An online inventory method, system, and medium

Abstract

This invention discloses an online inventory method, system, and medium, belonging to the field of inventory technology. The method includes: acquiring user behavior data based on data tracking; determining multi-dimensional indicators and comprehensive learning progress based on the behavior data; performing user behavior anomaly identification and generating identification results; performing dynamic feature filtering guided by learning behavior trends to generate a feature system; performing clustering on the feature system based on the K-means clustering algorithm to generate clustering results representing user group categories; wherein, the K value of the K-means clustering algorithm is determined by a dynamic adjustment rule guided by teaching weights; performing cluster feature anchoring processing and teaching matching on the clustering results to generate core features representing user learning details and teaching suggestions for guiding user learning; and generating an inventory report based on the core features, teaching suggestions, and clustering results. This invention has high inventory efficiency and data accuracy.

Description

Technical Field

[0001] This invention belongs to the field of inventory technology, specifically relating to an online inventory method, system, and medium. Background Technology

[0002] Against the backdrop of digital education transformation, online courses, with their advantages of flexibility and convenience, have been widely used in primary and secondary schools, universities, and various educational and training institutions, forming a large-scale online learning ecosystem.

[0003] The number of online education users in my country continues to grow, with the number of online courses increasing by an average annual rate of over 20%, and the proportion of learning time completed by students through online channels increasing year by year. However, with the increase in the number of online courses and the expansion of the learning group, educational administrators face severe challenges in inventory management: traditional manual inventory methods rely on teachers to collect statistics on students' course attendance, video viewing progress, and homework completion one by one. This is not only time-consuming and labor-intensive, but also prone to data omissions and statistical errors, making it difficult to achieve real-time tracking and dynamic inventory of learning progress.

[0004] Meanwhile, the increasing demand for personalized education requires educational administrators to accurately identify each student's learning weaknesses. Traditional assessment methods only provide basic progress data and cannot achieve in-depth analysis of learning behavior or timely warnings of anomalies. Furthermore, while some existing online education platforms have basic progress statistics functions, they lack a systematic assessment framework and cannot meet the professional needs of educational management, such as batch assessment, multi-dimensional analysis, and report export, thus hindering the optimization of teaching decisions. Summary of the Invention

[0005] The purpose of this invention is to provide an online inventory method, system, and medium to solve the problems of existing technologies in achieving real-time tracking and dynamic inventory of learning progress, as well as in achieving in-depth analysis of learning behavior and timely early warning of abnormal situations.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides an online inventory method, the method comprising:

[0008] Based on data tracking, user behavior data is obtained, and multi-dimensional indicators and comprehensive learning progress are determined based on the behavior data.

[0009] Based on multi-dimensional indicators and comprehensive learning progress, perform user behavior anomaly identification and generate identification results;

[0010] Dynamic feature selection guided by learning behavior trends is performed on multi-dimensional indicators, comprehensive learning progress, and recognition results to generate a feature system;

[0011] Clustering is performed on the feature system based on the K-means clustering algorithm to generate clustering results that characterize user group categories; wherein, the K value of the K-means clustering algorithm is determined by a dynamic adjustment rule guided by teaching weights;

[0012] Cluster feature anchoring and instruction matching are performed on the clustering results to generate core features that characterize user learning details and instructional suggestions that guide user learning. An inventory report is then generated based on the core features, instructional suggestions, and clustering results.

[0013] Preferably, the multi-dimensional indicators include: video completion rate, assignment pass rate, quiz accuracy rate, and interactive participation rate; the comprehensive learning progress is determined based on the multi-dimensional indicators, including:

[0014] Calculate the weight of each indicator in the multi-dimensional indicators;

[0015] Based on the entropy weight method and the proportion of each indicator, the fusion weight of each indicator in the multi-dimensional indicators is determined;

[0016] The multi-dimensional indicators are weighted and merged according to the fusion weight to generate a comprehensive learning progress.

[0017] Preferably, user behavior anomaly identification is performed based on multi-dimensional indicators and comprehensive learning progress, including:

[0018] Based on preset dynamic thresholds, abnormal behavior is judged for multi-dimensional indicators and comprehensive learning progress, abnormal types are generated, and the frequency of abnormal behavior is counted. The abnormal types include progress lag abnormality, homework abnormality, test abnormality and interaction abnormality.

[0019] Abnormal behavior is quantified to generate anomaly scores, and the anomaly type, frequency, and anomaly score of the abnormal behavior are used as the identification results.

[0020] Preferably, dynamic feature filtering guided by learning behavior trends is performed on multi-dimensional indicators, comprehensive learning progress, and recognition results, including:

[0021] A basic feature set is constructed based on multi-dimensional indicators and comprehensive learning progress. The basic feature set includes five basic features and two time-sensitive features. The five basic features are, in order, video completion rate, homework pass rate, test accuracy rate, interactive participation rate and comprehensive learning progress. The two time-sensitive features are the learning progress growth rate, which is used to quantify the dynamic changes in the user's learning progress, and the frequency of learning behaviors, which is used to characterize the user's learning coherence.

[0022] Extract the abnormal scores from the recognition results, associate the abnormal scores with the basic feature set, and generate an associated feature set;

[0023] Redundant features are removed from the associated feature set based on trend similarity analysis to generate a feature system.

[0024] Preferably, the trend similarity analysis method is as follows:

[0025] Select any two features from the set of associated features, calculate the time-series trend slope of the two features, and obtain the time-series trend slope corresponding to each feature;

[0026] Calculate the coefficient of variation of each feature in the associated feature set, which is used to characterize the feature dispersion, and extract the entropy value of each feature in the entropy weight method calculation process;

[0027] The discriminative power of each feature is determined based on its coefficient of variation and entropy value.

[0028] Calculate the trend consistency between any two features based on the time-series trend slope corresponding to each feature;

[0029] When the trend consistency between any two features is greater than a preset threshold, the feature with the lowest discriminative power among the two features will be removed.

[0030] Preferably, the dynamic adjustment rule for teaching weight orientation is as follows:

[0031] Based on the elbow rule, the silhouette coefficients of each K value in the preset K value range of the K-means clustering algorithm are determined, and at least two candidate K values ​​are selected based on the silhouette coefficients.

[0032] Construct a weighted approach to teaching intervention priorities, and determine the teaching suitability of each candidate K value based on the weighted approach to teaching intervention priorities.

[0033] Based on the teaching fit and silhouette coefficient of each candidate K value, the teaching guidance evaluation value of each candidate K value is determined, and the candidate K value corresponding to the largest teaching guidance evaluation value is taken as the K value of the K-means clustering algorithm.

[0034] Preferably, cluster feature anchoring and instructional matching are performed on the clustering results, including:

[0035] Calculate the mean of each feature in the feature system of each cluster in the clustering results;

[0036] Anchoring is performed on the mean of each feature based on the preset weights of course knowledge points, generating the anchoring score of each feature in the feature system of each cluster.

[0037] Based on the anchoring score, each feature in the feature system of each cluster is filtered to generate a core feature set for each cluster, wherein the core feature set of each cluster contains at most two core features;

[0038] The total anomaly score for each cluster is determined based on the identification results;

[0039] Based on the total anomaly score and core feature set of each cluster, the teaching matching degree is determined, and teaching suggestions are generated based on the teaching matching degree.

[0040] Preferably, the method further includes: incrementally updating the clustering results, including:

[0041] Real-time monitoring of changes and anomalies in users' learning behavior data;

[0042] Determine the update trigger value based on changes and abnormal changes in learning behavior data;

[0043] When the update trigger value reaches the preset trigger value, acquire the newly added learning behavior data;

[0044] Based on the anomaly-cluster association update rule, the center of the user's cluster is locally iteratively updated to generate a new clustering result.

[0045] Secondly, the present invention provides an online inventory system, the system comprising:

[0046] The management module is used to manage users and their learning objects;

[0047] The data acquisition module is used to collect students' learning behavior data in real time.

[0048] The inventory module is used to execute the online inventory method described above;

[0049] The early warning module is used to generate early warning information based on the identification results;

[0050] The display module is used to visualize early warning information and inventory results.

[0051] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described online inventory method.

[0052] The beneficial effects of this invention are:

[0053] This invention can collect student learning behavior data in real time and automatically complete the inventory and statistics of learning progress through technologies such as user behavior anomaly identification, group clustering, feature anchoring and teaching matching. At the same time, it can classify different users into different categories and generate teaching suggestions, which can significantly reduce the workload of teaching administrators and improve the efficiency and accuracy of inventory. It can also provide precise guidance for the allocation of teaching resources, optimization of teaching strategies, and evaluation of teaching quality. Attached Figure Description

[0054] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0055] Figure 1 This is a flowchart of an online inventory method provided by one embodiment of the present invention;

[0056] Figure 2 This is a block diagram of an online inventory system provided in one embodiment of the present invention. Detailed Implementation

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is 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. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0058] Example 1

[0059] Figure 1 This is a flowchart of an online inventory method provided by one embodiment of the present invention. Figure 1 As shown, this embodiment provides an online inventory method, which includes steps S100 to S500.

[0060] Step S100: Obtain user behavior data based on data tracking points, and determine multi-dimensional indicators and comprehensive learning progress based on the behavior data.

[0061] The users in this embodiment are mainly students, and the behavioral data includes, but is not limited to, video viewing data (viewing time, completion rate, number of invalid views), homework data (number of submissions, pass rate, distribution of wrong questions), quiz data (quiz score, accuracy rate, answering time), interaction data (number of discussion participations, number of questions asked), and abnormal behavior data (no submission, cheating, invalid learning), etc.

[0062] The inventory method in this embodiment uses a Browser / Server architecture. This architecture eliminates the need for dedicated software installation on the client side, allowing users to access the system directly through a browser. This reduces deployment and maintenance costs and supports access from multiple devices (computers, mobile phones, tablets), adapting to diverse usage scenarios for students and teachers. The data processing logic of this inventory method is implemented on the server side, while the client is only responsible for data display and user interaction. This facilitates system upgrades and iterations while effectively controlling data security and preventing client-side data leakage.

[0063] This embodiment embeds tracking code into the system's front-end pages (such as course video playback pages, assignment submission pages, and discussion forums) to collect students' learning behavior data in real time. This data includes video viewing time, fast-forward / pause actions, assignment submission time, answer accuracy, and number of comments in the discussion forum. The collected data is transmitted asynchronously to avoid affecting the user experience. Furthermore, data cleaning and standardization processes ensure the accuracy and completeness of the data, providing data support for subsequent learning progress calculation and intelligent analysis.

[0064] After acquiring behavioral data, the server selects a portion of the data as core data to calculate multi-dimensional indicators. In this embodiment, the multi-dimensional indicators mainly include four core indicators: video completion rate, assignment pass rate, test accuracy rate, and interactive participation rate. These four core indicators cover learning process and learning results, which are in line with the characteristics of online education scenarios.

[0065] Among them, video learning completion rate S1 is a core basic indicator that calculates the ratio of the actual time students watch course videos to the total course time, and removes invalid viewing (such as fast-forwarding more than 1.5 times or watching a single video segment for less than 30 seconds) to ensure the authenticity of the data.

[0066] Calculation formula:

[0067] .

[0068] Among them, the homework pass rate S2 is a core performance indicator, which is calculated as the ratio of the number of times a student submits a qualified homework assignment to the total number of homework assignments. Homework assignments not submitted on time are scored as 0 points, and the pass rate is calculated based on the actual score after the homework is completed and passed.

[0069] Calculation formula:

[0070] .

[0071] Among them, the test accuracy rate S3 is an auxiliary effect indicator, which calculates the average accuracy rate of students' unit tests and knowledge point tests. The average of multiple tests is taken, which mainly reflects the students' mastery of knowledge.

[0072] Calculation formula:

[0073] .

[0074] Among them, interactive participation S4 is a supplementary indicator that calculates the completion of students' interactive behaviors such as participating in course discussions, asking questions, and peer evaluations. The completion rate is calculated based on the preset interactive task volume to encourage students to actively participate in learning.

[0075] Calculation formula:

[0076] .

[0077] Furthermore, in this embodiment, the overall learning progress is determined by multi-dimensional indicators. The steps for determining the overall learning progress are as follows: calculate the weight of each indicator in the multi-dimensional indicators; determine the fusion weight of each indicator in the multi-dimensional indicators according to the entropy weight method and the weight of each indicator; and perform weighted fusion of each indicator in the multi-dimensional indicators according to the fusion weight to generate the overall learning progress.

[0078] This embodiment uses the entropy weight method to calculate the weights of each indicator, avoiding the limitations of subjective weights and ensuring that the weight allocation is consistent with students' actual learning. The specific calculation process is as follows:

[0079] First, calculate the weight of each indicator for each student (video completion rate, assignment pass rate, quiz accuracy, and interaction participation). The calculation formula is as follows:

[0080]

[0081] In the formula, Let j be the weight of the indicator for the i-th student. Let j be the j-th indicator of the i-th student, and I be the total number of students.

[0082] Then, the entropy value of each indicator is calculated using the entropy weight method, and the calculation formula is as follows:

[0083]

[0084] In the formula, e j Let the entropy value be the j-th index. It is a logarithmic function.

[0085] Next, the entropy value of each indicator is used to calculate the fusion weight of each indicator. The calculation formula is as follows:

[0086]

[0087] In the formula, The fusion weight for the j-th metric is denoted as fusion weight for video learning completion. The value is 0.4, which is the fusion weight for the job pass rate. The value is 0.3, which is the fusion weight for the test accuracy. The value is 0.2, which is the fusion weight of interaction participation. The value is 0.1.

[0088] Finally, a weighted average method is used to calculate the overall learning progress of each student, using the following formula:

[0089]

[0090] In the formula, P i The value represents the overall learning progress of the i-th student, ranging from [0,1]. The higher the value, the more advanced the learning progress.

[0091] After calculating the overall learning progress of each student, the overall learning progress of all students and the weight of each core indicator are output and stored synchronously in the system's database as the basis for subsequent calculations.

[0092] In this embodiment, to prevent a single abnormal indicator from affecting the overall progress, an outlier handling mechanism is incorporated. When the score of a certain indicator falls below 10%, a weighted correction (correction coefficient 0.8) is applied to that indicator. Simultaneously, teachers are supported in flexibly adjusting the fusion weights of each indicator based on the course type (theoretical course, practical course), improving the algorithm's adaptability. For example, for practical courses, the fusion weight of assignment pass rate can be increased to 0.4, and the fusion weight of video learning completion rate can be adjusted to 0.3.

[0093] Step S200: Perform user behavior anomaly identification based on multi-dimensional indicators and comprehensive learning progress, and generate identification results.

[0094] Specifically, user behavior anomaly identification is performed based on multi-dimensional indicators and comprehensive learning progress, including: judging abnormal behavior based on multi-dimensional indicators and comprehensive learning progress according to preset dynamic thresholds, generating anomaly types, and counting the frequency of abnormal behavior, wherein the anomaly types include progress lag anomaly, assignment anomaly, test anomaly, and interaction anomaly; quantifying the abnormal behavior, generating anomaly scores, and using the anomaly type, frequency, and anomaly score of the abnormal behavior as the identification results.

[0095] Among them, the dynamic threshold is the abnormal threshold for each indicator and learning progress, which can be flexibly set in combination with the overall learning level of the class and the difficulty of the course (such as learning progress being 30% lower than the class average or homework pass rate being <60%).

[0096] During the abnormal behavior identification process, the abnormality type (progress delay, homework abnormality, test abnormality, interaction abnormality) is determined by comparing each student's learning progress, core indicators, and corresponding abnormality thresholds. Simultaneously, the frequency of abnormal behavior is counted. After identification, the abnormal behavior is quantified using the following function expression:

[0097]

[0098] In the formula, Let be the abnormal score of the i-th student. Let M be the number of the m-th type of abnormal behavior of the i-th student, and M be the total number of abnormal behavior types. denoted as the weight of the m-th type of abnormal behavior.

[0099] This embodiment uses anomaly analysis to calculate learning progress and various core indicators, identify students with learning abnormalities, provide anomaly labels for cluster analysis in step S400, and simultaneously achieve real-time early warning, focusing on understanding which students need attention.

[0100] Step S300: Perform dynamic feature filtering based on learning behavior trends on multi-dimensional indicators, comprehensive learning progress, and recognition results to generate a feature system.

[0101] In this embodiment, after determining the learning progress and core indicator data of each user through step S100 and the abnormal scores and frequency of abnormal behaviors of each user through step S200, in order to implement targeted management and precise policies for different learning groups, this embodiment uses the K-means clustering algorithm to accurately and dynamically classify all users into groups, clarify the characteristic differences and core needs of different learning groups, and provide teaching administrators with specific basis for tiered supervision and personalized teaching.

[0102] K-means clustering is a static clustering algorithm, capable of clustering based only on static data at a specific point in time. It cannot capture real-time changes in student learning behavior, leading to a lag in updating clustering results. Since the system's learning progress tracking and anomaly alerts are real-time functions, the lag in clustering results prevents educational administrators from promptly grasping the dynamic changes in the student population, making it difficult to conduct targeted supervision and teaching adjustments.

[0103] To improve the clustering accuracy in step S400, this embodiment utilizes a feature dynamic screening rule guided by learning behavior trends to perform screening on multi-dimensional indicators, comprehensive learning progress, and recognition results, in order to focus on the dynamic trends and correlations of learning behaviors and construct a feature system that fits the teaching scenario.

[0104] Specifically, the steps for dynamic feature selection based on learning behavior trend are as follows: construct a basic feature set based on multi-dimensional indicators and comprehensive learning progress; extract abnormal scores from the identification results, associate the abnormal scores with the basic feature set to generate an associated feature set; and perform redundant feature removal on the associated feature set based on trend similarity analysis to generate a feature system.

[0105] The basic feature set of this embodiment includes five basic features and two time-sensitive features. The five basic features are, in order, video completion rate, homework pass rate, test accuracy rate, interactive participation rate, and overall learning progress. The two time-sensitive features are the learning progress growth rate and the frequency of learning behaviors.

[0106] The learning progress growth rate can be defined as the change in learning progress over the past 14 days, used to quantify the dynamic changes in a user's learning progress. The calculation formula is as follows:

[0107]

[0108] In the formula, r i P represents the growth rate of the learning progress of the i-th student. i,t Let P be the current overall learning progress of the i-th student. i,t-14 This represents the overall learning progress of the i-th student over the first 14 days.

[0109] The frequency of learning behavior can be a comprehensive indicator of the average daily study time and the number of assignments submitted per week, used to reflect the student's learning continuity. The calculation formula is as follows:

[0110]

[0111] In the formula, F i Let represent the frequency of the learning behaviors of the i-th student. S is the standardized value of the average daily study time of the i-th student. i Let be the standardized value of the number of assignments submitted per week by the i-th student.

[0112] The two time-sensitive features in this embodiment are used to compensate for the deficiency of conventional basic features in distinguishing between intensive learning and steady learning; the basic feature set in this embodiment can cover the process and results of student learning, and is the core foundation for reflecting the student's learning status, ensuring the relevance of the features.

[0113] After completing the construction of the basic feature set, the abnormal scores generated in step S200 are incorporated into the feature system to form a deep linkage between anomaly identification and cluster analysis, making the group characteristics of abnormal students more prominent.

[0114] Furthermore, this embodiment uses trend similarity analysis to remove redundant features from the associated feature set, avoiding the decrease in clustering efficiency caused by feature redundancy. The steps of the trend similarity analysis method in this embodiment are as follows:

[0115] S301: Select any two features from the associated feature set, calculate the time-series trend slope of the two features, and obtain the time-series trend slope corresponding to each feature. The expression for calculating the time-series trend slope is as follows:

[0116]

[0117] In the formula, k ij Let x be the slope of the time series trend of the j-th feature of the i-th student, t=14, s be the number of days in the time series, and x be the slope of the time series trend. ijs Let be the value of the j-th feature of the i-th student on the s-th day.

[0118] S302: Calculate the coefficient of variation of each feature in the associated feature set to characterize the feature dispersion, and extract the entropy value of each feature in the entropy weight method calculation process. The entropy value of each feature is calculated using the calculation method in step S100.

[0119] The expression for calculating the coefficient of variation is as follows:

[0120]

[0121] In the formula, Let be the coefficient of variation of the j-th feature. Let be the standard deviation of the j-th feature after standardization. Let the j-th feature be the standardized mean. If the standardized mean is... If the coefficient of variation of the j-th feature is zero, then the standard deviation after standardization of the j-th feature is directly used.

[0122] S303: Determine the discriminative power of each feature based on its coefficient of variation and entropy value. The expression for calculating the discriminative power of each feature is as follows:

[0123]

[0124] In the formula, D j Let J be the discriminant of the j-th feature, and J be the total number of features to be selected.

[0125] S304: Calculate the trend consistency between any two features based on the time-series trend slope corresponding to each feature. The expression for calculating trend consistency is:

[0126]

[0127] In the formula, To demonstrate the trend consistency between the j-th feature and the l-th feature, Let be the time-series trend slope of the j-th feature of the i-th student. Let be the time-series trend slope of the l-th feature of the i-th student. It is a constant, with a value of 0.001.

[0128] S305: When the trend consistency between any two features is greater than a preset threshold, the feature with the lowest discriminative power among the two features will be removed.

[0129] This embodiment selects features with "trend differentiation" based on the similarity of learning behavior trends, eliminates redundant trend features, and retains core features that can reflect changes in students' learning status, thereby improving the adaptability of clustering to dynamic learning behaviors.

[0130] Step S400: Perform clustering on the feature system based on the K-means clustering algorithm to generate clustering results for representing user group categories; wherein, the K value of the K-means clustering algorithm is determined by a dynamic adjustment rule guided by teaching weights.

[0131] This embodiment uses the K-means clustering algorithm to group students with similar learning behaviors into one category, clarifying the learning characteristics and weaknesses of each group. Targeted teaching strategies can be adopted for different groups, such as providing extended resources to high-achieving groups, arranging special tutoring for lagging groups, and conducting key supervision for abnormal groups, thereby achieving personalized teaching management.

[0132] K-means clustering is a widely used clustering algorithm. Its basic idea is to divide n objects into K clusters based on their attributes, ensuring that each object belongs to the cluster whose mean (centroid) is closest to it. However, a significant challenge with K-means is choosing the appropriate number of clusters, K, as this significantly impacts the quality of the clustering results. For example, if K is too small, some clusters may contain excessively dense clusters, while others may contain very few or no data points. This leads to inaccurate clustering results that fail to accurately reflect the inherent structure of the data. Conversely, if K is too large, each cluster may contain very few data points, making it difficult for cluster centers to accurately represent the characteristics of the entire cluster. In such cases, the clusters may become very sparse and difficult to interpret.

[0133] Existing technologies typically use the elbow rule and silhouette coefficient to adjust the specific K value. Since the K value determination relies solely on the clustering effect (profile coefficient) and does not consider teaching management needs, the clustering results become disconnected from teaching priorities. To address this, this embodiment employs a teaching weight-oriented dynamic adjustment rule to dynamically determine the specific K value, ensuring that the number of clusters matches the hierarchical requirements of teaching management.

[0134] Specifically, the dynamic adjustment rules for teaching weight orientation are as follows:

[0135] S401: Determine the silhouette coefficient of each K value in the preset K value range based on the elbow rule, and select at least two candidate K values ​​according to the silhouette coefficient; the feature system of this embodiment has 7-dimensional features, therefore, the preset K value range is 2≤K≤6; 2-3 K values ​​can be selected as basic candidate K values ​​through the elbow rule and silhouette coefficient, preferably 2-3 K values ​​with higher silhouette coefficients.

[0136] S402: Construct teaching intervention priority weights, and determine the teaching suitability of each candidate K value based on the teaching intervention priority weights.

[0137] Specifically, the priority weights for teaching interventions are as follows: Among them, w1 is the weight of the key supervisory group, with a value of 0.5; w2 is the weight of the regular management group, with a value of 0.3; and w3 is the weight of the optimization and improvement group, with a value of 0.2.

[0138] For each candidate K value, calculate its corresponding instructional fit. The formula for calculating instructional fit is as follows:

[0139]

[0140] In the formula, For the teaching fit of candidate K values, w u Let the priority weight of the u-th teaching intervention be... Let K be the number of students in the u-th teaching group, and n be the total number of students.

[0141] S403: Based on the teaching fit and silhouette coefficient of each candidate K value, determine the teaching guidance evaluation value of each candidate K value, and take the candidate K value corresponding to the largest teaching guidance evaluation value as the K value of the K-means clustering algorithm.

[0142] Specifically, the calculation expression for the teaching-oriented evaluation value of each candidate K value is as follows:

[0143]

[0144] In the formula, The candidate K value is the teaching-oriented evaluation value. The silhouette coefficient of candidate K values ​​is 0.4, the clustering effect weight is 0.4, and the teaching suitability weight is 0.6, emphasizing the principle of prioritizing teaching practicality. Then, the K value with the highest teaching orientation evaluation value is selected as the final number of clusters. Simultaneously, fine-tuning is made based on course type and student grade level (e.g., appropriately lowering the K value for lower-grade courses and appropriately increasing the K value for practical courses) to ensure that the K value both meets the algorithm's clustering requirements and adapts to actual teaching management practices.

[0145] After obtaining the final number of clusters, the initial K-means clustering calculation is performed. The specific operation is as follows: First, K feature vectors are randomly selected as initial cluster centers (to ensure that the initial centers are evenly distributed and to avoid local optima); second, the weighted Euclidean distance between the feature vector of each student and each initial cluster center is calculated, and then each student is assigned to the nearest cluster to form the initial group classification result (usually divided into excellent group, average group, lagging group, and abnormal group, and the specific cluster name is adjusted according to the K value and teaching scenario); finally, through iterative calculation, each cluster center is continuously updated until the cluster centers converge (the change in cluster centers between two iterations is <0.001), and the initial cluster centers and initial group classification results are output and stored synchronously in the system database.

[0146] Therefore, the teaching weight-oriented dynamic adjustment rule in this embodiment breaks the conventional logic of prioritizing clustering effect, incorporates the need for teaching intervention into the core indicator for determining the K value, and makes the number of clusters not only meet the algorithm requirements, but also fit the needs of hierarchical supervision of teaching management, thus solving the pain point that conventional K value optimization has the problem of being the best algorithm but useless for teaching.

[0147] Step S500: Perform cluster feature anchoring and instruction matching on the clustering results to generate core features for characterizing user learning details and instructional suggestions for guiding user learning, and generate an inventory report based on the core features, instructional suggestions and clustering results.

[0148] In this embodiment, the interpretation of the clustering results obtained in step S400 relies solely on the mean of features within each cluster, failing to correlate with specific teaching scenarios and making it difficult to apply the clustering results to teaching practice. This embodiment overcomes the above problems by performing cluster feature anchoring processing and teaching matching on the clustering results. Specifically, the steps of performing cluster feature anchoring processing and teaching matching on the clustering results include:

[0149] S501: Calculate the mean of each feature in the feature system of each cluster in the clustering results. The calculation formula is as follows:

[0150]

[0151] In the formula, Let be the mean of the j-th feature of the k-th cluster. Let be the number of students in the k-th cluster.

[0152] S502: Anchoring is performed on the mean of each feature based on the preset course knowledge point weights, generating the anchoring score for each feature in the feature system of each cluster. The formula for calculating the anchoring score is as follows:

[0153]

[0154] In the formula, The anchoring score for the j-th feature within the k-th cluster. The weight of the course knowledge point corresponding to the j-th feature is as follows: for example, the weight of mastering the course knowledge point corresponding to the homework pass rate is 0.4, and the weight of applying the course knowledge point corresponding to the interaction participation is 0.2.

[0155] S503: Based on the anchoring score, each feature in the feature system of each cluster is filtered to generate a core feature set for each cluster, wherein the core feature set of the cluster contains at most two core features.

[0156] In this embodiment, the two features with the highest anchor scores are preferably selected as the core features of the cluster to clarify the core learning characteristics of each group (such as the core features of the excellent group being high test accuracy and advanced learning progress, and the core features of the lagging group being low video completion rate and low homework pass rate).

[0157] S504: Determine the total anomaly score for each cluster based on the identification results.

[0158] S505: Determine the teaching matching degree based on the total anomaly score and core feature set of each cluster, and generate teaching suggestions based on the teaching matching degree.

[0159] In this embodiment, the expression for calculating the teaching matching degree is:

[0160]

[0161] In the formula, Let be the teaching matching degree of the j-th feature within the k-th cluster. The total score of all students in the k-th cluster is the abnormal score. Let be the abnormal score of the i-th student. Let be the global mean of the j-th feature.

[0162] This embodiment quantifies the severity of teaching problems corresponding to the core teaching features of each cluster by associating the core features of each cluster with the abnormal scores output by abnormal behavior identification. The higher the score, the more prominent the problem is for that group in that feature. At the same time, it combines the details of students' learning behavior (such as which course video was not completed or which assignment was unsatisfactory) to clarify the root cause of the problem.

[0163] After determining the severity of the teaching problems, a teaching suggestion matching formula is used to generate teaching suggestions with different priorities. The teaching suggestion matching formula is as follows:

[0164]

[0165] In the formula, Suggest kj This is a function for providing teaching suggestions.

[0166] This embodiment deeply integrates clustering results with course knowledge points, teaching objectives, and anomaly warnings. By anchoring core features and quantifying teaching problems, the clustering results can directly explain the "teaching problems of the student group" and output practical teaching suggestions. This solves the pain point of conventional clustering results being "visible but not usable" and truly serves personalized teaching management.

[0167] As a further optimization of this embodiment, the clustering also needs to be updated during the operation of the system. Traditional static clustering updates have the drawback of high computational burden. This embodiment adopts incremental clustering logic to adapt to the dynamic nature of students' learning behavior and reduce the system's computational consumption. That is to say, based on the initial clustering results, only the newly added learning behavior data (such as newly added video viewing and homework submission data on the same day) are subjected to local iterative calculations, without the need to re-cluster all student data, thereby reducing the amount of computation and improving the efficiency of clustering updates.

[0168] Specifically, incremental updates to the clustering results include: real-time monitoring of changes and anomalies in users' learning behavior data; determining update trigger values ​​based on changes and anomalies in learning behavior data; acquiring new learning behavior data when the update trigger value reaches a preset trigger value; and performing local iterative updates to the center of the cluster to which the user belongs based on anomaly-cluster association update rules to generate new clustering results.

[0169] In this embodiment, the update trigger condition is determined by: real-time monitoring of changes and abnormal changes in students' learning behavior data, and calculating the "abnormality level - feature change" update trigger value for each student. The calculation formula is as follows:

[0170]

[0171] In the formula, Update the trigger value for the "abnormality level - characteristic change" of the i-th student, w j Let the weight of the j-th feature be , Let j be the change in the j-th feature of the i-th student. The maximum change of the j-th feature is set to a preset trigger value of 0.4. When the value is greater than 0.4, it is determined that the student's learning status has changed significantly, triggering incremental clustering update; at the same time, a periodic triggering mechanism is set to automatically trigger full incremental clustering from 0:00 to 0:30 every day to ensure that the clustering results of all students can be updated in a timely manner.

[0172] For students who trigger the update, their newly added learning behavior data (such as video viewing, homework submission, quizzes, etc. on the same day) are collected, and the newly added data is added to the student's feature system using the processing method in step S100.

[0173] The anomaly-cluster association update rule in this embodiment adopts the anomaly-cluster association update formula, which only performs local iterative updates on the center of the cluster to which the student belongs, without having to re-cluster all student data. This ensures that the clustering characteristics of the anomaly group are more prominent, while reducing the amount of computation and lowering the system's computing power burden.

[0174] The anomaly-cluster association update formula is as follows:

[0175]

[0176] In the formula, The center of the updated k-th cluster, The center of the k-th cluster before the update. For new data The set of samples assigned to the k-th cluster.

[0177] Based on the updated cluster centers, the clustering operation is re-executed to determine the new group affiliation. The clustering results in the system database are updated synchronously to ensure that the clustering results are consistent with the students' real-time learning status, thus meeting the system's core requirements of "real-time inventory and dynamic management".

[0178] Example 2

[0179] Figure 2 This is a block diagram of an online inventory system provided in one embodiment of the present invention. Figure 2 As shown in the figure, this embodiment provides an online inventory system, which includes: a management module, a data acquisition module, an inventory module, an early warning module, and a display module.

[0180] The management module is used to manage users and their learning objects. In this embodiment, the management module is divided into a user management module and a course management module.

[0181] The user management module is responsible for managing all users in the system, including user registration, approval, login, password modification, role assignment, and access control. This module ensures that users with different roles can access the corresponding functions, prevents unauthorized operations, and guarantees system security.

[0182] The course management module is responsible for the full lifecycle management of online courses (i.e., learning objects), including course information entry, editing, deletion, and approval; uploading and management of course resources (videos, assignments, quizzes); and course categorization and retrieval. This module provides basic course data support for learning progress tracking.

[0183] The data collection module is used to collect students' learning behavior data in real time. Based on data tracking technology, it collects students' learning behavior data in real time, including video viewing time, fast-forward / pause operations, homework submission time, answer accuracy rate, and number of times they speak in the discussion forum. This module cleans and standardizes the collected data and stores it in the database, providing data support for learning progress calculation and intelligent analysis.

[0184] The inventory module is used to execute the online inventory method in Example 1. The data processing steps of the online inventory method have been described in detail in Example 1.

[0185] The early warning module is used to generate early warning information based on the identification results. When the inventory module identifies abnormal situations such as students lagging behind in learning, not submitting assignments, or failing tests during the online inventory process, it triggers the abnormal early warning mechanism and reminds students and teachers through system messages, SMS and other means. At the same time, it clusters abnormal information to facilitate teachers to carry out targeted supervision.

[0186] The visualization module is used to present early warning information and inventory results visually. It is responsible for the statistical analysis of system data, generating personalized learning and inventory reports. This module uses data visualization technology to display student learning progress, overall class learning status, and inventory results in chart form. It also supports report export and printing, providing data support for educational administrators.

[0187] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the online inventory method in Embodiment 1.

[0188] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the online inventory method in Embodiment 1.

[0189] This embodiment can collect student learning behavior data in real time, and through technologies such as user behavior anomaly identification, group clustering, feature anchoring and teaching matching, it can automatically complete the inventory and statistics of learning progress. At the same time, it can classify different users into different categories and generate teaching suggestions, which can significantly reduce the workload of teaching administrators, improve inventory efficiency and data accuracy, and provide precise guidance for teaching resource allocation, teaching strategy optimization and teaching quality assessment.

[0190] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0191] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0192] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An online inventory method, characterized by, The method includes: Based on data tracking, user behavior data is obtained, and multi-dimensional indicators and comprehensive learning progress are determined based on the behavior data. Based on multi-dimensional indicators and comprehensive learning progress, perform user behavior anomaly identification and generate identification results; Dynamic feature selection guided by learning behavior trends is performed on multi-dimensional indicators, comprehensive learning progress, and recognition results to generate a feature system; Clustering is performed on the feature system based on the K-means clustering algorithm to generate clustering results that represent user group categories; wherein, the K value of the K-means clustering algorithm is determined by a dynamic adjustment rule guided by teaching weights; Cluster feature anchoring and instruction matching are performed on the clustering results to generate core features that characterize user learning details and instructional suggestions that guide user learning. An inventory report is then generated based on the core features, instructional suggestions, and clustering results.

2. The online inventory method according to claim 1, characterized in that, The multi-dimensional indicators include: video completion rate, assignment pass rate, quiz accuracy rate, and interactive participation rate; the comprehensive learning progress is determined based on the multi-dimensional indicators, including: Calculate the weight of each indicator in the multi-dimensional indicators; Based on the entropy weight method and the proportion of each indicator, the fusion weight of each indicator in the multi-dimensional indicators is determined; The multi-dimensional indicators are weighted and merged according to the fusion weight to generate a comprehensive learning progress.

3. The online inventory method according to claim 2, characterized in that, User behavior anomaly identification is performed based on multi-dimensional indicators and comprehensive learning progress, including: Based on preset dynamic thresholds, abnormal behavior is judged for multi-dimensional indicators and comprehensive learning progress, abnormal types are generated, and the frequency of abnormal behavior is counted. The abnormal types include progress lag abnormality, homework abnormality, test abnormality and interaction abnormality. Abnormal behavior is quantified to generate anomaly scores, and the anomaly type, frequency, and anomaly score of the abnormal behavior are used as the identification results.

4. The online inventory method according to claim 3, characterized in that, Dynamic feature selection guided by learning behavior trends is performed on multi-dimensional indicators, comprehensive learning progress, and recognition results, including: A basic feature set is constructed based on multi-dimensional indicators and comprehensive learning progress. The basic feature set includes five basic features and two time-sensitive features. The five basic features are, in order, video completion rate, homework pass rate, test accuracy rate, interactive participation rate and comprehensive learning progress. The two time-sensitive features are the learning progress growth rate, which is used to quantify the dynamic changes in the user's learning progress, and the frequency of learning behaviors, which is used to characterize the user's learning coherence. Extract the abnormal scores from the recognition results, associate the abnormal scores with the basic feature set, and generate an associated feature set; Redundant features are removed from the associated feature set based on trend similarity analysis to generate a feature system.

5. The online inventory method according to claim 4, characterized in that, The trend similarity analysis method is as follows: Select any two features from the set of associated features, calculate the time-series trend slope of the two features, and obtain the time-series trend slope corresponding to each feature; Calculate the coefficient of variation of each feature in the associated feature set, which is used to characterize the feature dispersion, and extract the entropy value of each feature in the entropy weight method calculation process; The discriminative power of each feature is determined based on its coefficient of variation and entropy value. Calculate the trend consistency between any two features based on the time-series trend slope corresponding to each feature; When the trend consistency between any two features is greater than a preset threshold, the feature with the lowest discriminative power among the two features will be removed.

6. The online inventory method according to claim 2, characterized in that, The dynamic adjustment rule for the teaching weight orientation is as follows: Based on the elbow rule, the silhouette coefficients of each K value in the preset K value range of the K-means clustering algorithm are determined, and at least two candidate K values ​​are selected based on the silhouette coefficients. Construct a weighted approach to teaching intervention priorities, and determine the teaching suitability of each candidate K value based on the weighted approach to teaching intervention priorities. Based on the teaching fit and silhouette coefficient of each candidate K value, the teaching guidance evaluation value of each candidate K value is determined, and the candidate K value corresponding to the largest teaching guidance evaluation value is taken as the K value of the K-means clustering algorithm.

7. The online inventory method according to claim 2, characterized in that, Cluster feature anchoring and instructional matching are performed on the clustering results, including: Calculate the mean of each feature in the feature system of each cluster in the clustering results; Anchoring is performed on the mean of each feature based on the preset weights of course knowledge points, generating the anchoring score of each feature in the feature system of each cluster. Based on the anchoring score, each feature in the feature system of each cluster is filtered to generate a core feature set for each cluster, wherein the core feature set of each cluster contains at most two core features; The total anomaly score for each cluster is determined based on the identification results; Based on the total anomaly score and core feature set of each cluster, the teaching matching degree is determined, and teaching suggestions are generated based on the teaching matching degree.

8. The online inventory method according to claim 2, characterized in that, The method further includes: incrementally updating the clustering results, including: Real-time monitoring of changes and anomalies in users' learning behavior data; Determine the update trigger value based on changes and abnormal changes in learning behavior data; When the update trigger value reaches the preset trigger value, acquire the newly added learning behavior data; Based on the anomaly-cluster association update rule, the center of the user's cluster is locally iteratively updated to generate a new clustering result.

9. An online inventory system, characterized in that, The system includes: The management module is used to manage users and their learning objects; The data acquisition module is used to collect students' learning behavior data in real time. The inventory module is used to execute the online inventory method as described in any one of claims 1-8; The early warning module is used to generate early warning information based on the identification results; The display module is used to visualize early warning information and inventory results.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the online inventory method as described in any one of claims 1-8.

Images