Hybrid teaching mode-oriented online and offline course linkage recommendation method and system

Abstract

The invention discloses an online and offline course linkage recommendation method and system for a mixed teaching mode, relates to the technical field of course recommendation, and discloses the online and offline course linkage recommendation method and system for the mixed teaching mode. The method comprises the steps of generating a current track fingerprint representing teaching process characteristics for offline classroom knowledge points, calculating a reuse order and a track variation degree by matching historical fingerprints, and judging a high-order variation reuse scene; if yes, calculating a target cognition progressive level in the scene; based on the hierarchy and the trajectory variability, constructing a connection resource sequence composed of a plurality of progressive resources; calculating a dynamic cognitive absorption rate of the learner to determine an optimal pushing moment, and progressively pushing a resource sequence to the learner terminal; according to the method and the device, accurate identification of the differentiated offline teaching scene is realized, the personalized online preposed resource sequence matched with the differentiated offline teaching scene is dynamically generated, and the learning connection efficiency in a mixed teaching mode is improved.

Description

Technical Field

[0001] This invention relates to the field of course recommendation technology, and more specifically, to a method and system for recommending online and offline courses in a blended learning model. Background Technology

[0002] With the deep integration of information technology and education, blended learning has become the mainstream teaching paradigm. It aims to improve teaching flexibility and learning outcomes by organically integrating online self-study with offline classroom teaching. However, in practical application, this model faces a core challenge: a lack of deep, dynamic, and intelligent linkage between online resource recommendations and the actual offline teaching process. Existing recommendation technologies are mostly based on static knowledge point tags, learner historical scores, or single learning behavior data, and their recommendation logic is often independent of the specific teaching implementation process in the offline classroom. This leads to a serious disconnect between online preparatory resources and the dynamic details such as the teaching methods, explanation order, example selection, and level of abstraction used by the teacher in class. In particular, when the same knowledge point is reused in different teaching cycles, and the teacher's teaching trajectory, such as the explanation logic, case studies, and blackboard structure, changes significantly, the existing system cannot perceive these differences and still recommends content based on historically common patterns. This leaves learners facing the dilemma of insufficient cognitive preparation or resource mismatch, making it difficult for them to effectively adapt to classroom changes and weakening the collaborative advantages of online-offline support in blended learning. Therefore, there is an urgent need for an intelligent recommendation method that can perceive the dynamic characteristics of offline teaching in real time, quantify the differences in teaching trajectories, and dynamically generate personalized online resource sequences that are highly faithful to the classroom, so as to achieve a true closed loop and precise linkage between online and offline teaching links.

[0003] To address the above problems, this invention proposes a solution. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for recommending online and offline courses in a blended learning model, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The online and offline course recommendation method for blended learning includes the following steps: generating current trajectory fingerprints representing the characteristics of the teaching process for knowledge points in offline classrooms; performing matching retrieval in the offline classroom trajectory fingerprint history database that stores records of past teaching trajectory fingerprints; calculating the reuse order representing the cumulative number of occurrences of knowledge points and the trajectory variability representing the degree of difference between the current trajectory fingerprint and the most similar historical trajectory; and performing a high-order variation reuse scenario determination based on a preset reuse order threshold and trajectory variability threshold.

[0007] When it is determined to be a high-order variation reuse scenario, based on the reuse order and trajectory variation degree, the baseline cognitive level of the knowledge point and the single reuse standard level increment are called. The cumulative level contribution brought by the reuse order and the level compensation contribution brought by the trajectory variation are calculated and superimposed with the baseline cognitive level to obtain the target cognitive progression level representing the level of abstraction that the learner's cognition should reach.

[0008] Based on the progressive levels of target cognition and trajectory variability, the cognitive sensitivity and acceptance interval of the range of the level of preceding resources that the learner can effectively absorb is calculated, the connecting advanced level and the target level of online preceding resources are determined, the span of the connecting level is calculated and the number of steps is determined, and the connecting resource sequence is selected for each step to perform resource matching.

[0009] Calculate the learner's dynamic cognitive absorption rate of knowledge points and determine the push time for each step. At the push time, progressively push the sequence of connecting resources to the learner's terminal.

[0010] In a preferred embodiment, the current trajectory fingerprint is generated based on the extraction and encoding of four dimensions of features of the teaching process. The first dimension is the lecture handout page number sequence feature, which is generated by forming an ordered sequence of lecture handout page numbers related to knowledge points in the order of their appearance and encoding it to generate a lecture handout sequence feature vector. The second dimension is the example identifier sequence feature, which is generated by forming an ordered sequence of example identifiers related to knowledge points in the order of their explanation and encoding it to generate an example identifier sequence feature vector. The third dimension is the blackboard writing structure feature, which is generated by performing structured analysis on the blackboard writing content generated in the teaching process and encoding it to generate a blackboard writing structure feature vector. The fourth dimension is the duration distribution feature, which is generated by normalizing the duration of the teacher's explanation of each sub-topic of the knowledge point to generate a duration distribution feature vector. The feature vectors of the four dimensions are concatenated in a fixed order and normalized to obtain the current trajectory fingerprint vector.

[0011] In a preferred embodiment, the specific rule for determining a high-order variation reuse scenario based on a preset reuse order threshold and a trajectory variation threshold is as follows: if the reuse order is greater than or equal to the preset reuse order threshold and the trajectory variation is greater than or equal to the preset trajectory variation threshold, it is determined to be a high-order variation reuse scenario; wherein the reuse order is calculated by counting the total number of times the knowledge point appears in the current teaching week and all previous teaching weeks and adding 1; the trajectory variation is calculated by traversing the offline classroom trajectory fingerprint history database to calculate the cosine similarity between each historical trajectory fingerprint and the current trajectory fingerprint vector, selecting the maximum similarity as the maximum similarity, and subtracting the maximum similarity from 1 to obtain the trajectory variation.

[0012] In a preferred embodiment, the cumulative hierarchical contribution is calculated using a decay-accumulation method, and the calculation formula is as follows: ;in, The cumulative level contribution corresponding to knowledge point K, where j is the reuse sequence number accumulated from 2 to the current reuse level Stage(K). The increment of the standard level for single reuse is ξ, which is the preset marginal decay coefficient. The increment of the standard level for single reuse is determined based on the ratio of the average teaching time of the knowledge point to the standard class time of the curriculum.

[0013] In a preferred embodiment, the hierarchical compensation contribution is determined using a piecewise function: ;in, Contribution to tiered compensation For trajectory variability, The preset threshold for the initiation of moderate variation. The preset threshold for the onset of drastic mutation, For shape control parameters, For additional growth factor, It is the hyperbolic tangent function.

[0014] In a preferred embodiment, the logic for obtaining the cognitively sensitive acceptance interval width is as follows: using Calculations are performed, in which, To determine the width of the cognitively sensitive acceptance interval, and These are the preset lower and upper limits of the interval width, respectively; the connection of the advanced level is determined by the product of the cognitive sensitivity acceptance interval width and the dynamic modulation coefficient. The dynamic modulation coefficient is calculated using a periodic modulation formula based on the cosine function and is modulated according to the relative position of the target cognitive progression level within the cognitive cycle.

[0015] In a preferred embodiment, forming a bridging resource sequence includes: retrieving all completed resources related to the knowledge points from the learner's online learning records; extracting the cognitive progression level label for each resource; taking the maximum value as the learner's current cognitive progression level; defining the difference between the target level of the online prerequisite resource and the learner's current cognitive progression level as the bridging level span; determining the number of tiers based on the ratio of the bridging level span to the increment of the standard level for single reuse; allocating the bridging level span to each tier using a cumulative percentage coefficient in the form of a power function, satisfying the principle of increasing tier span; performing resource matching for the target cognitive progression level of each tier; calculating the cognitive bridging fit based on the target level deviation and bridging coherence; selecting the resource with the highest cognitive bridging fit as the bridging resource for each tier; and forming a bridging resource sequence.

[0016] In a preferred embodiment, the logic for obtaining the dynamic cognitive absorption rate is as follows: the learner's baseline cognitive absorption rate of the knowledge point is retrieved; the baseline cognitive absorption rate is amplified based on the trajectory variability according to a preset mutation activation coefficient to obtain the dynamic cognitive absorption rate; the total connection time window is the ratio of the connection level span to the dynamic cognitive absorption rate; the time proportion coefficient of each step push time is determined using an exponential decay function; and the push time of each step is allocated according to a preset interval decay principle based on the total connection time window and the number of steps.

[0017] In a preferred embodiment, the method further includes a step of performing an effectiveness check after the offline class ends. The effectiveness check includes two checks: the completion rate of the connecting resources and the accuracy rate of the offline quiz. If both of them reach a preset threshold, the linked recommendation is deemed effective; otherwise, the linked recommendation is deemed invalid and enters the offline optimization queue for subsequent offline updates of model parameters.

[0018] The online-offline course recommendation system for blended learning models includes the following modules: a scenario determination module, used to generate trajectory fingerprints of knowledge points, perform matching retrieval, calculate reuse order and trajectory variability, and determine high-order variation reuse scenarios based on corresponding thresholds; a target level calculation module, used to calculate the target cognitive progression level when a high-order variation reuse scenario is determined; a connecting resource sequence construction module, used to construct a connecting resource sequence based on the target cognitive progression level and trajectory variability; and a dynamic push tracking module, used to calculate the dynamic cognitive absorption rate, determine the push time for each level, execute progressive resource push, and track learners' resource learning completion status.

[0019] The technical effects and advantages of this invention's online-offline course linkage recommendation method and system for blended learning models are as follows: This invention generates teaching trajectory fingerprints and matches them with historical records to accurately identify high-order variation reuse scenarios; it dynamically calculates the target cognitive progression level based on the reuse order and trajectory variation degree, constructing personalized online resource connection goals for each learner; further, it combines a cognitive sensitivity acceptance interval and a tiered resource sequence mechanism to ensure that recommended resources smoothly connect with offline classrooms in terms of hierarchy and cognitive transition; it optimizes the push sequence based on the learner's dynamic cognitive absorption rate, achieving intelligent synchronization between resource push rhythm and learning progress; this method effectively solves the problems of disconnect between online and offline teaching, coarse resource recommendation, and insufficient personalization in blended learning models, improving learning connection efficiency and cognitive preparation effects, and supports continuous system optimization through a closed-loop verification mechanism. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the online and offline course recommendation method for a blended learning model according to the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of the online and offline course recommendation system for a blended learning model according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example

[0024] Please see Figure 1 As shown, this invention discloses a method for recommending online and offline courses in a blended learning model, comprising the following steps:

[0025] Step 1: Perform trajectory fingerprint matching and higher-order variation recognition on knowledge points in offline classes. The trajectory fingerprint is a feature vector formed after extracting and encoding features from multiple dimensions of the teaching process. The current trajectory fingerprint is matched with historical trajectory fingerprints in the offline class trajectory fingerprint history database to obtain the maximum similarity value. Based on the maximum similarity value, the trajectory variation degree, which represents the degree of difference between the current teaching trajectory and the most similar historical trajectory, is calculated. The cumulative occurrence frequency of knowledge points is counted to obtain the reuse order. Based on the reuse order and trajectory variation degree, it is determined whether it belongs to a higher-order variation reuse scenario.

[0026] This step involves the system performing trajectory fingerprint matching and higher-order mutation identification on each knowledge point involved in the upcoming or ongoing offline class, determining whether it belongs to a higher-order mutation reuse scenario, and outputting the reuse order and trajectory variability parameters required for subsequent steps.

[0027] It should be noted that this step is only performed if the learner has already established a basic grasp of knowledge point K. Before performing this step, the learner's current mastery level M(K) of knowledge point K is obtained from the learner's learning record database. The current mastery level M(K) is calculated as a weighted average of the learner's correct answer rate and resource completion rate related to knowledge point K in the most recent statistical window. The statistical window is the most recent D days, and in this embodiment, D=14. Only questions and resources covering knowledge point K are counted. If the current mastery level M(K) is lower than the preset mastery baseline threshold, it indicates that the learner has not yet established a basic understanding of knowledge point K. In this case, basic-level resources related to knowledge point K are directly retrieved from the online resource library for recommendation, and this step and subsequent steps are not performed. The initial value of the mastery baseline threshold is set to 0.6. The online resource library is a digital resource storage system maintained by the system, containing various course-related learning videos, documents, exercises, and other resources. Each resource is labeled with a corresponding knowledge point identifier and cognitive progression level tag.

[0028] Specifically:

[0029] First, an offline classroom trajectory fingerprint history database H(K) is established and maintained for each knowledge point K in the course. This offline classroom trajectory fingerprint history database H(K) stores the trajectory fingerprint records of knowledge point K in each offline teaching session. Each record contains four fields: the first field is the week of occurrence, which records the teaching week in which the teaching session is located; the second field is the trajectory fingerprint field, which stores the feature encoding vector of the teaching process; the third field is the abstraction level field, which records the cognitive abstraction level used in the teaching session; and the fourth field is the timestamp field, which records the creation time of the record.

[0030] It should be noted that the initial construction of the offline classroom trajectory fingerprint history database H(K) is completed before the start of each semester through offline analysis of historical teaching data. During the construction process, the system traverses all teaching records of the previous semester or historical semesters, extracts trajectory fingerprints for each teaching process involving knowledge point K, and stores them in the history database. During the semester, whenever a new offline teaching session for knowledge point K is completed, the system automatically adds the trajectory fingerprint and related information of this teaching session as a new record to the history database, realizing continuous incremental updates to the history database. Here, a knowledge point refers to the smallest identifiable unit of course teaching content, represented by the symbol K, and each knowledge point has a unique identifier in the system.

[0031] Furthermore, when the system detects that knowledge point K is being taught during an offline class, it initiates a real-time trajectory fingerprint generation process. The trajectory fingerprint is generated based on the extraction and encoding of four dimensions of features from the teaching process. The first dimension is the lecture note page number sequence feature, recording all lecture note page numbers related to knowledge point K in this lesson. These page numbers are arranged in an ordered sequence according to their chronological appearance during the teaching process. After normalization, this sequence is input into a pre-trained sequence encoder to generate a lecture note sequence feature vector of fixed length 64. The second dimension is the example identifier sequence feature, recording the unique identifiers of all examples related to knowledge point K in this lesson. These examples are arranged in an ordered sequence according to their chronological explanation. This sequence is then embedded and encoded. The code generates a fixed-length example sequence feature vector of 64 characters. The third dimension is the blackboard structure feature, which involves structurally analyzing the electronic blackboard content generated by the teacher during the lesson or the physical blackboard content acquired through image acquisition. This extracts the topological structure features of the blackboard, including but not limited to the number of elements, level depth, number of branches, and connections between elements, and encodes this into a fixed-length blackboard structure feature vector of 32 characters. The fourth dimension is the duration distribution feature, which records the teacher's explanation time on each sub-topic of knowledge point K, forming a duration distribution vector. After normalization, a fixed-length duration distribution feature vector of 32 characters is generated. The feature vectors from the above four dimensions are then concatenated in a fixed order to form the current trajectory fingerprint vector, denoted as... The four-dimensional vectors are preferably concatenated in the following order: lecture note sequence vector, example problem sequence vector, blackboard structure vector, and duration distribution vector. After concatenation, L2 normalization is performed again to obtain the current trajectory fingerprint vector. ;

[0032] Subsequently, trajectory fingerprint matching retrieval is performed in the offline classroom trajectory fingerprint history database H(K). Specifically, for each trajectory fingerprint record stored in the offline classroom trajectory fingerprint history database, the historical trajectory fingerprint is calculated. With the current trajectory fingerprint vector The cosine similarity between them; the cosine similarity is calculated as follows: The similarity value ranges from -1 to a closed interval of positive 1. In the application scenario of this invention, since the feature values ​​of each dimension of the trajectory fingerprint are all non-negative, the actual similarity value ranges from 0 to 1.

[0033] After traversing all records in the offline classroom trajectory fingerprint history database, the historical fingerprint with the highest similarity value is selected as the most similar historical trajectory fingerprint, denoted as . Record the maximum similarity value, denoted as . Simultaneously, perform statistical calculations on the reuse order, counting the total number of times knowledge point K has appeared in offline classes in the current teaching week and all previous teaching weeks. Increment this count by 1 to determine the reuse order of this occurrence. Its mathematical expression is: ;in r represents the counting function, and r represents the record in the offline classroom trajectory fingerprint history database. This indicates the teaching week corresponding to this record. Indicates the current teaching week;

[0034] Furthermore, subtract the maximum similarity value from the value of 1. The obtained value is defined as the trajectory variability V, which represents the degree of difference between the current trajectory fingerprint and the most similar historical trajectory fingerprint. Its value range is [0,1]. The larger the trajectory variability V value, the greater the degree of difference between the current teaching trajectory and the most similar historical trajectory; V=0 indicates complete consistency, and V close to 1 indicates a great difference.

[0035] The determination of high-order variation reuse scenarios is performed based on the reuse order Stage(K) and trajectory variability V(K) corresponding to the current teaching week. This determination adopts a dual-condition joint determination rule and uses a preset reuse order threshold. The reuse order threshold is used to define whether a knowledge point has entered a reuse state. The initial value is set to 2, indicating that a knowledge point must appear at least twice in an offline classroom to constitute a reuse scenario. A trajectory variability threshold is also preset. The preset trajectory variability threshold is used to define whether the variation of the teaching trajectory has reached a significant level, preferably 0.27 to 0.35, which can be determined based on the statistics of historical teaching data. In a preferred embodiment, it is set to 0.27.

[0036] Then the reuse level Stage and trajectory variability V are respectively compared with the preset reuse level threshold. and preset trajectory variability threshold The comparison and judgment are performed, and the execution logic of the judgment rule is as follows: If and If so, the output path marker Path(K)=C, indicating higher-order mutation reuse;

[0037] like and If so, the output path label Path(K)=B, indicating low-order mutation reuse;

[0038] Otherwise, output the path marker Path(K)=A, indicating non-reuse;

[0039] Where Path(K) is the path marker, which can take one of three values: A, B, or C;

[0040] When the path marker Path(K)=A, it indicates that knowledge point K is appearing for the first time in the offline classroom and has not yet formed a reuse scenario. Since the learner has a basic grasp of the knowledge point, the system retrieves pre-study and in-depth resources related to knowledge point K and connected with the content of this offline teaching from the online resource library. These resources include knowledge point application extensions and related knowledge point connections. The system pushes the list of pre-study and in-depth resources to the learner's terminal before the offline class begins, helping the learner to establish a connection between their existing online learning outcomes and the upcoming offline teaching.

[0041] When the path marker Path(K)=B, it indicates that knowledge point K is in a low-order variation reuse scenario but the degree of variation in the teaching trajectory has not reached a significant level. The offline teaching method is basically consistent with the historical teaching mode. Since the learners have already mastered the basics and the teaching mode is stable, the system retrieves resources related to knowledge point K and with the nature of review and consolidation from the online resource library, including knowledge point review, typical example reinforcement, and common mistake sorting, and pushes the list of reuse and consolidation resources to the learners' terminals before the offline class begins.

[0042] When Path(K)=C, it means that knowledge point K constitutes a high-order variation reuse scenario offline. The offline teaching method is significantly different from the historical model. Although learners have already mastered the basics, they need targeted cognitive preparation support to adapt to the teaching changes. The system enters the subsequent steps of this invention.

[0043] Step S2: When it is determined to be a high-order variation reuse scenario, based on the reuse order and trajectory variation degree, call the baseline cognitive level of the knowledge point and the single reuse standard level increment, calculate the cumulative level contribution brought by the reuse order and the level compensation contribution brought by the trajectory variation, and superimpose them with the baseline cognitive level to obtain the target cognitive progression level that represents the level of abstraction that the learner's cognition should reach.

[0044] This step, based on the reuse order Stage and trajectory variability V output in step S1, calculates and locks the target cognitive progression level corresponding to this offline high-order variation reuse scenario. This target cognitive progression level represents the level of abstract understanding that the learner should achieve regarding knowledge point K at the beginning of this offline class; specifically:

[0045] The system retrieves the specific cognitive progression level parameters for knowledge point K from the knowledge point parameter library. This library is constructed during system initialization through offline analysis of the course knowledge structure and historical teaching data. It stores two core parameters for each knowledge point; the first core parameter is the baseline cognitive level. This indicates that the baseline cognitive level reflects the initial cognitive progression level when knowledge point K first appears. Its value is determined based on the structural hierarchy of knowledge point K in the course knowledge graph; the higher the structural hierarchy, the larger the baseline cognitive level value. In a preferred embodiment, the rule for determining the baseline cognitive level is as follows: if knowledge point K is located at the first level of the knowledge graph, i.e., the basic concept level, the baseline cognitive level value is 1.0; if it is located at the second level, i.e., the basic application level, the value is 1.5; if it is located at the third level, i.e., the comprehensive application level, the value is 2.0; if it is located at the fourth level or above, i.e., the higher-order transfer level, the value is 2.5. The second core parameter is the single reuse standard level increment, which... This parameter indicates that it reflects the standard increase in cognitive progression level of knowledge point K with each offline high-level reuse, and its value is based on the average teaching time of knowledge point K. Class hours in accordance with curriculum standards The ratio is determined, and its corresponding expression is: ;in, This is the upper limit for hierarchical increments, initially set at 1.5, representing the standard course duration. According to the teaching syllabus, the initial value is set at 45 minutes, or 0.75 hours.

[0046] The cumulative hierarchical contribution resulting from the reuse order is then calculated. This calculation is based on the following teaching and cognitive principle: Each time a knowledge point undergoes a higher-level reuse, the learner's cognitive understanding level should correspondingly increase. However, the rate of increase diminishes marginally with increasing reuse order. This characteristic reflects the objective law that cognitive gains gradually slow down after multiple reuses. The system presets a marginal decay coefficient ξ, which controls the rate of marginal decay, with an initial value set at 0.35. The cumulative hierarchical contribution is calculated using a decay-accumulation method. The calculation formula is expressed as follows:

[0047] ; where j is the reuse sequence number, starting from 2 and counting up to the current reuse order Stage(K); the cumulative result of this formula reflects the cumulative hierarchical contribution of all orders from the 2nd reuse to the current reuse;

[0048] Furthermore, the hierarchical compensation contribution resulting from trajectory variation is calculated. When the degree of trajectory variation in this reuse is high, it indicates that the offline teacher's teaching methods have changed significantly compared to the historical model, possibly adopting new example types, new explanation structures, or new abstract expression methods. Learners need additional cognitive preparation to adapt to this teaching change, corresponding to additional hierarchical compensation. The calculation of the hierarchical compensation contribution adopts a piecewise function form, based on the trajectory variation. Different calculation rules are used for different intervals to achieve differentiated responses to different degrees of variation; therefore, the piecewise function form of the hierarchical compensation contribution is: ;

[0049] in To preset a moderate starting threshold for variation, the initial value for the required degree of variation to be compensated is defined as 0.27. The threshold for the start of drastic variation is preset, which defines the critical point of variability before entering the drastic variation range. The initial value is set to 0.55, and tanh is the hyperbolic tangent function. The initial value of the shape control parameter for the preset medium variation range is set to 1.8. This parameter controls the shape of the growth curve of the compensation function within the medium variation range. The initial value is set to 0.5 as an additional growth coefficient for the pre-defined drastic variation range. This parameter controls the additional growth rate of the compensation amount within the drastic variation range.

[0050] Based on the piecewise function above, we can obtain that when the trajectory variability V(K) is in the medium variability range, the preset medium variability threshold is reached. However, it did not reach the preset threshold for the initiation of drastic mutation. When the hyperbolic tangent function is used to calculate the hierarchical compensation contribution, the characteristics of the hyperbolic tangent function make the compensation amount grow faster at the beginning of the interval, and gradually slow down and tend to saturate as the variability increases, which is consistent with the marginal diminishing law of cognitive compensation.

[0051] When the trajectory variability V(K) is in the drastic variation range, that is, when it reaches or exceeds the preset drastic variation start threshold. At that time, the system adds a linear growth term to the end-point compensation value in the medium variation range to provide a more aggressive compensation response to drastic variation.

[0052] Finally, the baseline cognitive level will be determined. Cumulative hierarchical contributions Contribution to Hierarchical Compensation The three components are combined to obtain the progressive levels of target cognition for this offline reuse. The calculation method for this level value is the arithmetic sum of three components; Target cognition progressive levels This represents the level of abstraction that learners should achieve in their understanding of knowledge point K when entering this offline class. This level takes into account the basic difficulty of the knowledge point itself, the cumulative improvement brought about by repeated use, and the additional preparation requirements brought about by the variation in the teaching trajectory this time.

[0053] Step S3: Based on the progressive levels of target cognition and trajectory variability, calculate the cognitive sensitivity and acceptance interval width of the range of levels of preceding resources that the learner can effectively absorb, determine the connecting advanced level and the target level of online preceding resources, calculate the span of the connecting level and determine the number of steps, and select connecting resource sequences for each step to perform resource matching.

[0054] This step is based on the offline goal cognition progression level output in step S2. Based on the trajectory variability V(K), the target level of the preceding resources on the line is calculated, the advanced connection level is determined, and a connection resource sequence consisting of multiple progressive connection resources is constructed accordingly. Specifically:

[0055] First, the width of the cognitive sensitivity acceptance interval is calculated. In a blended learning model, learners have a cognitive sensitivity acceptance interval for pre-learning resources related to higher-order reusable knowledge points. Pre-learning resources received within this interval produce the best cognitive preparation effect; resources at too low a level result in insufficient preparation, while resources at too high a level result in cognitive leaps. The width of the cognitive sensitivity acceptance interval is positively correlated with trajectory variability; that is, the higher the trajectory variability, the greater the difference between the current offline teaching and the historical model, and the greater the cognitive adaptation challenge faced by learners. Therefore, a wider cognitive buffer zone is needed to accommodate different levels of preparatory resources. A lower limit for the interval width is preset. The initial value is set to 0.3 level units to ensure a basic buffer even under low variation conditions; and the upper limit of the interval width is preset. The initial value is set to 1.5 hierarchical units to prevent excessively wide intervals from causing resource delivery to be too scattered. This corresponds to the width of the cognitively sensitive acceptance interval. Cognitive sensitivity acceptance range width This is used to characterize the buffer range of pre-resource levels that learners can effectively absorb under blended learning conditions; to ensure that this range monotonically increases with the trajectory variability V(K) and does not exceed the implementable upper bound, the cognitive sensitivity acceptance interval width is defined. The calculation uses a bounded saturated scaling function, which ensures that the minimum buffer width is maintained when the trajectory variability V(K) is low, and the buffer width gradually expands and approaches the upper limit as the trajectory variability V(K) increases.

[0056] Subsequently, dynamic modulation coefficients are calculated. The connection between advanced levels is not a fixed value, but rather dynamically modulated based on the width of the cognitive sensitivity acceptance interval and the periodic position of the offline target level. The design principle of dynamic modulation is based on the periodic characteristics of cognitive learning: learners' cognitive understanding of a certain knowledge point exhibits a periodic deepening pattern. The sensitivity to prior resources varies at different stages of each cognitive cycle. The cognitive progression process is divided into several cognitive cycles, each corresponding to a complete stage of cognitive deepening. The cycle length is... This parameter reflects the number of levels a learner needs to traverse from initial understanding to deep mastery of a knowledge point, with an initial value set at 3.0 levels. It calculates the relative position of the offline target level within the current cognitive cycle using a progressive cognitive level approach. Regarding the period length Modulo operation yields the position within the period. Then, based on the obtained position within the period... The modulation coefficient γ(K) is calculated using a periodic modulation formula based on the cosine function. This formula ensures that the modulation coefficient takes a smaller value at the beginning of the period, corresponding to a conservative lead strategy; takes a maximum value at the midpoint of the period, corresponding to an aggressive lead strategy; and takes a smaller value again at the end of the period to avoid over-leading. The calculation formula is expressed as follows:

[0057] ;in, The lower limit of the preset modulation coefficient is initially set to 0.25. The initial value is set to 0.65 to preset the upper limit of the modulation step coefficient.

[0058] Further calculations connect the advanced levels It is the product of the cognitive sensitivity acceptance interval width W(K) and the modulation coefficient γ(K), representing the optimal advance of online resources relative to the offline target level; and it also represents the progressive level of target cognition. Connecting with advanced levels Adding them together, we get the target level of online pre-requisite resources. ;

[0059] Furthermore, the span of the connection levels and the number of steps are determined. All completed resources related to knowledge point K are retrieved from the learner's online learning records. The cognitive progression level label of each resource is extracted, and the maximum value is taken as the learner's current cognitive progression level. ;

[0060] And the target level of online front-end resources With respect to the learner's current cognitive progression level The difference is defined as the span of the connection level, Span(K);

[0061] If the span of the connection level Span(K) is less than or equal to 0, it means that the learner's current cognitive progression level has reached or exceeded the online target level. There is no need to push connection resources, and the termination mark is output without further analysis.

[0062] If the connection level span Span(K) is greater than 0, it indicates that there is a cognitive level gap that needs to be filled, and the connection resource sequence construction process continues. The connection resource sequence construction process specifically involves: calculating the connection level span Span(K) relative to the single reuse standard level increment. , and record as ;

[0063] According to this ratio Determine the number of ladder levels This determines the number of connection resources to be pushed, and the tier number. The determination adopts a segmented rule, that is, each standard level increment corresponds to approximately one step of connecting resources. When the span exceeds one standard increment, the number of steps is increased to ensure the smoothness of cognitive transition. For example: if the ratio In the interval If the step number is 1, and the ratio is in the interval... If the number of steps is 2, and the ratio is in the interval... The number of steps is 3 if the ratio is greater than 3, then the number of steps is 4.

[0064] Based on the determined number of ladder levels The span (Span(K)) of the transition level is allocated to each step to determine the target cognitive progression level for each step. The allocation rule follows the principle of increasing step span, that is, smaller transition level spans are allocated to early steps to achieve a smooth cognitive transition, and larger transition level spans are allocated to later steps to accelerate the approach to the target. For the i-th step, its target cognitive progression level is... The calculation method is based on the learner's current cognitive progression level. Add the span of the connecting levels, Span(K), multiplied by the corresponding cumulative percentage coefficient. ; where the cumulative percentage coefficient The calculation uses a power function form, and its formula is expressed as follows: Where i is the ladder index, ranging from 1 to... ρ is a distribution morphology parameter that controls the distribution pattern of the span of the connecting levels between each step. The value of the distribution morphology parameter ρ should be less than 1 to meet the requirement of the principle of increasing step span. When ρ is less than 1, the power function curve exhibits a concave characteristic, which makes the cumulative proportion of the early steps grow faster and the single step span smaller, while the cumulative proportion of the later steps grows slower and the single step span larger. The initial value of the distribution morphology parameter ρ is set to 0.7.

[0065] Finally, resource matching and selection are performed for each progressive level of target cognition, that is, all resources covering knowledge point K are selected from the online resource library to form a candidate resource set. For candidate resource set Each candidate resource Calculate its cognitive connectivity fit with the i-th step. Where j represents the candidate resource index, with values ​​ranging from 1 to the total number of candidate resource sets. Specifically, the calculation comprehensively considers two factors: the first factor is the target level deviation. Reflecting candidate resources Cognitive progression levels The degree of deviation from the target cognitive progression level of the i-th step, i.e., the candidate resources The absolute value of the difference between the cognitive progression level and the target cognitive progression level of the i-th step is divided by the increment of the standard level used in a single reuse; the second factor is coherence. Reflecting candidate resources The smoothness of transition with preceding cognitive states is calculated in two ways based on the ladder sequence number:

[0066] When i=1, which is the first step, the prior cognitive state is the learner's current cognitive progression level, and the level span of the current step is the first step's target cognitive progression level. With respect to the learner's current cognitive progression level The difference; then the coherence. for: ;

[0067] When i>1, it means the next step is being taken, and the preceding cognitive state is the previous step. Progressive levels of goal cognition The current ladder's level span is the i-th level of goal cognition progression. With the Step-by-step goal cognition progression levels The difference then becomes more coherent. for: ;

[0068] The formula for calculating cognitive congruence fit is expressed as follows: ;in The coherence weighting coefficient controls the relative importance of coherence in the fit calculation. In this embodiment, λ is set to 0.6. For each step i, the candidate resource with the highest cognitive coherence fit is selected as the coherence resource for that step. If multiple candidate resources have the same highest cognitive coherence fit, they are further sorted in descending order of their historical completion rates and the one with the highest completion rate is selected.

[0069] Step S4: Calculate the learner's dynamic cognitive absorption rate of knowledge points and determine the push time for each step. Push the sequence of connecting resources to the learner's terminal at the push time.

[0070] The learner's dynamic cognitive absorption rate of knowledge point K is calculated. This rate reflects the incremental increase in cognitive levels that the learner can digest per unit of time and is a key parameter for calculating the optimal intervention sequence. The baseline cognitive absorption rate of knowledge point K is retrieved from the knowledge point parameter database. This parameter is determined through statistical analysis of learners' historical online learning behavior data. Specifically, the analysis method involves extracting learning data from all completed resources related to knowledge point K from the learner's online learning records. For each completed resource, the difference between the cognitive progression level label of that resource and the learner's cognitive progression level of knowledge point K before learning that resource is calculated as the level improvement brought by that resource. The learning time spent by the learner to complete the resource is calculated, and the level improvement is divided by the learning time to obtain the level improvement rate per unit time for that resource. The arithmetic mean of the level improvement rates per unit time for all completed resources is taken to obtain the learner's baseline cognitive absorption rate of knowledge point K. The unit is per level per hour; if the learner has no historical learning data related to knowledge point K, the average absorption rate of the group for that knowledge point is used as the learner's baseline cognitive absorption rate.

[0071] The calculation of dynamic cognitive absorption rate is based on the following method: A higher trajectory variability indicates a greater difference between the current teaching and historical patterns, a greater cognitive challenge faced by learners, and correspondingly, a more fully activated cognitive system, leading to improved learning focus and absorption efficiency. A preset variability activation coefficient ψ is used, which controls the amplification effect of trajectory variability V(K) on the cognitive absorption rate, with an initial value set at 1.5. Dynamic cognitive absorption rate... The calculation formula is expressed as follows: ;

[0072] Then, the total connection time window is calculated in reverse. It is defined as the progressive cognitive level of learners from their first contact with the first bridging resource to their completion of all bridging resources and achievement of online learning goals. Total time required, in hours; total connection time window Defined as hierarchical span With cognitive absorption rate The ratio result; this ratio result represents the minimum time investment required to complete a given cognitive preparation;

[0073] Based on the total connection time window with ladder series Total connection time window The timing of resource pushes allocated to each tier follows a decreasing push interval principle. This means that the time interval between earlier and later tiers is larger, gradually decreasing until the final tier is pushed close to the start of the offline class. This principle is based on the following: earlier tiers help learners build a cognitive foundation, requiring sufficient time for digestion and understanding to ensure adequate absorption of knowledge; later tiers primarily serve as a final push, and should be pushed close to the start of the offline class to maintain cognitive activity and freshness of memory.

[0074] For the i-th step, its push time The calculation method is based on the start time of offline classes. Subtract the total connection time window Multiply by the corresponding time percentage coefficient Time percentage coefficient The calculation uses an exponential decay function, and its formula is expressed as follows:

[0075] Where m is an auxiliary variable used for summation and normalization, and i is the step number; η is the decay rate parameter, which controls the rate at which the push time decreases from the early steps to the later steps; when the decay rate parameter η is large, the time proportion coefficient of the later steps decreases faster, and the push time is closer to the start time of the offline class; when the decay rate parameter η is small, the push time of each step is more evenly distributed; the value of the decay rate parameter η must ensure that there is sufficient learning response time between the push time of the last step and the start time of the offline class; in a preferred embodiment, the decay rate parameter η is set to 1.6.

[0076] The sequential push of connecting resources is executed according to the calculated push times for each tier. Upon arrival, the following sequence of operations is performed: The system constructs a structured push message. The message content includes a push batch identifier in the format of the ratio of the current step number to the total number of steps; target offline classroom information including course name, teaching week, class date, class time, class location, knowledge point identifiers and names involved, current reused step; detailed information on the resources connecting to this step, including resource name, resource identifier, target cognitive progression level, estimated learning time of the resource, resource access link; learning suggestions including suggested completion time nodes and subsequent step announcements; and connection progress information including the number of steps completed, the number of steps remaining, and the remaining time until the start of the offline class. The system sends the constructed push message to the learner's terminal through a message push channel, prioritizing channels with strong real-time capabilities to ensure that learners receive the push in a timely manner. The system records complete log information for this push, including push timestamp, target learner identifier, push resource identifier, push channel type, and push result status, for subsequent effect tracking and anomaly analysis.

[0077] The system continuously tracks learners' completion status of resources at each level. Through a data interface with the online learning platform, the system obtains learners' resource learning progress information in real time and determines whether each level of resources has been completed based on preset completion criteria. These criteria include: the resource playback progress reaching a preset completion percentage of the total resource duration (e.g., 80% or more), or the accuracy rate of answers to embedded quizzes reaching a preset accuracy threshold (e.g., 70% or more). For cases of incomplete completion, the system employs the following strategy: if the system detects that the learner is not completing the next level of resources at the designated time... If a learner arrives at the next level of learning without completing the current level of learning, the system will include a reminder message when pushing the next level of learning, prompting the learner to prioritize completing the unfinished level of learning. If the system detects that a learner still has unfinished level of learning when the offline class starts, the system will record the number of the unfinished level and the resource information, and will push supplementary learning suggestions to the learner after the offline class ends. The suggestions will include access links to the unfinished resources and suggested supplementary learning time slots.

[0078] Preferably, this embodiment also includes a step of verifying the effectiveness of the collaborative recommendation after the offline class; the effectiveness test after the offline class includes only two items: the completion rate of the connecting resources and the accuracy rate of the offline quiz. If both items reach the preset threshold, the collaborative recommendation is deemed effective; otherwise, it is deemed invalid and enters the offline optimization queue; the effectiveness test in this step is not a prerequisite for the execution of the collaborative recommendation, but is only used for subsequent offline updates of model parameters.

[0079] Please see Figure 2 As shown, this invention discloses an online-offline course recommendation system for blended learning models, comprising the following modules:

[0080] The scenario determination module is used to generate trajectory fingerprints for knowledge points, perform matching retrieval, calculate reuse order and trajectory variability, and determine high-order variability reuse scenarios based on thresholds; the target level calculation module is used to calculate the target cognitive progression level when the scenario is determined to be a high-order variability reuse scenario; the connecting resource sequence construction module is used to construct connecting resource sequences based on the target cognitive progression level and trajectory variability; and the dynamic push tracking module is used to calculate the dynamic cognitive absorption rate, determine the push time for each level, execute progressive resource push, and track the learner's resource learning completion status.

[0081] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0082] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0083] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0084] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0085] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0086] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for recommending online and offline courses in a blended learning model, characterized in that: Includes the following steps: For knowledge points in offline classrooms, generate current trajectory fingerprints that represent the characteristics of the teaching process. Perform matching retrieval in the offline classroom trajectory fingerprint history database that stores records of past teaching trajectory fingerprints. Calculate the reuse order that represents the cumulative number of times knowledge points appear and the trajectory variability that represents the degree of difference between the current trajectory fingerprint and the most similar historical trajectory. Based on the preset reuse order threshold and trajectory variability threshold, perform a dual-condition joint high-order variation reuse scenario determination. When it is determined to be a high-order variation reuse scenario, based on the reuse order and trajectory variation degree, the baseline cognitive level of the knowledge point and the single reuse standard level increment are called. The cumulative level contribution brought by the reuse order and the level compensation contribution brought by the trajectory variation are calculated and superimposed with the baseline cognitive level to obtain the target cognitive progression level representing the level of abstraction that the learner's cognition should reach. Based on the progressive levels of target cognition and trajectory variability, the cognitive sensitivity and acceptance interval of the range of the level of preceding resources that the learner can effectively absorb is calculated, the connecting advanced level and the target level of online preceding resources are determined, the span of the connecting level is calculated and the number of steps is determined, and the connecting resource sequence is selected for each step to perform resource matching. Calculate the learner's dynamic cognitive absorption rate of knowledge points and determine the push time for each level. At the push time, progressively push the sequence of connecting resources to the learner's terminal.

2. The online and offline course recommendation method for blended learning models according to claim 1, characterized in that, The current trajectory fingerprint is generated based on the extraction and encoding of four dimensions of features in the teaching process. The first dimension is the lecture page number sequence feature, which forms an ordered sequence of lecture page numbers related to knowledge points in the order of their appearance and is encoded to generate a lecture page sequence feature vector. The second dimension is the example identifier sequence feature, which forms an ordered sequence of example identifiers related to knowledge points in the order of their explanation and is encoded to generate an example identifier sequence feature vector. The third dimension is the blackboard writing structure feature, which performs structured analysis on the blackboard writing content generated in the teaching process and encodes it to generate a blackboard writing structure feature vector. The fourth dimension is the duration distribution feature, which normalizes the duration of the teacher's explanation of each sub-topic of the knowledge point to generate a duration distribution feature vector. The feature vectors of the four dimensions are concatenated in a fixed order and normalized to obtain the current trajectory fingerprint vector.

3. The method for recommending online and offline courses in a blended learning model according to claim 1, characterized in that, The specific rules for determining high-order variation reuse scenarios based on the preset reuse order threshold and trajectory variation threshold are as follows: if the reuse order is greater than or equal to the preset reuse order threshold and the trajectory variation is greater than or equal to the preset trajectory variation threshold, it is determined to be a high-order variation reuse scenario. The reuse order is calculated by counting the total number of times the knowledge point appears in the current teaching week and all previous teaching weeks and adding 1. The trajectory variation is calculated by traversing the offline classroom trajectory fingerprint history database, calculating the cosine similarity between each historical trajectory fingerprint and the current trajectory fingerprint vector, selecting the maximum similarity as the maximum similarity, and subtracting the maximum similarity from 1 to obtain the trajectory variation.

4. The online and offline course recommendation method for blended learning models according to claim 1, characterized in that, The cumulative contribution at each level is calculated using a decaying accumulation method, and the calculation formula is as follows: ;in, The cumulative level contribution corresponding to knowledge point K, where j is the reuse sequence number accumulated from 2 to the current reuse level Stage(K). The increment of the standard level for single reuse is ξ, which is the preset marginal decay coefficient. The increment of the standard level for single reuse is determined based on the ratio of the average teaching time of the knowledge point to the standard class time of the curriculum.

5. The online and offline course recommendation method for blended learning models according to claim 4, characterized in that, The contribution of hierarchical compensation is determined using a piecewise function: ;in, Contribution to tiered compensation For trajectory variability, The preset threshold for the initiation of moderate variation, The preset threshold for the onset of drastic mutation, For shape control parameters, For additional growth factor, It is the hyperbolic tangent function.

6. The method for recommending online and offline courses in a blended learning model according to claim 1, characterized in that, The logic for obtaining the cognitive sensitivity acceptance interval width is as follows: using Calculations are performed, in which, To determine the width of the cognitively sensitive acceptance interval, and These are the preset lower and upper limits of the interval width, respectively; the connection of the advanced level is determined by the product of the cognitive sensitivity acceptance interval width and the dynamic modulation coefficient. The dynamic modulation coefficient is calculated using a periodic modulation formula based on the cosine function and is modulated according to the relative position of the target cognitive progression level within the cognitive cycle.

7. The method for recommending online and offline courses in a blended learning model according to claim 6, characterized in that, The formation of the connecting resource sequence includes: retrieving all completed resources related to the knowledge points from the learner's online learning records; extracting the cognitive progression level label for each resource; taking the maximum value as the learner's current cognitive progression level; defining the difference between the target level of the online prerequisite resource and the learner's current cognitive progression level as the connecting level span; determining the number of tiers based on the ratio of the connecting level span to the increment of the standard level for single reuse; allocating the connecting level span to each tier using a cumulative percentage coefficient in the form of a power function, satisfying the principle of increasing tier span; performing resource matching for the target cognitive progression level of each tier; calculating the cognitive connecting fit based on the target level deviation and connecting coherence; selecting the resource with the highest cognitive connecting fit as the connecting resource for each tier; and forming the connecting resource sequence.

8. The method for recommending online and offline courses in a blended learning model according to claim 7, characterized in that, The logic for obtaining the dynamic cognitive absorption rate is as follows: call the learner's baseline cognitive absorption rate of the knowledge point; amplify the baseline cognitive absorption rate based on the trajectory variability according to the preset variation activation coefficient to obtain the dynamic cognitive absorption rate; the total connection time window is the ratio of the connection level span to the dynamic cognitive absorption rate. The time proportion coefficient of each tier's push time is determined using an exponential decay function. Based on the total connection time window and the number of tiers, the push time of each tier is allocated according to the preset interval decay principle.

9. The method for recommending online and offline courses in a blended learning model according to claim 1, characterized in that, It also includes a step of validating the effectiveness after the offline classes have ended. The validity check includes two checks: the completion rate of connecting resources and the accuracy rate of offline in-class quizzes. If both of them reach the preset threshold, the linked recommendations are deemed to be effective; otherwise, the linked recommendations are deemed to be invalid and enter the offline optimization queue for subsequent offline updates of model parameters.

10. An online-offline course recommendation system for a blended learning model, used to implement the online-offline course recommendation method for a blended learning model as described in any one of claims 1-9, characterized in that, It includes the following modules: a scenario determination module, which is used to generate trajectory fingerprints of knowledge points, match and retrieve them, calculate the reuse order and trajectory variability, and determine high-order variability reuse scenarios based on the corresponding thresholds; The target level calculation module is used to calculate the target cognitive progression level when the scenario is determined to be a high-order mutation reuse scenario. The resource sequence construction module is used to construct a resource sequence based on the progressive levels of target cognition and trajectory variability; the dynamic push tracking module is used to calculate the dynamic cognitive absorption rate, determine the push time of each step, execute progressive resource push, and track the learner's resource learning completion status.

Images