Teaching object interaction method, system and device

Abstract

The invention discloses a teaching object interaction method, system and device. The teaching object interaction method comprises the steps of obtaining a plurality of interaction behavior characteristics of a real-time user at a current knowledge point; constructing a plurality of interaction behavior characteristics of the real-time user at the current knowledge point into a real-time interaction vector; inputting the real-time interaction vector into a learning recognition model, performing forward propagation after pre-training in the learning recognition model, and generating a mastering state label of a real-time user at the current knowledge point; according to the mastering state label of the real-time user at the current knowledge point, constructing a teaching interaction strategy of a plurality of to-be-learned knowledge points in a future time axis; wherein the mastering state labels of the plurality of knowledge points to be learned are inherited in a training sample of the learning recognition model; according to the method, the accuracy of learning path deduction and the individuation of learning strategies are improved, and the user learning experience of a teaching system is remarkably improved by combining real-time mastering state prediction and deduction based on a historical behavior mode.

Description

Technical Field

[0001] The present invention relates to the field of teaching interaction, and specifically to a teaching object interaction method, system and device. Background Art

[0002] In existing online education and teaching systems, the mastery status labels in the teaching object interaction process are usually predicted in real time based on learners' behavioral data (such as answering question accuracy, click frequency, stay time, etc.) to determine the learners' mastery of each knowledge point. The prior art generally adopts a classification prediction model based on a single learning behavior, and judges the mastery status of learners towards the knowledge points to be learned through this model. However, the prior art has the following defects:

[0003] Existing teaching systems usually rely on a single-point prediction model. Whenever a learner enters a new knowledge point, the system needs to re-predict the mastery status of this knowledge point. The limitation of this method is that the mastery status of each knowledge point needs to be independently evaluated, lacking a systematic consideration of the learner's overall learning path and behavioral trajectory. Therefore, the accuracy and stability of the prediction results are relatively low. Moreover, most of the prior art only focuses on the direct relationship between real-time learning behavioral data and the mastery status of the current knowledge point, ignoring the structured inheritance of historical user behaviors and learning patterns. Even in the case of similar learning paths, the system still needs to make predictions for each new knowledge point, failing to effectively utilize historical learning behavior patterns to accelerate the reasoning process. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides a teaching object interaction method, system and device, and solves the technical problems raised in the background art by inheriting the knowledge point mastery labels of the most similar users to construct the real-time learning interaction strategy of the current user.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0006] A teaching object interaction method, characterized in that the interaction method includes:

[0007] S1. Obtain several interaction behavior characteristics of the real-time user at the current knowledge point;

[0008] S2. Construct the several interaction behavior characteristics of the real-time user at the current knowledge point into a real-time interaction vector;

[0009] S3. Input the real-time interaction vector into the learning recognition model, and perform forward propagation after pre-training in the learning recognition model to generate the mastery status label of the real-time user at the current knowledge point;

[0010] S4. Construct teaching interaction strategies for a number of to-be-learned knowledge points on the future timeline according to the mastery status labels of real-time users for the current knowledge point; among them, the mastery status labels of the number of to-be-learned knowledge points are inherited from the training samples of the learning recognition model.

[0011] In some of the embodiments, the pre-training steps of the learning recognition model include:

[0012] A1. Construct an interactive training matrix for pre-training; among them, the matrix elements of the interactive training matrix are binary training samples corresponding to historical knowledge points, and the binary training samples include the historical interaction vectors of the historical knowledge points and their mastery status labels;

[0013] A2. Starting from the initial row vector of the interactive training matrix, frame the interactive training matrix with a rated framing width, and divide the interactive training matrix into an initial framed matrix and a to-be-framed matrix;

[0014] A3. Input the initial framed matrix into a supervised learning model for training, and through iterative training, obtain the learning recognition model.

[0015] In some of the embodiments, constructing an interactive training matrix for pre-training includes:

[0016] A1-1. Assign index numbers based on the learning order to T historical knowledge points to generate an index number sequence;

[0017] A1-2. Assign acquisition numbers to K historical users to generate an acquisition number sequence;

[0018] A1-3. Collect a number of interaction behavior characteristics of the k-th historical user for the t-th historical knowledge point, and construct them into a historical interaction vector; where t represents the index number of the historical knowledge point, and k represents the acquisition number of the historical user;

[0019] A1-4. Collect the unit test scores of the k-th historical user for the t-th historical knowledge point, and map the unit test scores to the mastery status labels of the historical knowledge point;

[0020] A1-5. Define the historical interaction vector of the historical knowledge point as the input feature and the mastery status label of the historical knowledge point as the target label, and construct the training sample of the historical knowledge point;

[0021] A1-6. Obtain the training samples of the k-th historical user for T historical knowledge points, and generate the training sample set of the k-th historical user; where T represents the total number of historical knowledge points;

[0022] A1-7. According to the index number of the historical knowledge point, sort the training sample set of the k-th historical user in sequence to generate the first row vector of this historical user;

[0023] A1-8. Obtain the first row vectors of K historical users, and align and splice them in the index number dimension to construct an interactive training matrix.

[0024] In some of these embodiments, input the initial box selection matrix into a supervised learning model for training, and through iterative training, obtain the learning recognition model, including:

[0025] A3-1. Perform matrix operations on the initial box selection matrix and the weight matrix of the supervised learning model to obtain the mastery estimations of a number of historical knowledge points;

[0026] A3-2. Convert the mastery estimations of a number of historical knowledge points into mastery estimation labels;

[0027] A3-3. Calculate the multi-class cross-entropy loss between the mastery estimation labels and the mastery status labels;

[0028] The expression of the multi-class cross-entropy loss is:

[0029]

[0030] Among them, represents the multi-class cross-entropy loss, N is the number of training samples in the initial box selection matrix, C is the number of classifications of the mastery status labels, y i,c represents the true label of the i-th training sample in the c-th class, p i,c represents the predicted probability of the i-th sample in the c-th class.

[0031] A3-4. Compare the multi-class cross-entropy loss of the initial box selection matrix with the minimum loss threshold;

[0032] A3-5. If the multi-class cross-entropy loss of the initial box selection matrix is less than or equal to the minimum loss threshold, export the supervised learning model as the learning recognition model; otherwise, iteratively update the model parameters of the supervised learning model to generate an updated model until the updated model is trained as the learning recognition model.

[0033] In some of these embodiments, iteratively update the model parameters of the supervised learning model to generate an updated model until the updated model is trained as the learning recognition model, including:

[0034] A3-5-1. Starting from the initial row vector of the matrix to be boxed, continue to box the matrix to be boxed with a rated box selection width to obtain the next box selection matrix;

[0035] A3-5-2. Input the next box selection matrix into the updated model to obtain the multi-class cross-entropy loss of the next box selection matrix;

[0036] A3-5-3. Compare the multi-class cross-entropy loss of the next box selection matrix with the minimum loss threshold;

[0037] A3-5-4. If the multi-class cross-entropy loss of the next box selection matrix is less than or equal to the minimum loss threshold, export the model being updated as the learning recognition model; otherwise, iteratively execute A3-5-1 to A3-5-3 until the learning recognition model is obtained.

[0038] In some of these embodiments, according to the mastery status label of the real-time user at the current knowledge point, a predefined teaching strategy for the knowledge point to be learned is constructed, including:

[0039] S4-1. Obtain the index number of the current knowledge point;

[0040] S4-2. According to the index number of the current knowledge point, extract the most similar user of the real-time user from the interactive training matrix;

[0041] S4-3. According to the most similar user of the real-time user, construct a predefined teaching strategy for the real-time user at the knowledge point to be learned.

[0042] In some of these embodiments, extracting the most similar user of the real-time user from the interactive training matrix according to the index number of the current knowledge point includes:

[0043] S4-2-1. According to the index number of the current knowledge point, determine the corresponding historical knowledge point;

[0044] S4-2-2. According to the corresponding historical knowledge point, select the historical column vector corresponding to the current knowledge point from the interactive training matrix; wherein, the historical column vector is characterized as the historical interaction vectors of K historical users at the corresponding historical knowledge points;

[0045] S4-2-3. Calculate the similarity between each historical interaction vector in the historical column vector and the real-time interaction vector to obtain K similarities of the real-time user in the historical column vector;

[0046] S4-2-4. Sort the K similarities in descending order to generate an ordered similarity sequence;

[0047] S4-2-5. Select the maximum similarity from the ordered similarity sequence;

[0048] S4-2-6. According to the maximum similarity, determine the historical user corresponding to the maximum similarity;

[0049] S4-2-7. Define the historical user corresponding to the maximum similarity as the most similar user.

[0050] In some of these embodiments, a predefined teaching strategy for the real-time user in the knowledge points to be learned is constructed according to the most similar user of the real-time user, including:

[0051] S4-3-1. Select the historical row vector corresponding to the real-time user from the interactive training matrix according to the most similar user; wherein, the historical row vector is characterized as a sequence of historical interaction vectors of the most similar user in T historical knowledge points;

[0052] S4-3-2. Starting from the index number of the current knowledge point, mark the index number of the knowledge point to be learned in the index number sequence;

[0053] S4-3-3. According to the index number of the knowledge point to be learned, mark the corresponding historical knowledge point to be learned from the historical row vector;

[0054] S4-3-4. Select a number of training samples corresponding to the historical knowledge point to be learned;

[0055] S4-3-5. Extract the mastery status label corresponding to the historical knowledge point to be learned from a number of training samples corresponding to the historical knowledge point to be learned;

[0056] S4-3-6. Define the mastery status label corresponding to the historical knowledge point to be learned as the mastery status label of the real-time user in the knowledge point to be learned.

[0057] S4-3-7. Sort the current knowledge point and the knowledge point to be learned on the time axis to generate a knowledge point time series;

[0058] S4-3-8. Mark the mastery status labels of the current time point and the knowledge point to be learned in the knowledge point time series;

[0059] S4-3-9. According to the mastery status label of each knowledge point, select the predefined teaching mode corresponding to the mastery status label;

[0060] S4-3-10. Define the combination of the predefined teaching modes of several knowledge points on the future time axis as the teaching interaction strategy on the future time axis.

[0061] Compared with the prior art, a teaching object interaction method of the present invention can inherit similar behavior patterns among multiple knowledge points for current learners based on the interaction behaviors and mastery status tags of historical users, thereby avoiding independent prediction of each knowledge point and improving the accuracy of learning path deduction and the personalization of learning strategies. Moreover, by combining real-time mastery status prediction and deduction based on historical behavior patterns, it can not only predict the mastery status of the current knowledge point, but also automatically deduce the learning path and teaching strategies of learners on future knowledge points. Through the "anchoring" of historical behaviors and the prediction of future knowledge points, it can provide a coherent teaching interaction strategy for the teaching system, making the teaching content push more in line with the cognitive rhythm of learners, thereby significantly enhancing the user learning experience of the teaching system.

[0062] In a second aspect, the present invention provides a teaching object interaction system, and the interaction system includes:

[0063] An interaction feature acquisition unit, configured to acquire a plurality of interaction behavior features of a real-time user at the current knowledge point

[0064] An interaction vector construction unit, configured to construct the plurality of interaction behavior features of the real-time user at the current knowledge point into a real-time interaction vector

[0065] A mastery status tag output unit, configured to input the real-time interaction vector into a learning recognition model, perform forward propagation after pre-training in the learning recognition model, and generate a mastery status tag of the real-time user at the current knowledge point

[0066] A teaching interaction strategy output unit, configured to construct teaching interaction strategies of a plurality of knowledge points to be learned on the future time axis according to the mastery status tag of the real-time user at the current knowledge point.

[0067] In a third aspect, the present invention provides a teaching object interaction device, including a memory and a processor, where the memory stores at least one computer executable instruction, and the processor is configured to run the computer executable instruction. When the computer executable instruction is run by the processor, the teaching object interaction method described in the first aspect is implemented.

[0068] Compared with the prior art, the beneficial effects of a teaching object interaction method and device of the present invention are the same as those of the teaching object interaction method described above, so they will not be elaborated here. Description of the Drawings

[0069] Figure 1 It is a flow chart of a teaching object interaction method of the present invention;

[0070] Figure 2 It is a flow chart of the construction of a predefined teaching strategy of the present invention;

[0071] Figure 3 Schematic diagram of the definition process of the most similar user according to the present invention;

[0072] Figure 4 Block diagram of the structure of an interactive system for teaching objects according to the present invention;

[0073] Figure 5 Electronic device diagram of an interactive device for teaching objects according to the present invention. Detailed implementation manners

[0074] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0075] Please refer to Figure 1 value Figure 3 , the present invention provides an interactive method for teaching objects, and the interactive method includes:

[0076] S1. Obtain several interactive behavior characteristics of the real-time user at the current knowledge point;

[0077] Among them, the real-time user is a learner in an online learning platform, and the interactive behavior characteristics include multi-modal behavior data related to the current knowledge point, specifically including but not limited to answer selection situations, mouse click times, page stay times, knowledge point query requests, voice question behaviors, etc., which are used to reflect the learning behavior state of the real-time user at the current knowledge point.

[0078] S2. Construct the several interactive behavior characteristics of the real-time user at the current knowledge point into a real-time interaction vector;

[0079] Among them, the real-time interaction vector is a structured feature expression, and the original interactive behaviors are standardized and encoded based on a preset behavior dimension specification. For example, numerical values such as answer accuracy rate, click frequency, and average stay duration are normalized and then composed into a vector to represent the multi-dimensional behavior characteristics of the current real-time user at this knowledge point.

[0080] S3. Input the real-time interaction vector into the learning recognition model, and perform forward propagation after pre-training in the learning recognition model to generate a mastery status label of the real-time user at the current knowledge point;

[0081] Among them, the learning and recognition model is a multi-classification model trained based on supervised learning. In its pre-training stage, training samples are constructed based on the real-time interaction vectors of historical users on various knowledge points and their actual learning results. The "interaction behavior matrix" refers to the knowledge point behavior-label mapping parameters learned inside the model. The model takes the current real-time interaction vector as the input and outputs the mastered status label closest to it. The forward label types include mastered, unmastered, vaguely understood, need to expand, etc.

[0082] In this embodiment, the learning and recognition model is used to predict the mastered status label of the real-time user on the current knowledge point according to the real-time interaction vector of the real-time user. Among them, the pre-training steps of the learning and recognition model include:

[0083] A1. Construct an interactive training matrix for pre-training. Among them, the matrix elements of the interactive training matrix are binary training samples corresponding to historical knowledge points. The binary training samples include the historical interaction vectors of the historical knowledge points and their mastered status labels.

[0084] Among them, the interactive training matrix is a two-dimensional matrix with a K×T structure. K represents the number of historical users, and T represents the number of historical knowledge points. Each cell in the matrix is a training sample, including the historical interaction vector of the historical user on the corresponding knowledge point and its corresponding mastered status label. This training matrix is used for structured modeling and batch processing of the behavior label data of real-time user-knowledge points.

[0085] A2. Starting from the initial row vector of the interactive training matrix, frame the interactive training matrix with a rated framing width, and divide the interactive training matrix into an initial framed matrix and a to-be-framed matrix.

[0086] Among them, the framing operation is used to select a sub-matrix with controllable dimensions from the interactive training matrix as the input batch of the model. The framing length of the sub-matrix is the column length of the interactive training matrix (i.e., the knowledge point dimension), and the framing width is the rated number of matrix row vectors (i.e., the real-time user dimension). Multiple sub-matrices are dynamically generated by sliding or sampling methods to improve the diversity and generalization ability of the training data.

[0087] A3. Input the initial framed matrix into the supervised learning model for training. After iterative training, the learning and recognition model is obtained.

[0088] Furthermore, in this embodiment, the step A1 specifically includes:

[0089] A1-1. Assign index numbers based on the learning order to T historical knowledge points to generate an index number sequence.

[0090] Among them, the index number is generated point by point based on the predefined learning order of each historical knowledge point in the curriculum system or syllabus;

[0091] A1-2. Assign a collection number to K historical users to generate a collection number sequence; among them, the collection number is generated by splicing the index number and the historical user identifier (ID field);

[0092] A1-3. Collect several interaction behavior characteristics of the kth historical user at the tth historical knowledge point and construct them into a historical interaction vector; where t represents the index number of the historical knowledge point, and k represents the collection number of the historical user;

[0093] During the process of the historical user learning this knowledge point, the online learning platform continuously collects its multi-modal interaction behavior data and performs standardized processing to form a behavior feature vector; the interaction behavior characteristics include but are not limited to click frequency, answer accuracy rate, page stay duration, voice interaction times, auxiliary resource usage records, etc.

[0094] A1-4. Collect the unit test score of the kth historical user at the tth historical knowledge point and map the unit test score to the mastery status label of the historical knowledge point;

[0095] After the learning task of this historical knowledge point is completed, count the score of this historical user in the first test or assessment of this historical knowledge point, quantify the mastery degree of this historical knowledge point, and map it to a discrete label according to the preset segmentation rules, for example:

[0096] Score ≥ 90 corresponds to the label: "Mastered";

[0097] Score 70–89 corresponds to the label: "Vaguely understood";

[0098] Score 50–69 corresponds to the label: "Not mastered";

[0099] Score < 50 corresponds to the label: "Needs expansion".

[0100] A1-5. Define the historical interaction vector of the historical knowledge point as the input feature and the mastery status label of the historical knowledge point as the target label to construct the training sample of the historical knowledge point;

[0101] That is, each training sample is an input-output pair, including the behavior feature vector and the corresponding mastery status label, serving as the basic training unit for training the supervised learning model.

[0102] A1-6. Obtain the training samples of the kth historical user at T historical knowledge points to generate the training sample set of the kth historical user; where T represents the total number of historical knowledge points;

[0103] That is to say, the training sample set represents a sequence set of the interaction behaviors and mastery status labels of the historical user under the complete learning path.

[0104] A1-7. According to the index number of the historical knowledge points, sort the training sample set of the k-th historical user in sequence to generate the first row vector of this historical user;

[0105] The first row vector represents the training sample set arranged according to the index number, and is used to unify the alignment of each real-time user sample in the time / knowledge point dimension.

[0106] A1-8. Obtain the first row vectors of K historical users, and align and splice them in the index number dimension to construct an interaction training matrix.

[0107] Among them, the interaction training matrix has a "K×T" structure, and each row corresponds to the complete training path of a historical user.

[0108] Furthermore, in this embodiment, the step A3 specifically includes:

[0109] A3-1. Perform matrix operations on the initial box selection matrix and the weight matrix of the supervised learning model to obtain the mastery estimations of several historical knowledge points;

[0110] A3-2. Convert the mastery estimations of several historical knowledge points into mastery estimation labels;

[0111] A3-3. Calculate the multi-class cross-entropy loss between the mastery estimation labels and the mastery status labels;

[0112] The expression of the multi-class cross-entropy loss is:

[0113]

[0114] Among them, represents the multi-class cross-entropy loss, N is the number of training samples in the initial box selection matrix, C is the number of classifications of the mastery status labels (that is, the number of types of mastery status labels, for example: "mastered", "vaguely understood", "not mastered", "need to expand", then C = 4, y i,c represents the true label (one-hot encoding) of the i-th training sample in the c-th class, p i,c represents the predicted probability of the i-th sample in the c-th class.

[0115] A3-4. Compare the multi-class cross-entropy loss of the initial box selection matrix with the minimum loss threshold;

[0116] A3-5. If the multi-class cross-entropy loss of the initial box selection matrix is less than or equal to the minimum loss threshold, export the supervised learning model as the learning and recognition model; otherwise, iteratively update the model parameters of the supervised learning model to generate an in-update model until the in-update model is trained as the learning and recognition model.

[0117] In this embodiment, the initial box selection matrix represents a set of behavior samples of several historical users on multiple knowledge points selected according to a preset rule. After being input into the supervised learning model, the mastery estimations of each knowledge point are generated through matrix operations with the weight matrix and converted into discrete mastery status labels. The estimated label is compared item by item with the true label in the sample, and the cross-entropy loss value is calculated to evaluate the classification accuracy in the current training stage. Then, the loss value is compared with the set minimum threshold to dynamically determine whether the current model meets the convergence requirement of the teaching label classification task. If not converged, parameter update is performed. This training control mechanism can ensure that the classification model for knowledge point mastery recognition in the teaching scenario has a stable output ability, avoid the deviation of the model training result from the knowledge point mastery rule due to the fluctuation of behavior characteristics, and effectively improve the mastery prediction quality in the teaching content response process.

[0118] Among them, the step A3-5 specifically includes:

[0119] A3-5-1. Starting from the initial row vector of the to-be-boxed matrix, continue to box the to-be-boxed matrix with a rated box selection width to obtain the next box selection matrix;

[0120] A3-5-2. Input the next box selection matrix into the in-update model to obtain the multi-class cross-entropy loss of the next box selection matrix;

[0121] A3-5-3. Compare the multi-class cross-entropy loss of the next box selection matrix with the minimum loss threshold;

[0122] A3-5-4. When the multi-class cross-entropy loss of the next box selection matrix is less than or equal to the minimum loss threshold, export the in-update model as the learning and recognition model; otherwise, iteratively execute A3-5-1 to A3-5-3 until the learning and recognition model is obtained.

[0123] In this embodiment, by continuously extracting the next batch of historical user behavior samples from the matrix to be boxed to construct the next box selection matrix and inputting it into the supervised learning model being updated, the corresponding multi-class cross-entropy loss value is obtained. This loss value is compared with the set minimum threshold to dynamically determine the convergence state of the current model, and based on this, it is decided whether to export the model or continue with parameter updates. This solution controls the sample input rhythm during the training process in a sliding batch manner, can maintain the phased stability of the training sample distribution in the teaching label classification training, reduce the risk of accuracy fluctuations caused by sample differences in the critical convergence state of the model, and thus improve the training robustness and stability of the mastery prediction model in teaching tasks.

[0124] The interaction method further includes:

[0125] S4. Construct teaching interaction strategies for several knowledge points to be learned on the future time axis according to the mastery status labels of the real-time user for the current knowledge point; among them, the mastery status labels of the several knowledge points to be learned are inherited from the training samples of the learning recognition model.

[0126] In an embodiment of the present invention, the set course is "Junior High School Mathematics - Linear Function". When a real-time user is learning on an online learning platform, for multiple knowledge points such as "function graph", "function properties", and "relationship between linear function and equation", the interaction behavior characteristics of the real-time user are collected in real time. According to the collected interaction behavior characteristics, the corresponding real-time interaction vector is constructed and input into the pre-trained learning recognition model. The recognition results show that the real-time user has mastered "function graph", has a vague understanding of "function properties", and has not mastered "relationship between linear function and equation". Based on the above mastery status labels, teaching modes are matched for different knowledge points, and finally, teaching interaction strategies including concept explanation videos, key example explanations, and comparison-type practice questions are generated to achieve personalized teaching responses for different knowledge points.

[0127] Further, in this embodiment, the step S4 specifically includes:

[0128] S4-1. Obtain the index number of the current knowledge point; the index number is the same in both historical knowledge points and the current knowledge point;

[0129] S4-2. According to the index number of the current knowledge point, extract the most similar user of the real-time user from the interaction training matrix;

[0130] S4-3. According to the most similar user of the real-time user, construct a predefined teaching strategy for the real-time user for the knowledge points to be learned.

[0131] In this embodiment, based on the mastery status label of the current knowledge point of the real-time user, first extract the index number of this knowledge point, and locate the interaction performance of historical users under this number in the interaction training matrix. Then anchor the historical user whose interaction behavior with the real-time user on this knowledge point is the closest, and then use the mastery status label of this most similar user on subsequent knowledge points as the basis for path deduction. Based on this path label information, the predefined teaching strategy corresponding to the subsequent knowledge point can be directly matched, realizing the rapid generation of strategies without having to reason item by item for each knowledge point.

[0132] Among them, the specific steps of step S4-2 include:

[0133] S4-2-1. Determine the corresponding historical knowledge point according to the index number of the current knowledge point;

[0134] S4-2-2. Select the historical column vector corresponding to the current knowledge point from the interaction training matrix according to the corresponding historical knowledge point; among them, the historical column vector is characterized as the historical interaction vectors of K historical users at the corresponding historical knowledge point;

[0135] S4-2-3. Calculate the similarity between each historical interaction vector in the historical column vector and the real-time interaction vector to obtain K similarities of the real-time user in the historical column vector;

[0136] The similarity can be calculated by means such as cosine similarity, Euclidean distance, Mahalanobis distance or a distance metric model based on a neural network.

[0137] S4-2-4. Sort the K similarities in size to generate an ordered similarity sequence;

[0138] S4-2-5. Select the maximum similarity from the ordered similarity sequence;

[0139] S4-2-6. Determine the historical user corresponding to the maximum similarity according to the maximum similarity;

[0140] S4-2-7. Define the historical user corresponding to the maximum similarity as the most similar user.

[0141] In this embodiment, first, based on the index number of the current knowledge point, its column vector in the interactive training matrix is determined, and the historical interaction behavior representations of all historical users on this knowledge point are extracted. Subsequently, the similarity between each historical interaction vector in this column vector and the real-time interaction vector of the current real-time user on this knowledge point is calculated, obtaining K similarity scores representing the degree of behavioral proximity. The similarity calculation can adopt vector measurement methods such as cosine similarity, Euclidean distance, and Mahalanobis distance, or a pre-trained behavior comparison network can be used for non-linear similarity modeling. The similarity results are sorted, and the historical user with the largest similarity is selected as the most similar user. Through the structural alignment of the historical behavior distribution and the current interaction characteristics, this process constructs an individual-group mapping mechanism for single-point behavior, which can accurately anchor historical samples with common learning patterns without the need for full-path recalculation.

[0142] Further, step S4-3 specifically includes:

[0143] S4-3-1. Select the historical row vector corresponding to the real-time user from the interactive training matrix according to the most similar user; where the historical row vector is characterized as a sequence of historical interaction vectors of the most similar user on T historical knowledge points;

[0144] S4-3-2. Starting from the index number of the current knowledge point, mark the index numbers of the knowledge points to be learned in the index number sequence;

[0145] S4-3-3. According to the index numbers of the knowledge points to be learned, mark the corresponding historical knowledge points to be learned from the historical row vector;

[0146] S4-3-4. Select several training samples corresponding to the historical knowledge points to be learned;

[0147] S4-3-5. Extract the mastery status labels corresponding to the historical knowledge points to be learned from several training samples corresponding to the historical knowledge points to be learned;

[0148] S4-3-6. Define the mastery status label corresponding to the historical knowledge point to be learned as the mastery status label of the real-time user on the knowledge point to be learned.

[0149] S4-3-7. Sort the current knowledge point and the knowledge points to be learned on the time axis to generate a knowledge point time series;

[0150] S4-3-8. Mark the mastery status labels of the current time point and the knowledge points to be learned in the knowledge point time series;

[0151] S4-3-9. According to the mastery status label of each knowledge point, select the predefined teaching mode corresponding to this mastery status label;

[0152] S4-3-10. Define the combination of predefined teaching modes of several knowledge points on the future time axis as the teaching interaction strategy on the future time axis.

[0153] In this embodiment, based on the identified most similar user, extract the historical row vector of this user from the interaction training matrix, and use the index number of the current knowledge point as the starting position, and sequentially mark its subsequent knowledge points as the learning targets to be learned. Further, select the training samples corresponding to these knowledge points to be learned, and extract the mastered status labels recorded therein, and use this as the predicted mastered status of the real-time user on the corresponding knowledge points. This process establishes an individual behavior and collective phase mapping relationship between the current behavior characteristics of the real-time user and the historical trajectory of the most similar user, and can complete the teaching state migration without model re-pushing. Subsequently, arrange the current knowledge point and the knowledge points to be learned in sequence to form a time series of knowledge points with a time sequence logic, and assign the teaching mode corresponding to the mastered status label to each knowledge point in the sequence. Finally, combine multiple teaching modes to form a set of structured future teaching interaction strategies, realizing the historical inheritance and dynamic adaptation of the teaching path, constructing a personalized teaching strategy sequence with a time sequence, and using it to complete the push of phased teaching content in a path-based manner in subsequent teaching interactions. This step can directly deduce a personalized learning path that conforms to the cognitive rhythm based on the structured behavior trajectory without performing per-knowledge-point reasoning, greatly improving the response efficiency of the teaching system and the coherence of the user experience.

[0154] Exemplarily, the predefined teaching mode can be expressed as that for each knowledge point, based on its mastered status label, a learning interaction method corresponding to the mastered status label will be preset, such as a concept explanation video, a key example explanation, or a comparison type exercise question, etc.

[0155] The embodiment of the present invention also provides a teaching object interaction system, which is used to implement the above method embodiments, and those that have been described will not be repeated here. The following terms such as "module", "unit", "sub-unit", etc. can be a combination of software and / or hardware that can achieve a predetermined function. Although the system described in the following embodiments is preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0156] As Figure 4 shown, Figure 4 is a structural block diagram of a teaching object interaction system of the present invention, and the system includes:

[0157] An interaction feature acquisition unit, which is used to acquire several interaction behavior characteristics of the real-time user at the current knowledge point

[0158] An interaction vector construction unit, which is used to construct the several interaction behavior characteristics of the real-time user at the current knowledge point into a real-time interaction vector

[0159] A mastery status label output unit, configured to input a real-time interaction vector into a learning and recognition model, perform forward propagation after pre-training in the learning and recognition model, and generate a mastery status label of the real-time user at the current knowledge point.

[0160] A teaching interaction strategy output unit, configured to construct teaching interaction strategies of several knowledge points to be learned on the future time axis according to the mastery status label of the real-time user at the current knowledge point.

[0161] In the above device, the interaction behavior features are obtained through the interaction feature acquisition unit, the real-time interaction vector is obtained through the interaction vector construction unit, the mastery status label of the current knowledge point is generated through the mastery status label output unit, and the teaching interaction strategies of the knowledge points to be learned on the future time axis are obtained through the teaching interaction strategy output unit, thereby solving the problem that the historical learning behavior pattern cannot be effectively utilized to accelerate the inference process in the current teaching interaction.

[0162] As Figure 5 shown, an embodiment of the present invention further provides a teaching object interaction device. The electronic device of the interaction device includes a memory 230 and a processor 210. The memory 230 stores at least one computer-executable instruction, and the processor 210 is configured to run the computer-executable instruction. When the computer-executable instruction is run by the processor 210, it is used to implement the above-mentioned teaching object interaction method.

[0163] The electronic device may include a processor 210, a communication interface 220, a memory 230, and a communication bus 240. Among them, the processor 210, the communication interface 220, and the memory 230 complete communication with each other through the communication bus 240. The processor 210 can call the logical instructions in the memory 230 to execute a teaching object interaction method disclosed in this embodiment.

[0164] In addition, when the logical instructions in the above-mentioned memory 230 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0165] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disks, or optical discs, and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A teaching object interaction method, characterized in that, The interactive method includes: S1. Obtain several interactive behavior characteristics of the real-time user at the current knowledge point; S2. Construct the several interactive behavior characteristics of the real-time user at the current knowledge point into a real-time interaction vector; S3. Input the real-time interaction vector into the learning and recognition model, perform forward propagation after pre-training within the learning and recognition model, and generate a mastery status label of the real-time user at the current knowledge point; S4. According to the mastery status label of the real-time user at the current knowledge point, construct teaching interaction strategies for several knowledge points to be learned on the future time axis; wherein, the mastery status labels of the several knowledge points to be learned are inherited from the training samples of the learning and recognition model.

2. The interactive method for teaching objects according to claim 1, wherein, The pre-training steps of the learning and recognition model include: A1. Construct an interactive training matrix for pre-training; wherein, the matrix elements of the interactive training matrix are binary training samples corresponding to historical knowledge points, and the binary training samples include historical interaction vectors of the historical knowledge points and their mastery status labels; A2. Starting from the initial row vector of the interactive training matrix, frame the interactive training matrix with a rated framing width, and divide the interactive training matrix into an initial framed matrix and a matrix to be framed; A3. Input the initial framed matrix into a supervised learning model for training, and through iterative training, obtain the learning and recognition model.

3. The interactive method for teaching objects according to claim 2, characterized in that, A1. Construct an interactive training matrix for pre-training, including: A1-1. Assign index numbers based on the learning order to T historical knowledge points to generate an index number sequence; A1-2. Assign collection numbers to K historical users to generate a collection number sequence; A1-3. Collect several interactive behavior characteristics of the k-th historical user at the t-th historical knowledge point, and construct them into a historical interaction vector; wherein, t represents the index number of the historical knowledge point, and k represents the collection number of the historical user; A1-4. Collect the unit test scores of the k-th historical user at the t-th historical knowledge point, and map the unit test scores to the mastery status labels of the historical knowledge points; A1-5. Define the historical interaction vector of the historical knowledge point as the input feature, and the mastery status label of the historical knowledge point as the target label, and construct a training sample of the historical knowledge point; A1-6. Obtain the training samples of the k-th historical user at T historical knowledge points, and generate a training sample set of the k-th historical user; wherein, T represents the total number of historical knowledge points; A1-7. According to the index number of the historical knowledge point, sort the training sample set of the k-th historical user in sequence to generate the first row vector of this historical user; A1-8. Obtain the first row vectors of K historical users, and align and splice them in the index number dimension to construct an interactive training matrix.

4. The interactive method for teaching objects according to claim 1, wherein Input the initial framed matrix into a supervised learning model for training, and through iterative training, obtain the learning and recognition model, including: A3-1. Perform matrix operations on the initial framed matrix and the weight matrix of the supervised learning model to obtain the mastery estimations of several historical knowledge points; A3-2. Convert the mastery estimations of several historical knowledge points into mastery estimation labels; A3-3. Calculate the multi-class cross-entropy loss between the mastery estimation labels and the mastery status labels; The expression of the multi-class cross-entropy loss is as follows: Among them, represents the multi-class cross-entropy loss, N is the number of training samples in the initial bounding box matrix, C is the number of classifications of the mastery status labels, and y i,c represents the true label of the i-th training sample in the c-th class, and p i,c represents the predicted probability of the i-th sample in the c-th class; A3-4. Compare the multi-class cross-entropy loss of the initial bounding matrix with the minimum loss threshold; A3-5. If the multi-class cross-entropy loss of the initial bounding matrix is less than or equal to the minimum loss threshold, export the supervised learning model as the learning and recognition model; otherwise, iteratively update the model parameters of the supervised learning model to generate an in-update model until the in-update model is trained as the learning and recognition model.

5. A teaching object interaction method according to claim 2, characterized in that, Iteratively updating the model parameters of the supervised learning model to generate an in-update model until the in-update model is trained as the learning and recognition model includes: A3-5-1. Starting from the initial row vector of the matrix to be bounded, continue to bound the matrix to be bounded with a rated bounding width to obtain the next bounding matrix; A3-5-2. Input the next bounding matrix into the in-update model to obtain the multi-class cross-entropy loss of the next bounding matrix; A3-5-3. Compare the multi-class cross-entropy loss of the next bounding matrix with the minimum loss threshold; A3-5-4. When the multi-class cross-entropy loss of the next bounding matrix is less than or equal to the minimum loss threshold, export the in-update model as the learning and recognition model; otherwise, iteratively execute A3-5-1 to A3-5-3 until the learning and recognition model is obtained.

6. The teaching object interaction method according to claim 5, characterized in that Construct a predefined teaching strategy for the knowledge points to be learned according to the mastery status label of the real-time user at the current knowledge point, including: S4-1. Obtain the index number of the current knowledge point; S4-2. According to the index number of the current knowledge point, extract the most similar user of the real-time user from the interactive training matrix; S4-3. Construct a predefined teaching strategy for the real-time user at the knowledge points to be learned according to the most similar user of the real-time user.

7. A teaching object interaction method according to claim 6, characterized in that, Extracting the most similar user of the real-time user from the interactive training matrix according to the index number of the current knowledge point includes: S4-2-1. Determine the corresponding historical knowledge point according to the index number of the current knowledge point; S4-2-2. Select the historical column vector corresponding to the current knowledge point from the interactive training matrix according to the corresponding historical knowledge point; wherein, the historical column vector is characterized as the historical interaction vectors of K historical users at the corresponding historical knowledge points; S4-2-3. Calculate the similarity between each historical interaction vector in the historical column vector and the real-time interaction vector to obtain K similarities of the real-time user in the historical column vector; S4-2-4. Sort the K similarities in size to generate an ordered similarity sequence; S4-2-5. Select the maximum similarity from the ordered similarity sequence; S4-2-6. Determine the historical user corresponding to the maximum similarity according to the maximum similarity; S4-2-7. Define the historical user corresponding to the maximum similarity as the most similar user.

8. A teaching object interaction method according to claim 7, wherein Constructing a predefined teaching strategy for the real-time user at the knowledge points to be learned according to the most similar user of the real-time user includes: S4-3-1. Select the historical row vector corresponding to the real-time user from the interactive training matrix according to the most similar user; wherein, the historical row vector is characterized as a sequence of historical interaction vectors of the most similar user at T historical knowledge points; S4-3-2. Starting from the index number of the current knowledge point, mark the index numbers of the knowledge points to be learned in the index number sequence; S4-3-3. According to the index numbers of the knowledge points to be learned, mark the corresponding historical knowledge points to be learned from the historical row vectors; S4-3-4. Select a number of training samples corresponding to the historical knowledge points to be learned; S4-3-5. Extract the mastery status labels corresponding to the historical knowledge points to be learned from a number of training samples corresponding to the historical knowledge points to be learned; S4-3-6. Define the mastery status label corresponding to the historical knowledge point to be learned as the mastery status label of the real-time user for the knowledge point to be learned; S4-3-7. Sort the current knowledge point and the knowledge points to be learned on the time axis to generate a knowledge point time sequence; S4-3-8. Mark the mastery status labels of the current time point and the knowledge points to be learned in the knowledge point time sequence; S4-3-9. According to the mastery status label of each knowledge point, select the predefined teaching mode corresponding to the mastery status label; S4-3-10. Define the combination of the predefined teaching modes of several knowledge points on the future time axis as the teaching interaction strategy on the future time axis.

9. An interactive system for teaching objects, characterized in that, The interaction system includes: An interaction feature acquisition unit for acquiring a number of interaction behavior features of the real-time user at the current knowledge point An interaction vector construction unit for constructing the number of interaction behavior features of the real-time user at the current knowledge point into a real-time interaction vector A mastery status label output unit for inputting the real-time interaction vector into the learning recognition model, performing forward propagation after pre-training in the learning recognition model, and generating the mastery status label of the real-time user at the current knowledge point A teaching interaction strategy output unit for constructing the teaching interaction strategies of several knowledge points to be learned on the future time axis according to the mastery status label of the real-time user at the current knowledge point.

10. An interactive device for teaching objects, characterized in that, It includes a memory and a processor. The memory stores at least one computer-executable instruction. The processor is configured to run the computer-executable instruction. When the computer-executable instruction is run by the processor, it implements a teaching object interaction device according to any one of claims 1 to 8.

Images