and risk management strategies in previous interactions.

[0047] Step 202: Quantify the user's risk preference data to obtain the user's risk preference score, and classify the user according to the user's risk preference score to obtain the user's risk preference category.

[0048] A user's risk preference score is calculated based on their risk preference data (including initial assessment data, ongoing behavioral data, and historical analysis data). For example, the risk preference score can be calculated using the following formula: RPSt = wQ Initial behavioral data t+wB f(continuous behavioral data t) + wH g(historical analysis data t). Where wQ, wB, wH are weighting factors that sum to 1 (e.g., wQ=0.4, wB=0.4, wH=0.2). f() and g() are normalization and aggregation functions (e.g., weighted averages or pre-trained regression models) used to map continuous behavioral data and historical analysis data to normalized risk scores (e.g., 0 to 1), respectively. For example, f(behavioral data t) can be a linear model, such as f=(c1 Average holding time + c2 Stop loss frequency + c3 Asset volatility preference +c4 Leverage usage / Normalization factor.

[0049] Based on the aforementioned risk preference score calculation formula, a comprehensive risk preference score (RPS) can be calculated, which is dynamically updated based on observed user behavior. Users are then categorized according to their risk preference scores to obtain risk preference types, such as "conservative" (e.g., RPS < 0.3), "moderate" (e.g., 0.3 ≤ RPS < 0.7), and "high-risk" (e.g., RPS ≥ 0.7). The thresholds for each category can be predefined or dynamically adjusted based on the distribution of the user group.

[0050] In one possible implementation, after obtaining the user's risk preference category, the system can dynamically adjust the components and weights of the reward function based on that category. For example, for users with a high risk preference, the weight of challenge rewards (wchall) can be increased, particularly for high-risk, high-reward scenarios. This could also include increasing strategic freedom (e.g., successfully completing open-ended challenges with multiple effective solutions), encouraging innovative exploration (e.g., trying unconventional but successful strategies in simulations), and providing additional rewards. For conservative users, the weight of time efficiency rewards (wtime) and completion rewards (wcomp) can be increased to encourage stable and continuous progress. Higher rewards and positive feedback can be provided for risk control indicators (e.g., successfully avoiding significant losses in simulations, even with moderate rewards) and prudent decision-making and adherence to best practices.

[0051] Step 203: Provide differentiated rewards based on the user's risk preference category.

[0052] Differentiated rewards are offered based on users' risk tolerance categories. For example, high-risk users can be rewarded with virtual currency and community recognition. Virtual currency rewards can be used to unlock advanced or experimental content; community recognition rewards showcase users' professional achievements within the learning community, for bold accomplishments. Conservative users can be rewarded with achievement badges and practical application opportunities. Achievement badges record users' achievements and progress in different areas, for continuous improvement and mastery; practical application opportunities provide opportunities to apply learned knowledge to real-world scenarios, for tangible and consistent results.

[0053] The aforementioned differentiated reward module emphasizes strategic freedom for users with high-risk appetites, encouraging innovation and exploration, providing more high-risk, high-reward challenge options, and offering greater sense of accomplishment and recognition upon success. For conservative users, the system's rewards focus on risk control indicators, encouraging sound decision-making, providing more low-risk, stable-return options, and offering positive feedback and affirmation upon successful risk mitigation. By linking specific reward formats with specific strategies based on different risk appetites through a differentiated reward mechanism, and by meticulously understanding user motivations, the system can better meet the psychological needs of diverse users, thereby enhancing their learning motivation and engagement.

[0054] like Figure 4 As shown, when the multimodal feedback module 300 dynamically adjusts the user's learning path during the learning process based on the user's failure event data, it specifically includes the following steps: Step 301: Capture user failure event data.

[0055] When a user fails a level, makes a major mistake (e.g., a wrong trading decision, fails a quiz, or is inactive for a long time in a critical task), or exhibits suboptimal learning behavior, the system captures all relevant interaction data, i.e., failure event data, including user input, specific wrong choices, partial progress, time taken, past learning history, and the context of the failed task (e.g., the knowledge points involved and the difficulty level).

[0056] Step 302: Perform multi-dimensional failure cause analysis on the user's failure event data to obtain analysis data.

[0057] The failure event data captured in step 301 is subjected to multi-dimensional failure cause analysis to obtain analysis data. In this embodiment, multi-dimensional failure cause analysis may include error type identification, knowledge point association analysis, and learning pattern analysis. Error type identification refers to analyzing the specific reasons for user failure. Specifically, errors are classified based on failure event data to obtain error types, and a confidence score Cerrortype (e.g., 0-1) is assigned to the identified error types. Error types may include conceptual misunderstanding errors, operational errors, data interpretation errors, risk assessment biases, lack of concentration, etc. For example, a pre-trained classifier (e.g., a neural network trained based on historical error patterns, or a rule-based expert system) can be used to analyze failure event data to classify errors. Knowledge point association analysis refers to identifying key knowledge points and skills related to failure. Specifically, the system maps the identified errors to specific knowledge points (KP_id) related to the failed task in the financial technology knowledge graph, and calculates a relevance score RKP for each associated knowledge point, representing its direct contribution to the failure event. Learning pattern analysis refers to assessing whether a user's learning style is suitable for the current content. Specifically, based on user profiles and observed behaviors (e.g., past preferences for video and text content, and interaction speed with visual elements), the system infers the user's preferred learning style (visual, auditory, or hands-on), and represents the user's preferred learning style as a learning style preference vector LSP=[wvisual, wauditory, whands-on].

[0058] Step 303: Generate personalized multimodal feedback for user failure events based on the analyzed data.

[0059] Personalized multimodal feedback for user failure events is generated based on the analysis data obtained in step 302. For example, the analysis data obtained in step 302 can be input into the Big Oracle model to obtain personalized multimodal feedback for user failure events. For instance, the identified error type, related knowledge points, the user's current comprehension level (inferred from past performance), and preferred learning style can be input as structured prompts into the Big Oracle model. These prompts instruct the Big Oracle model to generate detailed error text analysis and actionable improvement suggestions, and to customize the multimodal feedback content according to the user's context.

[0060] Multimodal feedback can include text analysis, short video generation, chart visualization, and interactive demonstrations. Text analysis uses a large oracle model to generate core explanations, detailed error breakdowns, and step-by-step corrective suggestions. Short video generation automatically generates targeted analytical videos to demonstrate correct concepts and operations to users. Specifically, if a user has a high preference for visual or hands-on activities, or if the error is due to operational mistakes, the system will trigger an automated video generation module (e.g., using pre-recorded video clips, animated charts, or text-to-video synthesis) to demonstrate the correct operation or concept. Chart visualization uses data visualization to show the impact of user decisions and correct solutions. Specifically, for issues of data interpretation errors or risk assessment biases, the system generates interactive charts or graphs to compare the user's decision with the optimal outcome, highlighting financial impacts and showcasing alternative solutions. Interactive demonstrations provide an interactive simulation environment for users to practice correct operations. Specifically, for operational errors or complex decision failures, an interactive simulation environment is provided, allowing users to practice correct operations under guidance.

[0061] In one possible implementation, the depth and complexity of the aforementioned multimodal feedback content can be personalized based on user characteristics. For example, the depth and complexity of the analysis can be adjusted according to the user's level of understanding (based on test scores, learning progress rewards, and prior knowledge assessments). For instance, simpler, more basic explanations can be provided to users who have difficulty understanding basic concepts, while more detailed analysis can be offered to advanced users. Alternatively, the presentation of feedback can be adjusted based on the user's learning style (visual, auditory, or hands-on). Relevant application scenarios and examples can also be emphasized based on the user's interests.

[0062] Step 304: Dynamically adjust the user's subsequent learning path based on the effect of multimodal feedback.

[0063] After providing personalized multimodal feedback, the system dynamically adjusts the user's subsequent learning path based on the effectiveness of the feedback. For example, the system monitors the user's engagement with the feedback (e.g., time spent on feedback, retrying tasks) and the user's performance in subsequent related tasks. It then evaluates the effectiveness of the feedback based on this evaluation and dynamically adjusts the user's subsequent learning path accordingly. This dynamic adjustment may include dynamically adjusting the content and difficulty of subsequent levels, and inserting supplementary learning modules as needed to reinforce weak knowledge points.

[0064] For example, the system can calculate an metric called Efeedback to quantify the effectiveness of multimodal feedback in improving user understanding or performance. This could be expressed as Efeedback = (Score after feedback - Score before feedback) / Maximum possible improvement, or based on a reduction in the recurrence rate of a specific error type in subsequent levels. The system then dynamically adjusts the learning path based on the Efeedback. If the Efeedback is low, the system may re-emphasize problematic knowledge points through different resources, simpler levels, or alternative teaching methods. If the Efeedback is high, the system can move on to more advanced topics or related concepts, indicating that the user has mastered them.

[0065] The aforementioned multimodal feedback module clearly quantifies the effectiveness of feedback and its direct causal relationship with dynamic path adjustment, creating a "closed-loop" adaptive system. Through the multimodal feedback closed loop, the system can "learn how to teach better" by evaluating its own feedback strategies, providing timely and effective guidance when users encounter difficulties, and promoting deep learning and knowledge internalization.

[0066] like Figure 5 As shown, when the level path generation module 400 plans the user's learning path based on the user's initial knowledge state and performance in the learning level, it specifically includes the following steps: Step 401: Build and maintain a knowledge graph in the fintech field.

[0067] Build and continuously update a comprehensive fintech knowledge graph. Nodes in the knowledge graph represent knowledge points (e.g., "blockchain basics," "derivatives pricing," "risk management models"), and edges represent relationships, characterizing the strength of associations and learning order between knowledge points (e.g., prerequisites, relevance, yes / no, examples, required skills). Each knowledge point node is labeled with key attributes, including difficulty level (e.g., a 1-10 scale), importance weight (e.g., 0-1, indicating the degree of criticality to understanding core fintech), estimated learning time, and relevant content resources (e.g., videos, articles, quizzes, practice exercises). The knowledge graph is regularly updated by domain experts and may incorporate new fintech concepts, market developments, and regulatory changes through automated content analysis.

[0068] Step 402: Based on the user's initial knowledge state, plan the initial learning path on the knowledge graph.

[0069] Based on the user's initial knowledge state, an initial learning path is planned on the knowledge graph. In this embodiment, the Monte Carlo Tree Search (MCTS) algorithm can be used to plan the initial learning path on the knowledge graph. The user's initial knowledge state can be the user's knowledge state at the time of joining the company or setting a new learning goal. For example, when a user joins the company or sets a new learning goal, the Monte Carlo Tree Search (MCTS) algorithm is initialized. The "state" in MCTS represents the user's current knowledge state (inferred from pre-assessment, diagnostic tests, or initial learning performance) and learning goal (e.g., "become a certified fintech analyst," "understand decentralized finance"). The MCTS algorithm explores possible learning paths through the knowledge graph, aiming to find a path that maximizes cumulative rewards, balances knowledge coverage and learning depth, and ensures that key knowledge points are fully mastered. Each knowledge point or learning module is considered a potential "state" in the MCTS tree, and the transitions between them (e.g., completing a module, mastering a concept) are "actions." The “reward” in MCTS is defined by a multi-objective function that takes into account factors such as: progress toward the target (e.g., coverage of the required knowledge points), knowledge acquisition rate, user engagement metrics (e.g., completion rate, time spent) and adherence to time limits.

[0070] Step 403: When a specific event is triggered, dynamically adjust the learning path based on the user's performance in the learning level.

[0071] In this embodiment, a dynamic learning path adjustment event can be triggered by a specific event to ensure responsiveness. For example, a specific event could be completing a level or module, a significant performance deviation (e.g., repeated failures in understanding a concept, or unusually fast completion indicating mastery), successful application of multimodal feedback (indicated by a high or low feedback effectiveness score), or a user actively changing their learning goals or interests. When the dynamic learning path adjustment event is triggered, the MCTS algorithm dynamically adjusts the learning path based on the user's performance in the level. For example, it might rerun the algorithm using an updated current knowledge state (reflecting newly acquired knowledge or identified weaknesses), user engagement metrics, and feedback effectiveness score to explore a new optimal learning path from the current state. The learning path is continuously re-evaluated and adjusted based on real-time user performance and feedback. The modification logic for the learning path can include the following: Addressing weaknesses: If a knowledge point is identified as weak (e.g., low score, high error rate, low feedback), MCTS will prioritize planning a path that includes review modules, alternative explanations, reinforcement exercises, or simpler prerequisite modules for that specific knowledge point.

[0072] Leveraging strengths: For knowledge points already mastered, MCTS explores advanced challenges, in-depth research, or related extensions to maintain engagement and accelerate learning.

[0073] Integrating Risk Preferences: For users with a high risk appetite, MCTS may prioritize pathways that include more exploratory modules, high-volatility case studies, or scenario-based learning involving higher financial risks. For conservative users, MCTS tends to favor structured sequential pathways that include key risk management content, simulated low-risk scenarios, and an emphasis on fundamental concepts.

[0074] Step 404: Personalize content recommendations within the user's learning path.

[0075] In this embodiment, personalized content recommendations are made within the user's learning path to supplement learning materials and enrich the user's learning experience. Personalized content recommendations can include recommending relevant topics and cases based on the user's interests and career goals, providing timely learning content based on current hot topics and market dynamics, or recommending teaching resources and tools that match the user's learning style.

[0076] Personalized recommendation logic can include the following points: Interest-based: Match user-stated interests and implicitly inferred interests (derived from content consumption patterns) with topic tags and keywords in the knowledge graph.

[0077] Based on career goals: Based on the user's stated aspirations, content related to specific fintech roles (e.g., "blockchain developer", "quantitative analyst", "financial cybersecurity") is recommended.

[0078] Timeliness: Integrates real-time market data and news sources, suggesting relevant current events or emerging trends as case studies or supplementary reading to ensure that the content is always up-to-date and practical.

[0079] By generating adaptive level paths, the system can provide each user with a customized learning experience, optimizing learning efficiency and effectiveness.

[0080] The aforementioned learning path generation module, based on a knowledge graph in the fintech field, responds to user behavior in real time, considers personalized user needs, and adaptively generates targeted learning paths, providing each user with a customized learning path and improving user satisfaction. Through intelligent recommendation and dynamic adjustment, it optimizes the allocation and use of educational resources, balancing the coverage and depth of educational content, avoiding repetitive learning or knowledge gaps, and improving the efficiency of educational resource utilization.

[0081] The challenge generation system in the financial education field provided in this application provides a system that, through dynamic difficulty matching, keeps learners in a "fluid zone," avoiding boredom caused by overly easy challenges or frustration caused by overly difficult ones, thereby improving learner efficiency. Through a differentiated reward mechanism, it meets the psychological needs of learners with different risk appetites, enhancing user motivation, increasing user participation, and improving learning persistence. Through a multimodal feedback loop, it helps learners gain a deeper understanding of the causes of errors and improvement methods, improving their mastery and application of knowledge points. Finally, through adaptive challenge path generation, it provides a customized learning path for each user, improving user learning satisfaction.

[0082] To facilitate understanding by those skilled in the art, embodiments of this application provide a data processing flow for a checkpoint generation system in the field of financial education, specifically including the following steps: Step 501: The data acquisition module obtains the user's initial information, such as learning objectives, risk preferences, and initial knowledge status.

[0083] Step 502: The level path production module (based on MCTS) receives the user's initial information, plans the user's initial learning path, and inputs the user's initial learning path into the level presentation and user interaction module.

[0084] Step 503: The level presentation and user interaction module receives the user's current learning path, collects user behavior data in real time, and inputs it into the dynamic difficulty matching module and the multimodal feedback module.

[0085] Step 504: The dynamic difficulty matching module receives real-time user behavior data, user historical performance, and user personal characteristics, adjusts level parameters in real time, and inputs the training data of the DQN network (including state, action, reward, and next state) into the differentiated reward module.

[0086] Step 505: The differentiated reward module receives the training data, user performance, and risk preference of the DQN network, calculates the instant reward, and feeds it back to the dynamic difficulty matching module for DQN model optimization. At the same time, it outputs the learning progress index to the level path generation module.

[0087] Step 506: The multimodal feedback module (based on GPT-4) receives real-time user behavior data, analyzes the reasons for failure, generates feedback content, and inputs the feedback validity score into the level path generation module.

[0088] Step 507: The level path generation module (based on MCTS) receives the learning progress index, feedback effectiveness score, and updated knowledge status, and re-plans the subsequent learning path. This path is fed back to the level presentation and user interaction module to form a continuously optimized learning loop.

[0089] For example, the level generation system in the field of financial education provided in this application embodiment can be deployed in the following hardware environments: Server-side: Processor: Intel Xeon or AMD EPYC series multi-core processor.

[0090] Memory: At least 64GB RAM, 128GB or more recommended.

[0091] Storage: SSD storage, with capacity determined based on user scale and data volume.

[0092] GPU: NVIDIA Tesla V100 or A100 series GPUs, used for deep learning model training and inference.

[0093] Client: Supports PCs or mobile devices with mainstream web browsers.

[0094] At least 4GB of RAM is recommended to ensure a smooth multimedia experience.

[0095] A browser environment that supports HTML5 and WebGL.

[0096] The software architecture of this system adopts a front-end and back-end separation design, and mainly includes the following components: front end: Build responsive user interfaces using React.js or Vue.js.

[0097] Implement 3D interactive scenes using Three.js or Babylon.js.

[0098] Use ECharts or D3.js to visualize data.

[0099] rear end: Python is used as the primary development language.

[0100] Build RESTful API services using Django or Flask.

[0101] Implement deep reinforcement learning models using PyTorch or TensorFlow.

[0102] database: Use MongoDB to store user behavior data and learning records.

[0103] Use Redis to implement caching and real-time data processing.

[0104] Use Neo4j to store and query knowledge graphs.

[0105] middleware: Implement message queues and event-driven architectures using RabbitMQ or Kafka.

[0106] Implement full-text search and log analysis using Elasticsearch.

[0107] Implement containerized deployment and service orchestration using Docker and Kubernetes.

[0108] Based on the same technical concept, embodiments of this application provide a method for generating levels in the field of financial education, the method comprising: The learning stages of a user in the financial education process are adjusted based on the user's state space data; the user's state space data includes the user's learning behavior data, historical performance data, and personal characteristics in the financial education system; the action space includes adjustments to knowledge point density, interaction complexity, time pressure, and risk-reward ratio. Differentiated rewards are applied to users during their learning process based on their risk preference data; the user's risk preference data includes the user's initial assessment data, continuous behavioral data, and historical analysis data. Dynamically adjust the user's learning path during the learning process based on the user's failure event data; Plan the user's learning path based on the user's initial knowledge level and performance in the learning stages.

[0109] In one possible design, adjusting the user's learning stages in the financial education learning process based on the user's state space data includes: The user's current state space data is input into the DQN network to obtain the predicted Q-values ​​of all possible actions in the action space; The DQN network randomly selects an action from the action space with probability ε, selects the action with the highest predicted Q value with probability 1-ε, and dynamically adjusts the ε value. The system adjusts the user's subsequent learning levels based on the selected action and provides corresponding rewards based on the user's state space data at the next moment.

[0110] In one possible design, the DQN network is a dual DQN network; the DQN network employs experience replay technology to store and reuse historical interaction data.

[0111] In one possible design, the differentiated rewards based on user risk preference data during the user learning process include: Collect users' risk preference data; The user's risk preference data is quantified to obtain the user's risk preference score, and the user's risk preference category is classified according to the user's risk preference score; Differentiated rewards are given based on users' risk preference categories.

[0112] In one possible design, dynamically adjusting the user's learning path during the learning process based on the user's failure event data includes: Capture user failure event data; Multi-dimensional failure cause analysis is performed on user failure event data to obtain analytical data; Personalized, multimodal feedback is generated based on analyzed data for user failure events; The user's subsequent learning path is dynamically adjusted based on the effectiveness of multimodal feedback.

[0113] In one possible design, planning the user's learning path based on the user's initial knowledge state and performance in learning levels includes: Build and maintain a knowledge graph in the fintech field; Based on the user's initial knowledge state, plan the initial learning path on the knowledge graph; When a specific event is triggered, the learning path is dynamically adjusted based on the user's performance in the learning level; Personalized content recommendations are made within the user's learning path.

[0114] Based on the same technical concept, embodiments of this application provide a computing device, such as... Figure 6 As shown, it includes at least one processor 601 and a memory 602 connected to at least one processor. In this embodiment, the specific connection medium between the processor 601 and the memory 602 is not limited. Figure 6 Taking the connection between the processor 601 and the memory 602 via a bus as an example, the bus can be divided into address bus, data bus, control bus, etc.

[0115] In the embodiments of this application, the memory 602 stores instructions that can be executed by at least one processor 601. By executing the instructions stored in the memory 602, at least one processor 601 can execute the checkpoint generation system and method in the field of financial education listed above.

[0116] The processor 601 is the control center of the computing device. It can connect to various parts of the computer device through various interfaces and lines, and perform resource settings by running or executing instructions stored in the memory 602 and calling data stored in the memory 602.

[0117] Optionally, processor 601 may include one or more processing units. Processor 601 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into processor 601. In some embodiments, processor 601 and memory 602 may be implemented on the same chip; in some embodiments, they may be implemented separately on independent chips.

[0118] Processor 601 can be a general-purpose processor, such as a central processing unit (CPU), digital signal processor, application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0119] Memory 602, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory 602 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic memory, magnetic disk, optical disk, etc. Memory 602 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. In the embodiments of this application, memory 602 may also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0120] Based on the same technical concept, embodiments of this application also provide a computer-readable storage medium storing a computer-executable program, which is used to cause a computer to execute the checkpoint generation system and method in the field of financial education listed in any of the above embodiments.

[0121] Based on the same technical concept, embodiments of this application also provide a computer program product, which includes computer program code. When the computer program code is run on a computer, it causes the computer to execute the checkpoint generation system and method in the field of financial education as listed in any of the above methods.

[0122] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0127] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A level generation system for financial education, characterized in that, The system includes: The dynamic difficulty matching module is used to adjust the learning levels of users in the financial education learning process based on the user's state space data; The differentiated reward module is used to provide differentiated rewards based on users' risk preference data during the learning process; The multimodal feedback module is used to dynamically adjust the user's learning path during the learning process based on the user's failure event data; The level path generation module is used to plan the user's learning path based on the user's initial knowledge state and performance in the learning level.

2. The system according to claim 1, characterized in that, The dynamic difficulty matching module is specifically used for: The user's current state space data is input into the DQN network to obtain the predicted Q-values ​​for all possible actions in the action space. The user's state space data includes the user's learning behavior data in the financial education system, historical performance data, and personal characteristics. The action space includes adjustments for knowledge point density, interaction complexity, time pressure, and risk-reward ratio. The DQN network randomly selects an action from the action space with probability ε, selects the action with the highest predicted Q value with probability 1-ε, and dynamically adjusts the ε value. The system adjusts the user's subsequent learning levels based on the selected actions and provides corresponding rewards based on the user's state space data at the next moment.

3. The system according to claim 2, characterized in that, The DQN network is a dual DQN network; the DQN network uses experience replay technology to store and reuse historical interaction data.

4. The system according to claim 1, characterized in that, The differentiated reward module is specifically used for: Collect users' risk preference data; the users' risk preference data includes users' initial assessment data, continuous behavior data, and historical analysis data; The user's risk preference data is quantified to obtain the user's risk preference score, and the user's risk preference category is classified according to the user's risk preference score; Differentiated rewards are given based on users' risk preference categories.

5. The system according to claim 1, characterized in that, The multimodal feedback module is specifically used for: Capture user failure event data; Multi-dimensional failure cause analysis is performed on user failure event data to obtain analytical data; Personalized, multimodal feedback is generated based on analyzed data for user failure events; The user's subsequent learning path is dynamically adjusted based on the effectiveness of multimodal feedback.

6. The system according to claim 1, characterized in that, The checkpoint path generation module is specifically used for: Build and maintain a knowledge graph in the fintech field; Based on the user's initial knowledge state, plan the initial learning path on the knowledge graph; When a specific event is triggered, the learning path is dynamically adjusted based on the user's performance in the learning level; Personalized content recommendations are made within the user's learning path.

7. A method for generating levels in the field of financial education, characterized in that, The method includes: The learning stages of a user's financial education learning process are adjusted based on the user's state space data; the user's state space data includes the user's learning behavior data, historical performance data, and personal characteristics in the financial education system; the action space includes adjustments to knowledge point density, interaction complexity, time pressure, and risk-reward ratio. Differentiated rewards are applied to users during their learning process based on their risk preference data; the user's risk preference data includes the user's initial assessment data, continuous behavioral data, and historical analysis data. Dynamically adjust the user's learning path during the learning process based on the user's failure event data; Plan the user's learning path based on the user's initial knowledge level and performance in the learning stages.

8. A computing device, characterized in that, include: Memory, used to store program instructions; A processor is configured to invoke program instructions stored in the memory and execute the system or method as described in any one of claims 1 to 7 according to the obtained program instructions.

9. A computer-readable storage medium, characterized in that, Includes computer-readable instructions that, when read and executed by a computer, cause the system or method as described in any one of claims 1 to 7 to be implemented.

10. A computer program product, characterized in that, The computer program product includes: computer program code, which, when run on a computer, causes the computer to perform the system or method described in any one of claims 1 to 7.

Images