A teaching behavior data modeling method based on federated learning

Abstract

The application relates to the technical field of educational big data processing, and discloses a teaching behavior data modeling method based on federated learning, which comprises the following steps: collecting multi-terminal learning behavior sequences and extracting active features, identifying repeated key nodes to construct a core feature set, summarizing various terminal data to generate a multi-terminal aggregation table, analyzing node correlations among students to form a cross-sequence correlation table, integrating the table through time matching and superposition, and finally summarizing to generate a complete teaching behavior modeling result set. The application constructs multi-source behavior sequences and introduces time sequence features for description, forms a stable structure through behavior compression and node screening, integrates multi-terminal data to construct a unified set to comb and combine relationships, forms a path structure through cross-sequence matching, and converts fragmented information into a continuous system. The innovation lies in systematizing node relationships, and the advantage lies in strengthening sequence dependence and combination expression, so that the behavior level performance is improved, and the adaptability of modeling to complex scenes is enhanced.

Description

Technical Field

[0001] This invention relates to the field of educational big data processing technology, specifically to a method for modeling teaching behavior data based on federated learning. Background Technology

[0002] The field of educational big data processing technology encompasses the technical aspects of collecting, storing, analyzing, and modeling the vast amounts of data generated during the educational process. The core of this technology involves the structured organization and feature extraction of student learning behavior data, teacher teaching behavior data, curriculum resource data, and educational evaluation data. It utilizes a data management platform to achieve unified management and indexing of educational data, preprocessing, classifying, and formatting the data, while establishing data relationship models for subsequent analysis. Overall, this technology covers data acquisition and cleaning, data representation and encoding, behavioral feature extraction, educational data modeling, and analysis methods, aiming to form a systematic educational information management and decision support system.

[0003] One of the methods for modeling teaching behavior data based on federated learning refers to a method for jointly modeling teaching behavior data distributed across different educational institutions or teaching scenarios through federated learning. The technical aspects addressed by this patent include feature extraction from teacher classroom behavior data, student interaction data, and homework grading data; establishing a unified data representation format; and training a global teaching behavior model through federated learning while ensuring local data storage and privacy protection. Specifically, each participating party performs data standardization and behavior encoding locally, uploads the training results or model parameters to a federated server for aggregation, and updates the model parameters through multiple iterations to form a global teaching behavior data model.

[0004] Existing technologies revolve around data aggregation and unified representation. In practical applications, they rely heavily on standardized processing and overall modeling, lacking detailed descriptions of continuous behavioral processes. This is especially true when multiple sources of behavior coexist, making it difficult to reflect the sequential dependencies and combination structures between behaviors. For example, while classroom interactions and learning operations can be uniformly coded, they lack path-based associations, resulting in a fragmented state of behavioral relationships. Consequently, it is difficult to extract stable structural features and key change patterns during the analysis phase. Furthermore, multi-subject data integration focuses on the aggregation of overall model parameters, lacking constraints on the correspondence between specific behavioral nodes. This can easily lead to analysis results that are biased towards overall trends while weakening structural hierarchical information, thus affecting the precision of subsequent teaching analysis. Summary of the Invention

[0005] This invention aims to provide a teaching behavior data modeling method based on federated learning, specifically involving multi-terminal teaching behavior data collection, feature extraction, cross-terminal aggregation, and sequence modeling techniques combined with a federated learning framework. It is applicable to student learning behavior analysis and teaching effectiveness evaluation in scenarios such as smart classrooms and online education platforms.

[0006] The basic solution provided by this invention is: a method for modeling teaching behavior data based on federated learning, comprising: S1: Collect student behavior sequences in tablet answering, computer note taking, smart classroom textbook browsing, and classroom interaction events. Sort the behaviors by occurrence time, extract the type, duration, operation sequence, and time interval of each behavior, organize the behaviors by type and sequence order, mark and classify repeated behaviors, and generate a behavior activity feature table. S2: Based on the behavioral activity feature table, identify key behavioral nodes that appear repeatedly in the sequence, analyze the type, order of occurrence and positional relationship of the behavioral nodes, arrange and group the selected nodes according to the sequence order, and superimpose and organize behavioral nodes of the same type to form a core teaching behavior feature set; S3: Call the core teaching behavior feature set uploaded by each terminal, classify and merge nodes with the same behavior type, organize the node order and occurrence combination in the behavior sequence of each terminal, integrate the node sequence according to time order and behavior type, organize all nodes into a unified set, and generate a multi-terminal teaching behavior aggregation table. S4: Based on the multi-terminal teaching behavior aggregation table, analyze the order of occurrence and combination relationship of the same behavior nodes in different student behavior sequences, organize and adjust the key nodes within the sequence, match and associate the corresponding nodes in each sequence, and integrate the processed nodes to form a cross-sequence teaching behavior association table. S5: Based on the cross-sequence teaching behavior association table, match and overlay the same behavior nodes in each terminal behavior sequence in chronological order, organize and classify the types, order and combination relationships of the nodes, integrate all sequence nodes into a complete behavior sequence set, and generate a teaching behavior modeling result set.

[0007] The behavioral activity feature table includes behavioral activity level, behavioral distribution status, behavioral change trend, and behavioral stability features. The core teaching behavior feature set includes key behavioral categories, behavioral combination forms, behavioral structural features, and behavioral weight attributes. The multi-terminal teaching behavior aggregation table includes terminal distribution structure, behavioral mapping relationship, behavioral consistency degree, and behavioral combination pattern. The cross-sequence teaching behavior association table includes node association relationship, behavioral path structure, sequence correspondence relationship, and node dependency structure. The teaching behavior modeling result set includes behavioral pattern classification, learning behavior profile, teaching interaction model, and behavioral trend features.

[0008] The working principle and advantages of this invention are as follows: This invention constructs a time-ordered sequence of multi-source learning behaviors and introduces duration and interval features to form a dynamic characterization. It forms stable structural units through repetitive behavior compression and key node screening, integrates behavioral combination relationships by merging cross-terminal nodes into a unified set, and forms a path-based structure through cross-sequence node matching and association. This transforms behavioral information from scattered fragments into a continuous system with related relationships. Its novelty lies in the systematic organization of behavioral node relationships and sequence structures. Its advantage lies in enhancing the sequential dependence and combination expression ability of nodes. The achieved effect is to improve the hierarchical performance of behavioral structures and enhance the adaptability of modeling results to complex teaching scenarios. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of step S1 provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of step S2 provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of step S3 provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of step S4 provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of step S5 provided in an embodiment of the present invention. Detailed Implementation

[0010] The following detailed explanation illustrates the specific implementation methods: The basic implementation examples are as follows: Figure 1 As shown: A method for modeling teaching behavior data based on federated learning, including: S1: Collect student behavior sequences in tablet answering, computer note taking, smart classroom textbook browsing, and classroom interaction events. Sort the behaviors by occurrence time, extract the type, duration, operation sequence, and time interval of each behavior, organize the behaviors by type and sequence order, mark and classify repetitive behaviors, and generate a behavior activity feature table.

[0011] Specifically, such as Figure 2 As shown, S1 includes: S101: Collects behavioral events generated by students during tablet answering, computer note-taking, smart classroom textbook browsing, and classroom interaction. Extracts the behavior type, occurrence time, end time, and collection sequence number. Converts each behavioral event into a unified event row, arranges the unified event rows in ascending order based on occurrence time, and arranges them in ascending order based on collection sequence number if the occurrence times are the same. Assigns consecutive operation bit numbers to each unified event row and generates a behavioral sequence event chain; details are as follows. In a smart classroom environment, student terminal devices continuously generate multi-source behavioral data. This includes data on student touch gestures from tablet-based question-and-answer systems, data on note-taking from computer clients, logs of textbook browsing from the smart classroom server, and data on interactive events such as raising hands and answering questions from the classroom interaction platform. Each type of behavioral event carries raw fields including a behavior type code, a millisecond-level timestamp of the behavior occurrence time, a millisecond-level timestamp of the behavior end time, and a collection sequence number assigned by the system.

[0012] The data collection program polls each terminal according to a preset sampling period. When a new behavioral event is detected, it extracts the aforementioned four-tuple information. Taking a tablet-based quiz scenario as an example, a student uses a stylus to complete a checkmark operation in the answer area at time 38562. This operation lasts for 0.8 seconds, and the system assigns it the collection sequence number 1267. The data collection program converts this event into a standard event format: the behavior type is mapped to "tablet quiz," the occurrence time is recorded as 38562, the end time is recorded as 38562.8, and the collection sequence number is recorded as 1267.

[0013] The unified event line generation process adopts an event-driven architecture. When the event buffer receives event data from different terminals, a sorting thread is triggered to execute a multi-level sorting algorithm. First, the occurrence time field value of all event lines is extracted and sorted in ascending order of value. When two event lines have the same occurrence time value, their acquisition sequence number field is further compared, with the smaller acquisition sequence number value ranked first. The sorted event sequence satisfies the constraints of temporal monotonicity and the uniqueness of acquisition sequence numbers.

[0014] The assignment of consecutive operation bit numbers uses an auto-incrementing sequence generation method. The first event line after sorting is assigned bit number 1, and subsequent event lines are assigned bit number 2, bit number 3, etc., in sequence. The binding relationship between consecutive operation bit numbers and event lines constitutes the core data structure of the behavior time-series event chain. The final output behavior time-series event chain contains N unified event lines, each carrying five fields: behavior type, occurrence time, end time, acquisition sequence number, and consecutive operation bit number.

[0015] Taking a classroom learning activity as an example, 10 behavioral event records were collected. After sorting and assigning positions, a behavioral chronological event chain was formed. The first record in this event chain is: behavior type "textbook browsing", occurrence time 38560.5, end time 38562.3, collection sequence number 1001, and continuous operation position number 1; the second record is: behavior type "tablet answering questions", occurrence time 38562.8, end time 38563.6, collection sequence number 1267, and continuous operation position number 2. The event chain completely records the various learning behaviors of students during this period in chronological order.

[0016] S102: Based on the behavioral timing event chain, read the behavior type, occurrence time, end time, and operation tag number from each unified event line. Map the behavior type to a type label, determine the duration based on the difference between the end time and the occurrence time, determine the operation sequence based on the operation tag number, and determine the time interval based on the difference between the next occurrence time and the previous end time. Establish an index relationship between the type label, duration, operation sequence, and time interval according to the operation tag number to generate a behavior attribute index matrix; specifically as follows: The construction of the behavioral attribute index matrix is ​​based on the behavioral time-series event chain as the data source. The matrix generation program reads the data of each field in the event chain line by line and performs three core processes: field mapping, time calculation and index association.

[0017] The field mapping phase establishes the correspondence between behavior types and type labels. The program maintains a behavior type mapping table, where the key is the original behavior type code and the value is the standardized type label. When the "Tablet Quiz" type is read, the mapping table is queried to obtain the corresponding label value, such as "B1"; when "Computer Notes" is read, the label value, such as "B2", is obtained. The mapping table is designed to be extensible; when a new behavior type appears, an entry can be added to the mapping table without affecting the existing processing logic.

[0018] The duration is calculated by subtracting the occurrence time from the end time. For an event line with an occurrence time of 38562.8 and an end time of 38563.6, the duration is calculated as Δt = 38563.6 - 38562.8 = 0.8 seconds. The program performs this subtraction operation on each record in the event chain, generating a sequence of duration values.

[0019] The operation order is determined directly from the consecutive operation bit number field. The consecutive operation bit number itself is the time sequence index of the events, with bit number 1 corresponding to the first operation and bit number N corresponding to the last operation. The operation order is consistent with the value of the consecutive operation bit number.

[0020] The time interval is calculated using the time difference method between adjacent events. Specifically, the time interval between the i-th event and the (i-1)-th event is equal to the occurrence time of the i-th event minus the end time of the (i-1)-th event. For the third event in the behavioral sequence event chain, its occurrence time is 38565.2, and the end time of the preceding event is 38563.6, so the calculated time interval ΔT = 38565.2 - 38563.6 = 1.6 seconds. When an event is the first event in the sequence, its time interval with the preceding event is set to a special value of -1, indicating that there is no preceding event.

[0021] The indexing mechanism establishes a mapping between four attributes—type label, duration, operation order, and time interval—and consecutive operation bit numbers. In the structured array generated by the program, each row corresponds to a consecutive operation bit number, and the row contains the four attribute values ​​corresponding to that bit number. This index matrix supports fast retrieval of complete attribute information for any event by bit number.

[0022] S103: Call the behavior attribute index matrix, aggregate similar event rows by type label, compare the type label, operation order, and time interval in adjacent event rows, mark event rows with consistent type labels and consecutive operation orders as duplicate behaviors, establish a classification key between the duplicate behaviors and their corresponding collection sequence numbers, associate the classification key, type label, duration, operation order, and time interval, map the duplicate markings and classification results by table fields, and generate a behavior activity feature table; details are as follows: The generation of the behavioral activity feature table includes four processing stages: event aggregation, duplicate detection, classification key generation, and table field mapping.

[0023] The event aggregation operation first groups the records in the behavior attribute index matrix according to the type label field. The program uses a hash table structure to store the aggregation results, where the key is the type label value and the value is the set of all event rows corresponding to that type label. After aggregation, events with type label "B1" form aggregation group A, and events with type label "B2" form aggregation group B.

[0024] The duplicate detection process iterates through adjacent event rows to perform consistency checks. The detection criteria include three sub-conditions: first, the type label values ​​are equal; second, the operation sequence values ​​are consecutive (i.e., the difference between adjacent bit numbers is equal to 1); and third, the time interval meets the continuity threshold requirement. Let the continuity threshold be 3 seconds; when the time interval is less than or equal to 3 seconds, it is considered time-continuous. In this step, the time interval determination uses a specific numerical comparison operation: read the interval value of the current event row, compare it with the threshold 3, and if the interval value is ≤ 3, the type labels are equal, and the operation bit numbers are consecutive, then the current event behavior is marked as a duplicate.

[0025] The classification key generation associates repetitive events with their corresponding acquisition sequence numbers. The program generates a unique classification key for each repetitive event, in the format "type label_first event acquisition sequence number". For example, for consecutive repetitive events with the type label "B1", if the first event acquisition sequence number is 1267, the generated classification key will be "B1_1267". The classification key is stored in association with four pieces of information: type label, duration, operation order, and time interval.

[0026] The table field mapping maps the processing results to a standardized table structure. The behavioral activity feature table contains the following fields: the duplicate flag field records whether the event is judged as a duplicate, with a value of 0 or 1; the category key field records the category identifier to which the event belongs; the type label field records the behavior type code of the event; the duration field records the duration of the event; the operation sequence field records the consecutive operation number; and the time interval field records the time interval between the event and the preceding event.

[0027] If the behavior attribute index matrix contains 12 event records, they are aggregated into 3 type groups. After duplicate detection, events 3 to 5 with type label "B1" meet the duplicate condition and are marked as duplicate behaviors. The corresponding records in the generated behavior activity feature table are: Duplicate Marker = 1, Classification Key = B1_1275, Type Label = B1, Duration = [0.8, 0.6, 0.9], Operation Order = [3, 4, 5], Time Interval = [1.2, 0.5]. This feature table comprehensively records the occurrence characteristics and repetition patterns of various learning behaviors.

[0028] S2: Based on the behavioral activity feature table, identify key behavioral nodes that recur in the sequence, analyze the type, order of occurrence and positional relationship of the behavioral nodes, arrange and group the selected nodes according to the sequence order, and superimpose and organize behavioral nodes of the same type to form a core teaching behavioral feature set.

[0029] Specifically, such as Figure 3 As shown, S2 includes: S201: Obtain the behavioral activity feature table, extract type labels, operation order, sequence position, repetition identifiers, and classification keys. Arrange the feature rows according to the operation order, count the number of times the same type label appears at each sequence position, and determine the feature rows with more than one occurrence and a repetition identifier as key behavioral nodes. Establish the correspondence between node numbers and type labels, operation order, and sequence position, and generate a key behavioral node index table; details are as follows: The construction of the key behavior node index table is based on the behavior activity feature table, and performs three core operations: feature extraction, frequency statistics, and node determination.

[0030] The feature extraction phase reads five fields from the behavioral active feature table: type label, operation order, sequence position, duplicate identifier, and classification key. The program creates a temporary data container and reads the field values ​​sequentially according to the row order of the original table, forming a feature row sequence. Each feature row contains an operation order value and a sequence position value, where the sequence position is defined as the ordinal number of the feature row in the table.

[0031] The frequency of occurrence statistics calculates the frequency of occurrence of the same type of label at each sequence position. The program constructs a two-dimensional statistical matrix of type label and sequence position, traverses all feature rows, and counts the type label value and sequence position value for each record. For example, if the event with the type label "Textbook Browsing" occurs at sequence positions 1, 3, and 5, then "Textbook Browsing" occurs once at sequence position 1, once at sequence position 3, and once at sequence position 5.

[0032] The key behavior node is determined using a threshold comparison method, with a threshold of 1. The program iterates through the statistical results and marks feature lines that appear more than once and have a duplicate flag value of 1 as key behavior nodes. The determination logic is expressed as follows: when the number of occurrences of a feature line is cnt>1 and the duplicate flag is flag=1, the feature line is determined to be a key behavior node.

[0033] Node numbers are generated using an auto-incrementing encoding method, starting from 1 and increasing sequentially. The program establishes a correspondence table between node numbers and type labels, operation order, and sequence position; this correspondence table is the core structure of the key behavior node index table.

[0034] Assume the behavioral activity feature table contains 20 feature rows. The feature row with the type label "Tablet Answering" appears at sequence positions 2, 4, and 7, with a repetition flag of 1 for each. Statistics show that "Tablet Answering" appears once at sequence position 2, once at sequence position 4, and once at sequence position 7; there are no instances where the frequency is greater than 1. Therefore, this type label does not generate a key node. The feature row with the type label "Classroom Interaction" appears three times at sequence position 5, with a repetition flag of 1 for each occurrence. Since the frequency of 3 occurrences exceeds the threshold of 1, it is determined to be a key behavioral node. The generated key behavioral node index table records: Node Number 1, Type Label "Classroom Interaction", Operation Sequence 5, Sequence Position 5.

[0035] S202: Based on the key behavior node index table, compare the difference in operation order and sequence position between adjacent node numbers. Nodes with an operation order difference of one are classified as consecutive nodes, and nodes with a sequence position difference greater than one are classified as interval nodes. Consecutive and interval nodes are arranged in ascending order of operation order. Nodes with the same classification key are grouped into the same node group, generating a behavior node order grouping table; specifically as follows: The construction of the behavior node sequential grouping table takes the key behavior node index table as input and performs three processes: node comparison, group merging, and sorting output.

[0036] The node comparison operation uses adjacent node numbers as the processing unit. The program sequentially reads two adjacent node records from the index table and extracts their respective operation order values ​​and sequence position values. Let the operation order of the preceding node be O1 and the sequence position be P1, and the operation order of the following node be O2 and the sequence position be P2. Calculate the operation order difference ΔO = O2 - O1, and calculate the sequence position difference ΔP = P2 - P1.

[0037] Continuous node determination rule: When ΔO=1, the subsequent node is determined to be a continuous node of the preceding node. Interval node determination rule: When ΔP>1, the subsequent node is determined to be an interval node of the preceding node. For example, if the preceding node's operation order is 5 and its sequence position is 5, and the subsequent node's operation order is 6 and its sequence position is 5, then ΔO=1 and ΔP=0 are calculated. ΔO=1 satisfies the continuous determination condition, and the subsequent node is marked as a continuous node. As another example, if the preceding node's sequence position is 5 and the subsequent node's sequence position is 8, then ΔP=3 is calculated. ΔP>1 satisfies the interval determination condition, and the subsequent node is marked as an interval node.

[0038] The sorted output arranges all node records in ascending order according to the operation sequence. The program uses a stable sorting algorithm to maintain the relative order of nodes with the same operation sequence. Consecutive nodes and interval nodes are arranged in a uniform order in the sorted result.

[0039] Grouping by category key groups all nodes with the same category key into the same node group. The program reads the category key field value from each node record and groups and aggregates according to the category key value. Nodes 1, 3, and 5 with the category key "Classroom Interaction_101" are grouped into node group A, and nodes 2 and 4 with the category key "Tablet Answering_203" are grouped into node group B.

[0040] The output data forms a behavior node sequence grouping table. The table structure includes five fields: node number, operation order, sequence position, node type flag (continuous or interval), and category key. Example of a record in the table: Node number 3, operation order 8, sequence position 10, node type "interval", category key "classroom interaction_101".

[0041] S203: Call the behavior node sequence grouping table, aggregate node groups by type label, accumulate the occurrence count of the same type label, summarize the corresponding operation sequence and sequence position sequence, map node groups with the same type label to the same feature item, associate node number, type label, occurrence count, operation sequence, sequence position sequence, and classification key to generate the core teaching behavior feature set; as detailed below: The core teaching behavior feature set is constructed using a behavior node sequential grouping table as the data source, and performs four processing steps: type aggregation, frequency accumulation, sequence summarization, and feature mapping.

[0042] The type aggregation operation merges node groups based on the type label field. The program iterates through the row-node sequential grouping table, extracts the type label value for each record, and uses a hash mapping structure to aggregate records with the same type label into the same set. The "Textbook Browsing" type aggregates node group C, and the "Classroom Interaction" type aggregates node group A.

[0043] The occurrence count is calculated by summing the occurrence counts of the same type of label across all node groups. The program reads the number of nodes in each node group and sums them to obtain the total occurrence count of that type of label. For example, if the "Classroom Interaction" type appears 3 times in node group A and 2 times in the node group corresponding to another categorization key, then the total occurrence count of "Classroom Interaction" is 5.

[0044] The operation order sequence summary sorts the operation order values ​​of all nodes corresponding to the same type of label in ascending order, forming an operation order sequence array. The program extracts the operation order field value of each node, stores it in a temporary list, performs a sorting operation on the list, and outputs an ordered sequence. The sequence position sequence summary uses the same method to summarize the sequence position values ​​of all nodes corresponding to the same type of label.

[0045] The feature mapping establishes a complete feature term for each type label, including node number, type label, occurrence count, operation order sequence, sequence position sequence, and classification key. The program generates a unique feature term identifier for each type label and populates the above six attribute values.

[0046] The core teaching behavior feature set stores all feature items in list form. Example of a feature set: The first feature item contains node number [1,3,5], type label "Classroom Interaction", frequency of occurrence 5, operation sequence [5,8,12], sequence position sequence [5,10,15], and classification key "Classroom Interaction_101"; the second feature item contains node number [2,4], type label "Tablet Answering", frequency of occurrence 3, operation sequence [6,9], sequence position sequence [6,9], and classification key "Tablet Answering_203". This feature set comprehensively describes the frequency and sequence distribution characteristics of various core teaching behaviors.

[0047] S3: Call the core teaching behavior feature set uploaded by each terminal, classify and merge nodes with the same behavior type, organize the node order and occurrence combination in the behavior sequence of each terminal, integrate the node sequence according to time order and behavior type, organize all nodes into a unified set, and generate a multi-terminal teaching behavior aggregation table.

[0048] Specifically, such as Figure 4As shown, S3 includes: S301: Obtain the core teaching behavior feature set uploaded by each terminal, extract the terminal identifier, behavior type, node number, node time, node order, and occurrence combination identifier, group nodes with the same behavior type into the same type bucket, compare the node number and node time within the same type bucket, retain the terminal identifier from which the node originated, establish a mapping relationship between behavior type and node number, node time, and node order, and generate a terminal node type merging table; details are as follows: The construction of the terminal node type merging table takes the core teaching behavior feature set as input and performs four processes: terminal identifier extraction, type bucket classification, node comparison, and mapping relationship establishment.

[0049] The terminal identifier extraction program reads the classification key field value of each feature item from the core teaching behavior feature set, and parses the classification key to obtain the corresponding terminal identifier information. The classification key format is "type label_terminal identifier_serial number", which is parsed to obtain the terminal identifier field. In this step, the program traverses all feature items, extracts the terminal identifier portion from the classification key, and builds a terminal identifier list.

[0050] Type buckets are categorized based on behavior type, grouping nodes of the same behavior type into their corresponding type buckets. The program creates a dictionary structure to store the type buckets, with the key being the behavior type name and the value being the set of all node records contained in that type. The core teaching behavior features are concentrated in three types: "textbook browsing," "classroom interaction," and "tablet answering," which are respectively created as type buckets A, B, and C.

[0051] Node comparisons are performed within buckets of the same type, checking the consistency of the node ID and node time fields. The program sequentially reads node records within buckets of the same type, extracting and comparing the node ID and node time field values. Records with the same node ID and node time are determined to be nodes from the same source. The comparison results are used to filter and retain unique node source terminal identifiers.

[0052] The mapping relationship is established by associating behavior types with node numbers, node times, and node sequences. The program generates a structured mapping table, where each row contains a list of behavior types, node numbers, node times, and node sequences. Example of a row in the terminal node type merging table: Behavior type "Classroom Interaction", Node number [101,102], Node time [38560,38565], Node sequence [1,2], Terminal identifier "Terminal A".

[0053] S302: Based on the terminal node type merging table, aggregate node rows by terminal identifier, compare the order difference between adjacent nodes and the node time sequence, determine nodes with an order difference of one as consecutive nodes, and determine nodes with an order difference greater than one as interval nodes, associate the combined identifier with consecutive nodes and interval nodes, arrange each terminal node chain in ascending order of node time, and generate a node sequence combination index; as detailed below: The construction of the node sequence combined index is based on the terminal node type merge table, and performs four processes: terminal aggregation, relationship determination, identifier association, and sequence sorting.

[0054] The terminal aggregation operation groups the records in the merge table by the terminal identifier field. The program iterates through all rows of the merge table, extracts the terminal identifier value, and uses a grouping aggregation algorithm to group records with the same terminal identifier into the same group. Terminal A corresponds to 3 node records, and terminal B corresponds to 4 node records.

[0055] Relationship determination is performed on adjacent nodes within the same terminal. The program reads node records within the same terminal, sorts them in ascending order by the node sequence field, calculates the sequence difference between adjacent nodes, and compares the node time order. Let the preceding node sequence be 1 and the following node sequence be 2. The sequence difference ΔS = 2 - 1 = 1. When ΔS = 1, the following node is determined to be a consecutive node. Let the preceding node sequence be 2 and the following node sequence be 5. The sequence difference ΔS = 5 - 2 = 3. When ΔS > 1, the following node is determined to be an intervening node.

[0056] The identifier association establishes associations between combined identifiers and consecutive or intervening nodes. The program reads the combined identifier field value from each node record. When a node is determined to be a consecutive node, its identifier value is associated with the consecutive node set; when a node is determined to be an intervening node, its identifier value is associated with the intervening node set.

[0057] The sequence sorting arranges the node chains of each terminal in ascending order of node time. The program sorts all node records within the same terminal by the node time field, generating a time-ordered node sequence. The output data is a node sequence composite index, and the index structure includes terminal identifier, node number, node time, node order, node type flag, and occurrence composite flag fields. Example of a terminal node sequence: Terminal identifier "Terminal A", node record sequence [(101,38560,1,continuous,ID1),(105,38563,2,continuous,ID1),(108,38568,5,interval,ID2)].

[0058] S303: Call the node sequence combination index, perform cross-sorting on each terminal node chain according to node time and behavior type. When node times are consistent, sort them in ascending order according to behavior type code. Merge nodes with the same behavior type and the same occurrence combination identifier, establish a unified node number association relationship with terminal identifier, behavior type, node order, node time, and occurrence combination identifier, map the association relationship to table fields, and generate a multi-terminal teaching behavior aggregation table; details are as follows: The construction of the multi-terminal teaching behavior aggregation table takes the node sequence combined index as input and performs three processes: cross sorting, node merging, and mapping writing.

[0059] Cross-sort performs a sorting operation on each terminal node chain based on two fields: node time and behavior type. The program first sorts all node records in ascending order by node time. When two records have the same node time value, they are further sorted in ascending order by behavior type code. The behavior type code uses a numeric encoding method: "Textbook browsing" is coded as 1, "Classroom interaction" as 2, and "Tablet answering" as 3. The sorted node sequence follows a time-priority, type-secondary arrangement rule.

[0060] Node merging combines adjacent nodes with the same behavior type and the same combination identifier in the sorted results. The program iterates through the sorted node sequence, checking whether the behavior type and combination identifier fields of the current node and its predecessor are completely identical. If both field values ​​are equal, the merge operation is performed, combining the predecessor node and the current node into a single record. The merged record retains the current node's time and the order of the nodes from their predecessors. This merging operation reduces redundant records and improves data aggregation.

[0061] The mapping process writes the merged node information into a multi-terminal teaching behavior aggregation table according to the table field format. Table fields include a unified node number, terminal identifier, behavior type, node order, node time, and occurrence combination identifier. The program assigns a unique unified node number to each merged record, following the rule of "UN" + a 6-digit serial number. The field mapping process fills the corresponding columns in the table structure with the values ​​of each field.

[0062] Table 1 shows an example of the generated results of the multi-terminal teaching behavior aggregation table: Table 1: Multi-Terminal Teaching Behavior Aggregation Table ; As shown in Table 1, the multi-terminal teaching behavior aggregation table fully records the unified number, source identifier, type classification, time characteristics and combination relationship of each terminal learning behavior, providing a unified data structure for subsequent cross-sequence analysis.

[0063] S4: Based on the multi-terminal teaching behavior aggregation table, analyze the order of occurrence and combination relationship of the same behavior nodes in different student behavior sequences, organize and adjust key nodes within the sequence, match and associate the corresponding nodes in each sequence, and integrate the processed nodes to form a cross-sequence teaching behavior association table.

[0064] Specifically, such as Figure 5 As shown, S4 includes: S401: Obtain the multi-terminal teaching behavior aggregation table, extract student identifier, unified node number, behavior type, node time, node order, and occurrence combination identifier. Split the behavior sequence by student identifier, arrange each behavior sequence in ascending order by node order, compare the order difference and node time relationship of adjacent nodes of the same student, mark nodes with an order difference of one as adjacent nodes, and mark nodes with an order difference greater than one as interval nodes. Establish the mapping relationship between student identifier, unified node number, behavior type, and node order, and generate a student node order matrix; details are as follows: The construction of the student node order matrix takes the multi-terminal teaching behavior aggregation table as input and performs four processes: sequence splitting, order arrangement, relationship determination, and mapping establishment.

[0065] Sequence splitting assigns the behavior records in the aggregation table to the corresponding student names based on the student identifier field. The program reads all records in the aggregation table, extracts the student identifier field value, and uses a grouping algorithm to group records with the same student identifier into the same sequence container. The program maintains the mapping relationship between student identifiers and behavior sequences, with each student corresponding to an independent node sequence.

[0066] The program sorts the student's action sequences in ascending order by the node order field. It iterates through each student's node sequence records, rearranging the records in ascending order of the node order values. The sorting result ensures that events within each student's node sequence are arranged in the chronological order of operations.

[0067] Relationship determination involves comparing the order difference and temporal sequence of adjacent nodes within the same student. The program reads two adjacent node records of the same student and calculates the node order difference ΔS = S_subsequent - S_previous. The determination rule is: if ΔS = 1, the subsequent node is marked as an adjacent node; if ΔS > 1, the subsequent node is marked as an interval node. Simultaneously, the temporal sequence of adjacent nodes is compared to determine the temporal order.

[0068] The mapping establishes an association index between student identifiers and unified node numbers, behavior types, and node order. The program generates a student node order matrix, where each row corresponds to a student identifier and contains all node record information for that student. Example of a student matrix record: Student identifier "Student 001", node record sequence [(UN000001, Classroom Interaction, 1), (UN000003, Classroom Interaction, 2), (UN000004, Tablet Answering, 3)], where the order difference between UN000001 and UN000003 is 1, indicating they are adjacent nodes.

[0069] S402: Based on the student node order matrix, aggregate different student node rows by behavior type, compare the node order and occurrence combination identifier of nodes of the same behavior type in each behavior sequence, determine the corresponding nodes as nodes with the same behavior type and the same occurrence combination identifier, mark the nodes with the same behavior type and non-zero node order difference as order offset nodes, associate the corresponding nodes, order offset nodes and student identifiers to generate a cross-sequence node matching table; the details are as follows: The construction of the cross-sequence node matching table is based on the student node order matrix, and performs four processes: type aggregation, sequence comparison, node labeling, and identifier association.

[0070] Type aggregation groups records of the same behavior type from different student node rows. The program iterates through the student node sequence matrix, extracts the behavior type field value of each node record, and stores all student node records according to behavior type. The "Classroom Interaction" type aggregates the node records of students 001, 002, and 003, while the "Textbook Browsing" type aggregates the node records of students 001 and 004.

[0071] Sequence comparison performs matching analysis on student node sequences within the same behavior type. The program reads all student node records under the same behavior type and compares the node order value and occurrence combination identifier field value of each record in its respective sequence. If student 001's "Classroom Interaction" node order is 2 and occurrence combination identifier is ID1, and student 002's "Classroom Interaction" node order is 2 and occurrence combination identifier is ID1, then these two nodes are determined to be corresponding nodes. If student 001's "Classroom Interaction" node order is 3 and occurrence combination identifier is ID1, and student 003's "Classroom Interaction" node order is 5 and occurrence combination identifier is ID1, then these two nodes are determined to be sequence offset nodes.

[0072] The node labeling adds a type label to each node based on the comparison results. The program labels nodes that meet the conditions of consistent behavior type and consistent combined identifier as "corresponding nodes", and labels nodes that meet the conditions of consistent behavior type but non-zero node order difference as "order offset nodes".

[0073] The identifier association establishes a relationship between the corresponding node, the sequence offset node, and the student identifier. The program generates a cross-sequence node matching table, the table structure of which includes behavior type, node status flag, student identifier list, unified node number list, and node order list. Example of a matching record: Behavior type "Classroom Interaction", Node status "Corresponding Node", Student identifier list [Student 001, Student 002], Unified node number list [UN000001, UN000003], Node order list [2,2].

[0074] S403: Call the cross-sequence node matching table, aggregate student identifiers and unified node numbers according to the corresponding nodes, rearrange the order offset nodes in ascending order according to the node order, merge the same node groups according to behavior type, establish the mapping relationship between the association category and student identifier, unified node number, behavior type, node order, occurrence combination identifier, and order offset marker, write the mapping relationship into the field column, and generate a cross-sequence teaching behavior association table; as detailed below: The construction of the cross-sequence teaching behavior association table takes the cross-sequence node matching table as input and performs four processes: node aggregation, order reordering, type merging, and mapping writing.

[0075] Node aggregation groups the records in the matching table according to the corresponding node field. The program reads all rows in the matching table, extracts the corresponding node field values, and groups records with the same corresponding node identifier into the same aggregation group. Each corresponding node aggregation group contains several student identifiers and corresponding unified node number information.

[0076] The sequential rearrangement process reorders the sequential offset nodes within each aggregation group in ascending order. The program extracts the sequential offset node information from each record and reorders them according to their numerical values ​​from smallest to largest. The rearrangement ensures that the sequential offset nodes are arranged in chronological or operational order.

[0077] Type grouping merges nodes with the same behavior type into a unified node set. The program performs secondary grouping of aggregate groups according to the behavior type field, grouping aggregate groups with the same behavior type into the same node set. After merging, each behavior type corresponds to a set containing multiple node groups.

[0078] The mapping write process writes the association processing results into the cross-sequence teaching behavior association table according to the table field format. The table fields include association category, student identifier, unified node number, behavior type, node order, occurrence combination identifier, and order offset flag. The program assigns a unique association category number to each record, in the format "CR" + 4-digit serial number. The field mapping process fills the corresponding columns in the table structure with the values ​​of each field.

[0079] Table 2 shows an example of the generated results of the cross-sequence teaching behavior association table: Table 2: Cross-Sequence Teaching Behavior Association Table ; As shown in Table 2, the cross-sequence teaching behavior association table establishes the cross-sequence correspondence between teaching behaviors of different students. The node records with corresponding relationships are linked by the association category number, providing an association data structure for teaching behavior pattern analysis.

[0080] S5: Based on the cross-sequence teaching behavior association table, match and overlay the same behavior nodes in each terminal behavior sequence in chronological order, organize and classify the types, order and combination relationships of the nodes, integrate all sequence nodes into a complete behavior sequence set, and generate a teaching behavior modeling result set.

[0081] Specifically, such as Figure 6 As shown, S5 includes: S501: Obtain the cross-sequence teaching behavior association table, extract terminal identifier, sequence identifier, node number, behavior type, node time, node order, and combination identifier; aggregate similar nodes by behavior type; compare time sequence and order difference; determine nodes in ascending time order as matching nodes; mark nodes with zero order difference as overlapping nodes; mark nodes with non-zero order difference as misaligned nodes; establish the relationship between matching nodes and terminal identifier and sequence identifier; and generate a time matching node table. Details are as follows: The construction of the time matching node table takes the cross-sequence teaching behavior association table as input and performs four processes: type aggregation, time comparison, node marking, and relationship establishment.

[0082] Type aggregation groups records based on behavior type, grouping records with the same behavior type in the associated tables into the same type bucket. The program iterates through all records in the associated tables, extracts the behavior type field value, and assigns the records to the corresponding type bucket according to the type value. The classroom interaction type bucket contains 10 records, the textbook browsing type bucket contains 8 records, and the tablet quiz type bucket contains 6 records.

[0083] The time comparison performs ascending time order determination and time difference calculation for node records within the same type of bucket. The program reads the node time field value of each record in the bucket and arranges the record sequence in ascending order of time value. The time difference ΔT between adjacent records is calculated as T_subsequent - T_previous. Assuming the previous node time is 38560 and the subsequent node time is 38562, the calculated ΔT = 2 seconds.

[0084] The matching node determination rule is based on the ascending time order: when the time value of a node is greater than the time of its predecessor node, the node is determined to be a matching node. This determination condition is expressed as: if T_current > T_previous, then the current node is marked as a matching node.

[0085] Misaligned node marking is based on order difference determination: Calculate the order difference ΔN between the current node and its corresponding associated record, where ΔN = Ncurrent - Ncorresponding. When ΔN ≠ 0, mark the current node as a misaligned node. When ΔN = 0, mark the current node as a coincident node.

[0086] The relationship establishment process establishes a correspondence between matching nodes, overlapping nodes, misaligned nodes, and terminal identifiers and sequence identifiers. The program generates a time-matching node table, whose structure includes fields for node number, behavior type, node time, matching status flag, terminal identifier, and sequence identifier.

[0087] S502: Based on the time-matching node table, arrange matching nodes by terminal identifier, and overlay overlapping and misaligned nodes according to the ascending order of node time. Count the number of occurrences of the same node number, compare corresponding combination identifiers, group nodes with the same identifier into the same combination group, and group nodes with different identifiers into the difference combination group. Associate node number, behavior type, node order, number of occurrences, and combination group to generate a node overlay relationship matrix; as detailed below: The construction of the node overlay relationship matrix is ​​based on the time-matched node table, and performs four processes: terminal arrangement, node overlay, frequency statistics, and combination grouping.

[0088] The terminal arrangement rearranges the matching node records according to the terminal identifier field. The program reads the time matching node table, extracts the terminal identifier field value, and sorts the records in ascending order according to the terminal identifier alphabetical order or number order. In the arrangement result, all node records of the same terminal are arranged consecutively.

[0089] Node overlay is performed by stacking overlapping and misaligned nodes in ascending order of node time. The program iterates through the sorted node records and overlays nodes with the same or similar times to the same time layer based on their time field values. The overlay operation merges nodes from different terminals that have the same or similar times. For example, if terminal A generates node UN000001 at time 38560 and terminal B generates node UN000003 at time 38560, then these two nodes will be overlaid to the same time layer.

[0090] The program counts the occurrences of each node number after overlay. It iterates through the overlay results and counts the number of times each unified node number appears across all terminals. For example, if node number UN000001 appears once in terminal A, once in terminal B, and once in terminal C, then the node's occurrence count is 3.

[0091] The program groups nodes into different combination groups based on the consistency of their corresponding combination identifiers. It reads the value of the combination identifier field from each node record, and groups two records together if their identifier values ​​are identical. For example, if node UN000001 has the identifier ID1, node UN000002 has the identifier ID1, and node UN000003 has the identifier ID2, then the first two nodes are grouped into combination group A, and the last node is grouped into combination group B.

[0092] The node overlay relationship matrix output includes five dimensions of information: node number, behavior type, node order, frequency of occurrence, and group affiliation. Example of a record in a row of the matrix: Node number UN000001, Behavior type "Classroom Interaction", Node order 2, Frequency of occurrence 3, Group "Group A".

[0093] S503: Call the node overlay relationship matrix, aggregate combination groups by behavior type, sort terminal sequence nodes in ascending order of node time, compare the order difference between adjacent combination groups, mark combination groups with an order difference of one as consecutive combination groups, and mark combination groups with an order difference greater than one as intermittent combination groups. Establish the mapping relationship between modeling entries and terminal identifiers, sequence identifiers, behavior types, node times, node order, frequency of occurrence, and combination groups, and generate a teaching behavior modeling result set; details are as follows: The construction of the teaching behavior modeling result set takes the node superposition relationship matrix as input and performs four processes: type aggregation, order sorting, combination marking, and modeling item generation.

[0094] Type aggregation groups groups by behavior type, categorizing combinations in the matrix according to their behavior type. The program iterates through all records in the matrix, extracts the behavior type field value, and assigns the combinations to the corresponding type sets based on the type value. The classroom interaction type set contains combination group A and combination group C, and the textbook browsing type set contains combination group B.

[0095] The program sequentially organizes the combination group nodes in each terminal sequence in ascending order of node time. It reads all combination group records under the same behavior type set and sorts the combination group sequence in ascending order of node time field value. The resulting sequence is ordered by time.

[0096] The combination label is determined based on the order difference between adjacent combination groups. The program calculates the node order difference ΔN between adjacent combination groups as ΔN = N_subsequent - N_previous. The determination rule is: when ΔN = 1, the subsequent combination group is marked as a continuous combination; when ΔN > 1, the subsequent combination group is marked as an intermittent combination. For example, if the node order of the preceding combination group is 2 and the node order of the subsequent combination group is 3, the calculated ΔN = 1, and it is marked as a continuous combination. If the node order of the preceding combination group is 3 and the node order of the subsequent combination group is 6, the calculated ΔN = 3, and it is marked as an intermittent combination.

[0097] The modeling entry generation process outputs the compiled results as a teaching behavior modeling result set according to a standard field format. Table fields include modeling entry number, terminal identifier, sequence identifier, behavior type, node time, node order, frequency of occurrence, and combination group information. The program assigns a unique modeling entry number to each record, in the format "MB" + 6-digit serial number.

[0098] Table 3 shows an example of the generated results of the teaching behavior modeling result set: Table 3: Results Set of Teaching Behavior Modeling ; As shown in Table 3, the teaching behavior modeling result set comprehensively records the temporal distribution, sequential characteristics, and frequency of occurrence of various teaching behaviors. The terminal source, sequence identifier, behavior type, and time series information are linked and integrated through modeling item numbers. This result set can support various application scenarios such as teaching behavior pattern mining, learning trajectory analysis, and teaching effectiveness evaluation.

[0099] This embodiment provides a teaching behavior data modeling method based on federated learning, specifically involving multi-terminal teaching behavior data collection, feature extraction, cross-terminal aggregation, and sequence modeling techniques combined with the federated learning framework. It is applicable to student learning behavior analysis and teaching effectiveness evaluation in scenarios such as smart classrooms and online education platforms.

[0100] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. A method for modeling teaching behavior data based on federated learning, characterized in that, include: S1: Collect student behavior sequences in tablet answering, computer note taking, smart classroom textbook browsing, and classroom interaction events. Sort the behaviors by occurrence time, extract the type, duration, operation sequence, and time interval of each behavior, organize the behaviors by type and sequence order, mark and classify repeated behaviors, and generate a behavior activity feature table. S2: Based on the behavioral activity feature table, identify key behavioral nodes that appear repeatedly in the sequence, analyze the type, order of occurrence and positional relationship of the behavioral nodes, arrange and group the selected nodes according to the sequence order, and superimpose and organize behavioral nodes of the same type to form a core teaching behavior feature set; S3: Call the core teaching behavior feature set uploaded by each terminal, classify and merge nodes with the same behavior type, organize the node order and occurrence combination in the behavior sequence of each terminal, integrate the node sequence according to time order and behavior type, organize all nodes into a unified set, and generate a multi-terminal teaching behavior aggregation table. S4: Based on the multi-terminal teaching behavior aggregation table, analyze the order of occurrence and combination relationship of the same behavior nodes in different student behavior sequences, organize and adjust the key nodes within the sequence, match and classify the corresponding nodes in each sequence, and integrate the processed nodes to form a cross-sequence teaching behavior association table.

2. The method for modeling teaching behavior data based on federated learning according to claim 1, characterized in that, The behavioral activity feature table includes behavioral activity level, behavioral distribution status, behavioral change trend, and behavioral stability features. The core teaching behavior feature set includes key behavioral categories, behavioral combination forms, behavioral structural features, and behavioral weight attributes. The multi-terminal teaching behavior aggregation table includes terminal distribution structure, behavioral mapping relationship, behavioral consistency degree, and behavioral combination pattern. The cross-sequence teaching behavior association table includes node association relationship, behavioral path structure, sequence correspondence relationship, and node dependency structure.

3. The method for modeling teaching behavior data based on federated learning according to claim 1, characterized in that, The steps for obtaining S1 include: S101: Collects behavioral events generated by students in tablet answering, computer note taking, smart classroom textbook browsing, and classroom interaction events. Extracts the behavior type, occurrence time, end time, and collection sequence number. Converts each behavioral event into a unified event line. Arranges the unified event lines in ascending order according to the occurrence time. If the occurrence times are the same, arranges the unified event lines in ascending order according to the collection sequence number. Assigns a continuous operation bit number to each unified event line and generates a behavioral time sequence event chain. S102: Based on the behavior time sequence event chain, read the behavior type, occurrence time, end time and operation number in each unified event line, map the behavior type to a type label, determine the duration based on the difference between the end time and the occurrence time, determine the operation order based on the operation number, determine the time interval based on the difference between the next occurrence time and the previous end time, establish an index relationship between the type label, duration, operation order and time interval according to the operation number, and generate a behavior attribute index matrix; S103: Call the behavior attribute index matrix, aggregate similar event rows by type label, compare the type label, operation order and time interval in adjacent event rows, mark event rows with consistent type labels and continuous operation order as repeated behaviors, establish a classification key with the corresponding collection sequence number for repeated behaviors, associate the classification key, type label, duration, operation order and time interval, map the repeated marking and classification results by table fields, and generate a behavior activity feature table.

4. The teaching behavior data modeling method based on federated learning according to claim 1, characterized in that, The steps for obtaining S2 include: S201: Obtain the behavioral activity feature table, extract type labels, operation order, sequence position, repetition identifier and classification key, arrange feature rows according to operation order, count the number of times the same type label appears in each sequence position, determine the feature rows with more than one occurrence and with repetition identifier as key behavioral nodes, establish the correspondence between node number and type label, operation order and sequence position, and generate key behavioral node index table. S202: Based on the key behavior node index table, compare the operation order difference and sequence position difference of adjacent node numbers, determine the node with an operation order difference of one as a continuous node, and determine the node with a sequence position difference greater than one as an interval node. Arrange the continuous nodes and interval nodes in ascending order of operation order, and group the nodes with the same classification key into the same node group to generate a behavior node order grouping table. S203: Call the behavior node sequence grouping table, aggregate node groups by type label, accumulate the occurrence count of the same type label, summarize the corresponding operation sequence and sequence position sequence, map the node groups with the same type label to the same feature item, associate the node number, type label, occurrence count, operation sequence, sequence position sequence and classification key, and generate the core teaching behavior feature set.

5. The method for modeling teaching behavior data based on federated learning according to claim 1, characterized in that, The steps for obtaining S3 include: S301: Obtain the core teaching behavior feature set uploaded by each terminal, extract the terminal identifier, behavior type, node number, node time, node order and occurrence combination identifier, group nodes with the same behavior type into the same type bucket, compare the node number and node time in the same type bucket, retain the terminal identifier of the node source, establish the mapping relationship between behavior type and node number, node time and node order, and generate a terminal node type merging table. S302: Based on the terminal node type merging table, aggregate node rows by terminal identifier, compare the order difference of adjacent nodes and the order relationship of node time, determine the nodes with an order difference of one as continuous nodes, determine the nodes with an order difference greater than one as interval nodes, associate the combined identifier with continuous nodes and interval nodes, arrange each terminal node chain in ascending order of node time, and generate a node sequence combination index. S303: Call the node sequence combination index, perform cross sorting on each terminal node chain according to node time and behavior type, sort the nodes in ascending order according to behavior type code when the node time is the same, merge nodes with the same behavior type and the same occurrence combination identifier, establish a unified node number and terminal identifier, behavior type, node order, node time, and occurrence combination identifier association relationship, map the association relationship to table fields, and generate a multi-terminal teaching behavior aggregation table.

6. The method for modeling teaching behavior data based on federated learning according to claim 1, characterized in that, The steps for obtaining S4 include: S401: Obtain the multi-terminal teaching behavior aggregation table, extract student identifier, unified node number, behavior type, node time, node order and occurrence combination identifier, split the behavior sequence according to student identifier, arrange each behavior sequence in ascending order according to node order, compare the order difference of adjacent nodes of the same student and the order relationship of node time, mark the node with an order difference of one as an adjacent node, mark the node with an order difference greater than one as an interval node, establish the mapping relationship between student identifier and unified node number, behavior type, node order, and generate student node order matrix; S402: Based on the student node order matrix, aggregate different student node rows according to behavior type, compare the node order and occurrence combination identifier of the same behavior type in each behavior sequence, determine the corresponding node as the node with the same behavior type and the same occurrence combination identifier, mark the node with the same behavior type and the node order difference as the order offset node, associate the corresponding node, the order offset node and the student identifier to generate a cross-sequence node matching table. S403: Call the cross-sequence node matching table, aggregate student identifiers and unified node numbers according to the corresponding nodes, rearrange the order offset nodes in ascending order according to the node order, merge the same node groups according to the behavior type, establish the mapping relationship between the association category and student identifier, unified node number, behavior type, node order, occurrence combination identifier, and order offset mark, write the mapping relationship into the field column, and generate the cross-sequence teaching behavior association table.

7. The method for modeling teaching behavior data based on federated learning according to claim 1, characterized in that, It also includes S5: Based on the cross-sequence teaching behavior association table, matching and superimposing the same behavior nodes in each terminal behavior sequence in chronological order, sorting and classifying the types, order and combination relationships of the nodes, integrating all sequence nodes into a complete behavior sequence set, and generating a teaching behavior modeling result set.

8. The method for modeling teaching behavior data based on federated learning according to claim 7, characterized in that, The steps for obtaining S5 include: S501: Obtain the cross-sequence teaching behavior association table, extract terminal identifier, sequence identifier, node number, behavior type, node time, node order, and combination identifier, aggregate similar nodes by behavior type, compare time sequence and order difference, determine nodes in ascending time order as matching nodes, mark nodes with zero order difference as overlapping nodes, mark nodes with non-zero order difference as misaligned nodes, establish the relationship between matching nodes and terminal identifier and sequence identifier, and generate a time matching node table; S502: Based on the time matching node table, arrange matching nodes according to terminal identifier, superimpose overlapping nodes and misaligned nodes according to the ascending order of node time, count the number of times the same node number appears, compare the corresponding combination identifiers, group nodes with the same identifier into the same combination group, group nodes with different identifiers into the difference combination group, associate node number, behavior type, node order, number of occurrences and combination group, and generate a node superimposition relationship matrix. S503: Call the node overlay relationship matrix, aggregate combination groups by behavior type, sort terminal sequence nodes in ascending order by node time, compare the order difference of adjacent combination groups, mark combination groups with an order difference of one as continuous combination, mark combination groups with an order difference greater than one as interval combination, establish the mapping relationship between modeling items and terminal identifier, sequence identifier, behavior type, node time, node order, occurrence frequency, and combination group, and generate a teaching behavior modeling result set.

9. A method for modeling teaching behavior data based on federated learning according to claim 7, characterized in that, In S5, the teaching behavior modeling result set includes behavior pattern classification, learning behavior profile, teaching interaction model, and behavior trend characteristics.

Images