Method and device for generating personalized learning feedback in educational memory operation system

Abstract

The invention discloses a personalized learning feedback generation method and device in an education memory operation system, and relates to the technical field of auxiliary education, and the method mainly comprises the steps: calling the learning data of a target student from a multi-level memory library containing short-term, middle-term and long-term memories; identifying a specific feedback demand type based on the data and the current learning scene; extracting corresponding multi-dimensional features from the multi-layer data according to the demand type; calling a feedback template library based on the features to generate personalized feedback content; and according to the feedback preference characteristics in the long-term memory of the student, adjusting the content presentation form and outputting the content presentation form to the terminal, thereby realizing accurate and adaptive personalized learning guidance. According to the invention, a closed-loop feedback generation mechanism based on the multi-level memory bank is constructed, so that the intelligent level and the personalized effect of the learning feedback system are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of assisted education technology, specifically to a method and apparatus for generating personalized learning feedback in an educational memory operating system. Background Technology

[0002] Although existing learning feedback systems are widely used in education, they still suffer from a number of inherent flaws, resulting in less than ideal feedback effectiveness. First, the timeliness of feedback is severely lacking. Most systems rely solely on static data from a single scenario (such as the final error result in an assignment), failing to incorporate short-term learning behaviors reflecting students' real-time thinking processes (such as answer trajectories and correction records), making the feedback delayed and superficial. Second, feedback from different learning scenarios (such as classroom, assignments, and exams) is fragmented and lacks coherence. It fails to utilize interim learning data to connect different scenarios and form a coherent diagnosis and guidance for students' learning journey. Third, the depth of existing feedback is limited, often stopping at immediate correction of knowledge points, failing to provide forward-looking suggestions and strategic guidance for students' ability development based on long-term accumulated data on ability characteristics and learning styles. Finally, the presentation format of feedback is monotonous and fixed, usually using uniform text descriptions, which cannot adapt to students' personalized cognitive preferences (such as charts, text, or videos), reducing students' acceptance and willingness to adopt feedback information. These shortcomings collectively make it difficult for existing learning feedback systems to make the leap from "passive notification" to "active adaptation". Summary of the Invention

[0003] To achieve a leap from "passive notification" to "active adaptation" in learning feedback systems, this invention proposes a method for generating personalized learning feedback in an educational memory operating system, comprising the following steps: S1: Retrieve short-term, medium-term, and long-term memory learning data of the target student from the multi-level memory bank of the educational memory operating system; S2: Based on the retrieved learning data and the current learning scenario, identify the types of feedback needs of the target student for immediate error correction, knowledge consolidation, or ability improvement. S3: Based on the identified feedback demand type, extract multidimensional features corresponding to the feedback demand type from short-term, medium-term and long-term memory learning data; S4: Based on the extracted multidimensional features, call the preset feedback template library to generate personalized feedback content that is adapted to the type of feedback requirement; S5: Based on the feedback preference characteristics in the target student's long-term memory learning data, adjust the presentation format of the generated personalized feedback content and output the adjusted feedback content to the terminal.

[0004] This invention significantly improves the intelligence level and personalization effect of the learning feedback system by constructing a closed-loop feedback generation mechanism based on a multi-level memory bank.

[0005] Furthermore, in step S2, the feedback request type is dynamically adjusted by monitoring the target student's learning interruption frequency, incorrect question marking behavior, or help request operation in real time.

[0006] Furthermore, in step S3, when extracting multidimensional features from intermediate memory learning data, a time-weighted algorithm is used for processing, wherein the weight of recent data is higher than the weight of distant data.

[0007] Furthermore, in step S4, when generating personalized feedback content, the tone intensity of the personalized feedback content is adjusted based on the emotional characteristics in the short-term memory learning data.

[0008] Furthermore, in step S4, the personalized feedback includes: error location and immediate prompts for immediate error correction needs, source tracing and reinforcement learning paths for knowledge consolidation needs, and capability development reports and learning strategy suggestions for capability enhancement needs.

[0009] Furthermore, in step S5, the feedback preference features include preferences for text, charts, or video formats, as well as preferences for the level of detail in the content.

[0010] This invention also proposes a personalized learning feedback generation device in an educational memory operating system, comprising: A memory data retrieval module is configured to perform step S1. The feedback requirement identification module is configured to perform step S2. The feature dimension extraction module is configured to perform step S3. The feedback content generation module is configured to perform step S4. The feedback adapter output module is configured to perform the S5 step.

[0011] Furthermore, the feedback requirement identification module is further configured to dynamically correct the feedback requirement type by monitoring the target student's learning interruption frequency, incorrect question marking behavior, or help request operation in real time.

[0012] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention proposes a personalized learning feedback generation method and device in an educational memory operating system. By integrating short-term, medium-term and long-term memory learning data for comprehensive analysis, it breaks through the limitation of traditional feedback that only focuses on immediate errors, and realizes a panoramic diagnosis and guidance from point-like error correction to periodic ability development, thereby enhancing the depth and foresight of feedback. (2) Introducing a dynamic demand identification mechanism based on real-time behavior monitoring can proactively sense changes in students' status and adaptively adjust feedback strategies, thereby effectively solving the problems of delayed feedback and scenario fragmentation in traditional systems; (3) By using a time-weighted algorithm to process mid-term memory data, we can capture students’ learning trends and stability characteristics, and ensure the timeliness and accuracy of phased feedback suggestions. At the same time, we can adjust the tone of feedback by combining the emotional characteristics in short-term memory, and enhance the emotional adaptability of human-computer interaction. Attached Figure Description

[0013] Figure 1 A flowchart illustrating the steps involved in generating personalized learning feedback in an educational memory operating system. Figure 2 This is a module diagram of a personalized learning feedback generation device in an educational memory operating system. Detailed Implementation

[0014] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0015] In current educational informatization practices, the construction of learning feedback systems often relies on single-dimensional student data. These systems typically only capture isolated learning outcomes in specific learning scenarios, such as incorrect answers in homework or lost points on exams, and generate fixed-format, homogeneous corrective prompts based on this. Because they fail to connect the continuous behavioral trajectory of students formed through classroom interaction, after-class exercises, and periodic assessments, their feedback often remains at the level of immediate correction of superficial errors. They fail to reveal the stability of knowledge mastery behind the errors, nor can they correlate them with long-term learning styles and ability development tendencies. This static and fragmented feedback model results in limited guidance, making it difficult to adapt to the dynamically evolving personalized needs of students throughout the learning cycle, and even more so, failing to provide teachers with coherent insights into learning progress across different teaching scenarios. Therefore, such as Figure 1 As shown, this invention proposes a method for generating personalized learning feedback in an educational memory operating system, including the following steps: S1: Retrieve short-term, medium-term, and long-term memory learning data of the target student from the multi-level memory bank of the educational memory operating system; S2: Based on the retrieved learning data and the current learning scenario, identify the types of feedback needs of the target student for immediate error correction, knowledge consolidation, or ability improvement. S3: Based on the identified feedback demand type, extract multidimensional features corresponding to the feedback demand type from short-term, medium-term and long-term memory learning data; S4: Based on the extracted multidimensional features, call the preset feedback template library to generate personalized feedback content that is adapted to the type of feedback requirement; S5: Based on the feedback preference characteristics in the target student's long-term memory learning data, adjust the presentation format of the generated personalized feedback content and output the adjusted feedback content to the terminal.

[0016] To better describe the technical content of this invention, this embodiment uses the learning process of a junior high school student in the "quadratic function" unit of mathematics as an application scenario. This educational memory operating system is deployed in the cloud, and its multi-level memory database consists of a high-performance caching database Redis, a relational database MySQL, and a big data storage system HDFS, used to store short-term, medium-term, and long-term learning data, respectively. Students conduct daily learning and interaction through personal tablet terminals, while teachers view learning data and intervene through a web-based backend management system.

[0017] During the system initialization phase, the target student began practicing the homework for the "Quadratic Functions" unit. The homework system recorded the student's answering process in real time, including the order in which they answered each question, the time spent on each question, the calculation and modification records on the scratch paper, and the final submission of the answer. This high-granularity real-time learning behavior data was synchronously transmitted and stored in the short-term memory of the educational memory operating system. At the same time, the system retrieved all unit test scores, knowledge point diagnostic reports, and error notebook records related to functions from the student's mid-term memory over the past two months; and retrieved stable characteristic data formed over the past semester from the long-term memory, such as the "visual learning style" preference derived from multiple behavioral analyses (e.g., understanding charts and animations is about 30% more efficient than pure text), and the development trend line in the dimensions of logical reasoning and calculation ability. These multi-level memory data together constitute the comprehensive data foundation for the generation of this personalized feedback.

[0018] As a student solved an application problem on a tablet about finding the maximum and minimum values ​​of a quadratic function within a specific interval, the system's feedback generation process was activated. The student spent more than average time on this problem, erasing and revising their solution multiple times, ultimately submitting an incorrect answer. The system captured this series of actions in real time and combined it with historical data recently recorded in the student's mid-term memory bank—the student's accuracy rate on "practical applications of quadratic functions" questions had consistently hovered between 60% and 70% in the past three unit tests, and there were two previous instances of incorrectly solving similar problems due to ignoring the "domain" constraint. Based on this, the system intelligently identified the feedback need. It determined that the current error was not an accidental calculation mistake, but rather highly correlated with a persistent learning weakness involving incomplete application of concepts. Therefore, instead of simply categorizing it as an "immediate error correction need" requiring immediate prompts, the system comprehensively assessed the student's performance and accurately identified it as a deeper "knowledge consolidation need."

[0019] After clarifying the feedback needs, a refined extraction process of multi-dimensional features is initiated. Targeting the core objective of "knowledge consolidation," the system extracts key features from memory banks across different time spans. From the short-term memory bank, it extracts the "error type" of this answer as "omission of problem-solving steps," specifically manifested as failure to consider the range of values ​​for the independent variable in the actual problem. Simultaneously, it extracts the "answer duration" feature, showing that the time taken was approximately 1.5 times the average time for similar questions, suggesting hesitation and confusion in understanding. Next, the system focuses on medium-term memory data. To obtain the most accurate trend judgment, it uses a time-series weighted algorithm to analyze test data from the past two months. This algorithm assigns a higher weight (e.g., 0.7) to the most recent test score and a lower weight (e.g., 0.3) to earlier test scores. Through this weighted calculation, it derives the stability coefficient of the student's mastery of the "quadratic function application" knowledge point, revealing the stability of their knowledge structure. Finally, the system extracted stable, instructive features from the long-term memory bank, namely the "visual learning style" label and the strategy preference of "high acceptance of step-by-step, prompted exercises." These features from different time dimensions collectively outline the student's learning profile.

[0020] Based on the aforementioned multidimensional features, the system enters the personalized feedback content generation stage. The system kernel calls a feedback template library specifically designed for the "middle school math knowledge consolidation" scenario. The selected template provides the structural framework for the feedback content, while the specific content is dynamically filled in by the extracted feature data. The system generates a detailed feedback text, whose core content includes two parts: first, "weakness identification," which clearly points out that "when solving quadratic function application problems, you repeatedly overlooked the analysis of the actual domain of the independent variable, which is the key reason for the fluctuations in your recent test scores"; second, "reinforcement learning path," which proposes specific action suggestions, namely, "We recommend that you first watch an animated video of about 4 minutes, which specifically explains how to determine the domain of a function in practical problems, and then complete 3 step-by-step practice questions selected by the system, which include step-by-step hints." The entire content generation process is not a simple text splicing, but an intelligent deduction based on feature data, ensuring that every suggestion is targeted and effective.

[0021] After the feedback content is generated, the final adaptation and output steps are performed. Considering the "visual learning style" and "preference for moderate level of detail" feedback characteristics extracted from long-term memory, the system intelligently transforms the presentation of the generated text content. It transforms the dry textual description of "domain analysis" into a clear comparison chart showing the correct and incorrect steps, and directly highlights and embeds recommended learning resource links within the feedback. After this adaptation, the originally plain text feedback becomes more visually appealing, highlighting key points, and better suited to the student's cognitive habits. Finally, the personalized feedback content is pushed to the student's tablet terminal and presented in a user-friendly pop-up window. Simultaneously, a feedback summary, including the identified core issue "the need to focus on domain analysis in quadratic function application problems" and the system's recommended learning path, is also synchronized to the teacher's web management backend, providing data support for offline intervention.

[0022] Once the student receives feedback on their tablet, all subsequent interactions become part of the evaluation of the system's effectiveness. The system recorded that the student clicked and watched the recommended animated video in its entirety and successfully completed all the practice questions. These positive behaviors were quantified as a "100% feedback acceptance rate." More importantly, in the following week's in-class quizzes, the student successfully solved all the quadratic function application problems involving domain analysis. By comparing the accuracy rates before and after the feedback, the system calculated the "effectiveness score" for this feedback. This evaluation data—high acceptance rate and high effectiveness score—was automatically uploaded and updated in the long-term memory. Based on this, the system reinforced the weight of the "video explanation + step-by-step prompt practice" feedback combination for the student. This means that when generating similar feedback for this student in the future, the system will prioritize this proven and efficient format, thereby achieving a personalized service upgrade that becomes increasingly intelligent with use.

[0023] Furthermore, to better understand the technical content of this invention, this invention also proposes a personalized learning feedback generation device in an educational memory operating system to further elaborate on the technical content of this invention. For example... Figure 2 As shown, this personalized learning feedback generation device, serving as the core processing engine of the educational memory operating system, is physically hosted by one or more high-performance servers and achieves a high degree of functional integration and collaboration through software modularization. The five precisely coordinated functional modules built within the device work together to complete the entire process from data perception to intelligent feedback output.

[0024] The memory data retrieval module serves as the entry point for interaction between the device and external multi-level memory banks. Internally, it encapsulates a series of adaptable interfaces and data access protocols for different storage media. For example, this module retrieves real-time student operation logs updated in seconds from the short-term memory via a high-efficiency Redis client interface; it retrieves phased test and practice records from the medium-term memory, composed of a MySQL relational database, using optimized SQL queries; and it reads pre-computed student ability trend models and learning style tags from the HDFS distributed file system of the long-term memory via a big data access component configured with a Hadoop connector. This module does not simply call all data at once, but rather incorporates a data scheduling strategy that intelligently retrieves, combines, and filters data on demand based on the initial needs determined by subsequent modules. For instance, when the initial signal points to immediate error correction, this module prioritizes and fully loads the short-term memory data, while only performing a lightweight pre-read of the medium- and long-term data. This significantly improves data processing efficiency and response speed while ensuring comprehensive information.

[0025] The feedback demand identification module in the device receives a mixed data stream from the data retrieval module. Using a scenario-demand mapping model trained on a large number of samples, it maps the current learning activity (such as "timed in-class practice," "homework assignments," or "mock exams") to an initial demand priority. Simultaneously, the module integrates a series of real-time behavior monitors that continuously analyze the data stream from the student's terminal, including but not limited to: whether the student's dwell time on the current page or question is abnormal (frequency of learning interruptions), whether they actively trigger the "mark as not understood" button (error marking behavior), and whether they frequently use the "help" function within a short period. When the monitoring logic identifies a strong signal such as "three consecutive errors on the same knowledge point accompanied by active marking," the module dynamically overwrites the initial mapping result, decisively upgrading the demand type from "instant error correction" to "knowledge consolidation." This dynamic decision-making mechanism based on real-time behavior streams enables the device to overcome the limitations of fixed rules and achieve proactive perception and precise intervention of students' learning difficulties.

[0026] The feature extraction module receives pre-classified learning data and explicit requirements from upstream sources. It is equipped with diverse feature calculation engines tailored to different requirement types and data characteristics. For mid-term memory data requiring trend analysis, the module invokes its time-series processing submodule, which incorporates a time-series weighting algorithm. This algorithm does not simply average historical data but assigns different weights to data at different points in time. For example, it sets the weight of data from the most recent month to 0.7, while setting the weight of data from one to three months ago to 0.3. Feature values ​​calculated through this weighting method (such as the stability coefficient of knowledge point mastery) more realistically reflect students' recent, dynamic learning status, effectively filtering out interference from outdated historical data. Furthermore, for unstructured behavioral sequence data (such as long-term question-answering time series), the module activates its deep learning analysis submodule. Utilizing network structures such as CNN-LSTM hybrid models, it automatically extracts high-order abstract features from complex time-series patterns, such as the "slope of change in logical reasoning ability" and "success rate of knowledge transfer," thereby transforming the raw data into quantitative indicators with profound guiding significance.

[0027] The feedback content generation module retrieves and matches data from a locally stored, rigorously categorized feedback template library based on the structured feature vectors output by the feature extraction module. This extensive library contains hundreds of templates finely divided according to different subjects, grade levels, knowledge points, and needs. Each template is a text structure with variable placeholders. The module's job is to "translate" the feature data into personalized guidance language. For example, when it receives feature inputs such as "Error Type: Conceptual Confusion (Vertex Coordinate Formula)," "Related Knowledge Point: Quadratic Function Graph," or "Learning Style: Visual," it accurately locates the template under the "Middle School Mathematics - Instant Error Correction - Function" category and dynamically fills the template with the feature data, generating feedback content such as "The vertex coordinate formula of a quadratic function may have been confused here; its graphical representation is shown in the figure below..." Furthermore, the module integrates affective computing capabilities, which can extract students' "emotional characteristic" signals from short-term memory data (such as changes in answering speed). When an "anxiety" signal is detected (a sudden drop in answering speed), the module automatically adds encouraging statements before the generated feedback and adopts a gentler tone of advice; while when the student is calm (answering speed remains consistent), it uses direct and efficient directive language, thereby achieving adaptive adjustment of the intensity of the feedback tone.

[0028] Ultimately, the feedback adaptation and output module, as the final stage of interaction between the device and the user, is responsible for personalizing the generated feedback content in terms of format. This module is tightly connected to a long-term memory bank, reading records of the student's "feedback preference characteristics." These characteristics may include preferences for media formats such as plain text, mixed text and graphics, or short videos, as well as specific requirements for the level of detail in the content. The corresponding rendering engine within the module formats the feedback content in real time based on these preferences. For example, for students who prefer charts, it transforms key points of the text description into intuitive mind maps or comparison tables; for students who prefer concise information, it automatically hides lengthy background analysis, highlighting only core errors and action suggestions; for students who prefer videos, it replaces text guidance with or adds concise video tutorial links. After format adaptation, the module pushes the final feedback product to the designated student's terminal device through a stable communication interface, and simultaneously transmits a version with the same reliability, focusing more on learning analysis and teaching suggestions, to the teacher's backend system.

[0029] In summary, the technical solution of this invention significantly improves the intelligence and personalization of learning feedback systems by constructing a closed-loop feedback generation mechanism based on a multi-level memory bank. First, by integrating short-term, medium-term, and long-term memory learning data for comprehensive analysis, it overcomes the limitations of traditional feedback that only focuses on immediate errors, achieving a panoramic diagnosis and guidance from point-based error correction to periodic ability development, greatly enhancing the depth and foresight of feedback. Second, by introducing a dynamic demand identification mechanism based on real-time behavior monitoring, it proactively senses changes in student status and adaptively adjusts feedback strategies, effectively solving the problems of delayed feedback and scenario fragmentation in traditional systems. Furthermore, by using a time-series weighted algorithm to process medium-term memory data, it accurately captures students' learning trends and stability characteristics, ensuring the timeliness and accuracy of phased feedback suggestions; simultaneously, it combines emotional characteristics from short-term memory to adjust the tone of feedback, enhancing the emotional adaptability of human-computer interaction. At the output level, based on personalized preferences in long-term memory, it adaptively adjusts the media format and level of detail of feedback content, making it easier for students to accept and understand, thereby significantly improving the adoption rate of feedback and learning efficiency. Ultimately, the solution constructs a complete technological closed loop from data perception and intelligent analysis to personalized output, which not only achieves continuous support for students throughout their entire learning cycle, but also provides teachers with a basis for understanding and intervening in students' learning progress.

[0030] 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 positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0031] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0032] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0033] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

Claims

1. A method for generating personalized learning feedback in an educational memory operating system, characterized by, The method comprises the steps of: S1: retrieving short-term, medium-term and long-term memory learning data of a target student from a multi-level memory library of an educational memory operating system; S2: identifying the feedback demand type of the target student in real time based on the retrieved learning data and the current learning scene; S3: extracting multi-dimensional features corresponding to the feedback demand type from the short-term, medium-term and long-term memory learning data according to the identified feedback demand type; S4: generating personalized feedback content adapted to the feedback demand type based on the extracted multi-dimensional features by calling a preset feedback template library; S5: adjusting the presentation form of the generated personalized feedback content according to the feedback preference features in the long-term memory learning data of the target student, and outputting the adjusted feedback content to a terminal.

2. The method for generating personalized learning feedback in an educational memory operating system of claim 1, wherein, In the S2 step, the feedback demand type is dynamically corrected by real-time monitoring of the learning interruption frequency, wrong question marking behavior or help operation of the target student.

3. The method for generating personalized learning feedback in an educational memory operating system of claim 1, wherein, In the S3 step, when extracting multi-dimensional features of medium-term memory learning data, a time sequence weighting algorithm is used for processing, in which the weight of recent data is higher than that of long-term data.

4. The method for generating personalized learning feedback in an educational memory operating system of claim 1, wherein, In the S4 step, when generating personalized feedback content, the emotional features in the short-term memory learning data are also used to adjust the emotional intensity of the personalized feedback content.

5. The method for generating personalized learning feedback in an educational memory operating system of claim 1, wherein, In the S4 step, the personalized feedback content includes error positioning and immediate prompt for immediate error correction demand, weak point tracing and reinforcement learning path for knowledge consolidation demand, and ability development report and learning strategy suggestion for ability improvement demand.

6. The method for generating personalized learning feedback in an educational memory operating system of claim 1, wherein, In the S5 step, the feedback preference features include preference for text, chart or video form, and preference for content detail level.

7. A device for applying a personalized learning feedback generation method to the educational memory operating system according to claims 1 to 6, characterized in that, The method comprises: a memory data retrieval module configured to perform the S1 step; a feedback demand identification module configured to perform the S2 step; a feature dimension extraction module configured to perform the S3 step; a feedback content generation module configured to perform the S4 step; a feedback adaptation output module configured to perform the S5 step.

8. The apparatus of claim 7, wherein, The feedback demand identification module is further configured to dynamically correct the feedback demand type by real-time monitoring of the learning interruption frequency, wrong question marking behavior or help operation of the target student.

Images