Multi-modal interactive Chinese grammar mastering method

Abstract

The embodiment of the invention provides a multi-modal interactive Chinese grammar mastering method, which comprises the following steps of: generating an initial input sentence complexity regulation and control parameter on the basis of historical learning data of a user and a real-time semantic understanding evaluation result; selecting a target sentence pattern and a grammar rule matched with the current cognitive level of the user according to the complexity regulation and control parameter; generating an example sentence containing the target sentence pattern, and displaying the example sentence to the user in a multi-mode form for interactive learning; and adjusting the complexity regulation and control parameter based on the feedback of the user, and realizing the output of the dynamically optimized learning content. Through the scheme of the embodiment of the invention, the problem of how to regulate and control the complexity of the input sentence according to the real-time semantic understanding level of the user so as to solve the cognitive overload problem of a learner in Chinese grammar mastering can be solved.

Description

Technical Field

[0001] This application relates to the technical field of natural language processing, and specifically relates to a multimodal interactive Chinese grammar mastery method. Background Art

[0002] A multimodal interactive Chinese grammar mastery method provides rich learning resources and interactive experiences for learners by integrating various media such as text, images, and sounds, aiming to help learners master Chinese grammar more efficiently. This method is user-centered and emphasizes the design of personalized learning paths. However, one problem it faces is how to dynamically adjust the complexity of the input sentences according to the user's real-time semantic understanding level, so as to avoid the phenomenon of cognitive overload caused by overly complex content for learners. The key to this problem lies in the need to accurately evaluate the learner's current level and flexibly adjust the difficulty of the learning materials through intelligent means to ensure the efficiency and smoothness of the learning process. Summary of the Invention

[0003] In view of this, the embodiments of the present disclosure provide a multimodal interactive Chinese grammar mastery method, which at least partially solves the problems existing in the prior art.

[0004] A multimodal interactive Chinese grammar mastery method includes: Generating an initial input sentence complexity regulation parameter based on the user's historical learning data and real-time semantic understanding evaluation results; Selecting a target sentence pattern and grammar rules that match the user's current cognitive level according to the complexity regulation parameter; Generating an example sentence containing the target sentence pattern and presenting it to the user in a multimodal form for interactive learning; Adjusting the complexity regulation parameter based on the user's feedback to achieve dynamic optimization of the output of learning content.

[0005] In a specific embodiment, the generating an initial input sentence complexity regulation parameter based on the user's historical learning data and real-time semantic understanding evaluation results further includes: Obtaining the user's historical correct rate data R and error type distribution E; Setting the initial regulation parameter α0 to the standard level 5; Adjusting α = α0 * (R / R_max) based on the following formula, where R_max represents the historical highest correct rate, α represents the current regulation parameter, and R is the user's historical average correct rate; If α exceeds the preset upper limit value α_max, then take α = α_max to ensure reasonable and effective regulation.

[0006] In a specific embodiment, the weight calculation for grammar categories is added in the said steps: Define a set G of grammar rule categories; According to historical data, count the proportion P(g) of each grammar type in errors, where g ∈ G; Use the weight update control parameter, α = α * Σ(P(g) * β(g)), where β(g) represents the difficulty coefficient of category g; If P(g)>Th_g (threshold), adjust the control influence of this category to a higher priority.

[0007] In a specific embodiment, the dynamic time weighting mechanism is added to the said steps: Introduce a duration variable t to represent the timestamp of the most recent training interval; Define the time discount function w(t) = e^(λt), where λ controls the decay rate; Incorporate the discount weight into the calculation, α = α * w(t); Ensure that α automatically decreases as t increases to adapt to the situation of long-term non-review.

[0008] In a specific embodiment, the criteria for classifying the user's cognitive level are refined in the said steps: Output the real-time correct rate R_real and the degree of understanding deviation δ through the semantic parsing module; If δ is greater than the deviation tolerance limit δ_threshold, increase the control strength coefficient η; Use the new control formula γ = η * α / (1 + δ)^2 to update the final control strength γ; Here, γ determines the complexity of the subsequent sentence patterns.

[0009] In a specific embodiment, a cognitive model correction rule is added: Assume that the cognitive model error ε follows a distribution N(μ,σ^2); Introduce the Bayesian estimation method to calculate the corrected complexity parameter α_new = α * exp(ε^2 / (2σ^2)); Evaluate the model stability by comparing the difference between α_new and the original control parameter; Set the critical point condition, |α - α_new| ≤ Th_diff to determine whether the model needs to be recalibrated.

[0010] In a specific embodiment, the way of paying attention to low-frequency error types in the adjustment strategy is improved: Extract the proportion φ of low-frequency wrong questions and the average score S_low in historical data; Calculate the comprehensive regulation factor F = k*(S_low / φ), where k represents the importance weight; Use the hybrid formula Δ = F*γ + γ to record the additional regulation amount; Verify that |F - F_old| does not exceed the specified limit to ensure the effect of gradual adjustment.

[0011] In a specific implementation, extend the dynamic adjustment logic to support context-dependent grammar rules: For the contextualized sentence S, record the number cnt of the target grammar rule Cx that appears in the context; According to the rule density cnt, adjust the scaling factor θ = exp(b / cnt) of γ, where b is used to adjust the sensitivity; Update the total regulation effect γ = θ * γ to ensure adaptation to the regular changes of complex sentences; Check the γ value range [γ_min, γ_max] and enforce the constraints.

[0012] In a specific implementation, further refine the assessment of the interaction effect between the target grammar rule and the user behavior data: Statistically calculate the cognitive load ci corresponding to each grammar rule Ri and normalize it to C = [c_1,..., c_k]; Define the load weight adjustment formula Ci = Ci^(1 - ρ*Ri), where ρ represents the behavior response rate; Combine the historical accuracy rate R_i to update the regulation item as Di = Ci*Ri^τ (τ > 0); Use the summation term ΣDi to modify the total dynamic control index and perform convergence optimization processing.

[0013] In a specific implementation, add constraint conditions for personalized recommendation to avoid overloading: Calculate the cognitive burden pressure p = p_base(z*p_rate) for the real-time progress z of each user; Apply the inequality to judge the constraint condition. If p ≥ p_limit to prevent further increasing the load, then limit γ ≤ γ_reduce; At the same time, combine the learning mode m of the user and select the most suitable regulation step A = min{A, m_A}, where A and m_A are the fixed and mode-adapted parameters respectively; Achieve adaptive control during the learning process through the above process.

[0014] An embodiment of the present disclosure provides a multimodal interactive Chinese grammar mastering method, including: generating an initial input sentence complexity adjustment parameter based on the user's historical learning data and real-time semantic understanding evaluation results; selecting a target sentence pattern and grammar rules that match the user's current cognitive level according to the complexity adjustment parameter; generating an example sentence containing the target sentence pattern and presenting it to the user in a multimodal form for interactive learning; adjusting the complexity adjustment parameter based on the user's feedback to achieve dynamic optimization of the output of learning content. Through the solution of the embodiment of the present disclosure, it is possible to solve the problem of how to adjust the complexity of the input sentence according to the user's real-time semantic understanding level to address the cognitive overload problem in the learner's Chinese grammar mastering. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions of the exemplary embodiments of the present disclosure, the drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present disclosure and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 is a flowchart of a multimodal interactive Chinese grammar mastering method; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In the following, only some exemplary embodiments are briefly described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present application. Therefore, the drawings and descriptions are considered to be exemplary rather than restrictive in nature.

[0018] Next, with reference to the drawings, a multimodal interactive Chinese grammar mastering method of the present invention is described. This method provides learning content suitable for the user's current semantic understanding level by dynamically adjusting the complexity of the input sentence and the output of information in a multimodal form, effectively reducing the cognitive burden on the learner. The following are the specific operation steps and detailed analysis.

[0019] First, the method extracts key metrics from the user's historical learning data and their real-time semantic understanding assessment results to generate an initial input sentence complexity regulation parameter. This process relies on big data analysis and artificial intelligence algorithms to deeply mine the user's historical behavior, combined with data such as the latest test scores and types of wrong questions to judge the learner's weak grammar points and the existing cognitive ability level. For example, in practical applications, after completing a set of exercises, the user may show more misunderstandings of the ba-construction sentences and be relatively proficient in simple sentence patterns. The system thus converts these performances into specific scoring dimensions and sets the basic threshold of the complexity parameter accordingly. In addition, to ensure real-time performance, short quizzes are arranged before and after each interaction session as a dynamic monitoring mechanism. For example, the system asks the user questions like "Which of the following sentences do you think is more natural?" to quickly capture their current grasp of language use and immediately feedback it into the next selection.

[0020] Secondly, after determining the basic threshold, according to the aforementioned complexity regulation parameter, it filters and adapts the target sentence patterns at the user's current stage and the related grammar rules hidden behind them. This is the core step because it determines the quality and effect of all subsequent example demonstrations. The selection criteria usually include two aspects: one is to be close to the edge of the existing knowledge structure - that is, the difficulty is slightly higher than the current level but still within the acceptable range; the other is to take into account comprehensiveness - ensuring that important or frequently used items are covered. At this stage, rule library matching technology may be used to automatically select eligible items, and manual review is used to ensure the content quality is correct. For example, for a beginner in Chinese at the elementary level, if the system recognizes that he has just learned the basic subject-verb-object structure, it can further recommend a slightly higher-level version involving the combination of locative words: "I put the book on the table", rather than jumping into the completely unfamiliar passive sentence pattern "bei".

[0021] The third step is to construct an example model and use multi-sensory media to deliver it to the user for an immersive learning experience. This means not only limited to text descriptions, but also multimedia means such as pictures, audio, and even interactive games will be used to comprehensively strengthen the memory points. Specifically, assuming the sentence "I put the book on the table" mentioned above, which expresses the action of position movement, is selected, it can show the visual picture of a book moving from the hand to the table while accompanied by the real-person voice pronunciation, and then let the user drag the icons to operate and confirm whether the order is correct to test the internalization effect. The advantage of doing this is that it can mobilize multiple regions of the brain to work together, thereby improving efficiency and maintaining the interest.

[0022] The last but equally important step is to adjust the originally set control values based on a continuous feedback loop mechanism to achieve the final fine-tuning goal. Make corresponding changes after considering various data collected by the methods introduced above, such as factors like the percentage of correct answers or the average difference in completion time. For example, when the number of consecutive incorrect answers to questions of the same type exceeds the defined tolerance value, immediately roll back to the simplified process and re-consolidate the foundation. Conversely, if the progress is smooth, gradually increase the challenge level until new uncharted areas are reached and continue to explore the direction of progress. In this way, a truly personalized teaching service model for each individual can be realized, meeting the needs of people with different backgrounds and levels.

[0023] In summary, a multimodal interactive Chinese grammar mastery method successfully solves the problem of cognitive overload that may be faced during the learning process by scientifically refining and decomposing the operation mechanism of each step and introducing advanced technology to ensure the seamless connection of all links, enabling everyone to find the most suitable growth path for themselves.

[0024] Next, the content of generating the initial input sentence complexity adjustment parameter based on the user's historical learning data and real-time semantic understanding evaluation results of the present invention is described. Specifically, it includes the following steps: First, obtain the user's historical correct rate data R and error type distribution E. Subsequently, set the initial adjustment parameter α0, and adjust the current adjustment parameter α through a formula. If α exceeds the preset upper limit α_max, its value is limited to α_max.

[0025] Obtaining the user's historical correct rate data R and error type distribution E is the first key step. Among them, R represents the correct rate statistical value in the learning task calculated from the user's historical data. This value reflects the user's overall learning situation, usually in the form of a percentage (range from 0 to 1). At the same time, extracting the user's error type distribution E can deeply understand the specific areas of the user's common grammar problems.

[0026] Setting the initial adjustment parameter α0 to a standard level is the next operation. In this method, the initial value is usually defined as a standard starting value, for example, set to 5. This value represents a balanced state and is the benchmark adjustment strength applicable in the absence of other references.

[0027] Subsequently, the current regulation parameter value is calculated using the formula α = α0 * (R / R_max). The parameters in this formula have clear meanings: α0 represents the standard starting value mentioned above, while R and R_max represent the user's average historical correct rate and the historical highest correct rate respectively. Incorporating the correct rate as a variable into the formula aims to ensure that the regulation parameter changes dynamically according to the user's actual situation. When the correct rate approaches the user's own best record, the intensity of adjustment tends to relax or remain the same, otherwise the difficulty increases.

[0028] If the value of α exceeds the preset upper limit α_max, the regulation result is set to be equal to this limit value. This setting aims to prevent the complexity from deviating too high or too low from the reasonable range.

[0029] In one embodiment, taking Chinese grammar learning as an example. Suppose a student's historical correct rate data is R = 0.8, and the highest correct rate the student has achieved historically is R_max = 0.95. The initial regulation parameter is set to the standard value α0 = 5. According to the formula α = α0 * (R / R_max), it is calculated that α = 4.21. Additionally, assume the upper limit value is α_max = 6. Then at this time, since α is not greater than α_max, no additional processing is required. If R is smaller, it will result in a lower α or even close to zero. However, due to the need for a certain training difficulty in the actual scenario, the effectiveness is ensured by setting a minimum lower limit.

[0030] Therefore, in the specific application process, the complexity change of grammar questions can be effectively controlled through such precise quantification methods. The ultimate goal is to optimize the personalized teaching path so that each student can continuously improve their skills at an appropriate level.

[0031] Next, the calculation of the weight for grammar categories is added in the steps of the present invention. Define the set G of grammar rule categories, which is used to represent various types of Chinese grammar rules to be considered. This set includes but is not limited to categories such as sentence component analysis, word order rules, punctuation usage, and tense collocations. Statistically analyze the historical data to determine the proportion P(g) of each grammar g in the user's error records, that is, the frequency of each grammar category in the errors. Introduce the regulation parameter α and set its value to the result of multiplying the original regulation parameter by the formula Σ(P(g) * β(g)). Here, β(g) represents the difficulty coefficient of the grammar category, reflecting the general difficulty of users when learning specific rules. Finally, determine whether the error proportion P(g) of a certain grammar category g exceeds the preset threshold Th_g. If it exceeds, the influence intensity of the regulation on this category is increased through an adjustment method.

[0032] Specifically, in one embodiment, assume there is a set of grammar rules G = {subject-verb agreement, article usage, clause construction}, and through the analysis of the historical practice data of thousands of students on the multimodal interactive Chinese grammar mastery platform, the error ratios for each category are obtained as P(subject-verb agreement) = 0.3, P(article usage) = 0.25, P(clause construction) = 0.45. At the same time, the difficulty coefficient is defined according to questionnaire surveys or platform statistics data. For example, β(subject-verb agreement) = 1.0, β(article usage) = 0.8, β(clause construction) = 1.2. Based on the above data, the formula α = α * Σ(P(g) * β(g)) is calculated, where α is a dynamically updated variable. The calculation of this formula can be interpreted as an adjustment of the influence value after comprehensively weighing the probability of grammar errors occurring and their actual difficulty. Assume the initial regulation parameter value is α = 1.0. Substituting these data into the formula, we get α = 1.0 * (0.3 * 1.0 + 0.25 * 0.8 + 0.45 * 1.2), and the calculated new α value is approximately 1.18.

[0033] For example, when the error frequency of a certain grammar category exceeds a specific threshold Th_g (such as Th_g is set to 0.4), then in this step, a higher-priority training content or a reminder feedback mechanism for this category will be triggered. Such an adjustment helps to focus on solving high-frequency problems and optimize the user experience. In addition, the selection of the parameters β(g) and the threshold Th_g should comprehensively consider the learning habits of the target users and the characteristics of the platform. Generally speaking, the range of α can vary between [0.8, 2], the value of the difficulty coefficient β(g) is more suitable in the range of [0.5, 1.5] for normal situations, and the error ratio threshold is usually set to [0.2, 0.6] to effectively capture important learning problem areas. The design logic of the formula is based on the scientific prediction of the error distribution and the learning bottlenecks of users to form a reasonable regulation basis, making the learning process more efficient and personalized.

[0034] Next, the content of adding a dynamic time weighting mechanism to the steps of the present invention is described. The duration variable t is introduced to represent the time stamp of the interval of the most recent training; the time discount function w(t) = e^(λt) is defined, where λ controls the decay rate; the discount weight is incorporated into the calculation, α = α * w(t); ensure that α automatically decreases as t increases to adapt to the situation of not reviewing for a long time.

[0035] The first step is to introduce a duration variable \(t\) representing the time interval since the last training. This variable records the number of seconds, days, or any reasonable unit of time that has passed since the user's last learning activity with the multimodal interaction system. For example, if the user's last learning time was March 6, 2025, the value calculated on March 13 would be 7 days.

[0036] The second step is to define the time discount function \(w(t)=e^{\lambda t}\). In this function, \(\lambda\) is a parameter that controls the rate of time decay, and its typical range is between -1 and -0.1. The default value can be set to -0.01 (optimal value) based on experience. The design of this formula is based on the idea of the time forgetting model, where the memory of content naturally decreases over time. The use of \(e\) ensures that the exponential function is smooth and predictable. By adjusting the value of \(\lambda\), the process of human memory retention of knowledge can be accurately simulated: a smaller \(\lambda\) value indicates a slower forgetting rate, while a larger \(\lambda\) value corresponds to a faster forgetting situation.

[0037] The third step involves applying the discount weight to the current learning factor \(\alpha\). The original learning rate \(\alpha\) represents the intensity of strengthening the mastery of grammar knowledge, but it is changed to a new calculation form \(\alpha=\alpha * w(t)\) to re - calibrate the weight according to the value of \(t\). This design aims to reflect the changing trend of the learning effort required after different lengths of intervals. When the time interval is short (such as just after reviewing a knowledge point), \(\alpha\) is close to the maximum effective value at this time; on the contrary, if the corresponding information has not been accessed for a long time, \(\alpha\) gradually decreases towards zero due to the exponential decay law.

[0038] Taking an example, in an embodiment, in a Chinese grammar learning platform, a student is learning the structure of object complements. Suppose they were previously familiar with this sentence pattern and left the application for five days without returning to consolidate the relevant concepts. When they enter the system again, a lower new learning efficiency \(\alpha\) is automatically evaluated according to the above formula to adapt to this situation. Therefore, more intensive or repetitive prompts may be arranged to help them better re - understand this grammar knowledge point. Specifically, if the default basic intensity for each course is initially set at 10 points, then after five days, according to the aforementioned rules, it may be adjusted down to a level of about 8 or 9 as a compensation strategy, thus optimizing the overall learning plan configuration process.

[0039] Next, the criteria for classifying the user's cognitive level in the steps of the present invention are described. Through step - by - step calculations and analyses in multiple links, the classification of the user's cognitive level can be dynamically adjusted, and the sentence complexity of interactive Chinese grammar learning can be optimized. The following are the specific steps and their meanings: First, the semantic parsing module outputs the real-time accuracy rate R_real and the degree of understanding deviation δ. This step is used to evaluate the user's current understanding of specific Chinese grammar rules. The real-time accuracy rate reflects the user's answering accuracy in the multimodal interaction task, and its value range is [0, 1]; the degree of understanding deviation δ represents the gap between the user's answer and the standard answer, and the higher its value, the greater the deviation of the user's understanding from the standard answer. For example, in one embodiment, after the semantic parsing module analyzes the user's answer regarding the subject-predicate-object structure in the sentence "He likes to eat apples", it is found that the deviation degree δ = 0.2, and R_real = 0.85.

[0040] Secondly, if the deviation degree δ is greater than the set deviation tolerance limit value δ_threshold, increase the regulation strength coefficient η. The deviation tolerance limit value here is a threshold, usually set within the range of 0.1 to 0.3, and the default optimal value is 0.2. This condition is set to ensure that when the user's understanding is significantly incorrect, more intervention is carried out to quickly correct the user's deviation. Specifically, in the above example, if the system-set δ_threshold = 0.15, then δ = 0.2 > δ_threshold meets the condition for increasing regulation.

[0041] In the third step, use the new regulation formula γ = η * α / (1 + δ)^2 to update the final regulation intensity γ. In this formula, α is the initial adjustment parameter, and its value range is [0.5, 2], and its default value is 1. The design of the formula takes into account the importance of the deviation degree for sentence pattern design: as the understanding deviation increases, the denominator (1 + δ)^2 of the formula increases faster, causing the final result γ to decrease rapidly, so that the output sentence pattern is more simplified and easy to understand, in line with the concept of reducing complexity in educational principles to relieve the user's understanding burden. For example, if the initial parameter α = 1, the regulation coefficient η = 1.5, and the deviation δ = 0.2, substituting into the formula can obtain γ = 1.5 * 1 / (1 + 0.2)^2 = 0.976.

[0042] The final regulation intensity γ is used to determine the adjustment plan for the complexity of the subsequent output sentence pattern. When the user gradually improves their understanding level, the system can recalculate the regulation intensity based on a lower deviation degree and generate more complex exercise content. This dynamic process ensures that the user can steadily improve their mastery of grammar knowledge at an appropriate difficulty level.

[0043] Next, describe the addition of the cognitive model correction rule of the present invention. Assume that a multimodal interactive Chinese grammar mastery method has been constructed and it is necessary to optimize the system's stability and accuracy by adding a cognitive model correction rule.

[0044] First, clarify the distribution characteristics of the cognitive model error ε. Assume that the error follows a normal distribution N(μ,σ^2). This step aims to quantify the gap between the current model prediction and the real data and lay a foundation for subsequent parameter adjustment. Here, ε represents the cognitive error of the model, usually obtained from experiments or simulations; the distribution parameter μ is the mean of the error, representing the average deviation degree of the system, and its value range can be any real number; σ^2 is the variance of the error, reflecting the degree of model output fluctuation, and it needs to be a non-negative real value, and its optimal value depends on the fault tolerance of the specific application scenario.

[0045] Next, introduce the Bayesian estimation method to calculate the corrected complexity parameter α_new = α * exp(ε^2 / (2σ^2)). Here, α is the complexity parameter of the original model, used to control the smoothness of the model fitting curve, and its value is greater than zero; the formula combines the model error ϵ with its uncertainty (i.e., σ in the distribution) to affect the change amplitude of the complexity parameter in an exponential form. This setting is based on the theoretical basis: when the model error increases, the complexity parameter α will change significantly to adapt to a larger error amplitude, thereby dynamically adjusting the model structure. Conversely, a smaller error will result in a slight change in the complexity parameter, keeping the system stable to a certain extent.

[0046] Then, evaluate the stability of the model by comparing the complexity parameters before and after correction. Calculate the distance between the two by the absolute value difference |α - α_new|, and use this to judge whether the correction effect is obvious. For example, in a multi-modal interactive Chinese grammar mastery model, if the user inputs a set of language samples with dialect pronunciation interference, the model error may temporarily increase. If the complexity parameter adjusted by this formula changes too much, it means that the current training parameters have low adaptability to the new samples.

[0047] Subsequently, define the critical point condition, that is, |α - α_new| ≤ Th_diff to judge whether the model needs to be recalibrated. Among them, the threshold Th_diff is a boundary condition set artificially, used to limit the acceptable deviation amplitude of the complexity parameter. Specifically, in the Chinese grammar teaching scenario, if a certain module focuses on identifying the subject-predicate-object sentence pattern and suddenly encounters a large number of passive voice sentence inputs, the original parameters may deviate from the optimal solution. At this time, if |α - α_new| exceeds the preset threshold, the recalibration program is triggered to update the model training strategy to cope with new grammar rules or data patterns. In the example, if Th_diff is set to 0.15 (assumed to be the best empirical value derived from historical data), and it is found that the parameter difference reaches 0.2 after a certain correction, it is determined that the model needs to be comprehensively readjusted to ensure the effectiveness and reliability of long-term operation.

[0048] Next, describe the way of paying attention to low-frequency error types in the improved adjustment strategy of the present invention: Extract the proportion φ and the average score S_low of low-frequency wrong questions in the historical data. This step analyzes the exercise records of previous learners, counts the proportion φ of the occurrence times of low-frequency error types in the total number of errors, and further calculates the average score S_low of these error types to evaluate their difficulty and importance. The range of φ is (0, 1], and the range of S_low is set according to the scoring rules. In the scenario with a full score of 100, the value range is [0, 100].

[0049] Calculate the comprehensive regulation factor F = k*(S_low / φ), where k is the importance weight. This formula indicates that the attention to low-frequency errors is positively correlated with the inverse ratio of the average score of the error type to its occurrence proportion φ. The value range of k is usually adjusted in [0.5, 2] to meet the requirements of different situations. When the best setting k = 1, it ensures that the focus is balanced and moderate; since low-frequency wrong questions are easily overlooked but often involve relatively complex or special grammar points, it is necessary to increase the attention to make up for the distribution deviation.

[0050] Use the mixed formula Δ = F*γ + γ to record the additional regulation amount. γ represents a dynamic parameter (usually taking values between 0 and 1), which is used to balance immediacy and stability. Δ, as the final regulation increment, reflects the change in the learning resources to be allocated. For example, in multi-modal interactive Chinese grammar teaching, Δ generally indicates that for relatively complex and less frequently used knowledge points such as the conjunction structure "both... and...", more intensive training is needed to make up for the neglect probability.

[0051] Verify that |F - F_old| does not exceed the specified limit to ensure the effect of gradual adjustment. This operation aims to limit the excessive oscillation caused by each strategy iteration and promote smooth evolution to adapt to the actual environmental requirements. Assume that in an embodiment, the usage rule of the special interrogative pronoun "which" in the division of a certain Chinese sentence component is initially set as the object of investigation. Specifically, it is found that the difference in the regulation factor maintains a reasonable level during several consecutive iterations, which realizes the above conditional constraint function.

[0052] Generally speaking, in the process of formulating the optimization plan, data analysis means are fully utilized, and reasonable quantitative indicators are used to guide the selection and implementation of adjustment measures, so as to better serve the core goal of improving the understanding effect of Chinese grammar.

[0053] Next, the extended dynamic adjustment logic of the present invention is described to support context-dependent grammar rules: For a contextual sentence S, the number cnt of the target grammar rules Cx included in the context of this sentence is recorded; in this step, the system identifies and tracks the application frequency of grammar rules in each specific context. For example, in a specific embodiment, assuming that the sentence involves the dialogue of characters in a historical scene, archaic syntax or diction patterns under specific etiquette may be frequently used as grammar rules.

[0054] Then, the scaling factor θ = exp(b / cnt) of γ is adjusted according to the rule density cnt, where the parameter b can be understood as a setting parameter for the sensitivity of this dynamic adjustment strategy, which is used to affect the degree of the final adjustment factor; generally speaking, b can be taken within the range of positive real numbers, and the preferred range is (0, 5]. The significance of this formula is to flexibly change the scaling ratio according to the different occurrence frequencies of the target grammar rule Cx: the smaller the rule density (i.e., the smaller cnt), it means that the current grammar element is relatively special or rare. At this time, increasing the divisor in the exp function can increase the scaling ratio to ensure sufficient attention, so that the overall control is more inclined to reflect these few but significant key grammar regulations. Specifically, if a certain special expression appears only a few times in a certain context (when the count cnt is very low), θ will have an obvious increasing trend, indicating that the influence of such situations should be increased to accurately reflect its importance. At the same time, in the optimal case, when the effect of neither exaggerating nor ignoring is desired, a more reasonable parameter value b can be determined based on a large number of experimental results, and it is generally possible to set it between 1 and 3.

[0055] After that, an update process is performed to summarize the control effect γ = θ * γ, ensuring that the new adjustment factor θ obtained through the previous calculation is multiplied by the original control parameter γ, thereby adapting to the changes in the more complex internal logic and external manifestation forms of sentences, ensuring that the system can still effectively adjust its own judgment mechanism when facing various sentence structure combinations, and improving the accuracy of Chinese semantic understanding and generation in a multi-modal interaction environment. For example, considering a teaching dialogue system containing classical literary fragments, for some obscure and rare traditional word collocations and syntax, the newly generated adjustment coefficient is appropriately increased, making the entire learning process not only scientific but also more in line with the actual needs of the educational object.

[0056] It is crucial to check and keep the adjusted γ value within the predetermined range [γ_min, γ_max] and implement this constraint measure to maintain the stability and controllability of the system and prevent the adjustment work from getting out of control beyond the predetermined boundaries. For example, setting the minimum value γ_min can avoid the risk of overlooking some grammar rules due to overly relaxed regulation; similarly, setting the maximum value γ_max is to prevent the problem of excessive tightening that may occur under extreme conditions. In an example, if it is found that the γ value in a certain situation exceeds the specified upper limit due to specific reasons, actions need to be taken to bring this value back to the normal range to prevent misleading the students' understanding of Chinese grammar rules during the teaching process.

[0057] Next, the interactive influence assessment between the further refined target grammar rules and the user behavior data of the present invention will be described.

[0058] The first step is to count the cognitive load ci corresponding to each grammar rule Ri and normalize it to C = [c_1,..., c_k]. This process refers to analyzing each Chinese grammar rule item by item and quantifying the cognitive complexity ci of each rule based on existing literature or experimental data (for example, the passive sentence has a higher cognitive difficulty compared to the declarative sentence). The range of the cognitive load ci is usually set to [0, 1], where 1 indicates that the rule is the most cognitively challenging. Subsequently, all cis are linearly normalized to obtain the normalized load list C. The purpose of normalization is to make the comparison of loads more intuitive and easier to control in subsequent calculations.

[0059] The second step defines the load weight adjustment formula Ci = Ci^(1 - ρ*Ri), where ρ represents the behavior response rate. This step introduces behavior data to dynamically correct the cognitive load weight Ci of the grammar rule. In the setting of this formula, the range of ρ is usually [0, 1], and its optimal value depends on the actual system response rate. The smaller the parameter ρ, the weaker the influence of the change in user behavior on the grammar rule weight; on the contrary, when ρ is larger, the load weight Ci will adjust more quickly with the change of the value of Ri. The significance of this setting is to dynamically adjust the importance of grammar rules through behavior data.

[0060] The third step combines the historical accuracy rate R_i to update the regulation item as Di = Ci*Ri^τ (τ > 0). This step introduces the user's long-term learning behavior as a key variable. Here, R_i is the historical correct answering ratio of a certain user under a specific grammar rule, and the exponent τ controls the influence strength of the historical record. Specifically, if τ increases, the accumulation effect of long-term learning achievements will be more emphasized. The parameter Di can be understood as the actual difficulty level of the rule, which is the weighted result of integrating the immediate cognitive complexity and the historical accuracy rate.

[0061] In the fourth step, the total dynamic control index is modified using the summation term ΣDi, and convergence optimization processing is performed. That is, the dynamic comprehensive control value of the system is obtained through the summation operation of each grammar rule regulation term Di. In addition, to ensure the stability of the algorithm, a convergence optimization method is also adopted to ensure that the interaction between the load and behavior data can quickly adapt to the new environment. The optimization process usually adjusts hyperparameters according to the actual situation, such as the values of τ and ρ, to obtain a more stable dynamic control system.

[0062] For example, in one embodiment, specifically assume that a beginner in Chinese grammar practices using the passive sentence rule (R3) and the subject-verb-object structure rule (R4), and the initial loads of these two rules have been measured as c_3 = 0.85 and c_4 = 0.35 respectively. After normalization, the loads are C3 = 0.76 and C4 = 0.22 respectively. By monitoring the learner's behavioral response time, ρ≈0.7 is determined, and combined with R_i, the historical correct rates R3 = 0.4 and R4 = 0.9 are obtained, and setting τ≈1.2, D3≈0.5 and D4≈0.2 can be calculated. By comparing Di, it is clearly shown that the more difficult passive sentence rule, rather than the relatively proficient subject-verb-object rule, currently requires more reinforcement, thus making the learning resource allocation more efficient.

[0063] Next, the process of the present invention for adding personalized recommendation constraint conditions to avoid overloading is described.

[0064] First, list the steps. First, calculate the cognitive burden pressure for each user's real-time progress. The formula is p = p_base(z * p_rate), where z represents the user's current learning progress, usually in the range of [0,1] (representing the percentage of learning completion), and p_rate represents the stress ratio coefficient, and the default value is preferably 1.2 to 1.8. Second, apply inequality judgment constraint conditions according to the cognitive burden pressure. Specifically, if the pressure value p≥p_limit (the optimal value of p_limit is about 0.7 to 1.0), then the regulation parameter γ is restricted so that γ does not exceed γ_reduce. Third, calculate the optimal regulation step A = min{A, m_A} in combination with the user's inherent learning mode m, where A is usually a fixed regulation step (such as a preset value of 0.1 or 0.2), and m_A is related to a specific learning mode. Fourth, achieve adaptive control in the multi-modal learning process by synthesizing the above results.

[0065] The significance of each step is as follows: In the first step, the cognitive burden of the user is calculated, and the base pressure value p_base is corrected based on the product of their learning depth z and the pressure multiplication factor p_rate. This can dynamically estimate the changing pattern of learning pressure and prepare for adjustment. The second step is the core judgment mechanism, which ensures that the learning burden does not grow excessively to a level that may cause fatigue or inefficiency, and stabilizes the volatility of the overall learning curve through the threshold judgment limit γ. The third step minimizes the two regulation parameters A and m_A to maintain both the general adjustment intensity and personal preferences. The last step integrates this information to ensure that Chinese grammar teaching progresses step by step under appropriate challenges without imposing too much psychological burden on the user.

[0066] In one embodiment, for example, when a beginner is using this method to learn the structure of Chinese compound sentences. If the system detects that the user has mastered half of the knowledge of the main-subordinate complex sentence (z = 0.5) and has accumulated a certain amount of pressure (such as setting p_rate = 1.5, and the resulting pressure value may be p > p_limit). At this time, the system will immediately reduce the frequency of difficulty increase, and at the same time optimize the adjustment amplitude A each time through the learning habit m to ensure that the amount of practice gradually decreases without overly disturbing the coherence and scientific nature of the learning plan. The specific numerical settings in this process are selected from the optimal range based on the feedback of educational psychology experiments and the verification results of big data analysis to accurately balance the relationship between efficiency and personal comfort.

[0067] A multi-modal interactive Chinese grammar mastery method of the present invention includes: In the context of globalization today, more and more non-native Chinese speakers hope to master Chinese more efficiently and accurately, especially for the Chinese language with its unique grammar structure and rich expression forms. To help learners effectively overcome the cognitive burden and learning difficulties in the process of mastering Chinese grammar, the present invention provides a brand-new multi-modal interactive learning solution. Through scientific and rigorous methodological guidance, combined with big data and advanced algorithm technologies, it precisely adapts to each user's learning path and rhythm to achieve twice the result with half the effort.

[0068] In terms of the overall process, the first step of the multi-modal interactive Chinese grammar mastery method is to base on the individual differences of users - specifically referring to the historical learning data accumulation here, such as the previous wrong question situation, answering speed, etc., which can indirectly reflect information about the user's level, and to evaluate the user's understanding of semantics in real time, generating the complexity regulation parameters (Complexity Regulation Parameters) for the initial input sentence, that is, the difficulty index used to define the sentence corresponding to the grammar knowledge points to be taught. This measure enables teachers to set an age-appropriate and personalized task starting point for students, thus better meeting the needs of personalized teaching and solving the specific pain points that are difficult to reach by traditional batch processing; in addition, at the initial stage of the whole process, this method can also quickly establish a relatively stable and reliable reference framework, providing a solid foundation for subsequent dynamic updates.

[0069] Subsequently, according to the pre-set parameters, sample sentences and grammar rules suitable for the target sentence patterns are carefully selected. These sentences should not only be strictly screened or created according to the adjusted complexity (such as simplifying vocabulary selection, controlling sentence length), but also take into account factors such as cultural background relevance to ensure their compliance with the expected teaching value and actual application scenarios. Then, they are presented to learners through diverse media methods: perhaps in the form of a combination of graphics, audio, or simulated interactions, aiming to strengthen understanding and memory from multiple perspectives, increasing the fun and sense of participation while enhancing the learning effect.

[0070] The most crucial step is to continuously collect and analyze the immediate responses of learners. In this two-way communication channel, the system can not only sense whether the user understands the current content, but more importantly, can capture those subtle but important non-structural data such as emotional tendencies or thinking bottlenecks, and then fine-tune the complexity of the input sentence set previously to ensure that it is always in the optimal state. The multi-modal interactive Chinese grammar mastery method precisely solves the challenging task of how to adjust the difficulty of grammar examples in real time according to the semantic understanding of different individuals through continuously optimizing such a self-learning mechanism. Specifically, it means avoiding cognitive overload caused by overly difficult materials or boredom caused by being too simple by flexibly adjusting the complex Chinese sentence input strategy, enabling each participant to obtain the best learning experience.

[0071] To sum up, this learning method designed based on users' history and real-time assessment can not only achieve the goal of personalized teaching, but also greatly enhance the educational quality and students' enthusiasm, ultimately promoting the construction of a more fair and efficient educational environment.

[0072] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another.

[0073] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0074] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of the features. In the description of the present application, "a plurality of" means two or more unless otherwise specifically defined.

[0075] Any process or method description represented in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of code including one or more executable instructions for implementing a specific logical function or process. And the scope of the preferred embodiments of the present application includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in a reverse order according to the involved functions, rather than in the order shown or discussed.

[0076] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a sequenced list of executable instructions for implementing a logical function, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatus, or devices.

[0077] It should be understood that each part of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the method in the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0078] In addition, each functional unit in various embodiments of the present application can be integrated into a processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a magnetic disk or an optical disc, etc.

[0079] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various changes or substitutions, and these should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multimodal interactive Chinese grammar mastering method, characterized in that: include: Generate the complexity control parameters of the initial input sentence based on the user's historical learning data and real-time semantic understanding evaluation results; According to the complexity control parameter, select the target sentence pattern and grammatical rules that match the user's current cognitive level; Generate sample sentences containing the target sentence pattern and present them to users in a multimodal form for interactive learning; The complexity control parameters are adjusted based on user feedback to achieve dynamic optimization of the output of learning content.

2. A multimodal interactive Chinese grammar mastering method according to claim 1, characterized in that: The generating of the initial input sentence complexity control parameter based on the user's historical learning data and real-time semantic understanding evaluation results further includes: Obtain the user's historical data on accuracy R and error type distribution E; The initial control parameter α0 is set to the standard level 5; Adjust α based on the following formula = α0 * (R / R_max), where R_max represents the highest historical accuracy, α represents the current control parameter, and R is the user's historical average accuracy; If α exceeds the preset upper limit α_max, then α = α_max is taken to ensure that the regulation is reasonable and effective.

3. A multimodal interactive Chinese grammar mastering method according to claim 2, characterized in that: The weight calculation of grammatical categories is added in the steps: Define a grammar rule category set G; According to historical data, calculate the proportion of each type of grammar in errors P(g), g∈G; Use the weight to update the control parameter, α = α * Σ(P(g) * β(g)), where β(g) represents the difficulty coefficient of category g; If P(g) > Th_g (threshold), adjust the regulatory impact of this category to a higher priority.

4. A multimodal interactive Chinese grammar mastering method according to claim 3, characterized in that: The steps described incorporate a dynamic time weighting mechanism: The duration variable t is introduced to represent the timestamp of the most recent training interval; Define the time discount function w(t) = e^(λt), where λ controls the decay rate; Incorporating the discount weight into the calculation, α = α * w(t); Make sure that α automatically decreases as t increases to accommodate long periods of no review.

5. A multimodal interactive Chinese grammar mastering method according to claim 4, characterized in that: The steps described refine the criteria for classifying user awareness levels: The semantic parsing module outputs the real-time accuracy R_real and the degree of understanding deviation δ; If δ is greater than the deviation tolerance limit δ_threshold, then increase the control intensity coefficient η; Use the new control formula γ = η * α / (1+δ)^2 to update the final control intensity γ; Here γ determines the complexity of the subsequent sentence.

6. A multimodal interactive Chinese grammar mastering method according to claim 5, characterized in that: Add cognitive model correction rules: Assume that the cognitive model error ε satisfies the distribution N(μ,σ^2); Introduce the Bayesian estimation method to calculate the corrected complexity parameter α_new = α * exp(ε^2 / (2σ^2)); The model stability is evaluated by comparing the difference between α_new and the original control parameters; Set the critical point condition, |α-α_new| ≤ Th_diff to determine whether the model needs to be recalibrated.

7. A multimodal interactive Chinese grammar mastering method according to claim 6, characterized in that: Improved the way the tuning strategy focuses on low-frequency error types: Extract the proportion φ and average score S_low of low-frequency wrong questions in historical data; Calculate the comprehensive control factor F = k*(S_low / φ), where k represents the importance weight; The hybrid formula Δ = F*γ + γ is used to record the additional control amount; Verify that |F F_old| does not exceed the specified bounds to ensure the effectiveness of the gradual adjustment.

8. A multimodal interactive Chinese grammar mastering method according to claim 7, characterized in that: Extended dynamic adjustment logic supports context-dependent grammar rules: For the contextualized sentence S, the number cnt of the target grammatical rule Cx that appears in the context of the situation is recorded; According to the rule density cnt, adjust the scaling factor θ=exp(b / cnt) of γ, where b is used to adjust the sensitivity; Update the total regulation effect γ = θ * γ to ensure adaptation to the regular changes of complex sentences; Check the gamma value range [γ_min, γ_max] and enforce the constraint.

9. A multimodal interactive Chinese grammar mastering method according to claim 8, characterized in that: Further refine the interaction impact assessment between the target grammatical rules and user behavior data: Count the cognitive load ci corresponding to each grammatical rule Ri and normalize it into C=[c_1,...,c_k]; Define the load weight adjustment formula Ci=Ci^(1 - ρ*Ri), where ρ represents the behavior response rate; Combined with the historical accuracy R_i, the updated control term is Di=Ci*Ri^τ (τ>0); The cumulative term ΣDi is used to modify the overall dynamic control index and perform convergence optimization.

10. A multimodal interactive Chinese grammar mastering method according to claim 9, characterized in that: Added constraints for personalized recommendations to avoid overload: Calculate cognitive load pressure p=p_base(z*p_rate) for each user's real-time progress z; Apply inequalities to determine constraints. If p ≥ p_limit prevents further load increase, then constrain γ ≤ γ_reduce. At the same time, combined with the user's learning mode m, the most suitable control step A=min{A,m_A} is selected, where A and m_A are fixed and mode-adaptive parameters respectively; Adaptive control is achieved during the learning process through the above process.

Images