Personalized learning path planning method and system based on LSTM fused Attention

Abstract

The invention relates to the technical field of learning path planning, in particular to a personalized learning path planning method and system based on LSTM fused Attention, and the method comprises the steps: building an answer feature time sequence based on the information of an answer record, a wrong question table and a knowledge point set, generating an answer expression feature in a hidden state through a time sequence correlation trend extraction model, and carrying out the prediction of the answer expression feature in the hidden state; wherein the time sequence association trend extraction model is constructed based on an LSTM architecture; calculating a dynamic attention weight of the answer performance characteristics through an Attention model of a time decay item, a continuous answer difficulty feedback item and a knowledge point association item, and adjusting iteration in a hidden state based on the dynamic attention weight; generating a predicted knowledge point mastering degree through a fusion mapping model, wherein the fusion mapping model is constructed based on a full connection layer and an activation function; and generating a personalized recommended learning path based on the predicted knowledge point mastery degree. According to the invention, personalized learning path planning meeting individual requirements of students is realized, and the intelligence of a college teaching system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of learning path planning, and in particular to a personalized learning path planning method and system based on LSTM fusion Attention. Background Art

[0002] With the rapid development of artificial intelligence technology, higher education and teaching are gradually transforming towards intelligence and personalization. This is because if the course learning path adopts a standardized model, it is difficult to adapt to the differentiated needs of individual students in terms of knowledge base, learning habits and weak links, resulting in some students not being able to keep up and reducing their learning efficiency, which has become one of the key issues restricting the improvement of teaching quality.

[0003] In college teaching, in order to solve the above problems, the field of educational technology has gradually introduced personalized learning path planning methods in recent years, dynamically adjusting learning content by analyzing students' learning behavior data.

[0004] Currently, student response data on online learning platforms is mostly stored in databases formatted as knowledge graphs. This is because the knowledge points learned in most subject areas are not isolated and unrelated, often requiring learners to master one or more points before moving on to other points. Knowledge graphs (KGs), a core AI technology, can reflect the knowledge hierarchy of a subject area and represent the predecessor and successor relationships between knowledge point entities, including logical relationships such as parallelism.

[0005] Currently, some methods build student knowledge mastery models based on knowledge graphs, identifying weak links by mining the correlation between wrong questions and knowledge points, and more accurately tracking the evolution of learning status. However, existing methods still have the following limitations:

[0006] First, there's insufficient utilization of temporal features. Students' answering behavior is significantly time-dependent. The impact of early answering on current knowledge mastery diminishes over time, while performance on consecutively challenging questions can reflect the dynamic improvement or fluctuation of a student's ability. Therefore, existing LSTM models don't fully utilize the temporal features of student answers.

[0007] Second, the adaptability of the attention mechanism needs to be improved. While some methods have introduced attention mechanisms to enhance focus on key learning behaviors, the calculation of attention weights is often based on a single dimension, such as the accuracy of answering questions, without comprehensively considering factors such as time decay. This results in insufficient adaptability of attention allocation to actual learning scenarios, ultimately affecting the accuracy of personalized path planning.

[0008] Therefore, how to achieve accurate modeling of students' learning status and personalized learning path planning through a model based on LSTM fusion Attention is a technical problem that needs to be solved. Summary of the Invention

[0009] To this end, the present invention provides a personalized learning path planning method and system based on LSTM fusion Attention, which uses the LSTM fusion Attention model to predict the mastery of knowledge points using the knowledge graph data of the online learning platform, and conducts personalized learning path planning that fits the individual needs of students, thereby realizing the intelligence of university teaching systems.

[0010] To achieve the above objectives, the present invention proposes a personalized learning path planning method based on LSTM fused with Attention, comprising:

[0011] Constructing a time series of answer features based on the answer records, wrong answer list, and knowledge point set in the student answer knowledge graph, wherein the time series correlation trend extraction model is constructed based on the LSTM architecture, and the answer performance features are used as hidden states for iterative calculation;

[0012] In the process of generating the answer performance features, the answer performance features are calculated using an attention model that sets a time decay term, a continuous answer difficulty feedback term, and a knowledge point association term, and the iteration of the hidden state of the temporal correlation trend extraction model is adjusted based on the dynamic attention weight;

[0013] The answer performance features at multiple time steps are used to generate predicted knowledge point mastery through a fusion mapping model, wherein the fusion mapping model is constructed based on a fully connected layer and an activation function;

[0014] Based on the predicted knowledge point mastery, a personalized recommended learning path is generated in the student's answer knowledge graph.

[0015] Furthermore, the student answer knowledge graph also includes a graph relationship between answer records and questions. The process of calculating the dynamic attention weight includes:

[0016] The answer feature time series is passed through the multi-head attention mechanism of the Attention model to calculate the original attention score;

[0017] Calculate the time decay term based on the proximity of the current time step to the set recent time step and the learning rhythm trainable parameters;

[0018] Calculating a continuous answer difficulty feedback item based on the difficulty coefficients, trainable adjustment parameters, and activation functions of the questions corresponding to the consecutive correct answers and the consecutive incorrect answers in the answer records stored in the student answer knowledge graph;

[0019] Calculate the knowledge point association item based on the association between the knowledge point at the current time step and the target knowledge point in the knowledge point set;

[0020] The dynamic attention weight is determined based on a fusion calculation of the original attention score, the time decay item, the continuous question answering difficulty feedback item and the knowledge point association item.

[0021] Furthermore, the process of calculating the original attention score includes:

[0022] The original attention score is calculated by sequentially applying the tanh activation function and the softmax activation function to the splicing variables of the target knowledge point and the hidden state of the current time step, the trainable weighted parameters, and the trainable bias parameters.

[0023] Furthermore, the process of calculating the time decay term includes: calculating the difference between the current time step and the set recent time step, calculating the difference multiplied by the negative of the learning rhythm trainable parameter and then performing an exponential function operation to determine the time decay term;

[0024] The process of calculating the continuous answering difficulty feedback item includes: calculating the continuous answering difficulty feedback item through the sigmoid activation function of the current answer result vector of the answer record, the current continuous error mark vector and the splicing vector of the difficulty coefficient, and the trainable adjustment parameter.

[0025] Furthermore, the process of determining the dynamic attention weight by performing fusion calculation based on the original attention score, the time decay item, the continuous answering difficulty feedback item, and the knowledge point association item includes:

[0026] The product of the original attention score, the time decay item, the continuous answering difficulty feedback item and the knowledge point association item is used as a temporary attention score;

[0027] The temporary attention score is normalized by the softmax activation function to determine the dynamic attention weight.

[0028] Furthermore, the iterative process of adjusting the hidden state of the temporal correlation trend extraction model based on the dynamic attention weight includes:

[0029] Calculate the hidden state of the set recent time step and the corresponding weighted sum vector of the dynamic attention weight, and multiply the hidden state of the current time step and the concatenated value of the weighted sum vector and the gated trainable parameter by the sigmoid activation function with the hidden state of the current time step to determine the hidden state of the next time step of the iteration.

[0030] Furthermore, the training optimization process of the learning rhythm trainable parameters, trainable adjustment parameters, trainable weighting parameters, trainable bias parameters and gating trainable parameters includes:

[0031] The product of the time decay term, the continuous answering difficulty feedback term, and the knowledge point association term is used as the event importance weight, the mean square error of the real knowledge point mastery and the training knowledge point mastery of the sample data set is calculated, and the weighted mean square error term is constructed by multiplying the event importance weight and the mean square error;

[0032] Obtain the dependency relationship between the mastery of training knowledge points and historical knowledge points in the student's answering knowledge graph to construct a knowledge point relationship loss item;

[0033] The weighted mean square error term and the knowledge point relationship loss term are weightedly summed to generate a comprehensive loss function, and the Attention model and the fusion mapping model including the learning rhythm trainable parameters, trainable adjustment parameters, trainable weighting parameters, trainable bias parameters and gated trainable parameters are trained and optimized through the comprehensive loss function.

[0034] In the above solution, the temporal features of LSTM are refined and weighted, which is more in line with the knowledge mastery contained in students' actual answering behavior. The Attention model is also combined with the correlation between knowledge points in the knowledge graph to accurately locate key knowledge points in the learning path. By combining the model loss function of the knowledge graph, the model training is more in line with the knowledge progression rules in actual teaching, significantly improving the accuracy of predicting knowledge point mastery.

[0035] Furthermore, the process of constructing the time series of answering features includes:

[0036] Normalizing the answering time, answering duration, and question difficulty contained in the answering records stored in the student answering knowledge graph as elements of the answering temporal features;

[0037] The knowledge point dimensions and knowledge point IDs contained in the knowledge point set stored in the student answer knowledge graph are encoded as elements of the answer time series feature through one-hot encoding.

[0038] Furthermore, the process of generating a personalized recommended learning path in the student's answer knowledge graph based on the predicted knowledge point mastery includes:

[0039] Calculating candidate learning path scores for the student's answering knowledge graph based on the predicted knowledge point mastery, cognitive load, and student preference adaptation;

[0040] A personalized recommended learning path is selected from the candidate learning paths according to the ranking of the candidate learning path scores.

[0041] The present invention also provides a system for applying the personalized learning path planning method based on LSTM fusion Attention, comprising:

[0042] An LSTM answer performance feature generation module is connected to the student answer knowledge graph to construct an answer feature time series based on the answer records, wrong answer list, and knowledge point set in the student answer knowledge graph, and to generate answer performance features from the answer feature time series through a time series correlation trend extraction model, wherein the time series correlation trend extraction model is built based on the LSTM architecture and iteratively calculates the answer performance features as hidden states;

[0043] An attention adjustment module is connected to the LSTM answer performance feature generation module, and is used to calculate dynamic attention weights of the answer performance features through an attention model that sets a time decay term, a continuous answer difficulty feedback term, and a knowledge point association term during the process of generating the answer performance features, and adjust the iteration of the hidden state of the temporal correlation trend extraction model based on the dynamic attention weights;

[0044] A mapping output module, connected to the LSTM answer performance feature generation module, is used to generate a predicted knowledge point mastery degree by fusing the answer performance features of multiple time steps through a fusion mapping model, wherein the fusion mapping model is constructed based on a fully connected layer and an activation function;

[0045] A learning path planning module is connected to the student answer knowledge graph and the mapping output module to generate a personalized recommended learning path in the student answer knowledge graph based on the predicted knowledge point mastery.

[0046] Compared with the prior art, the present invention has the following advantages:

[0047] 1. The knowledge graph data of the online learning platform is used to predict the mastery of knowledge points through the LSTM-Attention model, and personalized learning path planning is carried out to meet the individual needs of students, realizing the intelligence of the university teaching system.

[0048] 2. We have implemented refined weighting of LSTM's temporal features to better reflect the knowledge mastery contained in students' actual answering behaviors. We have also implemented the Attention model, combined with the associations between knowledge points in the knowledge graph, to accurately locate key knowledge points in the learning path. By combining the model loss function of the knowledge graph, we have made model training more consistent with the knowledge progression rules in actual teaching, significantly improving the accuracy of predicting knowledge point mastery.

[0049] 3. The standardized construction of the time series of answer features is achieved, which improves the quality of model input. Through the collaborative modeling of LSTM and dynamic Attention, the temporal dynamics of the learning state are accurately captured, and personalized recommended learning paths are generated more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flowchart of a personalized learning path planning method based on LSTM fusion Attention in an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of the usage structure of the personalized learning path planning system based on LSTM fusion Attention in an embodiment of the present invention;

[0052] Figure 3 This is a flow chart of adjusting the hidden state of the Attention model in the personalized learning path planning method based on LSTM fusion Attention in an embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of the student answering knowledge graph structure of the personalized learning path planning method based on LSTM fusion Attention in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0055] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0056] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside", and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0057] Furthermore, it should be noted that, in the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0058] like Figures 1 to 4 As shown, the present invention provides a personalized learning path planning method and system based on LSTM fusion Attention, which uses the LSTM fusion Attention model to predict the mastery of knowledge points using the knowledge graph data of the online learning platform, and performs personalized learning path planning that fits the individual needs of students, thereby realizing the intelligence of the university teaching system.

[0059] like Figures 1 to 4 As shown, this embodiment proposes a personalized learning path planning method based on LSTM fusion Attention. The online learning platform integrates a student answer knowledge graph. The student answer knowledge graph includes a graph relationship between answer records corresponding to a wrong answer table and answer records corresponding to a knowledge point set. The personalized learning path planning method includes:

[0060] Constructing a time series of answer features based on the answer records, wrong answer tables, and knowledge point sets in the student answer knowledge graph, and applying the time series of answer features to a time series correlation trend extraction model to generate answer performance features, wherein the time series correlation trend extraction model is built based on an LSTM architecture and uses the answer performance features as hidden states for iterative calculation;

[0061] In the process of generating the answer performance features, the answer performance features are calculated using an attention model that sets a time decay term, a continuous answer difficulty feedback term, and a knowledge point association term, and the iteration of the hidden state of the temporal correlation trend extraction model is adjusted based on the dynamic attention weight;

[0062] The answer performance features at multiple time steps are used to generate predicted knowledge point mastery through a fusion mapping model, wherein the fusion mapping model is constructed based on a fully connected layer and an activation function;

[0063] Based on the predicted knowledge point mastery, a personalized recommended learning path is generated in the student's answer knowledge graph.

[0064] It is understandable that the model that integrates the LSTM (Long Short-Term Memory) and attention mechanism makes up for the shortcomings of static mining of knowledge graph information, making the prediction of knowledge point mastery more accurate and in line with students' actual situation.

[0065] Specifically, the temporal correlation trend extraction model outputs a sequence of hidden states across multiple time steps, reflecting information about a student's learning up to a set recent time step T, such as recent test performance and the number of knowledge point repetitions. An attention mechanism assigns a weight to the hidden state at each time step, focusing on the learning events most relevant to the current prediction.

[0066] Specifically, the dimension of the hidden layer output vector of the time series correlation trend extraction model is 4, the number of input nodes of the fusion mapping model is 4, the hidden nodes can be set to 3, the output node is 1, and the learning rate is 0.05 to avoid gradient explosion or training instability.

[0067] like Figure 3 As shown, further, the student answer knowledge graph also includes a graph relationship between answer records and questions, and the process of calculating the dynamic attention weight includes:

[0068] The answer feature time series is passed through the multi-head attention mechanism of the Attention model to calculate the original attention score;

[0069] Calculate the time decay term based on the proximity of the current time step to the set recent time step and the learning rhythm trainable parameters;

[0070] Calculating a continuous answer difficulty feedback item based on the difficulty coefficients, trainable adjustment parameters, and activation functions of the questions corresponding to the consecutive correct answers and the consecutive incorrect answers in the answer records stored in the student answer knowledge graph;

[0071] Calculate the knowledge point association item based on the association between the knowledge point at the current time step and the target knowledge point in the knowledge point set;

[0072] The dynamic attention weight is determined based on a fusion calculation of the original attention score, the time decay item, the continuous question answering difficulty feedback item and the knowledge point association item.

[0073] As you can understand, the time decay term quantifies the diminishing impact of earlier answering behavior on the current state. For example, a wrong answer from a week ago has less impact on current mastery than a wrong answer from yesterday. The difficulty feedback term for continuous answering reflects the incentive or inhibitory effect of changing difficulty on learning. The knowledge point association term integrates the graph relationships of a knowledge point set to identify weak links across knowledge points. Therefore, the Attention model can make LSTM-based models more accurate in predicting knowledge point mastery.

[0074] like Figure 3 As shown, further, the process of calculating the original attention score includes:

[0075] The original attention score is calculated by sequentially applying the tanh activation function and the softmax activation function to the splicing variable of the target knowledge point and the hidden state of the current time step, the trainable weight parameter, and the trainable bias parameter.

[0076] Specifically, the calculation process of the original attention score is:

[0077] score1=tanh(W1[h t ;e k ]+b1)

[0078] Where score1 represents the original attention score, tanh represents the tanh activation function, W1 and b1 represent the trainable weight parameter and the trainable bias parameter respectively, [h t ;e k ] represents the hidden state h at the current time step t t Embedding e of target knowledge point k splicing variables.

[0079] like Figure 3 As shown, further, the process of calculating the time decay term includes: calculating the difference between the current time step and the set recent time step, calculating the difference multiplied by the negative of the learning rhythm trainable parameter and then performing an exponential function operation to determine the time decay term;

[0080] The process of calculating the continuous answering difficulty feedback item includes: calculating the continuous answering difficulty feedback item by using the current answer result vector of the answer record, the current continuous error mark vector and the splicing vector of the difficulty coefficient, and the trainable adjustment parameter through the sigmoid activation function.

[0081] Specifically, the calculation process of the time decay term is:

[0082] decay t =exp(-γ·(Tt))

[0083] Where decay t represents the time decay term, γ represents the training parameter of the learning rhythm, T and t represent the set recent time step and the current time step respectively.

[0084] Specifically, the calculation process of the difficulty feedback item of continuous answering questions is:

[0085] β t =σ(W b [r t ;d t ;c t ]+b b )

[0086] Where, β t represents the difficulty feedback item of continuous answering questions, σ represents the sigmoid activation function, W b 、b b Represents two trainable adjustment parameters, [r t ;d t ;c t ] represents the current answer result vector r at the current time step t t , current continuous error mark vector d t and the difficulty coefficient c t The concatenated vector of the question, where the correct answer is r t =1, error = 0, difficulty coefficient is the normalized value, when the number of errors in the question exceeds the set number, d t The set number of times is determined according to the difficulty coefficient, for example, 3 times for more difficult questions and 5 times for easier questions.

[0087] like Figure 3 As shown, further, the process of determining the dynamic attention weight by performing fusion calculation based on the original attention score, the time decay item, the continuous answering difficulty feedback item and the knowledge point association item includes:

[0088] The product of the original attention score, the time decay item, the continuous answering difficulty feedback item and the knowledge point association item is used as a temporary attention score;

[0089] The temporary attention score is normalized by the softmax activation function to determine the dynamic attention weight.

[0090] Specifically, the knowledge point associated item is the knowledge point embedding vector e at the current time step t t and the embedding e of the target knowledge point k The cosine similarity of .

[0091] like Figure 3As shown, further, the iterative process of adjusting the hidden state of the temporal correlation trend extraction model based on the dynamic attention weight includes:

[0092] Calculate the hidden state of the set recent time step and the corresponding weighted sum vector of the dynamic attention weight, and multiply the hidden state of the current time step and the concatenated value of the weighted sum vector and the gated trainable parameter by the sigmoid activation function with the hidden state of the current time step to determine the hidden state of the next time step of the iteration.

[0093] Specifically, the calculation process of the weighted sum vector is:

[0094]

[0095] Where c represents the weighted sum vector, T and t represent the set recent time step and current time step respectively, α t represents the dynamic attention weight, h t Represents the hidden state at the current time step that belongs to the most recent time step.

[0096] Specifically, the calculation process of the hidden state of the next time step is:

[0097] h t+1 =σ(W g [h t ;c]+b g )·h t

[0098] Where h t+1 、h t They represent the hidden state of the next time step and the hidden state of the current time step, T represents the set recent time step, σ represents the sigmoid activation function, and W g 、b g represents two gated trainable parameters, [h t ; c] represents the concatenated value of the hidden state and weighted sum vector c at the current time step, and · represents the element-by-element product.

[0099] Furthermore, the training optimization process of the learning rhythm trainable parameters, trainable adjustment parameters, trainable weighting parameters, trainable bias parameters and gating trainable parameters includes:

[0100] The product of the time decay term, the continuous answering difficulty feedback term, and the knowledge point association term is used as the event importance weight, the mean square error of the real knowledge point mastery and the training knowledge point mastery of the sample data set is calculated, and the weighted mean square error term is constructed by multiplying the event importance weight and the mean square error;

[0101] Obtain the dependency relationship between the mastery of training knowledge points and historical knowledge points in the student's answering knowledge graph to construct a knowledge point relationship loss item;

[0102] The weighted mean square error term and the knowledge point relationship loss term are weightedly summed to generate a comprehensive loss function, and the Attention model and the fusion mapping model including the learning rhythm trainable parameters, trainable adjustment parameters, trainable weighting parameters, trainable bias parameters and gated trainable parameters are trained and optimized through the comprehensive loss function.

[0103] Specifically, the calculation process of the weighted mean square error term is:

[0104]

[0105] Where, L WMSE represents the weighted mean square error term, y t ′、y t They represent the mastery of training knowledge points and real knowledge points respectively, and η represents the importance weight of the event.

[0106] Specifically, the calculation process of the knowledge point relationship loss term is:

[0107]

[0108] Where, L prior represents the knowledge point relationship loss term, k1→k2∈ξ indicates that the current knowledge point k1 is the previous knowledge point of the historical knowledge point k2 in the student's answer knowledge graph ξ, y′ k1 、y k1 Respectively represent the training knowledge point mastery and the true knowledge point mastery of the current knowledge point at the current time step, and ε represents a small error value such as 0.1 to avoid absolute coercion. Therefore, the knowledge point relationship loss term uses the current knowledge point mastery to be less than the previous knowledge point mastery, making it consistent with the learning process relationship between knowledge points.

[0109] The calculation process of the comprehensive loss function is:

[0110] Loss=λL WMSE +(1-λ)L prior

[0111] In the formula, Loss represents the comprehensive loss function, L WMSE represents the weighted mean square error term, L prior represents the knowledge point relationship loss term, and λ represents the weighted value.

[0112] In the above solution, the temporal features of LSTM are refined and weighted, which is more in line with the knowledge mastery contained in students' actual answering behavior. The Attention model is also combined with the correlation between knowledge points in the knowledge graph to accurately locate key knowledge points in the learning path. By combining the model loss function of the knowledge graph, the model training is more in line with the knowledge progression rules in actual teaching, significantly improving the accuracy of predicting knowledge point mastery.

[0113] like Figure 4 As shown, further, the information of the answer record in the student answer knowledge graph includes the answer time, answer time and question difficulty, and the information of the knowledge point set includes the knowledge point dimension and knowledge point ID. The process of constructing the answer feature time series includes:

[0114] Normalizing the answering time, answering duration, and question difficulty contained in the answering records stored in the student answering knowledge graph as elements of the answering temporal features;

[0115] The knowledge point dimensions and knowledge point IDs contained in the knowledge point set stored in the student answer knowledge graph are encoded as elements of the answer time series feature through one-hot encoding.

[0116] Specifically, the information of the wrong question table includes the answer results and score rates, which are directly used as elements of the answer time series feature.

[0117] Furthermore, the process of generating a personalized recommended learning path in the student's answer knowledge graph based on the predicted knowledge point mastery includes:

[0118] Calculating candidate learning path scores for the student's answering knowledge graph based on the predicted knowledge point mastery, cognitive load, and student preference adaptation;

[0119] A personalized recommended learning path is selected from the candidate learning paths according to the ranking of the candidate learning path scores.

[0120] Specifically, the cognitive load is used to avoid concentrating on learning difficult content, and is the ratio of the current difficulty coefficient to the total content difficulty coefficient. The student preference adaptation is the student's preference for learning time for video resources, exercise resources, etc., and is the ratio of each type of resource to the total learning time.

[0121] like Figure 2 As shown, this embodiment also provides a system for applying the personalized learning path planning method based on LSTM fusion Attention. The online learning platform integrates a student answer knowledge graph. The student answer knowledge graph includes a graph relationship between answer records corresponding to a wrong answer table and answer records corresponding to a knowledge point set. The system applied to the online learning platform includes:

[0122] An LSTM answer performance feature generation module is connected to the student answer knowledge graph and is used to construct an answer feature time series based on the answer records, wrong answer list, and knowledge point set in the student answer knowledge graph, and to generate answer performance features from the answer feature time series through a time series correlation trend extraction model, wherein the time series correlation trend extraction model is built based on the LSTM architecture and iteratively calculates the answer performance features as hidden states;

[0123] An attention adjustment module is connected to the LSTM answer performance feature generation module, and is used to calculate dynamic attention weights of the answer performance features through an attention model that sets a time decay term, a continuous answer difficulty feedback term, and a knowledge point association term during the process of generating the answer performance features, and adjust the iteration of the hidden state of the temporal correlation trend extraction model based on the dynamic attention weights;

[0124] A mapping output module, connected to the LSTM answer performance feature generation module, is used to generate a predicted knowledge point mastery degree by fusing the answer performance features of multiple time steps through a fusion mapping model, wherein the fusion mapping model is constructed based on a fully connected layer and an activation function;

[0125] A learning path planning module is connected to the student answering knowledge graph and the mapping output module to generate a personalized recommended learning path in the student answering knowledge graph based on the predicted knowledge point mastery.

[0126] In the above solution, the standardized construction of the time series of answer features is achieved, the quality of model input is improved, and through the collaborative modeling of LSTM and dynamic Attention, the temporal dynamics of the learning state is accurately captured, achieving more accurate generation of personalized recommended learning paths.

[0127] In this embodiment, the knowledge graph data of the online learning platform is used to predict the mastery of knowledge points through the LSTM fusion Attention model, and personalized learning path planning that fits the individual needs of students is carried out, realizing the intelligence of the university teaching system. The refined weighting of the time series features of LSTM is achieved, which is more in line with the knowledge mastery contained in the students' actual answering behavior. The Attention model is also combined with the correlation relationship of knowledge points in the knowledge graph to accurately locate the key knowledge points in the learning path. By combining the model loss function of the knowledge graph, the model training is made more in line with the knowledge progression law in actual teaching, significantly improving the accuracy of predicting the mastery of knowledge points. The standardized construction of the time series of answering features is achieved, improving the quality of model input. Through the collaborative modeling of LSTM and dynamic Attention, the time series dynamics of the learning state are accurately captured, and more accurate generation of personalized recommended learning paths is achieved.

[0128] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0129] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A personalized learning path planning method based on LSTM fusion Attention, characterized by: include: Constructing a time series of answer features based on the answer records, wrong answer tables, and knowledge point sets in the student answer knowledge graph, and applying the time series of answer features to a time series correlation trend extraction model to generate answer performance features, wherein the time series correlation trend extraction model is built based on an LSTM architecture and uses the answer performance features as hidden states for iterative calculation; In the process of generating the answer performance features, the answer performance features are calculated using an attention model that sets a time decay term, a continuous answer difficulty feedback term, and a knowledge point association term, and the iteration of the hidden state of the temporal correlation trend extraction model is adjusted based on the dynamic attention weight; The answer performance features at multiple time steps are used to generate predicted knowledge point mastery through a fusion mapping model, wherein the fusion mapping model is constructed based on a fully connected layer and an activation function; Based on the predicted knowledge point mastery, a personalized recommended learning path is generated in the student's answer knowledge graph.

2. The personalized learning path planning method based on LSTM fusion Attention according to claim 1 is characterized in that: The process of calculating the dynamic attention weight includes: The answer feature time series is passed through the multi-head attention mechanism of the Attention model to calculate the original attention score; Calculate the time decay term based on the proximity of the current time step to the set recent time step and the learning rhythm trainable parameters; Calculating a continuous answer difficulty feedback item based on the difficulty coefficients, trainable adjustment parameters, and activation functions of the questions corresponding to the consecutive correct answers and the consecutive incorrect answers in the answer records stored in the student answer knowledge graph; Calculate the knowledge point association item based on the association between the knowledge point at the current time step and the target knowledge point in the knowledge point set; The dynamic attention weight is determined based on a fusion calculation of the original attention score, the time decay item, the continuous question answering difficulty feedback item and the knowledge point association item.

3. The personalized learning path planning method based on LSTM fusion Attention according to claim 2 is characterized in that: The process of calculating the raw attention score includes: The original attention score is calculated by sequentially applying the tanh activation function and the softmax activation function to the splicing variables of the target knowledge point and the hidden state of the current time step, the trainable weighted parameters, and the trainable bias parameters.

4. The personalized learning path planning method based on LSTM fusion Attention according to claim 2 is characterized in that: The process of calculating the time decay term includes: calculating the difference between the current time step and the set recent time step, calculating the difference multiplied by the negative of the learning rhythm trainable parameter and then performing an exponential function operation to determine the time decay term; The process of calculating the continuous answering difficulty feedback item includes: calculating the continuous answering difficulty feedback item through the sigmoid activation function of the current answer result vector of the answer record, the current continuous error mark vector and the splicing vector of the difficulty coefficient, and the trainable adjustment parameter.

5. The personalized learning path planning method based on LSTM fusion Attention according to claim 2 is characterized in that: The process of determining the dynamic attention weight by performing fusion calculation based on the original attention score, the time decay item, the continuous answering difficulty feedback item, and the knowledge point association item includes: The product of the original attention score, the time decay item, the continuous answering difficulty feedback item and the knowledge point association item is used as a temporary attention score; The temporary attention score is normalized by the softmax activation function to determine the dynamic attention weight.

6. The personalized learning path planning method based on LSTM fusion Attention according to claim 3 is characterized in that: The iterative process of adjusting the hidden state of the temporal correlation trend extraction model based on the dynamic attention weight includes: Calculate the hidden state of the set recent time step and the corresponding weighted sum vector of the dynamic attention weight, and multiply the hidden state of the current time step and the concatenated value of the weighted sum vector and the gated trainable parameter by the sigmoid activation function with the hidden state of the current time step to determine the hidden state of the next time step of the iteration.

7. The personalized learning path planning method based on LSTM fusion Attention according to claim 6 is characterized in that: The training optimization process of the learning rhythm trainable parameters, trainable adjustment parameters, trainable weighting parameters, trainable bias parameters and gating trainable parameters includes: The product of the time decay term, the continuous answering difficulty feedback term, and the knowledge point association term is used as the event importance weight, the mean square error of the real knowledge point mastery and the training knowledge point mastery of the sample data set is calculated, and the weighted mean square error term is constructed by multiplying the event importance weight and the mean square error; Obtain the dependency relationship between the mastery of training knowledge points and historical knowledge points in the student's answering knowledge graph to construct a knowledge point relationship loss item; The weighted mean square error term and the knowledge point relationship loss term are weightedly summed to generate a comprehensive loss function, and the Attention model and the fusion mapping model including the learning rhythm trainable parameters, trainable adjustment parameters, trainable weighting parameters, trainable bias parameters and gated trainable parameters are trained and optimized through the comprehensive loss function.

8. The personalized learning path planning method based on LSTM fusion Attention according to any one of claims 1 to 7, characterized in that: The process of constructing the time series of answering features includes: Normalizing the answering time, answering duration, and question difficulty contained in the answering records stored in the student answering knowledge graph as elements of the answering temporal features; The knowledge point dimensions and knowledge point IDs contained in the knowledge point set stored in the student answer knowledge graph are encoded as elements of the answer time series feature through one-hot encoding.

9. The personalized learning path planning method based on LSTM fusion Attention according to any one of claims 1 to 7, characterized in that: The process of generating a personalized recommended learning path in the student's answer knowledge graph based on the predicted knowledge point mastery includes: Calculating candidate learning path scores for the student's answering knowledge graph based on the predicted knowledge point mastery, cognitive load, and student preference adaptation; A personalized recommended learning path is selected from the candidate learning paths according to the ranking of the candidate learning path scores.

10. A system using the personalized learning path planning method based on LSTM fusion Attention according to any one of claims 1 to 9, characterized in that: include: An LSTM answer performance feature generation module is connected to the student answer knowledge graph to construct an answer feature time series based on the answer records, wrong answer list, and knowledge point set in the student answer knowledge graph, and to generate answer performance features from the answer feature time series through a time series correlation trend extraction model, wherein the time series correlation trend extraction model is built based on the LSTM architecture and iteratively calculates the answer performance features as hidden states; An attention adjustment module is connected to the LSTM answer performance feature generation module, and is used to calculate dynamic attention weights of the answer performance features through an attention model that sets a time decay term, a continuous answer difficulty feedback term, and a knowledge point association term during the process of generating the answer performance features, and adjust the iteration of the hidden state of the temporal correlation trend extraction model based on the dynamic attention weights; A mapping output module, connected to the LSTM answer performance feature generation module, is used to generate a predicted knowledge point mastery degree by fusing the answer performance features of multiple time steps through a fusion mapping model, wherein the fusion mapping model is constructed based on a fully connected layer and an activation function; A learning path planning module is connected to the student answering knowledge graph and the mapping output module to generate a personalized recommended learning path in the student answering knowledge graph based on the predicted knowledge point mastery.

Images