Analysis method and system for mastering condition of knowledge points based on answering behaviors

Abstract

The invention discloses a knowledge point mastering condition analysis method and system based on answering behaviors, and relates to the technical field of education data processing. The method comprises the steps of associating knowledge points according to problem solving process data and obtaining knowledge point mastering degrees; acquiring learning data corresponding to the knowledge points; training local prediction models according to the learning data and the knowledge point mastering degree, and updating a global prediction model according to all the local prediction models; acquiring local learning data in real time, and predicting the mastering degree of personal knowledge points through a local prediction model; learning data of a plurality of users are sampled, and the overall knowledge point mastering degree is predicted through the global prediction model. According to the scheme, teaching improvement is driven through data, learning of each student can be guided in a personalized mode, teaching content and strategies are continuously optimized from the macroscopic level, and therefore the intelligent level and adaptability of an education system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of educational data processing, and in particular to a method and system for analyzing the mastery of knowledge points based on question answering behavior. Background Art

[0002] With the rapid development of technology, the application of big data in the field of education has become increasingly widespread, bringing unprecedented opportunities and challenges. At present, educational institutions can collect a large amount of student learning data through various online platforms. These data not only include students' answers, but also may involve multiple dimensions such as learning habits, learning path selection, and time management, providing educators with valuable information resources to understand and improve students' learning effects.

[0003] However, existing data analysis methods often focus on outcome evaluation (such as scores) while ignoring process evaluation (such as problem-solving ideas, time allocation, etc.), which limits the in-depth understanding and intervention of problems in students' learning process, and makes it difficult to accurately evaluate the mastery of each student's knowledge points, resulting in the inability to provide targeted improvement plans and resources; at the same time, due to the lack of effective data analysis tools, it is difficult for teachers to grasp the overall learning status of students, and thus it is impossible to effectively adjust teaching strategies and content to meet the overall learning needs of students. Therefore, how to extract and analyze the mastery of knowledge points from these data to achieve guided teaching, personalized learning paths, and optimized teaching content has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide a method for analyzing the mastery of knowledge points based on answering behavior, comprising: Associate knowledge points based on problem-solving process data and obtain the mastery of knowledge points; Acquire learning data corresponding to the knowledge point; Train local prediction models based on learning data and knowledge mastery, and update the global prediction model based on all local prediction models; Acquire local learning data in real time and use local prediction models to predict the mastery of individual knowledge points; sample the learning data of several users and use global prediction models to predict the mastery of overall knowledge points; The acquisition of the mastery of knowledge points includes: Analyze the problem-solving process data to obtain problem-solving information and question information; the problem-solving information includes problem-solving knowledge points, knowledge point scores and step durations, and the question information includes the total score of the question, the question difficulty rating and the question knowledge points; Generate a problem-solving knowledge point vector according to the problem-solving knowledge point; generate a question knowledge point vector according to the question knowledge point; calculate the similarity between the problem-solving knowledge point vector and the question knowledge point vector to obtain a first evaluation index; Acquire the problem-solving knowledge point that matches the problem knowledge point, and calculate the ratio of the knowledge point score corresponding to the problem-solving knowledge point to the total score of the problem to obtain a second evaluation index; Obtaining a third evaluation indicator according to the duration of the step; The mastery degree of the knowledge point is obtained according to the question difficulty rating, the first evaluation index, the second evaluation index and the third evaluation index.

[0005] Preferably, associating knowledge points according to the problem-solving process data includes: Processing the problem-solving process data to obtain question text data; Segmenting the topic text data into words and phrases; Performing word frequency statistics on the words and phrases to obtain a problem-solving word frequency vector; each dimension of the problem-solving word frequency vector represents a specific word or phrase, and the value corresponding to each dimension of the problem-solving word frequency vector is the frequency of occurrence of the word or phrase in the entire text; Performing word frequency statistics on the words and phrases contained in each of the knowledge points to obtain a knowledge point word frequency vector; The vector similarity between the knowledge point word frequency vector and the problem-solving word frequency vector is calculated by a similarity algorithm; when the vector similarity is greater than a first threshold, the corresponding knowledge point is set as the problem knowledge point.

[0006] Preferably, the first evaluation index is expressed as: ; in, represents the question knowledge point vector, represents the problem-solving knowledge point vector, represents the modulus of the problem-solving word frequency vector, Indicates the modulus of the problem-solving knowledge point vector.

[0007] Preferably, the acquiring the learning data corresponding to the knowledge point includes: Obtaining the user ID and the knowledge point ID corresponding to the knowledge point; Constructing a query statement according to the user ID and the knowledge point ID; Filter the data in the learning record table, and record the query result data set obtained as the learning data corresponding to the knowledge point; The learning data includes learning behavior habits, resource usage information, learning interaction data and problem-solving process data.

[0008] Preferably, the training of the local prediction model according to the learning data and the mastery degree of the knowledge points, and updating the global prediction model according to the local prediction model include: The learning data is set as the input of the local prediction model; the mastery level of the knowledge point is set as the output of the local prediction model; the local prediction model is trained according to the learning data and the mastery level of the knowledge point; Extracting local prediction model parameters from the local prediction model and encrypting and sending them to a central server; The central server receives the local prediction model parameters, aggregates the local prediction model parameters using a federated averaging algorithm, and updates the global prediction model parameters of the global prediction model; The central server sends the updated global prediction model parameters back to each local device; the local device updates the local prediction model according to the global prediction model parameters; The updating of the local prediction model and the global prediction model is continuously iterated.

[0009] Preferably, after the local device updates the local prediction model according to the global prediction model parameters, it further includes: Detecting a prediction error of a local prediction model based on the learning data; When the prediction error is greater than a second threshold, a local optimization data set and a local verification data set are obtained according to the learning data and the mastery degree of the knowledge points; Using the local optimization data set to locally optimize the local prediction model updated according to the global prediction model parameters; The local prediction model is evaluated using a local validation dataset.

[0010] Preferably, the sampling of learning data of several users and predicting the overall knowledge point mastery by a global prediction model comprises the following steps: Obtain learning data of users of several local prediction models through random sampling, stratified sampling and feature-oriented sampling; Input the learning data into the global prediction model to predict the knowledge mastery level of each user; Conduct statistical analysis on each user's mastery of knowledge points to obtain the overall mastery of knowledge points.

[0011] A knowledge point mastery analysis system based on answering behavior, comprising: The association module is used to associate the problem-solving process data with knowledge points and predict the mastery of the knowledge points; A learning data module, used to obtain learning data corresponding to the knowledge point; A training module, used for training a local prediction model according to the learning data and the mastery degree of the knowledge points, and updating a global prediction model according to the local prediction model; The prediction module is used to obtain local learning data in real time, and predict the mastery of individual knowledge points and the mastery of overall knowledge points respectively through the local prediction model and the global prediction model.

[0012] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for analyzing the mastery of knowledge points based on question answering behavior is implemented.

[0013] A storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the method for analyzing the mastery of knowledge points based on question-answering behavior can be implemented.

[0014] The present invention discloses a method for analyzing the mastery of knowledge points based on answering behavior, the method comprising: associating problem-solving process data with knowledge points to predict the mastery of knowledge points; obtaining learning data corresponding to the knowledge points; training a local prediction model according to the learning data and the mastery of the knowledge points, and updating a global prediction model according to the local prediction model; obtaining local learning data in real time, and predicting the mastery of individual knowledge points through a local prediction model; sampling the learning data of several users, and predicting the mastery of the overall knowledge points through a global prediction model. The present invention predicts the mastery of individual knowledge points and the mastery of the overall knowledge points respectively through a local prediction model and a global prediction model, so as to achieve data-driven teaching improvement, which can not only guide the learning of each student in a personalized manner, but also continuously optimize the teaching content and strategy from a macro level, thereby improving the intelligence level and adaptability of the education system.

[0015] The present invention obtains problem-solving information and question information, calculates the similarity between the problem-solving knowledge point vector and the question knowledge point vector to obtain a first evaluation index, calculates the ratio of the knowledge point score corresponding to the problem-solving knowledge point to the total score of the question to obtain a second evaluation index, and obtains a third evaluation index according to the step duration, so that the evaluation method of the knowledge point mastery degree comprehensively covers various key elements of the problem-solving process, provides a comprehensive and accurate evaluation result, and improves the reliability and reference value of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 A flowchart of a method for analyzing knowledge point mastery based on question answering behavior provided by an embodiment of the present invention; Figure 2 A schematic diagram of a process for obtaining the degree of mastery of a knowledge point provided by an embodiment of the present invention; Figure 3 A schematic diagram of a process for associating knowledge points according to problem-solving process data provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0021] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0022] Machine learning provides a way to implement artificial intelligence. By training models to recognize patterns in data, machine learning algorithms enable computers to perform tasks such as image recognition, speech recognition, and recommendation systems. Federated learning is a machine learning method that allows multiple devices or servers to collaboratively train a shared model while keeping the data local and private. This approach has significant advantages in terms of data privacy and security because it does not require the data to be centrally stored or transmitted to a central server. Federated learning has applications in many fields such as healthcare, finance, and mobile devices. It provides an effective solution for handling sensitive data while also improving the performance of the model. However, federated learning also faces some challenges, such as the security of model updates, trust issues between participants, and the speed of model convergence.

[0023] Each learner's learning behavior and knowledge mastery are unique. Traditional knowledge mastery prediction methods lack flexibility, have single evaluation criteria, and are difficult to predict due to small amounts of data. They are unable to provide personalized learning experiences and teaching suggestions. Learning is a continuous process, which requires the system to be able to obtain learning data in real time and dynamically adjust teaching content or learning plans.

[0024] The technical effects of the present invention include: 1. Prediction of personal knowledge mastery: By analyzing the user's problem-solving process data, including user information, question information, and problem-solving information, the system can obtain each user's mastery of specific knowledge points. This evaluation takes into account factors such as the correctness of the problem-solving, the problem-solving time, the rationality of the problem-solving steps, and the individuality of each user. Based on the evaluation of the individual's knowledge mastery, the system can generate personalized reinforcement exercise recommendations to help users focus on reviewing weak links and improve learning efficiency.

[0025] 2. Prediction of overall knowledge mastery: By sampling and analyzing the learning data of multiple users, the system can predict the overall knowledge mastery. This helps to discover knowledge points that are generally poorly mastered, reflecting the overall level, which is different from the personalized data of each user, thus providing a basis for the optimization of teaching content. Dynamic adjustment of teaching content: By analyzing the overall knowledge mastery, the teaching content can be dynamically adjusted, such as increasing the number of exercises or providing more detailed answers, to meet the learning needs of different users.

[0026] 3. Evaluation method for the degree of mastery of knowledge points: The present invention obtains problem-solving information and question information, calculates the similarity between the problem-solving knowledge point vector and the question knowledge point vector to obtain a first evaluation index, calculates the ratio of the knowledge point score corresponding to the problem-solving knowledge point to the total score of the question to obtain a second evaluation index, and obtains a third evaluation index based on the step duration, so that the evaluation method for the degree of mastery of knowledge points comprehensively covers various key elements of the problem-solving process, provides a comprehensive and accurate evaluation result, and improves the reliability and reference value of the prediction.

[0027] 4. Application of federated learning architecture: In order to solve data privacy and security issues, the present invention adopts a federated learning architecture. Under this architecture, each local device independently trains the model and encrypts the model parameters and sends them to the central server. The central server aggregates these parameters using a federated averaging algorithm, updates the global prediction model, and then sends it back to each local device. Continuous iteration and optimization of the model: The local prediction model and the global prediction model gradually improve the accuracy and adaptability of the prediction through continuous iterative training and updating. The system can also further adjust and optimize the model based on real-time student feedback to ensure that the model better adapts to students' specific learning habits and characteristics.

[0028] Example 1 See also Figures 1 to 3 The present invention provides a method for analyzing the mastery of knowledge points based on answering behavior, comprising the steps of: S1. Associate knowledge points based on problem-solving process data and obtain the mastery of knowledge points; In one embodiment, the problem-solving process data is image data collected by a user image acquisition device, or data directly input by a user through an online question-answering platform; the problem-solving process data that can be extracted through data analysis include: User information: basic information of the user, including grade and learning stage.

[0029] Question information: including the subject of the question, the total score of the question, the difficulty rating of the question and the knowledge points of the question.

[0030] Problem-solving information: used to record the user's problem-solving operations, including the problem-solving step number, problem-solving steps, problem-solving knowledge points, knowledge point scores and step duration; the problem-solving step number is used as an identifier in the problem-solving information, and the problem-solving step number is associated with the problem-solving steps, problem-solving knowledge points, knowledge point scores and step duration.

[0031] According to the problem-solving process data, the system associates knowledge points and obtains the mastery of knowledge points, including two parts: associating knowledge points and obtaining the mastery of knowledge points. Among them, the associated knowledge points are the knowledge points involved in the problem-solving process analyzed according to the problem-solving process data. For example, if the learner is solving a problem related to trigonometric functions, the system can identify the knowledge point involved in this problem as "trigonometric functions" based on the problem-solving process data.

[0032] The step of associating knowledge points according to the problem-solving process data specifically includes the following steps: S111, processing the problem-solving process data to obtain problem text data; dividing the problem text data into words and phrases; This step is the data preprocessing and data cleaning stage. If the problem-solving process data is image data collected by the user's image acquisition device, the part of the image data involving the question information is converted into question text data through optical character recognition (OCR) technology during the data preprocessing process; the converted question text data is long text data, and the long text data is segmented into phrases or words through the existing word segmentation algorithm, including removing stop words, converting letter case, removing special symbols, etc.

[0033] S112. Perform word frequency statistics on words and phrases to obtain a problem-solving word frequency vector; each dimension of the problem-solving word frequency vector represents a specific word or phrase, and the corresponding value is the frequency of occurrence of the word or phrase in the entire text.

[0034] S113, calculating the vector similarity between the knowledge point word frequency vector and the problem-solving word frequency vector of each knowledge point by a similarity algorithm; The similarity algorithm is the cosine similarity algorithm, which is expressed as: ; in, represents the problem-solving word frequency vector, represents the knowledge point word frequency vector of the i-th knowledge point, represents the modulus of the problem-solving word frequency vector, represents the modulus of the knowledge point word frequency vector of the i-th knowledge point, It represents the cosine similarity between the problem-solving word frequency vector and the knowledge point word frequency vector. The smaller the cosine similarity, the higher the similarity between the first vector and the second vector.

[0035] S114, calculating the vector similarity between the knowledge point word frequency vector and the problem-solving word frequency vector through a similarity algorithm; when the vector similarity is greater than a first threshold, setting the corresponding knowledge point as the problem knowledge point. The first threshold is a preset value that can be adjusted dynamically. In this way, the problem-solving process data is mapped to a specific knowledge point, providing a basis for subsequent problem-solving analysis.

[0036] Obtaining the mastery of knowledge points is to evaluate the learner's mastery of each knowledge point based on the correctness of the learner's problem solving, the time it takes to solve the problem, and the rationality of the problem-solving steps. For example, if the learner correctly and quickly solves problems related to trigonometric functions without any help or prompts, the system may determine that the learner has a good grasp of the knowledge point "trigonometric functions". On the contrary, if the learner takes a long time to solve the problem, or makes multiple mistakes in the problem-solving process, the learner is judged to have a low mastery of the knowledge point. Obtaining the mastery of knowledge points specifically includes the following steps: S121, analyzing the problem-solving process data to obtain problem-solving information and question information; Analyze the problem-solving process data, obtain the problem-solving steps, problem-solving knowledge points, knowledge point scores and step duration associated with each problem-solving step number, and obtain the total score, problem difficulty rating and problem knowledge points based on the problem information; S122, generating a problem-solving knowledge point vector according to the problem-solving knowledge points, and generating a question knowledge point vector according to the question knowledge points; calculating the similarity between the problem-solving knowledge point vector and the question knowledge point vector to obtain a first evaluation index; Generate a problem-solving knowledge point vector based on all problem-solving knowledge points, generate a question knowledge point vector based on all question knowledge points, and calculate the cosine similarity between the problem-solving knowledge point vector and the question knowledge point vector to obtain a first evaluation index; For example, for a certain problem-solving process, the set of all problem-solving knowledge points is represented as k p ={ p1, p3, p4, p5}, the set of all question knowledge points is denoted as k t ={ p1, p2, p3}; where p1, p2, p3, p4, p5 represent the properties of parallel lines, the determination of similar triangles, the Pythagorean theorem, the basic properties of a circle, and the triangle interior angle theorem, respectively.

[0037] Then get the union of all problem-solving knowledge points and question knowledge points, denoted as k u ={ p1, p2, p3, p4, p5}; The question knowledge point vector and the question solution knowledge point vector are generated based on the union of all the problem-solving knowledge points and the question knowledge points; the question knowledge point vector is represented as V tkp =[1, 1, 1, 0, 0]; the problem-solving knowledge point vector is represented by V pkp =[1, 0, 1,1, 1]; Calculate the similarity between the problem-solving knowledge point vector and the question knowledge point vector to obtain a first evaluation index; Preferably, the first evaluation index is expressed as: ; in, represents the question knowledge point vector, represents the problem-solving knowledge point vector, represents the modulus of the problem-solving word frequency vector, Indicates the modulus of the problem-solving knowledge point vector.

[0038] S123, obtaining problem-solving knowledge points that match the problem-solving knowledge points, and calculating the ratio of the knowledge point scores corresponding to the problem-solving knowledge points to the total score of the problem to obtain a second evaluation index; For example, based on the above problem-solving process, the set of all problem-solving knowledge points is represented as k p ={ p1, p3, p4, p5}, the set of all question knowledge points is denoted as k t ={ p1, p2, p3}; where p1, p2, p3, p4, p5 represent the properties of parallel lines, the determination of similar triangles, the Pythagorean theorem, the basic properties of a circle, and the triangle interior angle theorem, respectively.

[0039] Get the problem-solving knowledge points that match the problem knowledge points, specifically get the intersection k of all problem-solving knowledge points and the problem knowledge points m ={p1, p3}; then the second evaluation index is expressed as: ; in, represents the knowledge point score corresponding to knowledge point p1, represents the knowledge point score corresponding to knowledge point p3, Indicates the total score of the question.

[0040] S124, obtaining a third evaluation index according to the step duration; The third evaluation index is expressed as: ; in, Indicates the total score of the question. Indicates the total duration of effective steps. Indicates the total time of solving the problem. and is the preset weight parameter. For example, for the above problem-solving process, the intersection of all problem-solving knowledge points and question knowledge points is k m ={p1, p3}, the set of problem-solving knowledge points is denoted as k p ={p1, p3, p4, p5}, then ; ; in, , , and They represent the step durations corresponding to the problem-solving knowledge points p1, p3, p4, and p5, respectively. By considering the total effective step duration, that is, the time students spend on the correct problem-solving steps, this is more detailed than using only the total problem-solving time, because the total effective step duration distinguishes the time students spend on the correct path and possible wrong attempts, reflecting students' learning strategies and problem-solving abilities, and accurately assessing students' mastery of specific knowledge points.

[0041] S125. Obtain the mastery of the knowledge points according to the question difficulty rating, the first evaluation index, the second evaluation index and the third evaluation index.

[0042] ; Where D represents the difficulty rating of the question. , and are the weight parameters of the evaluation indicators respectively.

[0043] This program can comprehensively and accurately evaluate students' learning outcomes by integrating multiple evaluation indicators. It not only helps teachers understand students' learning status, but also provides a data basis for students' targeted learning feedback, thereby promoting the improvement of learning efficiency.

[0044] S2, obtain the learning data corresponding to the knowledge point; Learning data includes learning behavior habits, resource usage information, learning interaction data and problem-solving process data.

[0045] Learning behavior habits, used to record the user's learning time, number of learning times within a preset period, and learning time period; Resource usage information is used to record the types of learning resources used by users, the frequency and duration of resource use; learning resource types include textbooks, video courses, exercises, etc.

[0046] Learning interaction data is used to record the number of questions users ask during the learning process, as well as the number and depth of their participation in learning discussions.

[0047] In one embodiment, learning data is managed through a user table, a learning record table, and a knowledge point table. The user table records the basic information of all users, and each user has a record in the table. The user's unique identifier (UserID) is used to distinguish different users. The knowledge point table records the information of all knowledge points, and each knowledge point has a unique identifier (KnowledgePointID). The main fields include KnowledgePointID (knowledge point ID), Subject (subject), Unit (unit), Chapter (chapter), KnowledgePoint (knowledge point name), etc. The learning record table records the user's specific learning data. Each record is associated with a user and a knowledge point to form a learning event. The main fields include RecordID (record ID), UserID (user ID), KnowledgePointID (knowledge point ID), StudyTime (learning time), Frequency (learning frequency), InteractionCount (number of interactions), etc.

[0048] UserID in the user table is a foreign key in the learning record table. This means that each learning record is associated with a specific user. This relationship is a "one-to-many" relationship, that is, one user (with one record in the user table) can have multiple learning records (with multiple records in the learning record table). KnowledgePointID in the knowledge point table is a foreign key in the learning record table. This means that each learning record is associated with a specific knowledge point. This relationship is also a "one-to-many" relationship, that is, one knowledge point (with one record in the knowledge point table) can be learned by multiple users (with multiple records in the learning record table). The learning record table acts as a "bridge" to connect user behavior (learning data) with knowledge point information. Through the learning record table, each user's learning status of each knowledge point can be tracked.

[0049] By querying the learning record table, you can pull information from the user table and the knowledge point table across tables. For example, to understand the detailed records of a user learning a specific knowledge point, you can find the corresponding record in the learning record table by UserID and KnowledgePointID.

[0050] Based on the above content, obtain the learning data corresponding to the knowledge points, specifically: Obtain the user ID and the knowledge point ID corresponding to the knowledge point; construct a query statement based on the user ID and the knowledge point ID to filter the relevant data in the learning record table, and the query result data set obtained is recorded as the learning data corresponding to the knowledge point.

[0051] S3: Train local prediction models based on learning data and knowledge mastery, and update the global prediction model based on all local prediction models; the local prediction model and the global prediction model adopt a federated learning architecture; Federated learning is a distributed machine learning architecture that pays special attention to data privacy protection and data security. It allows models to be trained independently on multiple nodes, while the original data is always kept locally, thus avoiding the privacy risks brought by centralized data storage. In the knowledge point mastery analysis method based on answering behavior, it is very valuable to collect a large amount of student learning data. These data include students' learning process, knowledge point mastery, and learning behavior. Through analysis and modeling, it is possible to optimize the allocation of educational resources, provide personalized learning suggestions, and improve the effectiveness of knowledge transfer. However, direct centralized data storage will cause privacy and security risks, as well as a large burden on communication, and the advantages of the federated learning solution are revealed at this time.

[0052] In one embodiment, the federated learning architecture includes: Local device: Data storage and preprocessing: Each user’s learning data is stored on the local device (e.g., mobile phone, tablet, personal computer). Each local device uses its own training data to update the local prediction model.

[0053] Central server: The central server does not receive local data, but only receives and processes the model parameters (such as weights and biases) uploaded by each client. When the global prediction model is updated, the model parameters of each local device are aggregated through weighted averaging and other methods to generate new global prediction model parameters, which are then broadcast back to each client.

[0054] Specifically, the local prediction model is trained according to the learning data and the degree of mastery of the knowledge points, and the global prediction model is updated according to all the local prediction models, including the following steps: S31. Construct a local prediction model with learning data as input and knowledge point mastery as output, and train the local prediction model based on the learning data and knowledge point mastery obtained in steps S1 and S2; perform model training on each local device using the collected learning data, and use local learning data as input to train the model to predict the mastery of each knowledge point.

[0055] S32, extracting local prediction model parameters from the local prediction model and encrypting and sending them to the central server; After completing local training, the local device extracts local prediction model parameters from the local prediction model. The local prediction model parameters include weight matrix and bias term. In order to protect privacy, the uploaded model parameters are encrypted and uploaded to the central server through a secure network protocol (such as SSL / TLS).

[0056] S33, the central server receives local prediction model parameters from each local device, aggregates the local prediction model parameters using a federated average algorithm, and updates the global prediction model parameters; The central server collects the model parameters uploaded by each client; if the uploaded model parameters are encrypted, the central server first decrypts them; then, the weighted average of each local prediction model parameter is calculated based on the federated average algorithm, and the weight can be determined based on the amount of data from each client. The federated average algorithm is expressed as: ; in, is the updated global prediction model parameter, is the i-th local prediction model parameter, and N is the number of local prediction models.

[0057] S34. The central server sends the updated global prediction model parameters back to each local device, and the local device updates the local prediction model according to the global prediction model parameters.

[0058] The central server sends the updated global prediction model parameters back to each local device through a secure network protocol. The local device receives the updated global prediction model parameters to replace or update the local prediction model parameters to ensure that the local prediction model remains consistent with the global prediction model.

[0059] S35, continuously iterate the updating of the local prediction model and the global prediction model. Continuously iterate steps S31-S34. With the addition of each round of new learning data, the model after each round of training will be continuously optimized, and the global prediction model will become more accurate and adaptable with multiple rounds of operations.

[0060] In this embodiment, as the learning activity progresses, new learning data is continuously collected; the model is repeatedly trained using new local data, and updated model parameters are uploaded; the central server repeatedly performs parameter aggregation operations to generate increasingly accurate global prediction models; the above process is repeated, so that the model is optimized round by round, gradually improving its prediction accuracy and adaptability, which not only effectively protects students' privacy, but also can provide more accurate predictions and personalized learning suggestions based on the ever-changing learning situation.

[0061] Preferably, in step S34, after the local device updates the local prediction model according to the global prediction model parameters, the step further includes: S341. Detecting a prediction error of a local prediction model according to the learning data; S342: when the prediction error is greater than a second threshold, obtaining a local optimization data set and a local verification data set according to the learning data and the mastery degree of the knowledge points; S343, using the local optimization data set to locally optimize the local prediction model updated according to the global prediction model parameters; After updating the local prediction model according to the global prediction model parameters, the received global prediction model is further locally optimized using local data, i.e., a local optimization data set, to enhance the adaptability to the local learning behavior.

[0062] Specifically, by continuing to use local SGD (stochastic gradient descent) for several rounds of small-step optimization, some targeted parameters and specific modules can be optimized locally through personalized data. The specific number of training rounds and learning rate can be flexibly set according to the amount of data and model complexity.

[0063] For example, if a student has a large prediction error on questions in a certain subject or involving a certain knowledge point, personalized features will be introduced or modified based on the local learning data of that knowledge point to make the model more adaptable to the student's specific learning habits and characteristics.

[0064] S344, evaluating the local prediction model using a local validation dataset; Use the local validation dataset to evaluate the locally optimized model to ensure that the performance of the personalized adjusted model is improved; based on the prediction results of the local prediction model, obtain student feedback in real time (for example, students' subjective ratings of question difficulty) to further adjust and optimize the model.

[0065] This embodiment prepares a local optimized data set by using local learning data and the degree of mastery of knowledge points, ensuring that the model fully considers the actual learning situation of specific students; according to personalized needs, the local prediction model parameters are optimized to make the model better adapt to the students' specific learning habits and characteristics, improve the accuracy and adaptability of the model prediction, and enable the model to assist learning in a more personalized and precise manner.

[0066] S4. Obtain local learning data in real time and predict the mastery of individual knowledge points through local prediction models; sample the learning data of several users and predict the mastery of the overall knowledge points through global prediction models.

[0067] Specifically, for predicting the mastery of individual knowledge points, local learning data is obtained in real time, and relevant features are extracted from the local learning data to predict the user's mastery of each knowledge point. Since the model is trained based on the user's historical data, it can accurately reflect the user's current mastery. The platform will feed back the prediction results to the user, such as using a mastery score or probability value to represent the mastery of individual knowledge points, recommending review plans or additional exercises, and providing relevant learning resources; and dynamically adjust the user's learning plan based on the user's learning performance and the evaluation results of the model, such as increasing the number of exercises for specific knowledge points or recommending more detailed answers.

[0068] For predicting the overall knowledge point mastery, the sampling of learning data of several users and predicting the overall knowledge point mastery through a global prediction model include the following steps: The learning data of users of several local prediction models are obtained through random sampling, stratified sampling and feature-oriented sampling. Among them, the sampled local prediction models are local prediction models that have been authorized for user data collection, and the data of all local prediction models that can be sampled are the overall user data set. Random sampling is to randomly extract samples from the overall user data set; stratified sampling is to sample according to grades, subjects, etc. to ensure the representativeness of the samples; feature-oriented sampling is to conduct targeted sampling based on specific features such as learning progress and learning behavior.

[0069] Input the learning data into the global prediction model to predict the knowledge mastery level of each user; Statistical analysis is performed on the knowledge point mastery of each user to obtain the overall knowledge point mastery; the overall knowledge point mastery includes the mean value, variance, and standard deviation of the knowledge point mastery of each user.

[0070] By looking at the overall mastery of knowledge points, we can find out which knowledge points are more difficult or easier for users to master overall; by identifying trends and patterns in the data, such as the overall low mastery of certain knowledge points among a certain type of user; the analysis results are fed back to the teaching content development team or the education platform as a basis for improving teaching content and methods. Adjust the teaching plan based on the mastery, such as increasing the number of exercises or providing detailed answers.

[0071] After implementing this application, the local device can capture the user's learning data offline in real time, and predict the degree of mastery of individual knowledge points through local prediction model analysis, so as to further discover the weak links in the mastery of probability and statistics knowledge points of a certain user. Based on the degree of mastery of individual knowledge points, personalized reinforcement exercise recommendations are generated to help the user focus on reviewing the weak parts, improve the efficiency and pertinence of user learning, reduce blind learning, and greatly improve user experience and learning effects.

[0072] For the central server, the learning data of several users are sampled for global prediction model analysis to predict the overall mastery of knowledge points. For example, if it is found that most users have a poor grasp of a certain chapter, the platform can adjust the difficulty of the video content of that chapter based on this result, increase targeted explanations, set more exercises, or pay more attention to teaching, thereby optimizing the course structure, improving the pertinence of teaching content and overall teaching effectiveness, and improving the reputation of the platform and user satisfaction.

[0073] Based on the knowledge point mastery analysis method based on answering behavior proposed in this application, a prediction mechanism for individual knowledge point mastery and overall knowledge point mastery can be implemented. By using data to drive teaching improvements, the online education platform can not only provide personalized guidance for each student's learning, but also continuously optimize teaching content and strategies from a macro level, thereby improving the intelligence level and adaptability of the education system.

[0074] Example 2 The present invention provides a knowledge point mastery situation analysis system based on question answering behavior.

[0075] A knowledge point mastery analysis system based on answering behavior, comprising: The association module is used to associate the problem-solving process data with knowledge points and predict the mastery of the knowledge points; Learning data module, used to obtain learning data corresponding to knowledge points; The training module is used to train the local prediction model according to the learning data and the degree of mastery of knowledge points, and update the global prediction model according to the local prediction model; The prediction module is used to obtain local learning data in real time, and predict the mastery of individual knowledge points and the mastery of overall knowledge points through local prediction models and global prediction models respectively.

[0076] Example 3 An electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, a method for analyzing the mastery of knowledge points based on question-answering behavior is implemented.

[0077] Example 4 A storage medium stores a computer program, the computer program includes program instructions, and when the program instructions are executed by a processor, a method for analyzing the mastery of knowledge points based on question answering behavior can be implemented.

[0078] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A method for analyzing the mastery of knowledge points based on answering behavior, characterized in that: include: Associate knowledge points based on problem-solving process data and obtain the mastery of knowledge points; Acquire learning data corresponding to the knowledge point; Train local prediction models based on learning data and knowledge mastery, and update the global prediction model based on all local prediction models; Acquire local learning data in real time and use local prediction models to predict the mastery of individual knowledge points; sample the learning data of several users and use global prediction models to predict the mastery of overall knowledge points; The acquisition of the mastery of knowledge points includes: Analyze the problem-solving process data to obtain problem-solving information and question information; the problem-solving information includes problem-solving knowledge points, knowledge point scores and step durations, and the question information includes the total score of the question, the question difficulty rating and the question knowledge points; Generate a problem-solving knowledge point vector according to the problem-solving knowledge point; generate a question knowledge point vector according to the question knowledge point; calculate the similarity between the problem-solving knowledge point vector and the question knowledge point vector to obtain a first evaluation index; Acquire the problem-solving knowledge point that matches the problem knowledge point, and calculate the ratio of the knowledge point score corresponding to the problem-solving knowledge point to the total score of the problem to obtain a second evaluation index; Obtaining a third evaluation indicator according to the duration of the step; The mastery degree of the knowledge point is obtained according to the question difficulty rating, the first evaluation index, the second evaluation index and the third evaluation index.

2. A method for analyzing knowledge point mastery based on question answering behavior according to claim 1, characterized in that: The associating knowledge points according to the problem-solving process data includes: Processing the problem-solving process data to obtain question text data; Segmenting the topic text data into words and phrases; Performing word frequency statistics on the words and phrases to obtain a problem-solving word frequency vector; each dimension of the problem-solving word frequency vector represents a specific word or phrase, and the value corresponding to each dimension of the problem-solving word frequency vector is the frequency of occurrence of the word or phrase in the entire text; Performing word frequency statistics on the words and phrases contained in each of the knowledge points to obtain a knowledge point word frequency vector; The vector similarity between the knowledge point word frequency vector and the problem-solving word frequency vector is calculated by a similarity algorithm; when the vector similarity is greater than a first threshold, the corresponding knowledge point is set as the problem knowledge point.

3. The method for analyzing the knowledge point mastery based on the answering behavior according to claim 1, characterized in that: The first evaluation index is expressed as: ； in, represents the question knowledge point vector, represents the problem-solving knowledge point vector, represents the modulus of the problem-solving word frequency vector, Indicates the modulus of the problem-solving knowledge point vector.

4. The method for analyzing the knowledge point mastery based on the answering behavior according to claim 1, characterized in that: The acquiring of learning data corresponding to the knowledge point includes: Obtaining the user ID and the knowledge point ID corresponding to the knowledge point; Constructing a query statement according to the user ID and the knowledge point ID; Filter the data in the learning record table, and record the query result data set obtained as the learning data corresponding to the knowledge point; The learning data includes learning behavior habits, resource usage information, learning interaction data and problem-solving process data.

5. The method for analyzing the knowledge point mastery based on question answering behavior according to claim 1, characterized in that: The training of the local prediction model according to the learning data and the mastery degree of the knowledge points, and updating the global prediction model according to the local prediction model include: The learning data is set as the input of the local prediction model; the mastery level of the knowledge point is set as the output of the local prediction model; the local prediction model is trained according to the learning data and the mastery level of the knowledge point; Extracting local prediction model parameters from the local prediction model and encrypting and sending them to a central server; The central server receives the local prediction model parameters, aggregates the local prediction model parameters using a federated averaging algorithm, and updates the global prediction model parameters of the global prediction model; The central server sends the updated global prediction model parameters back to each local device; the local device updates the local prediction model according to the global prediction model parameters; The updating of the local prediction model and the global prediction model is continuously iterated.

6. A method for analyzing knowledge point mastery based on question answering behavior according to claim 5, characterized in that: After the local device updates the local prediction model according to the global prediction model parameters, the method further includes: Detecting a prediction error of a local prediction model based on the learning data; When the prediction error is greater than a second threshold, a local optimization data set and a local verification data set are obtained according to the learning data and the mastery degree of the knowledge points; Using the local optimization data set to locally optimize the local prediction model updated according to the global prediction model parameters; The local prediction model is evaluated using a local validation dataset.

7. The method for analyzing knowledge point mastery based on question answering behavior according to claim 1, characterized in that: The method of sampling the learning data of a number of users and predicting the overall knowledge mastery by a global prediction model comprises the following steps: Obtain learning data of users of several local prediction models through random sampling, stratified sampling and feature-oriented sampling; Input the learning data into the global prediction model to predict the knowledge mastery level of each user; Conduct statistical analysis on each user's mastery of knowledge points to obtain the overall mastery of knowledge points.

8. A knowledge point mastery analysis system based on answering behavior, characterized in that: include: The association module is used to associate the problem-solving process data with knowledge points and predict the mastery of the knowledge points; A learning data module, used to obtain learning data corresponding to the knowledge point; A training module, used for training a local prediction model according to the learning data and the mastery degree of the knowledge points, and updating a global prediction model according to the local prediction model; The prediction module is used to obtain local learning data in real time, and predict the mastery of individual knowledge points and the mastery of overall knowledge points respectively through the local prediction model and the global prediction model.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, a method for analyzing the mastery of knowledge points based on question answering behavior as described in any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, a method for analyzing knowledge point mastery based on question answering behavior as described in any one of claims 1 to 7 can be implemented.

Images