An automated problem data calling method, system and device

Abstract

This invention discloses an automated method, system, and device for retrieving question data, comprising the following steps: Step 1: Obtaining user learning progress data and course metadata; Step 2: Constructing a data retrieval request based on JSON format and executing an intelligent elastic retrieval process; Initiating a data retrieval request to the question bank server, and identifying the error type when receiving an error response from the server; Intelligent elastic retry strategy: If the error type is a server-recoverable error, an exponential backoff retry engine is started for retrying, with the retry interval calculated based on an exponential function of the base interval and the number of retry attempts, plus a random jitter value; Client-side circuit breaker degradation: If the error type is a client-side error or the server error rate exceeds a preset threshold, a client-side circuit breaker is triggered, the circuit breaker enters the OPEN state, and the degradation content is returned directly within the circuit breaker time window; Step 3: Receiving successfully retrieved question data.

Description

Technical Field

[0001] This invention relates to the field of intelligent education technology, and in particular to a method, system and device for automating the retrieval of test data. Background Technology

[0002] In today's era of rapid advancements in educational informatization and intelligentization, and increasingly personalized and diversified learning needs, traditional manual problem-solving and data retrieval methods are no longer sufficient to meet the practical needs of learners and educators for efficient and intelligent learning support. These methods suffer from significant limitations, including low efficiency, cumbersome operation, and difficulty in adapting to the demands for rapid processing and precise analysis of large-scale data. Therefore, there is an urgent need for an innovative method that can automatically retrieve problem-solving data. This method aims to automate the entire process—from automatic question bank extraction and intelligent question matching to personalized learning path recommendations and real-time feedback on learning outcomes—by constructing a data retrieval system based on intelligent algorithms and big data technology. This fundamentally solves the problems inherent in traditional methods. This approach addresses core pain points such as excessive human intervention, delayed data retrieval, and insufficient learning adaptability, effectively improving learning efficiency and outcomes. It significantly reduces the time and effort costs for learners and educators in areas such as question selection, data processing, and progress tracking. Simultaneously, through precise data mining and analysis mechanisms, it provides scientific and data-driven support for educational decision-making, promoting the transformation and upgrading of educational evaluation models from experience-driven to data-driven. Ultimately, it achieves optimized allocation of educational resources, comprehensive improvement of educational quality, and further promotion of educational equity. It not only has practical application value in solving current educational problems but also contains long-term strategic significance in promoting the intelligent development of education and building a lifelong learning system.

[0003] In existing technologies, network request retry mechanisms are implemented using the duration+step parameter. However, the duration+step strategy may trigger a cascading failure effect when the server fails. When the failure rate of the question bank service exceeds 5%, the retry mechanism will cause a backlog of client requests, eventually triggering the system circuit breaker. Therefore, this paper proposes an automated method, system, and device for retrieving question-solving data. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an automated method, system, and device for retrieving problem-solving data.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An automated method for retrieving problem-solving data includes the following steps: Step 1: Obtain user learning progress data and course metadata; Step 2: Construct a data retrieval request based on JSON format and execute the intelligent elastic retrieval process; the intelligent elastic retrieval process includes: The data retrieval request is initiated to the question bank server. When an error response is received from the server, the error type is identified using a three-dimensional error code classifier. Intelligent elastic retry strategy: If the error type is a server-side recoverable error, the exponential backoff retry engine is activated to retry. The retry interval is calculated based on the exponential function of the base interval and the number of retries, and a random jitter value is added. Client-side circuit breaker degradation: If the error type is a client-side error or the server-side error rate exceeds a preset threshold, the client-side circuit breaker is triggered. The circuit breaker enters the OPEN state and directly returns the degradation content within the circuit breaker time window, stopping the sending of requests to the server. Step 3: Receive the successfully retrieved exercise data, combine it with the user feature vector, generate a personalized exercise list using the multimodal matching degree calculation model, and output it.

[0006] The above further includes: Further, step one includes: Integrated registration information parsing engine to extract metadata about users' professional background and learning goals; Deploy a knowledge graph scanner to build a network linking course knowledge points; Initialize a hybrid cold start matrix, which includes individual historical weights α, group characteristic weights β, and course progress weights γ, wherein the initial value of α is less than that of β; Based on the aforementioned metadata, knowledge graph, and cold start matrix, a three-dimensional user profile and course-independent feature vector library are generated.

[0007] Furthermore, in step two: The three-dimensional error code includes at least a system module identifier, an error type code, and an error detail identifier; In the intelligent elastic retry strategy, the calculation formula for exponential backoff is as follows: ,in To preset the basic interval time, This is the current number of retries. The range of random fluctuations; The client-side circuit breaker has three states: CLOSE, OPEN, and HALF-OPEN. Its state transition is driven by the request failure rate within a recent time window.

[0008] Furthermore, step three includes: The user feature vector and the exercise feature vector are input into the attention gating network to calculate the initial matching degree; By using the Ebbinghaus review weight injector, review weights based on the forgetting curve are added to the exercises; The difficulty gradient adaptive adjuster dynamically adjusts the weight of the exercise difficulty based on the user's fluency in solving the problem. By combining the initial matching degree, review weight, and difficulty weight, a final dynamic weight is generated to arrange the personalized exercise list.

[0009] Furthermore, it also includes step four: neural network optimization and trust decay process: User answering behavior data and matching results are fed back into a neural network model that integrates bidirectional LSTM and attention mechanism for training; Through dynamic trust decay function Incrementally adjust the individual historical weights ,in, is the initial weight, k is the decay coefficient, and n is the number of valid responses; By aggregating model parameter updates from multiple clients through a federated learning parameter server, cross-domain joint model optimization can be achieved.

[0010] An automated problem-solving data retrieval system includes: The data acquisition and initialization module is used to obtain user learning progress data and course metadata; An intelligent call execution module is used to construct a data call request based on JSON format and execute an intelligent elastic call process; the intelligent call execution module includes: The request sending unit is used to initiate the data retrieval request to the question bank server. The error handling unit, built into the client's three-dimensional error code classifier, is used to identify the type of error response returned by the server. The intelligent elastic retry unit is used to initiate an exponential backoff retry engine to retry when a server-side recoverability error is detected. The client-side circuit breaker degradation unit is used to trigger the client-side circuit breaker, stop the request, and return degradation content when a client error is detected or the server error rate exceeds the threshold. The exercise matching and output module is used to receive successfully invoked exercise data, and combine it with user feature vectors to generate and output a personalized exercise list through a multimodal matching degree calculation model.

[0011] Furthermore, the data acquisition and initialization module includes: The registration information parsing engine is configured to extract metadata about the user's professional background and learning objectives. A knowledge graph scanner, configured to build a network of connections between course knowledge points; The cold start matrix initialization unit is configured to initialize a hybrid matrix of individual historical weights α, group characteristic weights β, and course progress weights γ.

[0012] Furthermore, the exercise matching and output module includes: Attention-gated network units are used to calculate the initial match between the user and the exercise; The review weight injection unit is used to add review scheduling weights to exercises based on the Ebbinghaus forgetting curve; The difficulty adjustment unit is used to adaptively adjust the weight of exercises in the list based on the user's fluency in solving the problems.

[0013] An electronic device includes a processor and a memory, the memory storing a computer program, wherein the processor executes the computer program to implement the steps of an automated problem-solving data retrieval method.

[0014] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of an automated problem-solving data retrieval method.

[0015] The present invention has the following beneficial effects: In this invention, through the synergy of intelligent elastic retry strategy and client-side circuit breaker degradation, the system can automatically identify and distinguish fault types, thereby taking the most appropriate fault-tolerant measures. This directly avoids the avalanche effect caused by temporary overload of the question bank server or network jitter, ensuring the high availability of online education services in high-concurrency scenarios and providing users with a smooth and uninterrupted learning experience. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of an automated problem-solving data retrieval method proposed in this invention. Figure 2 This is a system block diagram of an automated question-solving data retrieval system proposed in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figures 1-2 As shown, an automated method for retrieving problem-solving data includes the following steps: Step 1: Obtain user learning progress data and course metadata; Step 2: Construct a data retrieval request based on JSON format and execute the intelligent elastic retrieval process; the intelligent elastic retrieval process includes: initiating the data retrieval request to the question bank server, and when receiving an error response from the server, identifying the error type through a three-dimensional error code classifier; Intelligent elastic retry strategy: If the error type is a server-side recoverable error, the exponential backoff retry engine is activated to retry. The retry interval is calculated based on an exponential function of the base interval and the number of retries, and a random jitter value is added. Client-side circuit breaker degradation: If the error type is a client-side error or the server-side error rate exceeds a preset threshold, the client-side circuit breaker is triggered. The circuit breaker enters the OPEN state and directly returns the degradation content within the circuit breaker time window, stopping the sending of requests to the server. Step 3: Receive the successfully retrieved exercise data, combine it with the user feature vector, generate a personalized exercise list using the multimodal matching degree calculation model, and output it.

[0019] In one embodiment, step one includes: Integrated registration information parsing engine to extract metadata about users' professional background and learning goals; Deploy a knowledge graph scanner to build a network linking course knowledge points; Initialize a hybrid cold start matrix, which includes individual historical weights α, group characteristic weights β, and course progress weights γ, wherein the initial value of α is less than that of β; Based on the aforementioned metadata, knowledge graph, and cold start matrix, a three-dimensional user profile and course-independent feature vector library are generated.

[0020] In one embodiment, in step two: The three-dimensional error code includes at least a system module identifier, an error type code, and an error detail identifier; In the intelligent elastic retry strategy, the calculation formula for exponential backoff is as follows: ,in, To preset the basic interval time, This is the current number of retries. The range of random fluctuations; The client-side circuit breaker has three states: CLOSE, OPEN, and HALF-OPEN. Its state transition is driven by the request failure rate within a recent time window.

[0021] In one embodiment, step three includes: The user feature vector and the exercise feature vector are input into the attention gating network to calculate the initial matching degree; By using the Ebbinghaus review weight injector, review weights based on the forgetting curve are added to the exercises; The difficulty gradient adaptive adjuster dynamically adjusts the weight of the exercise difficulty based on the user's fluency in solving the problem. By combining the initial matching degree, review weight, and difficulty weight, a final dynamic weight is generated to arrange the personalized exercise list.

[0022] In this embodiment: Ebbinghaus Review Weight Injector: Predicts the best time for users to review knowledge points using the memory decay model, and dynamically injects review weights into the exercise matching degree calculation.

[0023] include: Memory decay curve modeling unit: Built-in standard Ebbinghaus forgetting curve parameters: 20 minutes, 1 hour, 9 hours, 1 day, 2 days, 6 days, 31 days, adjusting the decay rate based on the user's historical memory performance.

[0024] Knowledge Point-Time Association Matrix: Records the last learning timestamp of each knowledge point, stores the user's historical mastery rating of each knowledge point, and maintains the prerequisite relationship graph between knowledge points.

[0025] Review urgency calculation engine: Review weight = base weight × time decay factor × knowledge point importance coefficient, where, The optimal review interval = f(historical mastery level, knowledge point difficulty, user memory characteristics), and the knowledge point importance coefficient = calculated based on the centrality of the knowledge graph.

[0026] enter: User ID + current time; user's historical learning records (knowledge point learning time series); set of knowledge point tags for the exercises to be matched.

[0027] Processing procedure: It iterates through the knowledge points that the user has learned, calculates the memory strength decay value of each knowledge point, identifies knowledge points whose memory strength is below a threshold (such as 60%), and assigns extra review weight to the exercises related to these knowledge points. The weight value increases with the degree of "missing the best review time".

[0028] Output: Exercise review weight vector: {Exercise ID: Review weight value}; Review urgency score: Indicates the user's overall need to consolidate knowledge.

[0029] In one embodiment, step four is also included: neural network optimization and trust decay process. User answering behavior data and matching results are fed back into a neural network model that integrates bidirectional LSTM and attention mechanism for training; Through dynamic trust decay function The individual historical weight α is adjusted incrementally, where is the initial weight, k is the decay coefficient, and n is the number of valid responses; By aggregating model parameter updates from multiple clients through a federated learning parameter server, cross-domain joint model optimization can be achieved.

[0030] An automated problem-solving data retrieval system includes: The data acquisition and initialization module is used to obtain user learning progress data and course metadata; An intelligent call execution module is used to construct a data call request based on JSON format and execute an intelligent elastic call process; the intelligent call execution module includes: The request sending unit is used to initiate the data retrieval request to the question bank server. The error handling unit, built into the client's three-dimensional error code classifier, is used to identify the type of error response returned by the server. The intelligent elastic retry unit is used to initiate an exponential backoff retry engine to retry when a server-side recoverability error is detected. The client-side circuit breaker degradation unit is used to trigger the client-side circuit breaker, stop the request, and return degradation content when a client error is detected or the server error rate exceeds the threshold. The exercise matching and output module is used to receive successfully invoked exercise data, and combine it with user feature vectors to generate and output a personalized exercise list through a multimodal matching degree calculation model.

[0031] In one embodiment, the data acquisition and initialization module includes: The registration information parsing engine is configured to extract metadata about the user's professional background and learning objectives. A knowledge graph scanner, configured to build a network of connections between course knowledge points; The cold start matrix initialization unit is configured to initialize a hybrid matrix of individual historical weights α, group characteristic weights β, and course progress weights γ.

[0032] In one embodiment, the exercise matching and output module includes: Attention-gated network units are used to calculate the initial match between the user and the exercise; The review weight injection unit is used to add review scheduling weights to exercises based on the Ebbinghaus forgetting curve; The difficulty adjustment unit is used to adaptively adjust the weight of exercises in the list based on the user's fluency in solving the problems.

[0033] In this embodiment: The data acquisition and initialization module starts: the registration information parsing engine extracts the user's key metadata. The knowledge graph scanner scans the "Python Programming" course to build a knowledge point network. A hybrid cold start matrix is ​​initialized: since the user is new, personal history weight α=0.3, group feature weight β=0.5, and course progress weight γ=0.2 are set. An initial 3D profile of "User A" is generated: ability vector (inferred from metadata), behavioral pattern (default), and learning objective ("Master Python basics"). A course-independent feature vector library is built: the BERT model is used to convert the stems of all questions in the question bank into embedding vectors, which are then associated with knowledge point tags, difficulty coefficients, etc. A unique trace_id, INIT_001, is generated for this initialization process and logged.

[0034] Intelligent elastic call and circuit breaker protection: When the system needs to call the exercises for "User A", the intelligent call execution module starts working: Constructing and Sending the Request: The intelligent call execution module constructs a data call request based on JSON format and sends it through the request sending unit. The request body contains information such as user ID, course ID, number of required exercises, and feature vector. The trace_id:REQ_001 is injected into the request header.

[0035] Receive and classify the response: The question bank server returned a "502 Bad Gateway" error due to a sudden surge in load. The error handling unit received this response.

[0036] Three-dimensional error code classification: The error handling unit generates a three-dimensional error code SYS-502-TIMEOUT based on the error type.

[0037] Intelligent elastic retry strategy decision and execution: Due to a 5xx error (server-side recoverability error), the intelligent elastic retry unit is triggered: Set the base interval to 500ms and the jitter to 30%.

[0038] First retry: Interval = 500 * 2^(0) + random(-150ms, +150ms) ≈ 350ms ~ 650ms before execution.

[0039] If it fails again, retry for the second time: interval = 500*2^(1) + random(-150ms,+150ms)≈850ms~1150ms.

[0040] Each retry request carries the same trace_id: REQ_001, making it easy to trace the complete chain in the ELK log system.

[0041] Client-side circuit breaker and degradation decision-making and execution: Meanwhile, the circuit breaker degradation unit continuously monitors the request failure rate over the past minute. It assumes that the failure rate has exceeded a preset 5% threshold during the retry period.

[0042] The circuit breaker immediately transitions from the CLOSE state to the OPEN state. During the next 5 seconds (the circuit breaker window), all requests to the question bank server are no longer actually sent; instead, the client directly and quickly returns a degraded set of content (e.g., a pre-set list of general exercises of moderate difficulty). After 5 seconds, the circuit breaker enters the HALF-OPEN state, allowing a small number of trial requests to pass. If the trial succeeds, the circuit breaker closes (CLOSE); if it fails, it returns to the OPEN state.

[0043] Successful call: The exercise data JSON package was successfully retrieved on the first retry.

[0044] Dynamic calculation of multimodal matching degree: The exercise matching and output module receives exercise data.

[0045] Feature input: Concatenate the feature vector of "User A" with the feature vector of each exercise called.

[0046] Attention Gated Network Calculation: The concatenated vector is input into the network consisting of the Transformer encoder and the GLU gate to calculate an initial matching score (e.g., between 0 and 1).

[0047] Dynamic weight injection: The Ebbinghaus review weight injector detects that "User A" has a question about "function definition" that was answered incorrectly three days ago. According to the forgetting curve, this is the best time to review, so a higher review weight (e.g., +0.3) is injected into the question about this knowledge point.

[0048] The difficulty gradient adaptive adjuster calculates "user A's fluency" based on the user's average time spent answering questions recently and the reference time for the current question. If the user's fluency is low (slow answering), the weight of high-difficulty questions is appropriately reduced (e.g., multiplied by a coefficient of 0.8).

[0049] Generate the final list: Taking into account the initial matching degree, review weight, and difficulty adjustment factor, calculate the final dynamic weight of each question. Sort the questions by weight from high to low to generate the final personalized exercise list, which is then presented to "User A".

[0050] Neural Network Optimization and Trust Decay: After User A completes this exercise: The neural network optimization uses the matching degree between its answer behavior (correct or incorrect, time taken) and the prediction as training samples to fine-tune the fusion network of bidirectional LSTM + attention mechanism.

[0051] As User A answers the questions (n) more times, the dynamic trust decay function α = 0.3 × e^(-0.1 × n) begins to take effect. After User A completes 5 questions (n=5), the weight of α increases from 0.3 to approximately 0.48, and the system begins to rely more on User A's personal behavioral data than on the group data at the time of the cold start.

[0052] Local model parameter updates are periodically uploaded to the federated learning parameter server, where they are aggregated with updates from other clients to achieve joint optimization of the model across the entire network, while protecting user privacy.

[0053] An electronic device includes a processor and a memory, the memory storing a computer program, wherein the processor executes the computer program to implement the steps of an automated problem-solving data retrieval method.

[0054] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of an automated problem-solving data retrieval method.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for automating the retrieval of problem-solving data, characterized in that, Includes the following steps: Step 1: Obtain user learning progress data and course metadata; Step 2: Construct a data retrieval request based on JSON format and execute the intelligent elastic retrieval process; The intelligent elastic invocation process includes: The data retrieval request is initiated to the question bank server. When an error response is received from the server, the error type is identified using a three-dimensional error code classifier. Intelligent elastic retry strategy: If the error type is a server-side recoverable error, the exponential backoff retry engine is activated to retry. The retry interval is calculated based on the exponential function of the base interval and the number of retries, and a random jitter value is added. Client-side circuit breaker degradation: If the error type is a client-side error or the server-side error rate exceeds a preset threshold, the client-side circuit breaker is triggered. The circuit breaker enters the OPEN state and directly returns the degradation content within the circuit breaker time window, stopping the sending of requests to the server. Step 3: Receive the successfully retrieved exercise data, combine it with the user feature vector, generate a personalized exercise list using the multimodal matching degree calculation model, and output it.

2. The automated problem-solving data retrieval method according to claim 1, characterized in that, Step one includes: Integrated registration information parsing engine to extract metadata about users' professional background and learning goals; Deploy a knowledge graph scanner to build a network linking course knowledge points; Initialize a hybrid cold start matrix, which includes individual historical weights α, group characteristic weights β, and course progress weights γ, wherein the initial value of α is less than that of β; Based on the aforementioned metadata, knowledge graph, and cold start matrix, a three-dimensional user profile and course-independent feature vector library are generated.

3. The automated problem-solving data retrieval method according to claim 1, characterized in that, In step two: The three-dimensional error code includes at least a system module identifier, an error type code, and an error detail identifier; In the intelligent elastic retry strategy, the calculation formula for exponential backoff is as follows: ,in, To preset the basic interval time, This is the current number of retries. The range of random fluctuations; The client-side circuit breaker has three states: CLOSE, OPEN, and HALF-OPEN. Its state transition is driven by the request failure rate within a recent time window.

4. The automated problem-solving data retrieval method according to claim 1, characterized in that, Step three includes: The user feature vector and the exercise feature vector are input into the attention gating network to calculate the initial matching degree; By using the Ebbinghaus review weight injector, review weights based on the forgetting curve are added to the exercises; The difficulty gradient adaptive adjuster dynamically adjusts the weight of the exercise difficulty based on the user's fluency in solving the problem. By combining the initial matching degree, review weight, and difficulty weight, a final dynamic weight is generated to arrange the personalized exercise list.

5. The automated problem-solving data retrieval method according to claim 4, characterized in that, It also includes step four: neural network optimization and trust decay process: User answering behavior data and matching results are fed back into a neural network model that integrates bidirectional LSTM and attention mechanism for training; Through dynamic trust decay function Incrementally adjust individual historical weights ,in, is the initial weight, k is the decay coefficient, and n is the number of valid responses; By aggregating model parameter updates from multiple clients through a federated learning parameter server, cross-domain joint model optimization can be achieved.

6. An automated problem-solving data retrieval system, characterized in that, include: Data acquisition and initialization module: used to acquire user learning progress data and course metadata; Intelligent Invocation Execution Module: Used to construct data invocation requests based on JSON format and execute the intelligent elastic invocation process; The intelligent invocation execution module includes: Request sending unit: used to initiate the data retrieval request to the question bank server; Error handling unit: A three-dimensional error code classifier built into the client, used to identify the type of error response returned by the server; Intelligent elastic retry unit: used to initiate the exponential backoff retry engine to retry when a server-side recoverability error is detected; Client-side circuit breaker and degradation unit: used to trigger the client-side circuit breaker, stop the request and return degradation content when a client error is detected or the server error rate exceeds the threshold; Exercise matching and output module: This module receives successfully retrieved exercise data and, in conjunction with the user's feature vector, generates and outputs a personalized exercise list using a multimodal matching degree calculation model.

7. The automated problem-solving data retrieval system according to claim 6, characterized in that, The data acquisition and initialization module includes: Registration information parsing engine: Extracts metadata about the user's professional background and learning goals; Knowledge graph scanner: Constructs a network of connections between course knowledge points; Cold start matrix initialization unit: Initializes a mixed matrix of individual historical weights α, group characteristic weights β, and course progress weights γ.

8. The automated problem-solving data retrieval system according to claim 6, characterized in that, The exercise matching and output module includes: Attention-gated network unit: used to calculate the initial match between the user and the exercise; Review weight injection unit: used to add review scheduling weights to exercises based on the Ebbinghaus forgetting curve; Difficulty adjustment unit: Used to adaptively adjust the weight of exercises in the list based on the user's fluency in solving the problems.

9. An electronic device comprising a processor and a memory, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 5.

Images