Learning efficiency improving method and device based on big data analysis and medium
By collecting learning behavior logs, constructing an interaction matrix and a temporal causal model, the problem of the dynamic relationship between learning motivation and efficiency in the teaching process was solved, enabling high-temporal-resolution intervention in the learning process and improving learning efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIANGNAN UNIV
- Filing Date
- 2026-01-12
- Publication Date
- 2026-04-28
AI Technical Summary
Existing technologies struggle to construct dynamic causal relationships between learning motivation and learning efficiency with high temporal resolution in real teaching processes, and cannot precisely characterize the contagion and diffusion effects of peer interaction on learning efficiency, thus failing to provide reliable evidence for process-based intervention.
By collecting learning behavior logs, dividing them into time slices, constructing a learning interaction relationship matrix, calculating learning motivation and efficiency indicators, establishing a time-series causal relationship model, calculating intervention candidate points, and implementing motivation enhancement interventions.
It achieves high temporal resolution tracking of fluctuations in learning motivation and changes in efficiency, accurately identifies the state of the learning process, and selects the minimum intervention point to improve overall learning efficiency.
Smart Images
Figure CN121937255A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of learning behavior mining technology, and in particular to a method, device and medium for improving learning efficiency based on big data analysis. Background Technology
[0002] With the popularization of educational informatization and online learning environments, multi-source data such as learning behavior logs, answer records, homework scores, and classroom interaction records are gradually being collected in a process-oriented manner. Big data-based learning behavior analysis and learning efficiency assessment are gradually becoming important means of monitoring and ensuring the quality of the teaching process. Related research, on the one hand, uses learning behavior logs to construct time series features to predict learning performance, risk of dropping out of classes, or task completion, assisting teachers in making macro-level teaching decisions. On the other hand, it introduces methods such as social network analysis and learning path mining into learning behavior research at the class or course level to reveal the interaction relationships among learners, the structure of the learning community, and the impact of different learning strategies on knowledge mastery.
[0003] However, existing technologies still have significant shortcomings in leveraging big data to improve learning efficiency. On the one hand, existing methods mostly aggregate and analyze learning behaviors and outcomes at a relatively coarse time granularity, such as weeks, units, or entire semesters. This makes it difficult to depict the fluctuations in learning motivation over short time scales and their immediate impact on learning efficiency. Furthermore, it is difficult to construct a dynamic causal relationship reflecting the evolution of learning motivation and efficiency over time in real teaching operations, thus failing to provide a reliable basis for high-time-resolution process interventions. On the other hand, existing methods often treat learners as independent individuals, at most statistically analyzing the number of interactions or the strength of social connections. They lack a detailed characterization of the structural characteristics formed within the class based on learning interactions, and cannot quantitatively measure the contagion and diffusion effects of changes in peer learning motivation on the learning efficiency of other learners through these interactions. Summary of the Invention
[0004] In view of the aforementioned existing problems, the present invention is proposed.
[0005] Therefore, this invention provides a learning efficiency improvement method based on big data analysis to solve the problems of existing technologies, such as the difficulty in constructing a dynamic causal relationship between learning motivation and learning efficiency with high temporal resolution in real teaching processes and the difficulty in accurately selecting the minimum intervention point that maximizes overall benefits while considering peer interaction structures.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for improving learning efficiency based on big data analysis, which includes collecting learners' learning behavior logs, dividing them into continuous time slices according to a uniform time interval, and combining the learning behavior logs of each learner in each time slice to generate a time slice-level behavior record set. Based on the behavioral record set, the interaction behavior between learners is identified, and a learning interaction relationship matrix is established with learners as nodes and interaction intensity as weights. Learning motivation indicators are calculated based on the duration of time spent on learning resources, the number of times active learning occurs, the completion rate of learning tasks, and non-learning operations. Learning efficiency indicators are calculated based on the course knowledge points, learner answer records, and homework scores within the teaching cycle. Within the time window, a temporal causal relationship model of the target learners is established by combining the learning interaction relationship matrix, learning motivation indicators, and learning efficiency indicators. Based on the temporal causal relationship model, the theoretical improvement value of implementing learning motivation enhancement interventions on learners within a time slice is calculated, and a set of intervention candidate points is generated. During the teaching process, the minimum intervention point with the highest theoretical improvement value is selected based on the intervention candidate points, and the matching intervention strategy is implemented for the learners.
[0007] As a preferred embodiment of the learning efficiency improvement method based on big data analysis described in this invention, the specific steps for generating a time-slice-level set of behavioral records are as follows: In a real teaching environment, collect learning behavior logs of each learner; The timestamps of the operations in the learning behavior log are time-standardized and converted into a unified local teaching time. The learning behavior logs are then sorted in chronological order to obtain the time-standardized learning behavior logs. The teaching process is divided into continuous time slices by setting time intervals. Within each time slice, a set of behavior records at the time slice level is generated for each learner based on the time-standardized learning behavior logs. The learning behavior log includes the learner identifier, the timestamp of the operation, the operation type, the resource identifier associated with the operation, and extended information fields for each log entry.
[0008] As a preferred embodiment of the learning efficiency improvement method based on big data analysis described in this invention, the establishment of a learning interaction relationship matrix with learners as nodes and interaction intensity as weights refers to identifying the operation types belonging to learning interactions in the behavior record set, identifying the interactive behaviors between learners, and constructing a learning interaction relationship matrix with all learners in the class as vertices and weighted relationships. The intensity of interaction among learners is used as the relation weight in the learning interaction relation matrix.
[0009] As a preferred embodiment of the learning efficiency improvement method based on big data analysis described in this invention, the calculation of the learning motivation index refers to the normalization and smoothing of the learning resource dwell time, the number of active learning sessions, the completion rate of learning tasks, and non-learning operations to obtain the learning motivation index. The computational learning efficiency index refers to the statistical analysis of course knowledge points within the teaching cycle, and the calculation of knowledge mastery for each learner, each course knowledge point, and each time slice based on learners' answer records and homework scores. Based on the knowledge point weights and corresponding knowledge mastery levels of all course knowledge points, a comprehensive knowledge level calculation is performed on all course knowledge points in the course. The comprehensive knowledge level is then subjected to differential calculation and smoothing of adjacent time slices to obtain a learning efficiency index.
[0010] As a preferred embodiment of the learning efficiency improvement method based on big data analysis described in this invention, the specific steps for establishing a temporal causal association model for the target learner are as follows: Based on the learning interaction relationship matrix, select 10 neighbor learners as stable interaction neighbors for each learner, and generate a stable interaction neighbor set; Based on learning motivation indicators, learning efficiency indicators, and a stable set of interactive neighbors, a temporal causal association model is constructed for the target learner at each time slice to predict the current learning efficiency. By utilizing the learning efficiency index of the target learner in each time slice, as well as the historical learning efficiency index, historical learning motivation index, and stable interactive neighbor learning motivation index of the corresponding time slice, the parameter vectors in the time-series causal association model are estimated based on the least squares method.
[0011] As a preferred embodiment of the learning efficiency improvement method based on big data analysis described in this invention, the specific steps of calculating the theoretical improvement value of implementing learning motivation enhancement intervention on learners within a time slice and generating a set of intervention candidate points are as follows: After the parameter vector estimation is completed, for each learner, the scenario in which peers in the stable interaction neighbor set experience changes in learning motivation during the candidate time slice is taken as the learning motivation enhancement hypothesis. Based on the temporal causal relationship model, the total influence strength of peers on learners' learning efficiency is calculated. By combining the overall influence of peers on learners’ learning efficiency, the theoretical increase in learning efficiency of the target learner and other learners over the predicted time period is calculated. Furthermore, the theoretical improvement values of each learner in each candidate time slice were summarized within the class, and a set of intervention candidate points indexed by learner identity and candidate time slice was obtained.
[0012] As a preferred embodiment of the learning efficiency improvement method based on big data analysis described in this invention, the candidate time slice refers to the time position within the time window that satisfies the condition that the time slice is not earlier than the minimum lag time, the target learner is in an online learning state and has at least one learning behavior record within the time slice, the target learner is in the learning interaction relationship matrix within the time slice, and the time slice plus the predicted duration does not exceed the end position of the current time window.
[0013] As a preferred embodiment of the learning efficiency improvement method based on big data analysis described in this invention, the step of selecting the minimum intervention point with the highest theoretical improvement value based on intervention candidate points during the teaching process, and then implementing a matching intervention strategy for the learner, specifically includes the following steps: During the teaching process, when the current time slice is reached, intervention candidate points that are no later than the current time and have not yet been implemented are selected from the intervention candidate point set; The theoretical improvement values of the intervention candidate points are sorted in descending order. The intervention candidate points are selected from the sorting results as the target intervention objects for the current time slice, and the corresponding learner and the time slice in which the learner is located are determined as the minimum intervention point. For each minimum intervention point, the learning motivation index is compared with the motivation threshold, and the learning efficiency index is compared with the efficiency threshold. Based on the comparison results, the current state of the target learners is divided into different types of learning states, and intervention is carried out according to the intervention strategies corresponding to the learning states.
[0014] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the learning efficiency improvement method based on big data analysis as described in the first aspect of the present invention.
[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the learning efficiency improvement method based on big data analysis as described in the first aspect of the present invention.
[0016] The beneficial effects of this invention are as follows: By dividing learning behavior logs into continuous time slices at uniform time intervals, high temporal resolution dynamic tracking of fluctuations in learning motivation and changes in learning efficiency is achieved, significantly improving the accuracy of identifying the state of the learning process; by constructing a learning interaction relationship matrix with learners as nodes and interaction intensity as weights, quantitative representation of the learning interaction structure within the class and the peer influence path is achieved; by calculating the theoretical improvement value of implementing learning motivation enhancement interventions for learners in each time slice based on temporal causal relationships, the minimum intervention target and intervention time with the greatest intervention benefit are selected under the meaning of the overall learning efficiency of the class, thereby achieving a higher overall efficiency improvement while keeping the number of interventions under control. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart of a method for improving learning efficiency based on big data analytics.
[0019] Figure 2 A flowchart for generating a set of behavior records.
[0020] Figure 3 A flowchart for calculating learning motivation indicators.
[0021] Figure 4 This is a flowchart for calculating the learning efficiency index. Detailed Implementation
[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for improving learning efficiency based on big data analysis, including the following steps: S1. Collect learners' learning behavior logs and divide them into continuous time slices according to a uniform time interval. Combine the learning behavior logs of each learner in each time slice to generate a set of behavior records at the time slice level.
[0026] In a real teaching environment, learners' learning behavior logs are collected.
[0027] Furthermore, each learner participating in the instruction is assigned a unique learner identifier, which remains unchanged throughout the entire instruction cycle.
[0028] A unified learning behavior log is defined for all learning-related operations. The learning behavior log includes the learner identifier corresponding to each log, the timestamp of the operation, the operation type, the resource identifier associated with the operation, and extended information fields. Among them, the operation type includes entering the course page, starting to do the questions, submitting the answer, sending discussion information, and browsing other people's answers. The resource identifier associated with the operation includes the question number, the discussion topic number, and the courseware number. The extended information fields include the time taken to answer the questions, the answer result, and the text length, etc., and are stored in the form of key-value pairs.
[0029] Classroom answering terminals, online course web pages, and mobile learning applications report logs according to a unified structure for learning behavior logs, and write the learning behavior logs into the original log table of the learning behavior database.
[0030] The timestamps of the operations in the learning behavior log are time-standardized and converted into a unified local teaching time. The learning behavior logs are then sorted in chronological order to obtain the time-standardized learning behavior logs.
[0031] Furthermore, when receiving the original log table, the local time reported by each terminal is uniformly converted to Coordinated Universal Time, and the corresponding time zone offset is configured according to the teaching location. The operation time of the learning behavior log is converted into a unified local teaching time to obtain a standardized timestamp.
[0032] For all learning behavior logs within the same teaching class and on the same date, sort them in ascending order according to the standardized timestamps to obtain the time-standardized learning behavior log sequence.
[0033] For multiple records with duplicate timestamps, they are sorted in a secondary order according to the priority sequence of "prioritizing page operations, then starting the quiz, and finally submitting the quiz".
[0034] By setting time intervals, the teaching process is divided into several consecutive time slices. Within each time slice, a set of behavior records at the time slice level is generated for each learner based on the time-standardized learning behavior log.
[0035] Furthermore, a uniform time interval is set, and the teaching time within a day is discretized into multiple adjacent non-overlapping time slices; for each standardized learning behavior log, the index of the time slice to which it belongs is calculated based on the timestamp, and each learning behavior log is assigned to the corresponding time slice; where the index of the time slice refers to the ratio of the timestamp to the time interval.
[0036] For each learner, all learning behavior logs are combined within each time slice to generate a time slice-level set of behavior records.
[0037] Data cleaning and missing time slice completion are performed on learning behavior logs with missing fields in the time slice level behavior records.
[0038] Furthermore, if the missing field of the behavior record is a key value in the extended information, it will be filled in according to the default value, such as setting the missing answer time to the average time of the question; if the missing field is the operation type or the resource identifier associated with the operation, the current behavior record will be marked as unavailable and stored in the exception log table.
[0039] For learners who have no behavioral records in multiple consecutive time slices, an empty behavioral record is created for each empty time slice in the learning behavior log sequence, denoted as an empty set, to maintain the integrity of the timeline.
[0040] The cleaned time-slice-level behavior records are partitioned and stored according to class, course, and date.
[0041] S2. Identify the interactive behaviors between learners based on the behavioral record set, and establish a learning interaction relationship matrix with learners as nodes and interaction intensity as weights; calculate the learning motivation index based on the duration of stay in learning resources, the number of times of active learning, the proportion of completion of learning tasks, and non-learning operations, and calculate the learning efficiency index based on the course knowledge points, learners' answer records, and homework scores within the teaching cycle.
[0042] The system identifies the types of operations that belong to learning interactions in the behavior record set, identifies the interactive behaviors between learners, and constructs a weighted learning interaction relationship matrix with all learners in the class as vertices; the interaction intensity between each learner is used as the relationship weight of the learning interaction relationship matrix.
[0043] Furthermore, the types of operations that belong to learning interactions can be identified from the set of behavior records. These types of operations refer to actions in teaching applications that have a directional influence on another learner's learning content, learning behavior, or shared tasks. These include, but are not limited to, posting content in discussion forums, replying to others' posts, peer-reviewing others' assignments or answers, submitting progress reports and marking collaborating members within project groups, and sending course-related messages in group chat tools.
[0044] For each interaction, extract the learner identifier of the initiator and the learner identifier of the interaction object. For example, if learner A replies to learner B's statement, it is considered that there is an interaction from A to B. In a group task, if learner A and learner B are labeled in the same group, an undirected interaction is established between them in the current time slice.
[0045] For each time slice, a weighted learning interaction relationship matrix is constructed with all learners in the class as vertices. If no interaction from learner A to learner B is detected in the time slice, the weight of the A→B relationship is 0. If an interaction is detected in the time slice, the interaction strength between the initiating learner and the interacting learner is calculated to obtain the interaction strength between the initiating learner and the interacting learner.
[0046] Specifically, the interaction strength is calculated by using a logarithmic compression method to perform a non-linear mapping on the number of interactions between learner A and learner B within a time slice. The interaction strength is obtained by summing the number of direct messages, replies, peer reviews, and group tasks, and then taking the logarithm, as follows: ; in, Indicates time slice The intensity of interaction between learner A and learner B This indicates that learner A is in the time slice The number of times introverted learner B sends direct messages. This indicates that learner A is in the time slice The number of responses to content from introverted learner B. This indicates that learner A is in the time slice The number of times introverted learner B had their assignments or answers peer-reviewed. This indicates that learner A and learner B are in the same time slice. The number of times you participate in the same group task.
[0047] The intensity of the interaction between the learner who initiates the interaction and the learner who is interacting with the other learner is used as the relation weight in the learning interaction relation matrix.
[0048] The study normalizes and smooths the duration of time spent on learning resources, the number of times of active learning, the completion rate of learning tasks, and non-learning operations to obtain learning motivation indicators.
[0049] Furthermore, for each learner and time slice, several basic learning characteristics are calculated based on the set of behavioral records. These basic learning characteristics include the duration of time spent on learning resources, the number of active learning operations, the completion rate of learning tasks, and the proportion of non-learning operations.
[0050] Among them, the duration of stay in learning resources refers to the total time spent accessing resources such as course pages, question pages, and courseware pages within the statistical time slice; the number of active learning operations refers to the number of learning-related operations initiated by learners within the statistical time slice, such as actively entering new resources, actively asking questions, and actively initiating discussions; the learning task completion rate refers to the ratio of the actual number of planned tasks completed to the number of tasks that should be completed within the time slice; and the non-learning operation rate refers to the proportion of operations unrelated to learning within the time slice to the total number of operations.
[0051] The basic learning features are normalized by linear scaling.
[0052] In linear scaling, a reference maximum value is defined for each basic learning feature within a given teaching cycle; the reference maximum value for each basic learning feature includes the reference maximum value for the duration of study resource stay, the reference maximum value for the number of active learning operations, and the reference value for the learning task completion rate.
[0053] It should be noted that within the pre-set teaching cycle, such as a week or a semester, the maximum values of each basic learning characteristic of all learners in all time slices are used as reference maximum values.
[0054] The average value of each normalized basic learning feature is calculated to obtain the original score of learning motivation.
[0055] It should be noted that the proportion of non-learning operations is a negative characteristic, and when calculating the average value, the value should be 1 minus the proportion of non-learning operations.
[0056] The learning motivation metric is obtained by averaging the raw learning motivation score over the application length.
[0057] It should be noted that the learning motivation index is used to characterize the learner's subjective engagement and willingness to learn in each time slice within a unified time dimension.
[0058] The course knowledge points within the teaching cycle are statistically analyzed, and the knowledge mastery is calculated for each learner, each course knowledge point, and each time slot based on learners' answer records and homework scores.
[0059] Furthermore, the course knowledge points that need to be learned within the pre-set teaching cycle are statistically analyzed and numbered. Let there be a total of... Each course knowledge point, regarding the course knowledge points Set weights, with each course knowledge point having a weight greater than 0, and the specific weight of each course knowledge point can be determined based on the teaching syllabus.
[0060] Based on learners' answer records and assignment scores, knowledge mastery is calculated for each learner, each course knowledge point, and each time slot. Specifically, the knowledge points involved in the course are numbered and managed. When creating questions and assigning assignments, each question and assignment is labeled with at least one course knowledge point involved and its weight percentage. In the learning behavior log, each answer and assignment submission record includes the learner identifier, question / assignment identifier, submission time, original score, and full score. When calculating knowledge mastery, for the target learner, target knowledge point, and target time slot, all answer records and assignment records of the target learner with a non-zero weight percentage for the target knowledge point within the target time slot are selected. For each record, a normalized score is calculated based on the original score and the full score, and then averaged according to the weight percentage of each record with the target knowledge point to obtain the knowledge mastery of the target knowledge point for each record.
[0061] Based on the knowledge point weights and corresponding knowledge mastery levels of all course knowledge points, a comprehensive knowledge level calculation is performed on all course knowledge points in the course. The comprehensive knowledge level is then subjected to differential calculation and smoothing of adjacent time slices to obtain a learning efficiency index.
[0062] Furthermore, based on all the course knowledge points involved, a comprehensive knowledge level score is calculated by assigning weights to each knowledge point and corresponding levels of knowledge mastery, as shown below: ; in, Indicates learners in time slices The score reflects the overall knowledge level within the subject. This indicates the total number of knowledge points in the course. Indicates course knowledge points Knowledge point weights Indicates learners in time slices Internal course knowledge points Knowledge mastery level.
[0063] Based on the comprehensive knowledge level score, the difference between adjacent time slices, i.e. the difference in comprehensive knowledge level scores between adjacent time slices, is used to describe learning efficiency and obtain the raw score of learning efficiency.
[0064] The raw learning efficiency score is smoothed to obtain a learning efficiency index.
[0065] S3. Within the time window, combine the learning interaction relationship matrix, learning motivation indicators, and learning efficiency indicators to establish a temporal causal relationship model for the target learners; based on the temporal causal relationship model, calculate the theoretical improvement value of implementing learning motivation enhancement intervention for learners within the time slice, and generate a set of intervention candidate points.
[0066] Based on the learning interaction relationship matrix, select several neighbor learners as stable interaction neighbors for each learner, and generate a stable interaction neighbor set.
[0067] Furthermore, for each learner, in each time slice, an interactive neighbor set is extracted based on the learning interaction relationship matrix, and the neighbor set of each learner is sorted from largest to smallest according to the relationship weight, and a certain number of interactive neighbors are retained, such as the top 50%, as the final neighbor set.
[0068] Divide the time slice into segments, and within a time window, take the union of the final neighbor sets to obtain a stable set of interacting neighbors.
[0069] Based on learning motivation indicators, learning efficiency indicators, and a stable set of interactive neighbors, a temporal causal association model is constructed for the target learner at each time slice to predict the current learning efficiency.
[0070] Furthermore, for each learner, at each time slice, a time-series causal relationship model between learning motivation indicators and learning efficiency indicators is constructed, represented as follows:
[0071] in, Indicates time slice For learners Predictive learning efficiency Indicate learner The bias constant, This represents the upper limit of the lag time for learning efficiency. The upper limit of the lag time for indicating learning motivation. Indicate learner Time lag The influence coefficient of historical learning efficiency on current learning efficiency at any given moment. Indicate learner Time lag The influence coefficient of historical learning motivation on current learning efficiency at any given moment. Indicates companion Time lag Motivation for learning at any time, for learners The impact coefficient of current learning efficiency Indicate learner In time slice The smoothed learning efficiency index Indicate learner In time slice Smoothed learning motivation indicators Indicate learner A stable set of interacting neighbors. Indicates that companion 𝑗 is in time slice Smoothed learning motivation indicators Indicate learner With companions The relation weights of the learning interaction relation matrix.
[0072] It should be noted that the upper limits of the lag time for learning efficiency and learning motivation are used to limit the maximum time span of historical learning efficiency and historical learning motivation considered when constructing a time-series causal relationship model of learning motivation and learning efficiency indicators. Specifically, within a time window, the learning efficiency and learning motivation indicators of all learners in the class are obtained and aligned according to the time slice index to obtain the learning efficiency time series and the learning motivation time series. Generally, the number of time slices within the time window is used as the candidate upper limits of the lag time for learning efficiency and the lag time for learning motivation. Autocorrelation analysis is performed on each candidate lag order in the candidate upper limits of the lag time for learning efficiency and the lag time for learning motivation to obtain the autocorrelation coefficients of learning motivation and learning efficiency. Autocorrelation thresholds for learning motivation and learning efficiency are set, and the candidate upper limits of the lag time for learning efficiency and the lag time for learning motivation with autocorrelation coefficients greater than or equal to the autocorrelation thresholds are used as the final upper limits of the lag time for learning efficiency and the lag time for learning motivation.
[0073] It should also be noted that the autocorrelation threshold is a parameter obtained by calibrating the candidate upper limit based on historical lag time.
[0074] By utilizing the learning efficiency index of the target learner in each time slice, as well as the historical learning efficiency index, historical learning motivation index, and stable interactive neighbor learning motivation index of the corresponding time slice, the parameter vectors in the time-series causal association model are estimated based on the least squares method.
[0075] For each learner's parameter vector , represented as: ; Collect all satisfaction data for each learner The time slices are used to organize the learning efficiency metrics and corresponding lag parameters into a matrix form, and the output vector is generated. , represented as: ; Input matrix Each row corresponds to the lag variable settings of the time slice, including , as well as The parameter vector is estimated using the least squares method. The parameter estimation process is expressed as follows: ; After parameter vector estimation, for each learner, the scenario of changes in learning motivation among peers in the stable interaction neighbor set during candidate time slices is used as the hypothesis for learning motivation enhancement. Based on the temporal causal association model, the total influence strength of peers on learner learning efficiency is calculated. Combining the total influence strength of peers on learner learning efficiency, the theoretical improvement value of learning motivation enhancement on the learning efficiency of the target learner and other learners within the predicted duration is calculated. The theoretical improvement values of each learner in each candidate time slice are summarized within the class to obtain a set of intervention candidate points indexed by learner identity and candidate time slice.
[0076] Furthermore, after parameter estimation, the overall influence of peers on learners' learning efficiency is calculated to evaluate the peer motivation contagion effect on learners. The total influence strength of peer motivation is defined as follows: ; in, Indicates companion For learners The overall intensity of the impact.
[0077] To derive the potential minimum intervention point from the results of temporal causality computation, it is assumed that in a candidate time slice... For learners The learning motivation index generates a unit increment And use the estimated parameters to calculate in the subsequent Approximate impact on the learning efficiency of other learners within the time slice.
[0078] Among them, candidate time slices refer to time slices A1-A4 that meet the following conditions within the time window: A1: The time slice is not earlier than the time position corresponding to the minimum lag time.
[0079] A2: Within the time slice, the target learner is in an online learning state and there is at least one learning-related behavior record.
[0080] A3: Within a time slice, the target learner has at least one effective interaction weight with an interaction neighbor in the learning interaction matrix, i.e., the learning interaction matrix.
[0081] A4: The time slice plus the forecast duration does not exceed the end of the current time window.
[0082] For every affected learner In time slice Due to the learner The change in learning efficiency caused by a change in learning motivation is approximately: ; in, Indicates time slice For learners The learning motivation hypothesis assumes that after improving by one unit, in a time slice Learners The amount of change in predicted learning efficiency. Indicate learner Time lag Motivation for learning at any time is important for learners The impact coefficient of current learning efficiency Indicate learner With learners The relation weights of the learning interaction matrix. Represents the time alignment factor, when It takes the value 1 when it is active, and 0 otherwise.
[0083] At a given prediction step size Below, for all learners The positive learning efficiency increments are accumulated to obtain the result in time slices. For learners The theoretical uplift value of implementing a learning motivation enhancement intervention is expressed as: ; in, Indicates when in time slice For learners When performing a hypothetical intervention that increases learning motivation by one unit, the predicted step size is... Then for all learners The theoretical improvement in learning efficiency.
[0084] For each learner and time slices calculate We obtain candidate intervention points for all learners and store the results in the intervention candidate point set.
[0085] Among them, candidate intervention points refer to those that have completed the calculation of the learning statistical improvement hypothesis and the theoretical improvement value. combination.
[0086] S4. During the teaching process, the minimum intervention point with the highest theoretical improvement value is selected based on the intervention candidate points, and the matching intervention strategy is implemented for the learners.
[0087] During the teaching process, when the current time slice is reached, intervention candidate points that are no later than the current time and have not yet been intervened are selected from the intervention candidate point set.
[0088] Furthermore, at the arrival of each new time slice, all entries that satisfy the time constraints are read from the set of intervention candidate points. Among them, time constraint refers to No later than the current time slice, and the corresponding intervention has not yet been implemented.
[0089] The theoretical improvement values of the intervention candidate points are sorted in descending order. A limited number of intervention candidate points are selected from the sorting results as the target intervention objects for the current time slice, and the corresponding learner and the time slice in which the learner is located are determined as the minimum intervention point.
[0090] For each minimum intervention point, the learning motivation index is compared with the motivation threshold, and the learning efficiency index is compared with the efficiency threshold.
[0091] Furthermore, for each student at the minimum intervention point, the student's learning motivation and learning efficiency indicators in the current time slice are read and compared with the motivation and efficiency thresholds to classify the student's status at the minimum intervention point into four intervention types.
[0092] The four intervention types include Intervention State 1, Intervention State 2, Intervention State 3, and Intervention State 4. When the learning motivation index is less than the motivation threshold and the learning efficiency index is less than the efficiency threshold, the student's state is classified as Intervention State 1. When the learning motivation index is less than the motivation threshold and the learning efficiency index is greater than or equal to the efficiency threshold, the student's state is classified as Intervention State 2. When the learning motivation index is greater than or equal to the motivation threshold and the learning efficiency index is less than the efficiency threshold, the student's state is classified as Intervention State 3. When the learning motivation index is greater than or equal to the motivation threshold and the learning efficiency index is greater than or equal to the efficiency threshold, the student's state is classified as Intervention State 4.
[0093] Based on the comparison results, the current state of the target learners is divided into different types of learning states, and intervention strategies are implemented according to the corresponding learning states to improve learning efficiency.
[0094] Furthermore, the intervention strategy corresponding to intervention state 1 specifically involves recommending a peer with a high relationship weight with the student and high current learning motivation and efficiency, generating a joint practice task in the learning application, and pushing it to both students simultaneously.
[0095] The intervention strategy corresponding to intervention state 2 is to automatically select a set of practice questions that are slightly less difficult than the student's current task and arrange a Q&A session with a peer with high learning motivation.
[0096] The intervention strategy corresponding to intervention state 3 is as follows: assign the student a new task with a slightly higher difficulty level, clearly define the student as the leader in the group in the task description, and send invitations to students in the student's neighborhood with moderate motivation to join the new task.
[0097] The intervention strategy corresponding to intervention state 4 is specifically to assign students extended tasks or peer tutoring tasks, and to open up a Q&A portal for students with low motivation in the learning application.
[0098] It should be noted that the acquisition of the motivation threshold and efficiency threshold is as follows: during the teaching cycle, the learning motivation indicators and learning efficiency indicators acquired by all learners in all time slices are statistically analyzed, and after sorting the learning motivation indicators and learning efficiency indicators from smallest to largest, the learning motivation indicators and learning efficiency indicators that are generally located at the 59th percentile are taken as the motivation threshold and efficiency threshold, with a value range of [0,1].
[0099] This embodiment also provides a computer device applicable to the learning efficiency improvement method based on big data analysis, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the learning efficiency improvement method based on big data analysis proposed in the above embodiment.
[0100] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.
[0101] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the learning efficiency improvement method based on big data analysis as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.
[0102] In summary, this invention achieves high temporal resolution dynamic tracking of fluctuations in learning motivation and changes in learning efficiency by dividing learning behavior logs into continuous time slices at uniform time intervals, significantly improving the accuracy of learning process state identification. By constructing a learning interaction relationship matrix with learners as nodes and interaction intensity as weights, it achieves quantitative representation of the learning interaction structure within the class and the peer influence path. Based on temporal causal relationships, it calculates the theoretical improvement value of implementing learning motivation enhancement interventions for learners in each time slice, enabling the selection of the minimum intervention target and intervention time with the greatest intervention benefit under the meaning of overall class learning efficiency, thereby achieving a higher overall efficiency improvement while keeping the number of interventions under control.
[0103] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for improving learning efficiency based on big data analysis, characterized in that: include, Collect learners' learning behavior logs and divide them into continuous time slices according to a uniform time interval. Combine each learner's learning behavior logs in each time slice to generate a time slice-level behavior record set. Based on the behavioral record set, the interaction behavior between learners is identified, and a learning interaction relationship matrix is established with learners as nodes and interaction intensity as weights. Learning motivation indicators are calculated based on the duration of time spent on learning resources, the number of times active learning occurs, the completion rate of learning tasks, and non-learning operations. Learning efficiency indicators are calculated based on the course knowledge points, learner answer records, and homework scores within the teaching cycle. Within the time window, a temporal causal relationship model of the target learners is established by combining the learning interaction relationship matrix, learning motivation indicators, and learning efficiency indicators. Based on the temporal causal relationship model, the theoretical improvement value of implementing learning motivation enhancement interventions on learners within a time slice is calculated, and a set of intervention candidate points is generated. During the teaching process, the minimum intervention point with the highest theoretical improvement value is selected based on the intervention candidate points, and the matching intervention strategy is implemented for the learners.
2. The learning efficiency improvement method based on big data analysis as described in claim 1, characterized in that: The specific steps for generating the time-slice-level behavior record set are as follows: In a real teaching environment, collect learning behavior logs of each learner; The timestamps of the operations in the learning behavior log are time-standardized and converted into a unified local teaching time. The learning behavior logs are then sorted in chronological order to obtain the time-standardized learning behavior logs. The teaching process is divided into continuous time slices by setting time intervals. Within each time slice, a set of behavior records at the time slice level is generated for each learner based on the time-standardized learning behavior logs. The learning behavior log includes the learner identifier, the timestamp of the operation, the operation type, the resource identifier associated with the operation, and extended information fields for each log entry.
3. The learning efficiency improvement method based on big data analysis as described in claim 2, characterized in that: The establishment of a learning interaction relationship matrix with learners as nodes and interaction intensity as weights refers to identifying the operation types that belong to learning interactions in the behavior record set, identifying the interactive behaviors between learners, and constructing a learning interaction relationship matrix with all learners in the class as vertices and weighted relationships. The intensity of interaction among learners is used as the relation weight in the learning interaction relation matrix.
4. The learning efficiency improvement method based on big data analysis as described in claim 3, characterized in that: The computational learning motivation index refers to the normalization and smoothing of learning resource dwell time, number of active learning sessions, learning task completion rate, and non-learning operations to obtain the learning motivation index. The computational learning efficiency index refers to the statistical analysis of course knowledge points within the teaching cycle, and the calculation of knowledge mastery for each learner, each course knowledge point, and each time slice based on learners' answer records and homework scores. Based on the knowledge point weights and corresponding knowledge mastery levels of all course knowledge points, a comprehensive knowledge level calculation is performed on all course knowledge points in the course. The comprehensive knowledge level is then subjected to differential calculation and smoothing of adjacent time slices to obtain a learning efficiency index.
5. The learning efficiency improvement method based on big data analysis as described in claim 4, characterized in that: The specific steps for establishing the temporal causal association model for the target learners are as follows: For each learner, a set of stable interactive neighbors is generated by selecting neighbor learners as stable interactive neighbors based on the learning interaction relationship matrix. Based on learning motivation indicators, learning efficiency indicators, and a stable set of interactive neighbors, a temporal causal association model is constructed for the target learner at each time slice to predict the current learning efficiency. By utilizing the learning efficiency index of the target learner in each time slice, as well as the historical learning efficiency index, historical learning motivation index, and stable interactive neighbor learning motivation index of the corresponding time slice, the parameter vectors in the time-series causal association model are estimated based on the least squares method.
6. The learning efficiency improvement method based on big data analysis as described in claim 5, characterized in that: The calculation of the theoretical improvement value of implementing learning motivation enhancement intervention on learners within a time slice, and the generation of a set of intervention candidate points, are as follows: After the parameter vector estimation is completed, for each learner, the scenario in which peers in the stable interaction neighbor set experience changes in learning motivation during the candidate time slice is taken as the learning motivation enhancement hypothesis. Based on the temporal causal relationship model, the total influence strength of peers on learners' learning efficiency is calculated. By combining the overall influence of peers on learners’ learning efficiency, the theoretical increase in learning efficiency of the target learner and other learners over the predicted time period is calculated. Furthermore, the theoretical improvement values of each learner in each candidate time slice were summarized within the class, and a set of intervention candidate points indexed by learner identity and candidate time slice was obtained.
7. The learning efficiency improvement method based on big data analysis as described in claim 6, characterized in that: The candidate time slice refers to the time position within the time window that satisfies the following conditions: the time slice is not earlier than the minimum lag time; the target learner is in an online learning state and has at least one learning behavior record within the time slice; the target learner is in the learning interaction relationship matrix within the time slice; and the time slice plus the predicted duration does not exceed the end position of the current time window.
8. The learning efficiency improvement method based on big data analysis as described in claim 7, characterized in that: During the teaching process, the minimum intervention point with the highest theoretical improvement value is selected based on the candidate intervention points, and a matching intervention strategy is implemented for the learner. The specific steps are as follows: During the teaching process, when the current time slice is reached, intervention candidate points that are no later than the current time and have not yet been implemented are selected from the intervention candidate point set; The theoretical improvement values of the intervention candidate points are sorted in descending order. The intervention candidate points are selected from the sorting results as the target intervention objects for the current time slice, and the corresponding learner and the time slice in which the learner is located are determined as the minimum intervention point. For each minimum intervention point, the learning motivation index is compared with the motivation threshold, and the learning efficiency index is compared with the efficiency threshold. Based on the comparison results, the current state of the target learners is divided into different types of learning states, and intervention is carried out according to the intervention strategies corresponding to the learning states.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the learning efficiency improvement method based on big data analysis as described in any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the learning efficiency improvement method based on big data analysis as described in any one of claims 1 to 8.