An AI-driven adaptive learning path recommendation method

Abstract

The application relates to the technical field of machine learning, in particular to an AI-driven adaptive learning path recommendation method, which comprises the following steps: acquiring an answer knowledge point access sequence record to generate a node sequence set, generating a closed loop structure based on position interval fitting degree, performing proportion conversion and node weight adjustment to generate a conflict association result, generating a path constraint structure according to the adjustment of directed edge connection strength, screening a candidate path and calculating a cumulative result to generate a path recommendation scheme. In the application, the closed loop structure is generated by analyzing the access sequence record to construct the structure and detecting the position interval, the conflict association result is generated by combining dynamic weight adjustment, the constraint relationship is established, the candidate path is screened in the knowledge graph, the fixed directed relationship and weight limitation are effectively broken through, the trial-and-error track in the individual cognitive backtracking process is mapped, the potential cognitive closed loop and path conflict in the complex learning situation are deeply analyzed, and it is ensured that the recommended path is highly consistent with the real individual knowledge acquisition law.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to an AI-driven adaptive learning path recommendation method. Background Technology

[0002] Machine learning technology primarily involves modeling and training data to enable computer systems to automatically extract features and form prediction or decision-making capabilities based on sample data. Core aspects include acquiring and labeling training data, constructing feature representation methods, iteratively optimizing model parameters, and decision-making mechanisms based on output results. It is widely applied in classification, regression, clustering, and personalized recommendation scenarios, and uses statistical analysis and pattern learning of historical behavioral data to infer unknown inputs. Traditional self-matching learning path recommendation methods address the differences in learners' knowledge acquisition processes by generating a learning content sequence for learners through pre-defined rules or statistical methods based on historical learning records. This involves determining the learning path by statistically analyzing learners' answer accuracy, constructing a directed structure based on the sequential dependencies between knowledge points, matching and filtering through the learning records of similar users, and calculating recommendation priorities according to fixed weights. The corresponding course units or knowledge nodes are then progressively pushed to learners in a predetermined order.

[0003] Existing technologies rely on preset rules and historical record statistics to plan the order of learning content. This model only focuses on the accuracy of answering questions and relies on the fixed sequential dependencies of knowledge points to build a directed framework. At the same time, it matches and filters through similar user behavior records and follows fixed weights to calculate recommendation priorities. This makes it impossible to accurately capture the cognitive backtracking behavior in the individual knowledge analysis process when facing complex learning situations. The rigid, predetermined push order easily ignores the potential cognitive loops and path conflicts in the dynamic learning process, making it difficult for the generated learning path to truly fit the individual's real knowledge acquisition patterns. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: an AI-driven adaptive learning path recommendation method, comprising the following steps: S1: The learning terminal device acquires and sorts the knowledge point access order records generated by the learner in the question-answering interaction interface, parses the knowledge point number to form path sequence data, analyzes the path sequence data and establishes the knowledge point order correspondence relationship, and generates a node order set. S2: Based on the node sequence set, obtain the numbers of the repeatedly occurring knowledge points and detect the position intervals, calculate the position interval fitting degree and determine the correlation of the repeated path structure, and generate a closed-loop structure. S3: Perform a proportional conversion calculation on the closed-loop structure, calculate the ratio of the closed-loop length to the total length of the path sequence data and accumulate the conflict ratio, perform weight adjustment analysis in combination with the node sequence set, and generate conflict association results; S4: Adjust the strength of directed edge connections according to the conflict association structure and establish the path constraint correspondence. Obtain the set of directed edges between adjacent knowledge points of the path sequence data through the knowledge graph data structure and read the edge connection relationship to generate the path constraint structure. S5: Perform a screening judgment on the path constraint structure, remove directed edges that do not conform to the path constraint structure and screen candidate paths, calculate the cumulative result of path connection, and generate an adaptive learning path recommendation scheme.

[0005] As a further embodiment of the present invention, the node sequence set includes the preceding knowledge point number, sequential inheritance relationship, and hierarchical position attribute; the closed-loop structure includes repeating node identifier, cyclic span distance, and first-to-last connected node pairs; the conflict association structure includes the accumulated conflict ratio value, attenuation weight coefficient, and contradictory link pointing; the path constraint structure includes the minimum connectivity threshold, edge exclusion rule, and node binding combination; and the adaptive learning path recommendation scheme includes the optimal learning sequence, alternative branch routes, and expected learning step size.

[0006] As a further aspect of the present invention, the path constraint correspondence, after being adjusted in the knowledge graph according to conflict and order weights, is a mapping relationship between the directed edge connection strength and feasible connection rules between knowledge points.

[0007] As a further aspect of the present invention, the path sequence data is an ordered set of paths composed of the access numbers of knowledge points recorded by the learner in chronological order during the question-answering process.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Obtain the knowledge point access order record generated by the learning terminal device, detect the timestamp and perform ascending sorting to form an access sequence, parse the knowledge point number and encode it according to the order of appearance, perform association judgment based on the adjacent positions of the number, and generate an access order encoding sequence. S102: Based on the access order encoding sequence, retrieve adjacent encoding pairs one by one and perform front and back matching operations, establish number pointing relationship according to the matching results, perform deduplication and screening to maintain a unique mapping structure, and establish a sequential correspondence mapping table; S103: Call the sequence mapping table, read the mapping pairs one by one and connect the subsequent numbers according to the starting number to form a chain relationship, perform closed loop detection to eliminate loop paths, establish topological associations between nodes according to the connection relationship, and generate a node sequence set.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Obtain the node sequence set, retrieve the repeated knowledge point numbers and record the corresponding position indexes, calculate the interval values ​​based on the index differences, perform normalization processing to form an interval sequence, and generate a position interval sequence. S202: Based on the position interval sequence, extract the interval values ​​to construct a comparison group, perform difference comparison according to the interval consistency benchmark, calculate the fitting degree of the comparison group and form a correlation result to obtain the interval fitting degree set; S203: Based on the interval fitting degree set, determine the relationship between the fitting degree value and the preset threshold and filter the numbered combination, call the numbered combination to perform the path backtracking connection operation, establish a closed path association relationship, and generate a closed loop structure.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Obtain the total length of the closed-loop structure and path sequence data, retrieve the closed-loop path length and perform ratio calculation to obtain the percentage value, accumulate the percentages one by one to form a ratio superposition result, and generate a conflict ratio sequence. S302: Based on the conflict ratio sequence, extract the ratio values ​​one by one and perform interval division, determine the conflict level according to the preset ratio threshold and establish the mapping relationship between ratio and level to obtain the conflict weight set; S303: Based on the conflict weight set and the node order set, match the node number and weight item by item and perform correction calculation. Based on the correction result, establish the conflict connection relationship between nodes and generate the conflict association.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Obtain the conflict association structure, retrieve the node conflict connection relationship and extract the connection weight value, perform a proportional correction operation based on the weight difference to adjust the edge connection strength, form an edge strength sequence, and generate a connection strength sequence. S402: Based on the connection strength sequence, extract the edge strength values ​​one by one, perform interval classification, determine the constraint level according to the preset strength threshold, and establish the correspondence between the edge and the level to obtain the edge constraint set; S403: Based on the edge constraint set and path sequence data, retrieve the set of directed edges between adjacent knowledge points and read the connection relationship, perform constraint mapping matching to establish the constraint connection relationship between nodes, and generate the path constraint structure.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Obtain the path constraint structure, retrieve directed edge connection relationships and extract edge constraint levels, perform edge-by-edge screening and judgment according to the preset constraint judgment criteria, eliminate directed edges that do not meet the constraint conditions and retain feasible connection relationships, and generate a candidate path set. S502: Based on the candidate path set, extract the node connection sequence for each path and perform path accumulation operation, calculate the connection accumulation value according to the node connection order and form a path score result to obtain a path connection accumulation value set. S503: Based on the cumulative value set of path connections, compare the multi-path score values ​​one by one and perform sorting and selection operations. Combine the path constraint structure to perform path matching verification, establish path priority association relationship, and generate an adaptive learning path recommendation scheme.

[0013] As a further aspect of the present invention, the path priority association is based on the path scoring results and constraint verification, which sorts the candidate learning paths and establishes a correspondence between them in terms of priority recommendation order.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the access order of knowledge points is parsed to construct a node order set, and the positional interval of repeated knowledge point numbers is detected to generate a closed-loop structure. The proportion conversion is performed to calculate the cumulative conflict ratio, and the weight is adjusted in combination with the node order set to generate conflict association results. Based on the conflict association structure, the strength of directed edge connections is adjusted to establish path constraint correspondence, and candidate paths are screened in the knowledge graph data structure. The cumulative results are calculated to generate an adaptive recommendation scheme, which effectively breaks through the inherent dependency relationship and fixed weight limitation, maps the trial and error trajectory in the individual cognitive backtracking process, and deeply analyzes the potential cognitive closed loop and conflict in the dynamic learning process when facing complex learning situations, ensuring that the path is highly consistent with the individual's real knowledge acquisition rules. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides an AI-driven adaptive learning path recommendation method, comprising the following steps: S1: The learning terminal device acquires and sorts the knowledge point access order records generated by the learner in the question-answering interaction interface, parses the knowledge point number to form path sequence data, analyzes the path sequence data and establishes the knowledge point order correspondence relationship, and generates a node order set. S2: Based on the node sequence set, obtain the numbers of repeatedly occurring knowledge points and detect the position intervals, calculate the position interval fitting degree and determine the correlation of repeated path structures to generate a closed-loop structure; S3: Perform a proportional conversion calculation on the closed-loop structure, calculate the ratio of the closed-loop length to the total length of the path sequence data and accumulate the conflict ratio, combine the node sequence set to perform weight adjustment analysis, and generate conflict association results; S4: Adjust the strength of directed edge connections based on the conflict association structure and establish the corresponding relationship of path constraints. Obtain the set of directed edges between adjacent knowledge points of the path sequence data through the knowledge graph data structure and read the edge connection relationship to generate the path constraint structure. S5: Perform filtering and judgment on the path constraint structure, remove directed edges that do not conform to the path constraint structure and filter candidate paths, calculate the cumulative result of path connection, and generate an adaptive learning path recommendation scheme.

[0020] The node sequence set includes the preceding knowledge point number, sequential inheritance relationship, and hierarchical position attribute. The closed-loop structure includes the identifier for repeating nodes, the cyclic span distance, and the first and last connected node pairs. The conflict association structure includes the cumulative conflict ratio, the attenuation weight coefficient, and the contradictory link direction. The path constraint structure refers to the minimum connectivity threshold, the edge exclusion rule, and the node binding combination. The adaptive learning path recommendation scheme includes the optimal learning sequence, alternative branch routes, and the expected learning step size.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Obtain the knowledge point access order record generated by the learning terminal device, detect the timestamp and perform ascending sorting to form an access sequence, parse the knowledge point number and encode it according to the order of appearance, perform association judgment based on the adjacent positions of the number, and generate an access order encoding sequence. The system retrieves user history interaction log files from the learning system database, extracting string-formatted unique identifiers for knowledge points and corresponding 13-bit millisecond-level timestamps of the operations. It parses the log data, verifying the structural integrity of each timestamp entry, and performs multi-dimensional cleanup on entries containing null values, non-numeric characters, or shorter than 13 characters. It reads all cleaned valid timestamp values ​​and uses the system's internal sorting logic component to perform ascending value rearrangement. The first valid timestamp value in the sequence is selected as the benchmark parameter. Left and right scanning cursors are set, and the cursors are continuously moved alternately, moving timestamps with values ​​less than the benchmark parameter to the left and those with values ​​greater to the right, generating a chronological benchmark time series. The corresponding string-formatted unique identifiers for knowledge points are re-aligned and bound according to the positional correspondence of the benchmark time series, establishing the initial access sequence data. Integerization mapping is performed on the string-formatted unique identifiers for knowledge points in the sequence, constructing a global graph mapping dictionary in the system memory area. The unique identifier of the first knowledge point in the initial access sequence is assigned and marked with the integer value 1. Then, the sequence is traversed backwards. If a new unique identifier of a knowledge point not yet appearing in the global graph mapping dictionary is encountered, the largest previously assigned integer value is extracted and incremented by 1 before allocation. If a unique identifier of a knowledge point already existing in the dictionary is encountered, its original integer value is directly retrieved and assigned. After reading the integer sequence code data set generated by the above mapping operation, any two integer sequence codes in adjacent physical positions are extracted from the sequence. The sequence position index parameter value corresponding to the current integer sequence code is obtained, as is the sequence position index parameter value corresponding to the next adjacent integer sequence code. The difference between the sequence position index parameter value corresponding to the next integer sequence code and the sequence position index parameter value corresponding to the current integer sequence code is calculated to obtain the distance between adjacent positions. This process involves extracting the sequence position index parameter value corresponding to the current integer sequence code (12), obtaining the sequence position index parameter value corresponding to the next integer sequence code (13), performing a difference operation between 13 and 12, and calculating the distance between adjacent positions as 1. The advantage of this operation logic is that it quantifies the closeness between knowledge points by directly comparing the differences in the index positions of adjacent nodes. All integer sequence codes with a distance of 1 between adjacent positions are then combined and packaged according to their order of appearance to construct an access order code sequence.

[0022] S102: Based on the access order encoding sequence, retrieve adjacent encoding pairs one by one and perform front and back matching operations. Establish number pointing relationship according to the matching results, perform deduplication and screening to maintain unique mapping structure, and establish sequential correspondence mapping table. Read the entire set of integer data in the access sequence encoding sequence, and extract the encoding pair structure containing two adjacent integer values ​​one by one, following a unidirectional reading order from the leftmost to the rightmost side of the sequence. Perform a bit-by-bit comparison operation at the numerical level on all extracted encoding pairs, retrieve the last integer value in the current first encoding pair, and simultaneously retrieve the first integer value in the immediately adjacent second encoding pair, performing an identity check on the two values. If the two values ​​are completely identical, generate numbered pointer record data including the correspondence between the first and last nodes. Traverse all the initially established numbered pointer record data, extract duplicate mapping records where the first and last node numbers are completely identical, and perform overwrite deletion on the identified duplicate mapping records, so that only one connection record with a unique direction is retained between any two target integer values. Obtain the first node number parameter value and the last node number parameter value after the deduplication operation, set the first node number parameter value as the storage key name, and the last node number parameter value as the storage value, and establish a hash mapping dictionary table structure. For example, if the extracted predecessor node number parameter is 5 and the extracted successor node number parameter is 8, then an associated storage item is created in the hash mapping dictionary structure, directly pointing from 5 to 8. After all the above key-value pair input operations, the hash mapping dictionary structure is output as a sequential mapping table. The advantage of this operation logic is that, through the construction and deduplication of the key-value pair dictionary, invalid and redundant edge data in massive interaction logs is removed.

[0023] S103: Call the sequential mapping table, read the mapping pairs one by one and connect the subsequent numbers according to the starting number to form a chain relationship, perform closed loop detection to eliminate loop paths, establish topological associations between nodes based on the connection relationship, and generate a node sequence set; The generated sequential mapping table is invoked, and the preceding and following node numbers in the dictionary mapping items are read item by item, starting from the first position. Using the first read preceding node number as the starting analysis point, its mapped following node number is found. This following node number is then used as the current preceding search condition to continue the depth-first traversal of the mapping table, appending the newly found node numbers to the end of the sequence, forming a chain-like connection path data composed of multiple nodes. During this downward traversal, a historical query record stack is established to save all existing node numbers in the current chain-like connection path in real time. Before appending the current node number, a comparison operation is performed between the node number to be appended and all values ​​in the historical query record stack. If the node number to be appended already exists in the record stack, a truncation operation is triggered, canceling the last number pointing to the record data that caused the closed loop. After the aforementioned anti-loop processing, all acyclic connection path data is extracted. The numerical values ​​of all independent node numbers in the entire path network are counted to form the total number of node sets. Simultaneously, the number of all directed connections between multiple nodes is counted to form the total number of edge sets. These two parameters are then substituted into the topology graph data structure matrix for instantiation and configuration. Specifically, the total number of node sets is calculated to be 24, and the total number of edge sets is calculated to be 30. The 24 independent nodes and 30 directed edges are then written into the memory structure according to the directed acyclic feature matrix rules. The advantage of this operation is that by introducing a historical record stack comparison mechanism, infinite loop anomalies during graph structure instantiation are prevented in advance. The final output of the configured matrix data set is a node sequence set.

[0024] Please see Figure 3 The specific steps of S2 are as follows: S201: Obtain the node sequence set, retrieve the repeated knowledge point numbers and record the corresponding position index, calculate the interval value based on the index difference, perform normalization processing to form an interval sequence, and generate a position interval sequence. Read the node sequence set data from the previous output and traverse all node number information within the structure matrix. Retrieve the counter component to count the total frequency of multiple node numbers, and extract duplicate node numbers with a current frequency parameter greater than or equal to 2. Extract the absolute order of these duplicate node numbers in their initial access sequence and summarize them into an absolute position array consisting of multiple integers. For a single duplicate node number, extract the two adjacent sequence numbers from its corresponding absolute position array. Calculate the difference between the subsequent and preceding sequence numbers to obtain the absolute interval span parameter value. Perform a linear scaling transformation on all calculated absolute interval span parameter values ​​using a range normalization mechanism. Retrieve the maximum and minimum interval limit parameter values ​​from all data. Subtract the minimum interval limit parameter value from the currently obtained absolute interval span parameter value to obtain the interval deviation data. Subtract the minimum interval limit parameter value from the maximum interval limit parameter value to obtain the global range data. Finally, divide the interval deviation data by the global range data to obtain the normalized position interval parameter value. The process includes obtaining the current absolute interval span parameter value of 15, retrieving the minimum interval limit parameter value of 3, retrieving the maximum interval limit parameter value of 27, performing a difference operation to obtain the interval deviation data of 12, performing a difference operation to obtain the global range data of 24, and performing a ratio division operation between 12 and 24 to obtain the normalized position interval parameter value of 0.50. The advantage of this operation logic is that it unifies the multi-dimensional absolute position spans to the same data evaluation scale through range ratio processing. All normalized position interval parameter values ​​generated by the operation are bound and stored with their corresponding node numbers, and the position interval sequence is integrated and output.

[0025] S202: Based on the positional interval sequence, extract the interval values ​​to construct a comparison group, perform difference comparison according to the interval consistency benchmark, calculate the degree of fit of the comparison group and form a correlation result to obtain the interval fit set; Read all normalized position interval parameter values ​​corresponding to the same node number in the position interval sequence, and integrate the sets of values ​​belonging to the same node number into independent test comparison group data. Set the expected optimal review frequency baseline parameter value, which is obtained by averaging the behavioral data of the top 10% of users in the historical overall pass rate, and fixed at 0.10. Call the variance calculation component to obtain the variance statistic parameter value of multiple normalized position interval parameter values ​​in the current test comparison group. Perform a difference operation between the variance statistic parameter value and the optimal review frequency baseline parameter value to obtain the baseline deviation difference parameter. Perform a mapping process on the baseline deviation difference parameter using the natural exponent transformation rule, that is, use the negative baseline deviation difference parameter value raised to the power of the natural constant as the comparison group fit parameter value for the current node number. The calculation process involves obtaining a variance statistic of 0.20 for a single test comparison group, reading a baseline parameter of 0.10 for the optimal review frequency, performing a difference operation between the two parameters to obtain a baseline deviation difference parameter of 0.10, and substituting 0.10 into the natural exponentiation model to calculate a comparison group fit parameter of approximately 0.904. The advantage of this calculation logic is that by combining variance with the natural exponential decay function, it amplifies the numerical representation of small deviations. The comparison group fit parameter values ​​calculated for all nodes are compiled into a table to generate an interval fit set.

[0026] S203: Based on the interval fit set, determine the relationship between the fit value and the preset threshold and filter the numbered combination. Call the numbered combination to perform the path backtracking connection operation, establish the closed path association relationship, and generate a closed loop structure. Extract all comparison group fit parameter values ​​stored within the interval fit set. Set a preset fit threshold parameter, specifically 0.75. Compare the extracted comparison group fit parameter values ​​with the preset fit threshold parameter of 0.75. Retain node numbers with comparison group fit parameter values ​​greater than or equal to 0.75 in the dataset, and remove node numbers less than 0.75. Retrieve the retained node number combinations and locate the initial position node identifier and subsequent recurrence position node identifiers in the original node sequence set. Establish a reverse directed edge data record between the recurrence position node identifier and the initial position node identifier, pointing from the recurrence position to the initial position. For example, if the retained node number is 19, its initial position node number parameter is 4, and its recurrence position node number parameter is 16, then write a reverse directed edge attribute parameter between node number 16 and node number 4 to construct a path backtracking loop structure. The advantage of this operational logic lies in its intuitive representation of the cross-time-period graph characteristics of repeated memorization of knowledge points by establishing connecting edges in reverse. The topological graph structure, after integrating all data records with added reverse-linked directed edges, outputs a closed-loop structure.

[0027] Please see Figure 4 The specific steps of S3 are as follows: S301: Obtain the total length of the closed-loop structure and path sequence data, retrieve the closed-loop path length and perform ratio calculation to obtain the percentage value, accumulate the percentages one by one to form the ratio superposition result, and generate the conflict ratio sequence. Extract the closed-loop structure data and the original node sequence set data from the previously created dataset. Use a graph traversal and statistical algorithm to obtain the total number of nodes in each independent closed loop of the closed-loop structure, thus obtaining the closed-loop path length parameter. Count the total number of all independent nodes in the original node sequence set to obtain the total path sequence data length parameter. Perform a ratio division operation between each closed-loop path length parameter and the total path sequence data length parameter to calculate the single-cycle ratio. Establish a global proportional accumulator parameter with an initial value of 0.00. Sequentially sum the calculated single-cycle ratio value with the global proportional accumulator parameter, overwriting the original global proportional accumulator parameter value with the new result of each summation, until all independent closed loop data within the closed-loop structure has been traversed and summed, yielding the ratio superposition result parameter value. For example, the closed-loop path length parameter for a single independent closed loop is 5, the total path sequence data length parameter is 50, and the ratio division operation between these two parameters yields a single-cycle ratio value of 0.10. The global proportional accumulator parameter value recorded in the preceding operation phase is extracted to be 0.25. Summing 0.10 and 0.25 yields an updated global proportional accumulator parameter value of 0.35. The advantage of this operational logic is that it quantifies the total scale of repeated learning behavior in the overall learning path by accumulating the proportion of each loop. All sequentially generated global proportional accumulator update results are compiled and recorded to generate a conflict proportion sequence.

[0028] S302: Based on the conflict ratio sequence, extract the ratio values ​​item by item and perform interval division, determine the conflict level according to the preset ratio threshold and establish the mapping relationship between ratio and level to obtain the conflict weight set; Table 1 Parameter Table for Conflict Level Ranges

[0029] Table 1 lists the numerical ranges and corresponding coefficients used to define different grade intervals. Read the multiple ratio superposition result parameter values ​​included in the conflict ratio sequence. Extract the lower and upper limits of the region values ​​from Table 1. Perform interval boundary inclusion judgment operation on the extracted multiple ratio superposition result parameter values ​​and the upper and lower limits values ​​in Table 1. If the extracted ratio superposition result parameter value is greater than or equal to 0.00 and less than or equal to 0.20, assign it a Level 1 conflict label and extract the corresponding penalty coefficient reference value of 1.10. If the extracted ratio superposition result parameter value is greater than 0.20 and less than or equal to 0.50, assign it a Level 2 conflict label and extract the penalty coefficient reference value of 1.30. If the extracted ratio superposition result parameter value is greater than 0.50, assign it a Level 3 conflict label and extract the penalty coefficient reference value of 1.60. Obtain the global baseline conflict constant parameter preset by the system. Perform a multiplication operation on the extracted penalty coefficient reference value and the global baseline conflict constant parameter to obtain the conflict weight numerical parameter. This process includes reading a single proportional summation result parameter value of 0.35, verifying that it falls within the range of 0.21 to 0.50, and extracting the corresponding penalty coefficient reference value of 1.30. The global baseline conflict constant parameter setting value is obtained as 2.00. Multiplying 1.30 and 2.00 yields a conflict weight parameter value of 2.60. The advantage of this operational logic is that, through the combination of interval segmentation comparison and constant multiplication amplification, continuous proportional distribution data is transformed into discrete weight indicators with definite discriminative power. All node numbers are bound to the conflict weight parameters calculated by multiple self-calculations as a pairwise mapping relationship, outputting a conflict weight set.

[0030] S303: Based on the conflict weight set and the node order set, match the node number and weight item by item and perform correction calculation. Based on the correction result, establish the conflict connection relationship between nodes and generate the conflict association result. The process retrieves mapping dictionary data from the conflict weight set and initial Euclidean distance parameters between multiple nodes from the node order set. It obtains the first conflict weight parameter value corresponding to the preceding node with a direct directed connection and the second conflict weight parameter value corresponding to the following node. The arithmetic mean of the first and second conflict weight parameters is calculated to obtain a weight correction parameter value. Then, the weight correction parameter value is multiplied by the initial Euclidean distance parameter between the two nodes to obtain a corrected conflict connection strength parameter. Specifically, the first conflict weight parameter value of the preceding node is 1.50, and the second conflict weight parameter value of the following node is 2.50. The arithmetic mean of these two parameters yields a weight correction parameter value of 2.00. The initial Euclidean distance parameter between the preceding and following nodes is retrieved as 3.00. Multiplying 2.00 by 3.00 yields a corrected conflict connection strength parameter of 6.00. The advantage of this computational logic lies in generating a composite edge strength index that simultaneously considers node weights and connection physical characteristics by multiplying the mean of node attributes with the initial edge distance. The modified inter-node conflict connection strength parameters obtained through the above calculation are then used to multidimensionally cover the distance data in the original node sequence set, reconstructing the graph mapping model and generating conflict association results.

[0031] Please see Figure 5 The specific steps of S4 are as follows: S401: Obtain the conflict association structure, retrieve the conflict connection relationship of the node and extract the connection weight value, perform a proportional correction operation based on the weight difference to adjust the edge connection strength, form an edge strength sequence, and generate a connection strength sequence. Extract the conflict association structure data set output from the previous stages, and traverse and read the attribute data of all independent directed edges in the graph structure model. Extract the value of the conflict connection strength parameter between the currently bound nodes for each directed edge. Perform a location search in the entire graph topology to find all other adjacent directed edges that share the same preceding source node as the current directed edge, and extract the association comparison strength parameter value bound to the first retrieved adjacent directed edge. Subtract the association comparison strength parameter value from the current directed edge's corrected node conflict connection strength parameter value, and perform a difference operation to obtain the strength deviation difference parameter. Extract the system's pre-configured smoothing convergence adjustment coefficient parameter, and multiply the strength deviation difference parameter and the smoothing convergence adjustment coefficient parameter to obtain the strength compensation correction value. Then, sum the strength compensation correction value with the original corrected node conflict connection strength parameter value to obtain the absolute edge connection strength parameter value. The process includes: reading the current directed edge's corrected inter-node conflict connection strength parameter value (6.00); locating and capturing the association comparison strength parameter value of adjacent directed edges originating from the same source node (4.00); performing a difference operation between these two values ​​to obtain a strength deviation difference parameter of 2.00; extracting the smoothing convergence adjustment coefficient parameter set to 0.30; multiplying 2.00 and 0.30 to obtain a strength compensation correction value of 0.60; and summing 0.60 with the base parameter value of 6.00 to obtain an absolute edge connection strength parameter value of 6.60. The advantage of this operational logic is that, through the mathematical processing mechanism of multiplicative compensation for adjacent edge differences, it neutralizes extreme abrupt changes in local node connection weights. All calculated absolute edge connection strength parameter values ​​are arranged and combined according to the edge index order of the graph structure to generate a connection strength sequence.

[0032] S402: Based on the connection strength sequence, extract the edge strength values ​​one by one, perform interval classification, determine the constraint level according to the preset strength threshold, and establish the correspondence between the edge and the level to obtain the edge constraint set; Read the absolute strength parameter values ​​of all edges in the connection strength sequence. The system presets two key constant parameters for segmentation strength: a lower constraint limit parameter value of 4.00 and an upper constraint limit parameter value of 8.00. Extract the absolute strength parameter value of a single edge in the sequence and perform a numerical comparison operation with both the lower and upper constraint limit parameter values. If the parameter value is determined to be strictly less than the lower constraint limit parameter value, assign a level 1 boundary constraint parameter to the edge. If the parameter value is determined to be greater than or equal to the lower constraint limit parameter value and strictly less than the upper constraint limit parameter value, assign a level 2 boundary constraint parameter to the edge. If the parameter value is determined to be greater than or equal to the upper constraint limit parameter value, assign a level 3 boundary constraint parameter to the edge. For example, if the absolute strength parameter value of a single edge is extracted to be 6.60, substitute it into the judgment logic, verify that 6.60 is greater than 4.00 and less than 8.00, and based on the above judgment result, directly assign a level 2 boundary constraint parameter to the connection edge. The advantage of this operational logic lies in its ability to convert continuously distributed intensity values ​​into structured hierarchical index data by setting double fixed boundaries to perform numerical range judgments. All connection edge identifier codes are bound one-to-one with the generated boundary constraint level parameters, and a key-value pair data table is created to output the generated edge constraint set.

[0033] S403: Based on the edge constraint set and path sequence data, retrieve the set of directed edges between adjacent knowledge points and read the connection relationship, perform constraint mapping matching to establish the constraint connection relationship between nodes, and generate the path constraint structure; The system retrieves data recorded within the edge constraint set and simultaneously obtains the original path sequence data document. A sliding data window of length 2 is used to perform a moving scan operation within the original path sequence data document, sequentially reading one-to-one directed connection combinations formed by adjacent start and end knowledge points. For each scanned directed connection combination, the start and end knowledge point identifier codes are extracted and merged to form a composite index string. A precise key-value matching operation is performed on this composite index string within the edge constraint set to retrieve the successfully matched boundary constraint level parameter values. The system obtains the preset cost scaling factor parameter from the global storage space and multiplies the captured boundary constraint level parameter value with the cost scaling factor parameter to obtain the inter-node network constraint cost parameter mapped to this connection combination. For example, if the captured boundary constraint level parameter value is 2, and the preset cost scaling factor parameter setting in the system is 2.50, multiplying 2 and 2.50 yields a mapped inter-node network constraint cost parameter of 5.00. The advantage of this operational logic lies in the fact that, through composite index string matching and multiplicative scaling operations, discrete levels are transformed back into continuous cost weights that can be directly called by graph computation algorithms. The weight values ​​on all directed edges in the structure are then uniformly updated and overlaid with the newly calculated inter-node network constraint cost parameters, and the output is saved to form a complete path constraint structure model.

[0034] Please see Figure 6 The specific steps of S5 are as follows: S501: Obtain the path constraint structure, retrieve directed edge connection relationships and extract edge constraint levels, perform edge-by-edge screening and judgment based on preset constraint judgment criteria, remove directed edges that do not meet the constraint conditions and retain feasible connection relationships, and generate a candidate path set. The complete path constraint structure model, overwritten with the network constraint cost parameters between nodes, is retrieved using the aforementioned steps. A memory traversal interface is called to read all independent directed edge data record objects stored within this structure model. A preset threshold parameter for bottom-line cost interception is extracted; this parameter is fixed at 3.50 based on the lower quartile of past system filtered log data. The network constraint cost parameter value for each directed edge data record object is extracted iteratively and compared with the preset threshold parameter. If the network constraint cost parameter value is greater than or equal to the preset threshold parameter, a structure persistence operation is performed, preserving the directed edge and its two endpoints' connections in the graph model dataset. If the network constraint cost parameter value is strictly less than the preset threshold parameter, a structure culling operation is performed, permanently deleting the directed edge object and its connection index from the graph model's memory stack. The process involves extracting the network constraint cost parameter value between nodes of the first directed edge, setting it to 5.00, and reading the preset threshold parameter value for bottom-line cost interception, which is set to 3.50. If the two parameters are compared and the logical condition that 5.00 is greater than 3.50, the system triggers a structure-retention operation to retain the edge. The advantage of this operation is that it forcibly filters and removes inefficient and redundant links within the model by setting an absolute value comparison rule. After completing the comparison and pruning operation for all directed edges in the entire graph, a connected component traversal operation is performed on the remaining node network to find all connected path data groups that can run from the starting point to the ending point, and these are packaged and output as a candidate path set.

[0035] S502: Based on the candidate path set, extract the node connection sequence for each path and perform path accumulation operation. Calculate the cumulative connection value according to the node connection order and form the path score result to obtain the path connection cumulative value set. The process involves unpacking and extracting each connected path sequence data group stored in the candidate path set. For a single path sequence data group, starting from the starting node and following the direction of the directed edges, all directed edge data objects associated with the path sequence are read one by one, and the network constraint cost parameter values ​​between nodes attached to these directed edge data objects are extracted. The initial cumulative total score parameter value for a single path is set to 0.00 in the calculation program. The first extracted network constraint cost parameter value between nodes is summed with the initial cumulative total score parameter value, and the calculated value overwrites the original initial cumulative total score parameter value. This process is repeated for the second parameter value and the current cumulative total score parameter value until all network constraint cost parameter values ​​between nodes included in the single path sequence are summed, thus obtaining the cumulative connection score value corresponding to a single connected path. For example, a single connected path sequence includes three directed edges, with network constraint cost parameter values ​​between nodes of 5.00, 4.20, and 6.10 respectively. The initial parameter value of the cumulative total score is extracted as 0.00. Three consecutive summation operations are performed sequentially, resulting in a cumulative connection score of 15.30 for the connected path. The advantage of this operation logic is that it integrates the comprehensive value of the global path in terms of learning smoothness through sequential iterative summation. After the calculation engine has completed the summation of all paths in the set, the system identifier codes of the multiple paths and their calculated final cumulative connection scores are paired into data pairs, which are then aggregated into a set of cumulative path connection values.

[0036] S503: Based on the cumulative value set of path connections, compare the multi-path score values ​​one by one and perform sorting and selection operations. Combine the path constraint structure to perform path matching verification, establish path priority associations, and generate an adaptive learning path recommendation scheme. The system accesses the cumulative connection value data storage area and reads all path system identifiers and their associated cumulative connection score data pairs. It then initiates a descending sorting algorithm engine, performing pairwise comparisons and position swaps on all cumulative connection scores, pushing the record with the highest score to the head of the sequence, and sorting the entire sequence in descending order of score. It extracts the top 3 candidate path record data pairs from the top of the sequence after the descending sort. It then extracts the pre-set dynamic priority decay coefficient parameter. Finally, it multiplies the cumulative connection scores of each record in the top 3 data pairs with the dynamic priority decay coefficient parameter, calculating and outputting the corrected multi-factor priority score parameters. For example, the cumulative connection score of the first path in the first position is 15.30, and the cumulative connection score of the second path in the second position is 14.80. The set dynamic priority decay coefficient parameter is read as 0.95. Multiplying 15.30 with 0.95 yields a reduced priority score parameter of 14.535 for the first path. Multiplying 14.80 with 0.95 yields a reduced priority score parameter of 14.060 for the second path. The advantage of this operation is that by extracting the top three high-order values ​​and applying a decay multiplication verification mechanism, it mitigates the randomness of recommendations induced by an extreme single highest score. Based on the newly generated reduced priority score parameters, the first three paths are again compared and rearranged in descending order to generate a sequence code. This sequence code is then combined with a node sequence document containing knowledge points from multiple paths and packaged for output, generating an adaptive learning path recommendation scheme data document.

[0037] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention 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 the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the technical solution.

Claims

1. An AI-driven adaptive learning path recommendation method, characterized in that, Includes the following steps: S1: The learning terminal device acquires and sorts the knowledge point access order records generated by the learner in the question-answering interaction interface, parses the knowledge point number to form path sequence data, analyzes the path sequence data and establishes the knowledge point order correspondence relationship, and generates a node order set. S2: Based on the node sequence set, obtain the numbers of the repeatedly occurring knowledge points and detect the position intervals, calculate the position interval fitting degree and determine the correlation of the repeated path structure, and generate a closed-loop structure. S3: Perform a proportional conversion calculation on the closed-loop structure, calculate the ratio of the closed-loop length to the total length of the path sequence data and accumulate the conflict ratio, perform weight adjustment analysis in combination with the node sequence set, and generate conflict association results; S4: Adjust the strength of directed edge connections according to the conflict association structure and establish the path constraint correspondence. Obtain the set of directed edges between adjacent knowledge points of the path sequence data through the knowledge graph data structure and read the edge connection relationship to generate the path constraint structure. S5: Perform a screening judgment on the path constraint structure, remove directed edges that do not conform to the path constraint structure and screen candidate paths, calculate the cumulative result of path connection, and generate an adaptive learning path recommendation scheme.

2. The AI-driven adaptive learning path recommendation method according to claim 1, characterized in that, The node sequence set includes the preceding knowledge point number, sequential inheritance relationship, and hierarchical position attribute. The closed-loop structure includes repeating node identifiers, cyclic span distance, and first-to-last connected node pairs. The conflict association structure includes the accumulated conflict ratio value, attenuation weight coefficient, and contradictory link direction. The path constraint structure refers to the minimum connectivity threshold, edge exclusion rule, and node binding combination. The adaptive learning path recommendation scheme includes the optimal learning sequence, alternative branch routes, and expected learning step size.

3. The AI-driven adaptive learning path recommendation method according to claim 1, characterized in that: The path constraint correspondence, after being adjusted according to conflict and order weights in the knowledge graph, represents the mapping relationship between the directed edge connection strength and feasible connection rules between knowledge points.

4. The AI-driven adaptive learning path recommendation method according to claim 1, characterized in that: The path sequence data is an ordered set of paths formed by the access numbers of knowledge points recorded by the learner in chronological order during the question-answering process.

5. The AI-driven adaptive learning path recommendation method according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain the knowledge point access order record generated by the learning terminal device, detect the timestamp and perform ascending sorting to form an access sequence, parse the knowledge point number and encode it according to the order of appearance, perform association judgment based on the adjacent positions of the number, and generate an access order encoding sequence. S102: Based on the access order encoding sequence, retrieve adjacent encoding pairs one by one and perform front and back matching operations, establish number pointing relationship according to the matching results, perform deduplication and screening to maintain a unique mapping structure, and establish a sequential correspondence mapping table; S103: Call the sequence mapping table, read the mapping pairs one by one and connect the subsequent numbers according to the starting number to form a chain relationship, perform closed loop detection to eliminate loop paths, establish topological associations between nodes according to the connection relationship, and generate a node sequence set.

6. The AI-driven adaptive learning path recommendation method according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Obtain the node sequence set, retrieve the repeated knowledge point numbers and record the corresponding position indexes, calculate the interval values ​​based on the index differences, perform normalization processing to form an interval sequence, and generate a position interval sequence. S202: Based on the position interval sequence, extract the interval values ​​to construct a comparison group, perform difference comparison according to the interval consistency benchmark, calculate the fitting degree of the comparison group and form a correlation result to obtain the interval fitting degree set; S203: Based on the interval fitting degree set, determine the relationship between the fitting degree value and the preset threshold and filter the numbered combination, call the numbered combination to perform the path backtracking connection operation, establish a closed path association relationship, and generate a closed loop structure.

7. The AI-driven adaptive learning path recommendation method according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Obtain the total length of the closed-loop structure and path sequence data, retrieve the closed-loop path length and perform ratio calculation to obtain the percentage value, accumulate the percentages one by one to form a ratio superposition result, and generate a conflict ratio sequence. S302: Based on the conflict ratio sequence, extract the ratio values ​​one by one and perform interval division, determine the conflict level according to the preset ratio threshold and establish the mapping relationship between ratio and level to obtain the conflict weight set; S303: Based on the conflict weight set and the node order set, match the node number and weight item by item and perform correction calculation. Based on the correction result, establish the conflict connection relationship between nodes and generate the conflict association.

8. The AI-driven adaptive learning path recommendation method according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Obtain the conflict association structure, retrieve the node conflict connection relationship and extract the connection weight value, perform a proportional correction operation based on the weight difference to adjust the edge connection strength, form an edge strength sequence, and generate a connection strength sequence. S402: Based on the connection strength sequence, extract the edge strength values ​​one by one, perform interval classification, determine the constraint level according to the preset strength threshold, and establish the correspondence between the edge and the level to obtain the edge constraint set; S403: Based on the edge constraint set and path sequence data, retrieve the set of directed edges between adjacent knowledge points and read the connection relationship, perform constraint mapping matching to establish the constraint connection relationship between nodes, and generate the path constraint structure.

9. The AI-driven adaptive learning path recommendation method according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Obtain the path constraint structure, retrieve directed edge connection relationships and extract edge constraint levels, perform edge-by-edge screening and judgment according to the preset constraint judgment criteria, eliminate directed edges that do not meet the constraint conditions and retain feasible connection relationships, and generate a candidate path set. S502: Based on the candidate path set, extract the node connection sequence for each path and perform path accumulation operation, calculate the connection accumulation value according to the node connection order and form a path score result to obtain a path connection accumulation value set. S503: Based on the cumulative value set of path connections, compare the multi-path score values ​​one by one and perform sorting and selection operations. Combine the path constraint structure to perform path matching verification, establish path priority association relationship, and generate an adaptive learning path recommendation scheme.

10. The AI-driven adaptive learning path recommendation method according to claim 9, characterized in that: The path priority association, based on path scoring results and constraint verification, sorts candidate learning paths and establishes a correspondence for priority recommendation order.

Images