Personalized learning method and system based on knowledge graph

Abstract

The application relates to a knowledge graph-based personalized learning method and system, and relates to the field of quality education technology, which comprises the following steps: step 100: collecting identity information of a user, and calling a knowledge node from a preset knowledge graph; step 101: determining an identity weight in response to the identity information and the knowledge node, and calling a user's answering parameter based on the knowledge node; step 102: determining a passing coefficient in response to the answering parameter, and calling a basic difficulty score from the preset knowledge graph based on the knowledge node; step 103: determining a dynamic difficulty score by combining the basic difficulty score, the identity weight and the passing coefficient; and step 104: updating the knowledge graph according to the dynamic difficulty score. The application has the effects of improving the learning efficiency of the user and quickly determining the difficulty of each knowledge point.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of quality education technology, and particularly to a personalized learning method and system based on a knowledge graph. BACKGROUND

[0002] A knowledge graph is a directed graph formed by connecting different knowledge points as nodes through relationships.

[0003] In the prior art, learning platforms generally use an algorithm tag (such as "breadth-first search"), a difficulty level (popular / improved) static classification of question banks, and require users to select practice content themselves. The platform provides a corresponding preset question list (such as "100 introductory questions") according to the user's selection and records the completion status of the questions.

[0004] When the user's selected knowledge content does not match the user's current learning situation, it is easy to cause the user to have difficulty learning the knowledge content, thereby resulting in a low learning efficiency of the user. SUMMARY

[0005] In order to improve the learning efficiency of the user and quickly determine the difficulty of each knowledge point, the present application provides a personalized learning method and system based on a knowledge graph.

[0006] In a first aspect, the present application provides a personalized learning method based on a knowledge graph, which adopts the following technical solution:

[0007] A personalized learning method based on a knowledge graph, comprising:

[0008] Step 100: Collect the identity information of the user and retrieve the knowledge nodes from the preset knowledge graph;

[0009] Step 101: Determine the identity weight in response to the identity information and the knowledge nodes, and retrieve the user's answer parameters based on the knowledge nodes;

[0010] Step 102: Determine the passing coefficient in response to the answer parameters, and retrieve the basic difficulty score from the preset knowledge graph based on the knowledge nodes;

[0011] Step 103: Determine the dynamic difficulty score by combining the basic difficulty score, the identity weight, and the passing coefficient;

[0012] Step 104: Update the knowledge graph according to the dynamic difficulty score.

[0013] By adopting the technical scheme, the system maintains a standardized difficulty score for each knowledge point according to the answering performance of the user at different knowledge points, so as to reflect the mastering difficulty of the user for different knowledge points, and then facilitate the user to filter out the knowledge points conforming to the situation of the user for learning, and improve the learning efficiency of the user.

[0014] Optionally, the learning path planning method further comprises:

[0015] Step 200: collecting a learning target;

[0016] Step 201: determining a necessary node in response to the learning target, and determining a starting node in response to the answering parameter;

[0017] Step 202: determining a learning path and a path difficulty score in combination with the starting node, the necessary node, and a preset knowledge graph;

[0018] Step 203: calculating an adjacent difficulty difference according to the path difficulty score;

[0019] Step 204: when the adjacent difficulty difference is greater than a preset buffer threshold, determining a step node in combination with the adjacent difficulty difference and the learning path;

[0020] Step 205: determining a preposition node and a preposition difficulty score from the preset knowledge graph based on the step node, and determining a step difficulty interval based on the step node and the path difficulty score;

[0021] Step 206: if the preposition difficulty score falls into the step difficulty interval, updating the learning path according to the preposition node.

[0022] By adopting the technical scheme, the path of learning knowledge points is automatically planned according to the learning target of the user, and the difficulty score difference between each two adjacent knowledge points is verified, so that other knowledge points are inserted as transitions to reduce the difficulty increasing amplitude when the difficulty score difference is too large, the situation that the user learns knowledge points is difficult due to too high difficulty increase is reduced, and the learning efficiency of the user is improved.

[0023] Optionally, the learning path planning method further comprises:

[0024] Step 207: if the preposition difficulty score falls into the step difficulty interval, determining a preposition difficulty difference according to the preposition difficulty score and the path difficulty score;

[0025] Step 208: when the preposition difficulty difference is greater than a preset buffer threshold, determining a recheck node and a recheck difficulty score from the preset knowledge graph based on the preposition node;

[0026] Step 209: If the recheck difficulty score falls into a step difficulty interval, determining a recheck difficulty difference according to the recheck difficulty score, the pre difficulty score and the path difficulty score;

[0027] Step 210: When the recheck difficulty difference is not greater than a preset buffer threshold, updating the learning path according to the recheck node and the pre node.

[0028] By using the above technical solution, when the difficulty gap is still too high after introducing other knowledge points, the number of transition knowledge points is increased, and the knowledge points are sequentially referenced in the direction close to the root node in the knowledge graph until a suitable difficulty score difference is reached, thereby improving the learning efficiency of the user.

[0029] Optionally, the learning path planning method further comprises:

[0030] Step 211: Determining an insertion number according to the pre node and the recheck node;

[0031] Step 212: If the insertion number is greater than 1, determining a difficulty variance according to the adjacent difficulty difference;

[0032] Step 213: Determining a learning number according to the learning path;

[0033] Step 214: Determining a learning efficiency in combination with the difficulty variance and the learning number, and determining a coverage number according to the necessary node and the learning path;

[0034] Step 215: Determining a node number according to the necessary node;

[0035] Step 216: Calculating a quotient of the coverage number and the node number, and defining the quotient as a coverage rate;

[0036] Step 217: Determining a path score in combination with the learning efficiency and the coverage rate, and selecting a learning path based on the path score.

[0037] By using the above technical solution, when different transition knowledge points can all achieve the effect of reducing the difficulty score difference, the learning paths formed by introducing different transition knowledge points are evaluated according to the distribution of knowledge points in the learning paths and the smoothness of the difficulty score improvement, so that the learning path with the highest knowledge point coverage rate and the most stable difficulty score improvement is selected.

[0038] Optionally, the method further comprises a question list generation method, and the question list generation method comprises:

[0039] Step 300: Retrieving a custom question list, and reading a collection amount, a pass rate and a freshness from the custom question list;

[0040] Step 301: determining the order form score according to the collection amount, passing rate and freshness;

[0041] Step 302: if the order form score is greater than the preset high-quality threshold, determining the order form proportion according to the passing coefficient;

[0042] Step 303: generating the personalized order form according to the order form proportion, the self-defined order form and the preset basic order form.

[0043] By using the above technical solution, the order forms uploaded by other users are evaluated from three dimensions of collection amount, passing rate and freshness, so as to filter out high-quality order forms, and appropriate order forms are selected according to the learning situation of the user to form a structure, and then the personalized order form suitable for the current user is formed by the basic order form and the high-quality order form uploaded by other users.

[0044] Optionally, the order form generation method further comprises:

[0045] Step 304: collecting the order form result based on the personalized order form;

[0046] Step 305: determining the answer node and the corresponding node accuracy rate based on the order form result;

[0047] Step 306: if the node accuracy rate is lower than the preset lag threshold, determining the proportion of the question type according to the node accuracy rate;

[0048] Step 307: updating the personalized order form according to the proportion of the question type, and determining the answer sequence according to the order form result;

[0049] Step 308: determining the sequence accuracy rate based on the answer sequence;

[0050] Step 309: if the sequence accuracy rate is lower than the preset increase point threshold, determining the insertion node according to the answer sequence and the preset knowledge graph;

[0051] Step 310: updating the personalized order form according to the insertion node.

[0052] By using the above technical solution, when the user's accuracy rate for a certain knowledge point is low, the proportion of related questions in the order form is reduced, and when the user makes some fixed error patterns, the basic knowledge related to the knowledge point is retrieved from the knowledge graph to supplement the user's understanding of the pre-knowledge of complex knowledge.

[0053] Optionally, it further comprises a shielding recognition method, and the shielding recognition method comprises:

[0054] Step 400: determining the error position in response to the test paper result and the personalized test paper if the node accuracy is lower than a preset falling behind threshold and / or the sequence accuracy is lower than a preset increasing point threshold;

[0055] Step 401: determining the error frequency based on the error position;

[0056] Step 402: calling the error text of the question in combination with the error position and the personalized test paper when the error frequency is greater than a preset repetition threshold;

[0057] Step 403: determining the overlapping area according to the error text;

[0058] Step 404: updating the personalized test paper in response to the overlapping area.

[0059] By using the above technical solution, when the user frequently makes mistakes at the same position, it is judged that the display device has a problem at the position, which leads to incomplete display of the text of the question. At this time, the text distribution of the error question at the position in the test paper is called to determine the overlapping area of the text as the range of the display device failure, and the text distribution of the questions in the test paper is dynamically adjusted to reduce the incomplete display of the questions.

[0060] Optionally, the occlusion identification method further comprises:

[0061] Step 405: collecting the answer image when the error frequency is greater than the preset repetition threshold;

[0062] Step 406: identifying the answer time from the answer image based on the error position, and determining the answer threshold according to the node accuracy and the sequence accuracy;

[0063] Step 407: calling the correct text of the question in combination with the error position and the personalized test paper if the answer time is lower than the answer threshold;

[0064] Step 408: determining the combination area according to the correct text;

[0065] Step 409: determining the occlusion area in combination with the combination area and the overlapping area;

[0066] Step 410: updating the personalized test paper in response to the occlusion area.

[0067] By using the above technical solution, when the user frequently makes mistakes at the same position, the time for the user to handle the error question is detected, and it is judged that the user directly skips the question when the time spent by the user is too low, i.e. the user cannot handle the question. At this time, it is judged that the text of the question is not completely displayed, so as to narrow the overlapping area and further reduce the degree of adjustment of the test paper.

[0068] Optionally, the occlusion identification method further comprises:

[0069] Step 411: If the answering duration is lower than the answering threshold, identifying the constraint feature from the correct text;

[0070] Step 412: determining the stem region based on the constraint feature, and identifying the correct answer and the incorrect answer from the correct text;

[0071] Step 413: defining the correct answer as the answer feature when the incorrect answer coincides with the overlapping region;

[0072] Step 414: updating the combined region in combination with the stem region and the answer feature.

[0073] By adopting the above technical solution, when the user frequently makes mistakes at the same position, the question that the user does not skip in the same position is identified, thereby locating the text feature that has a constraint effect on the answer result in the question that the user does not skip and the answer position in the question, and then the display region of the question is reduced, and the case that the occlusion range is too small is reduced.

[0074] In a second aspect, the application provides a personalized learning system based on a knowledge graph, which adopts the following technical solution:

[0075] A personalized learning system based on a knowledge graph, comprising:

[0076] A collection module for collecting identity information, learning goals, test results and answer images;

[0077] A memory for storing the program of any of the above personalized learning methods based on a knowledge graph;

[0078] A processor, the program in the memory can be loaded and executed by the processor.

[0079] By adopting the above technical solution, the system maintains a standardized difficulty score for each knowledge point according to the user's answering performance at different knowledge points, thereby reflecting the user's mastery difficulty of different knowledge points, and then facilitating the user to filter out the knowledge points that meet their own conditions for learning, and improving the user's learning efficiency.

[0080] In summary, the application includes at least one of the following beneficial technical effects:

[0081] 1. The system maintains a standardized difficulty score for each knowledge point according to the user's answering performance at different knowledge points, thereby reflecting the user's mastery difficulty of different knowledge points, and then facilitating the user to filter out the knowledge points that meet their own conditions for learning, and improving the user's learning efficiency;

[0082] 2. Automatically plan a learning knowledge point path according to the user's learning goal, and check the difficulty score gap between each two adjacent knowledge points, so as to insert other knowledge points as transition to reduce the difficulty score gap when the difficulty score gap is too large, reduce the case that the user learns knowledge points more difficult due to too high difficulty, and improve the user's learning efficiency;

[0083] 3. When different transition knowledge points can be introduced to achieve the effect of reducing the difficulty score gap, evaluate the learning path according to the distribution of knowledge points in the learning path formed by introducing different transition knowledge points and the smoothness of the difficulty score increase, so as to select the learning path with the highest knowledge point coverage and the most stable difficulty score increase. BRIEF DESCRIPTION OF DRAWINGS

[0084] Figure 1 is a flowchart of a personalized learning method based on a knowledge graph.

[0085] Figure 2 is a flowchart of a learning path planning method.

[0086] Figure 3 is a flowchart of a question list generation method. DETAILED DESCRIPTION

[0087] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0088] Referring to Figure 1 A personalized learning method based on a knowledge graph, comprising:

[0089] Step 100: Collect the user's identity information, and retrieve the knowledge nodes from the pre-set knowledge graph.

[0090] The identity information refers to the information used to show the user's benchmark level, which can use information such as age, grade, past competition award records, target event level (such as school competition, provincial competition, national competition, IOI, etc.), etc. The identity information can be input by the user, and the collection method of the identity information is selected by the staff according to the actual situation, which is not described here.

[0091] The knowledge graph refers to structuring the knowledge points in the field of programming algorithm, and constructing a directed graph form of knowledge graph. The graph takes "knowledge points" as the basic unit, and regards knowledge points such as "recursion", "BFS", "KMP algorithm" as nodes (Node) in the graph. The nodes are connected by directed edges (Edge) to represent the prerequisite relationship or knowledge dependency relationship, forming a knowledge network with strong logical depth and progression. Generally, Neo4j graph database is used to store the knowledge graph.

[0092] The construction of the knowledge graph follows the hierarchical design of the knowledge system in the algorithm competition field, and is divided into several core fields, including but not limited to: basic algorithms, data structures, dynamic programming, graph theory, search algorithms, string algorithms, number theory, computational geometry, mathematical theory and competition skills, etc. Under each core field, specific knowledge points are subdivided, for example: "basic algorithms" include simulation, enumeration, recursion, and greed; "data structures" include stack, queue, union-find set, and line segment tree.

[0093] The knowledge node is each knowledge point in the knowledge graph. The method for retrieving the knowledge node is well known in the art and will not be repeated here.

[0094] Step 101: Determine the identity weight in response to the identity information and knowledge node, and retrieve the user's answer parameters based on the knowledge node.

[0095] The identity weight refers to a numerical value that represents the importance of different knowledge nodes under the current user's identity. The more important the knowledge node, the higher the identity weight. The identity weight corresponding to the identity information and the knowledge node can be queried from the identity relationship table. The identity relationship table refers to a data table that records different identity information and knowledge nodes and their corresponding identity weights.

[0096] The answer parameter refers to the question answer record related to the knowledge node in the user's historical answer record. The answer parameter includes the user's answer record in daily practice and questions, and also includes the user's score in the stage simulation test or formal competition (such as NOI, ICPC, Blue Bridge Cup, etc.). The method for retrieving the answer parameter is well known in the art and will not be repeated here.

[0097] Step 102: Determine the passing coefficient in response to the answer parameter, and retrieve the basic difficulty score from the preset knowledge graph based on the knowledge node.

[0098] The passing coefficient refers to the correctness rate of the user's answers in the answer parameters. The passing coefficient includes the correctness rate of the user in daily practice and the questions, and the passing coefficient also includes the performance normalized score of the user in the stage simulation test or formal competition (such as NOI, ICPC, Blue Bridge Cup, etc.). The performance normalized score refers to the score obtained by normalizing the score of the user in different tests or competitions. The determination method of the passing coefficient and the performance normalized score is well known in the art, and will not be described here.

[0099] The basic difficulty score refers to scoring each knowledge according to the following multiple dimensions: 1. Concept complexity (for example, whether it involves multiple recursion, mathematical reasoning, etc.); 2. Coding difficulty; 3. Student's general mastery (according to the student's questionnaire); 4. Coupling degree with other knowledge points (as a basis for dependency); 5. Frequency and influence in competitions (number of times in domestic and international competitions, score ratio). The system will weight and integrate the above dimensions to form a standardized score (for example, 0-100 points). Each knowledge node in the knowledge graph contains its corresponding basic difficulty score.

[0100] The system can also analyze the actual dependency relationship between knowledge points according to the learning path data of a large number of users (such as the order of completing the questions, the order of learning time, and the knowledge point mastery record). For example, if a large number of users have generally mastered "tree array" before mastering "segment tree", the directed edge weight between the two can be enhanced; if "binary answer" and "greedy" are often alternately arranged in the user's learning path, the system can introduce a weak dependency edge and mark it as "recommended order" rather than "strong dependency"; the dependency strength can be modeled using conditional probability, such as P(mastery B | mastery A) being higher than a certain threshold, which is determined as an effective dependency.

[0101] Step 103: Determine the dynamic difficulty score in combination with the basic difficulty score, the identity weight, and the passing coefficient.

[0102] The dynamic difficulty score refers to the difficulty score after the basic difficulty score is corrected according to the identity weight and the passing coefficient. It can be calculated according to the formula D_new = α * D_prev + β * (1 - correctness rate) + γ * (performance normalized score) + δ * (identity weight), where D_new is the required dynamic difficulty score, D_prev is the basic difficulty score, and α, β, γ, and δ are system adjustment parameters, which can be initially set to 0.25 and can be optimized through continuous learning.

[0103] Step 104: Update the knowledge graph according to the dynamic difficulty score.

[0104] The system maintains a standardized difficulty score for each knowledge point according to the user's performance in answering questions at different knowledge points, reflecting the user's mastery of different knowledge points, and facilitating the user to select knowledge points suitable for their own situation for learning, improving the user's learning efficiency.

[0105] Referring to Figure 2 , the learning path planning method:

[0106] Step 200: Collect learning goals.

[0107] The learning goal refers to the user's learning goal (such as "through the NOIP popular group" "complete LeetCode classic 100 questions"), and the learning goal can be input by the user. The method of collecting identity information is selected by the staff according to the actual situation, which is not described here.

[0108] Step 201: Determine the necessary node in response to the learning goal, and determine the starting node in response to the answering parameter.

[0109] The necessary node refers to all knowledge nodes needed to achieve the learning goal, which can be queried from the target relationship table corresponding to the learning goal. The target relationship table refers to a data table recording different learning goals and their corresponding necessary nodes.

[0110] The starting node refers to all knowledge nodes that the user has mastered at present, that is, the user's current ability level. The knowledge nodes with a correct rate higher than 0.9 in the answering parameter can be used as the starting node. The method of determining the starting node is selected by the staff according to the actual situation, which is not described here.

[0111] Step 202: Determine the learning path and path difficulty score in combination with the starting node, necessary node and preset knowledge graph.

[0112] The learning path refers to the shortest path from the starting node to the necessary node in the knowledge graph. Generally, A* algorithm is used as the core algorithm for path planning. The planning method of the learning path is known to those skilled in the art, and is not described here.

[0113] The path difficulty score is the dynamic difficulty score corresponding to each knowledge node in the learning path. The method for determining the path difficulty score is known to those skilled in the art, and is not described here.

[0114] Step 203: Calculate the adjacent difficulty difference according to the path difficulty score.

[0115] The adjacent difficulty difference refers to the difference between the dynamic difficulty scores of adjacent knowledge nodes in the learning path. The adjacent difficulty difference shows the difficulty change of the learning path. The calculation method of the adjacent difficulty difference is known to those skilled in the art, and is not described here.

[0116] Step 204: When the adjacent difficulty difference is greater than the preset buffer threshold, a step node is determined in combination with the adjacent difficulty difference and the learning path.

[0117] The buffer threshold refers to the maximum adjacent difficulty difference in which the user can learn more efficiently. Generally, 0.3 is used as the buffer threshold, and the buffer threshold is selected by the staff according to the actual situation, which will not be described here.

[0118] The adjacent difficulty difference greater than the buffer threshold represents that the difficulty span in the learning path is too large, and the user is more difficult to learn at this time. The step node refers to the knowledge node located on the side away from the root node of the two knowledge nodes with the adjacent difficulty difference greater than the buffer threshold. The determination method of the step node is known to those skilled in the art, which will not be described here.

[0119] Step 205: Based on the step node, the prenode and the pre-difficulty score are determined from the preset knowledge graph, and the step difficulty interval is determined based on the step node and the path difficulty score.

[0120] The prenode refers to all knowledge nodes in the knowledge graph located on the side close to the root node of the step node. The pre-difficulty score is the dynamic difficulty score corresponding to the prenode. The determination method of the prenode and the pre-difficulty score is known to those skilled in the art, which will not be described here.

[0121] The step difficulty interval refers to the difficulty span generated by the step node, i.e. the range from D(A) to D(B), wherein B is the step node, A is the direct predecessor node of the step node in the learning path, and D(x) is the dynamic difficulty score of the knowledge node x.

[0122] Step 206: If the pre-difficulty score falls into the step difficulty interval, the learning path is updated according to the prenode.

[0123] The pre-difficulty score falling into the step difficulty interval represents that the difficulty span of the step node can be reduced after inserting the prenode. At this time, the prenode is inserted into the learning path as the direct predecessor node of the step node to form a new learning path. If there are multiple prenodes with pre-difficulty scores falling into the step difficulty interval, multiple learning paths are formed.

[0124] According to the user's learning goal, the learning knowledge point path is automatically planned, and the difficulty score difference between each two adjacent knowledge points is verified, so that other knowledge points are inserted as transitions when the difficulty score difference is too large to reduce the difficulty increase amplitude, reduce the case that the user learns knowledge points more difficult due to too high difficulty increase, and improve the user's learning efficiency.

[0125] The learning path planning method further comprises:

[0126] Step 207: If the pre-node difficulty score falls into the step difficulty interval, determining a pre-node difficulty difference according to the pre-node difficulty score and the path difficulty score.

[0127] The pre-node difficulty difference refers to the difference between the dynamic difficulty scores of the pre-node and its direct predecessor node and the difference between the pre-node and the step node. The calculation method of the pre-node difficulty difference is known to those skilled in the art, and will not be described here.

[0128] Step 208: When the pre-node difficulty difference is greater than the preset buffer threshold, determining a recheck node and a recheck difficulty score based on the pre-node from the preset knowledge graph.

[0129] The pre-node difficulty difference greater than the buffer threshold represents that the difficulty span of the learning path after inserting the pre-node is still too large. The recheck node refers to all knowledge nodes in the knowledge graph located on the side of the pre-node close to the root node. The recheck difficulty score is the dynamic difficulty score corresponding to the recheck node. The determination method of the recheck node and the recheck difficulty score is known to those skilled in the art, and will not be described here.

[0130] Step 209: If the recheck difficulty score falls into the step difficulty interval, determining a recheck difficulty difference according to the recheck difficulty score, the pre-node difficulty score and the path difficulty score.

[0131] The recheck difficulty score falling into the step difficulty interval represents that the difficulty span of the step node can be reduced after inserting the pre-node. The recheck difficulty difference refers to the difference between the dynamic difficulty scores of the recheck node and its direct predecessor node, the pre-node and the recheck, and the pre-node and the step node. The calculation method of the recheck difficulty difference is known to those skilled in the art, and will not be described here.

[0132] Step 210: When the recheck difficulty difference is not greater than the preset buffer threshold, updating the learning path according to the recheck node and the pre-node.

[0133] The recheck difficulty difference not greater than the buffer threshold represents that the difficulty span of the learning path is reduced after inserting the recheck node and the pre-node. At this time, the recheck node and the pre-node are inserted into the original learning path to form a new learning path. If the recheck difficulty difference is greater than the buffer threshold, the above steps are repeated to continue to increase the inserted knowledge nodes to reduce the difficulty span of the learning path.

[0134] When the difficulty difference is still too high after introducing other knowledge points, the number of transition knowledge points is increased, and the knowledge points are sequentially referenced in the direction close to the root node in the knowledge graph until a suitable difficulty score difference is reached, thereby improving the learning efficiency of the user.

[0135] The learning path planning method further comprises:

[0136] Step 211: determining the insertion number according to the pre-node and the check node.

[0137] The insertion number refers to the number of new learning paths formed after inserting the pre-node and / or the check node in the original learning path, i.e. the product of the number of pre-nodes and the number of check nodes. The determination method of the insertion number is well known to those skilled in the art, and will not be described here.

[0138] Step 212: if the insertion number is greater than 1, determining the difficulty variance according to the adjacent difficulty difference.

[0139] The insertion number greater than 1 represents that there are multiple learning paths. The difficulty variance refers to the variance value of the adjacent difficulty difference of each learning path. The difficulty variance shows the smoothness of the dynamic difficulty score increase in the learning path. The smaller the difficulty variance is, the more stable the dynamic difficulty score increase is. The calculation method of the difficulty variance is well known to those skilled in the art, and will not be described here.

[0140] Step 213: determining the learning number according to the learning path.

[0141] The learning number refers to the number of knowledge nodes contained in the learning path. The determination method of the learning number is well known to those skilled in the art, and will not be described here.

[0142] Step 214: determining the learning efficiency in combination with the difficulty variance and the learning number, and determining the coverage number according to the necessary node and the learning path.

[0143] The learning efficiency refers to a numerical value for showing the difficulty of the user learning according to the learning path. The smaller the difficulty variance and the learning number are, the greater the learning efficiency is. The learning efficiency corresponding to the difficulty variance and the learning number can be queried from the efficiency relationship table. The efficiency relationship table refers to a data table recording different difficulty variances, learning numbers and their corresponding learning efficiencies.

[0144] The coverage number refers to the number of necessary nodes covered in the learning path, i.e. the intersection of the knowledge nodes and the necessary nodes in the learning path. The determination method of the coverage number is well known to those skilled in the art, and will not be described here.

[0145] Step 215: determining the node number according to the necessary node.

[0146] The node number refers to the number of knowledge nodes contained in the necessary node. The determination method of the node number is well known to those skilled in the art, and will not be described here.

[0147] Step 216: calculating the quotient of the coverage number and the node number, and defining it as the coverage rate.

[0148] Coverage refers to a numerical value for showing the coverage degree of the learning path for the necessary nodes, and the calculation method of the coverage is known to those skilled in the art, which will not be repeated here.

[0149] Step 217: determine the path score in combination with the learning efficiency and the coverage, and select the learning path based on the path score.

[0150] The path score refers to a numerical value for showing the good or bad of the learning path, which can use the formula PathScore = 0.6*Coverage + 0.3*Efficiency + 0.1*Matching, wherein PathScore is the required path score, the higher the path score, the better the learning path, Coverage is the coverage rate described above, Efficiency is the learning efficiency described above, and Matching refers to the user ability matching degree, which can use the dynamic difficulty difference between the starting node and its direct successor node in the learning path as the user ability matching degree.

[0151] When different transition knowledge points can be introduced to achieve the effect of reducing the difficulty score gap, the learning paths are evaluated according to the distribution of knowledge points in the learning paths formed by introducing different transition knowledge points and the smoothness of the difficulty score improvement, so as to select the learning path with the highest knowledge point coverage and the most stable difficulty score improvement.

[0152] Referring to Figure 3 , the question set generation method comprises:

[0153] Step 300: call the custom question set, and read the collection quantity, pass rate and freshness from the custom question set.

[0154] The custom question set refers to the user contribution question set, i.e. the question set defined and shared by the user on the platform, and the calling method of the custom question set is selected by the staff according to the actual situation, which will not be repeated here.

[0155] The collection quantity refers to the number of times the custom question set is collected, the pass rate refers to the pass rate of the custom question set, and the freshness refers to a numerical value for measuring the freshness of the question set, which can be calculated using the formula T = max(1, 1 / log2(Δt+2)), wherein T is the required freshness, Δt is the update time of the question set from the current (unit: day), and the reading method of the collection quantity, pass rate and freshness is selected by the staff according to the actual situation, which will not be repeated here.

[0156] Step 301: determine the question set score in combination with the collection quantity, pass rate and freshness.

[0157] The question set score is a numerical value used to evaluate the quality of a custom question set. It can be calculated using the formula QualityScore=P^0.6×C^1.2×T, where QualityScore is the required question set score, P is the number of favorites, C is the pass rate, and T is the freshness.

[0158] Step 302: If the score of the question sheet is greater than the preset high quality threshold, determine the proportion of question sheets according to the passing coefficient.

[0159] The "quality threshold" refers to the minimum score required for a good custom question set, typically set to 4. The specific threshold is chosen by staff based on actual circumstances and will not be elaborated upon here. A question set score greater than the quality threshold indicates that the custom question set is relatively valuable. The question set ratio refers to the proportion of custom questions in the real-time generated question sets. A higher pass coefficient results in a higher ratio. For example, a pass coefficient below 40% uses a 20% ratio, a pass coefficient between 40% and 70% uses a 50% ratio, and a pass coefficient above 70% uses a 70% ratio. The selection range for the question set ratio is set by staff based on actual circumstances and will not be elaborated upon here.

[0160] Step 303: Generate a personalized question sheet based on the question sheet ratio, the custom question sheet, and the preset basic question sheet.

[0161] The basic question set refers to the question set pre-built by the platform based on the knowledge graph structure. It has the characteristics of clear hierarchy and wide coverage. Examples include the "Graph Traversal" special question set and the "State Compression DP" training set. The basic question set is pre-set by the staff and will not be described in detail here.

[0162] Personalized question sheets are question sheets generated according to the proportions of custom question sheets and basic question sheets, tailored to the user's situation. The method for generating personalized question sheets is common knowledge in the field and will not be elaborated here.

[0163] The question sets uploaded by other users are evaluated from three dimensions: number of collections, pass rate, and freshness. This allows us to select high-quality question sets and choose a suitable question set structure based on the user's learning progress. Finally, a personalized question set suitable for the current user is formed by combining basic question sets with high-quality question sets uploaded by other users.

[0164] The question paper generation method also includes:

[0165] Step 304: Collect the question sheet results based on the personalized question sheet.

[0166] Questionnaire results refer to the user's response records to personalized questionnaires. Questionnaire results can be collected through scanners or other means. The method of collecting questionnaire results is selected by the staff according to the actual situation, and will not be elaborated here.

[0167] Step 305: determining a reply node and its corresponding node accuracy rate based on the question list result.

[0168] The reply node refers to the knowledge node involved in each question in the personalized question list, and the node accuracy rate is the accuracy rate of the question involving the reply node. The determination method of the reply node and the node accuracy rate is well known to those skilled in the art, and will not be described here.

[0169] Step 306: if the node accuracy rate is lower than a preset lag threshold, determining a question type proportion in response to the node accuracy rate.

[0170] The lag threshold refers to the minimum node accuracy rate at which the user has a higher degree of mastery of the reply node. The lag threshold is selected by the staff according to the actual situation, and will not be described here. The node accuracy rate lower than the lag threshold represents that the user has a lower degree of mastery of the reply node. The question type proportion refers to the proportion value of the question about the reply node in the question list. The lower the node accuracy rate, the lower the question type proportion. The question type proportion corresponding to the node accuracy rate can be queried from the question type data table. The question type data table refers to a data table recording different node accuracy rates and their corresponding question type proportions.

[0171] Step 307: updating the personalized question list according to the question type proportion, and determining a question sequence according to the question list result.

[0172] The question sequence refers to the system regarding the user's question process as a time sequence, identifying the error behavior pattern by using pattern mining method. In simple terms, it is to find the key problems affecting the user's mastery by analyzing the "combination habits" of the user's question sequence. Generally, ApacheMahout is used to mine the sequential patterns and potential error rules in the user's learning behavior. The determination method of the question sequence is well known to those skilled in the art, and will not be described here.

[0173] For example, a user continuously makes mistakes in the question sequence of "greedy algorithm → interval covering problem" → "difference constraint"; and the system finds that other users usually complete "prefix sum" and "greedy strategy optimization" before mastering "difference constraint". The system judges that the user lacks basic knowledge support, automatically recommends to insert a "prefix sum reinforcement question list", and lowers the short-term recommendation priority of "difference constraint".

[0174] Step 308: determining a sequence accuracy rate based on the question sequence.

[0175] The sequence accuracy rate is the accuracy rate of the question involving the question sequence. The determination method of the sequence accuracy rate is well known to those skilled in the art, and will not be described here.

[0176] Step 309: If the sequence accuracy is lower than a preset inflection point threshold, determine an inserted node in combination with the answer sequence and a preset knowledge graph.

[0177] The inflection point threshold refers to the maximum sequence accuracy that requires insertion of other knowledge nodes to assist the user in learning. The inflection point threshold is selected by staff according to actual conditions, and will not be described here. The sequence accuracy lower than the inflection point threshold represents that the user lacks basic knowledge support, and the inserted node is the direct predecessor node of the answer sequence in the knowledge graph. The determination method of the inserted node is common knowledge in the art, and will not be described here.

[0178] Step 310: Update the personalized test paper according to the inserted node.

[0179] When the user's accuracy for a certain knowledge point is low, the proportion of related questions in the test paper is reduced, and when the user makes some fixed error patterns, the basic knowledge related to the related knowledge points is retrieved from the knowledge graph to supplement the user's understanding of the pre-knowledge of complex knowledge.

[0180] The occlusion recognition method comprises:

[0181] Step 400: If the node accuracy is lower than a preset lag threshold and / or the sequence accuracy is lower than a preset inflection point threshold, determine an error position in response to the test paper result and the personalized test paper.

[0182] The error position refers to the position of the answer error in the test paper result. The determination method of the error position is common knowledge in the art, and will not be described here.

[0183] Step 401: Determine an error frequency based on the error position.

[0184] The error frequency refers to the error frequency of the same position. The number of error positions in the approximate area centered on each error position can be counted as the number of errors, and the quotient of the number of errors and the number of test paper pages is calculated as the error frequency. The approximate area refers to a circular area near the error position centered on the error position. The number of test paper pages refers to the total number of test paper pages, i.e. the total number of test paper repetitions. The approximate area and the number of test paper pages can be input by staff in advance, and will not be described here.

[0185] Step 402: When the error frequency is greater than a preset repetition threshold, retrieve the error text of the question in combination with the error position and the personalized test paper.

[0186] The repetition threshold is the minimum error frequency of the display device failure leading to frequent errors at the same position, which is selected by the staff according to the actual situation, and will not be repeated here. The error frequency greater than the repetition threshold represents that the user frequently makes errors at the same position, and at this time, it is judged that the display device of the personalized test paper is faulty, leading to incomplete display of the question. The error text refers to the text range of the question located at the error position and the user's reply error, and the determination method of the error text is the common knowledge of the person skilled in the art, which will not be repeated here.

[0187] Step 403: determining the overlapping area according to the error text.

[0188] The overlapping area refers to the range of the overlap of the error text, i.e. the intersection of the error text, and the determination method of the overlapping area is the common knowledge of the person skilled in the art, which will not be repeated here.

[0189] Step 404: updating the personalized test paper in response to the overlapping area.

[0190] When the user frequently makes errors at the same position, it is judged that the display device has a problem at this position, leading to incomplete display of the text of the question, at this time, the text distribution of the error question located at this position in the test paper is called, so that the overlapping area of the text is taken as the range of the display device failure, and the text distribution of the questions in the test paper is dynamically adjusted to reduce the incomplete display of the questions.

[0191] The occlusion recognition method further comprises:

[0192] Step 405: collecting the answer image when the error frequency is greater than the preset repetition threshold.

[0193] The answer image refers to the picture when the user answers the personalized test paper, which can be collected by a fixed camera. The collection method of the answer image is selected by the staff according to the actual situation, which will not be repeated here.

[0194] Step 406: identifying the answering duration from the answer image based on the error position, and determining the answering threshold according to the node accuracy and the sequence accuracy.

[0195] The answering duration refers to the time spent by the user in answering the question at the error position, i.e. the time spent by the user from answering the question before the error position to answering the question at the error position. The determination method of the answering duration is the common knowledge of the person skilled in the art, which will not be repeated here.

[0196] The answer threshold refers to the minimum time length required by the user to answer the question. The smaller the node accuracy or sequence accuracy, the lower the user's mastery of the relevant knowledge node, and the more likely the user will skip the question. The lower the time length of the user answering the question, the lower the answer threshold corresponding to the node accuracy or sequence accuracy can be queried from the threshold relationship table. The threshold relationship table refers to a data table recording different node accuracies and sequence accuracies and their corresponding answer thresholds.

[0197] Step 407: If the answer time is less than the answer threshold, the correct text of the question is retrieved in combination with the error position and the personalized question sheet.

[0198] The answer time less than the answer threshold represents that the user answers the question too quickly, i.e., the user does not review the question completely. At this time, it is determined that the text of the question is blocked, and the text of the question is defined as the error text. The correct text is the text range of the other questions except the error text in the error position. The determination method of the correct text is well known to those skilled in the art, and will not be described here.

[0199] Step 408: Determine the combined area according to the correct text.

[0200] The combined area refers to the total range of the correct text, i.e., the union of the correct text. The determination method of the combined area is well known to those skilled in the art, and will not be described here.

[0201] Step 409: Determine the blocked area in combination with the combined area and the overlapping area.

[0202] The blocked area refers to the area blocked on the device for displaying the personalized question sheet, i.e., the remaining part of the overlapping area after subtracting the combined area. The determination method of the blocked area is well known to those skilled in the art, and will not be described here.

[0203] Step 410: Update the personalized question sheet in response to the blocked area.

[0204] When the user frequently makes mistakes at the same position, the time length of the user processing the question is detected, and when the user spends too little time, it is determined that the user directly skips the question, i.e., the user cannot process the question. At this time, it is determined that the text of the question is not completely displayed, thereby narrowing the overlapping area and reducing the degree of adjustment required by the question sheet.

[0205] The occlusion identification method further comprises:

[0206] Step 411: If the answer time is less than the answer threshold, identify the constraint feature from the correct text.

[0207] The constraint feature refers to a text description range in the stem of the question that has a constraint effect on the answer, which can be recognized by a text recognition technology. The determination method of the constraint feature is common knowledge to those skilled in the art, and will not be repeated here.

[0208] Step 412: Determine the stem area based on the constraint feature, and identify the correct answer and the incorrect answer from the correct text.

[0209] The stem area refers to the total range of the constraint feature, that is, the union set of the constraint feature. The determination method of the stem area is common knowledge to those skilled in the art, and will not be repeated here.

[0210] The correct answer refers to the text range of the correct answer of the question, and the incorrect answer refers to the text range of the incorrect answer of the question. The determination method of the correct answer and the incorrect answer is common knowledge to those skilled in the art, and will not be repeated here.

[0211] Step 413: When the incorrect answer coincides with the overlapping area, the correct answer is defined as the answer feature.

[0212] The fact that the incorrect answer does not coincide with the overlapping area means that at least one incorrect answer is blocked, and at this time, the correct answer can only be answered by directly observing the correct answer. The answer feature is the correct answer at this time.

[0213] Step 414: Update the combined area in combination with the stem area and the answer feature

[0214] When the user frequently makes mistakes in the same position, the question in the same position that the user does not skip is identified, so as to locate the text feature in the question that has a constraint effect on the answer result and the answer position in the question, and then narrow the display area of the question and reduce the situation that the blocking range is too small.

[0215] Based on the same invention concept, the embodiment of the present application provides a personalized learning system based on a knowledge graph, comprising:

[0216] The acquisition module is used to acquire identity information, learning goals, question list results and answer images.

[0217] The memory is used to store the program of any one of the above personalized learning methods based on the knowledge graph.

[0218] The processor can load and execute the program in the memory.

[0219] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional modules is exemplified, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0220] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall also be considered as the protection scope of the present application.

Claims

1. A personalized learning method based on knowledge graphs, characterized in that, include: Step 100: Collect the user's identity information and retrieve knowledge nodes from the preset knowledge graph; Step 101: In response to the identity information and knowledge node, determine the identity weight, and retrieve the user's answer parameters based on the knowledge node; Step 102: In response to the answer parameters, determine the pass coefficient and retrieve the basic difficulty score from the preset knowledge graph based on the knowledge node; Step 103: Determine the dynamic difficulty score by combining the basic difficulty score, identity weight, and pass coefficient; Step 104: Update the knowledge graph based on the dynamic difficulty score; It also includes a learning path planning method, which includes: Step 200: Collect learning objectives; Step 201: Determine necessary nodes in response to the learning objective, and determine the starting node in response to the answer parameters; Step 202: Determine the learning path and path difficulty score by combining the starting node, necessary nodes, and preset knowledge graph; Step 203: Calculate the difference in difficulty between adjacent paths based on the path difficulty score; Step 204: When the adjacent difficulty difference is greater than a preset buffer threshold, determine the step node by combining the adjacent difficulty difference and the learning path; Step 205: Determine the preceding node and the preceding difficulty score from the preset knowledge graph based on the step node, and determine the step difficulty range based on the step node and the path difficulty score; Step 206: If the prerequisite difficulty score falls into the step difficulty range, update the learning path according to the prerequisite node; The learning path planning method also includes: Step 207: If the prerequisite difficulty score falls into the step difficulty range, determine the prerequisite difficulty difference based on the prerequisite difficulty score and the path difficulty score; Step 208: When the difference in difficulty between the preceding nodes is greater than a preset buffer threshold, determine the selection node and the selection difficulty score from the preset knowledge graph based on the preceding nodes; Step 209: If the multiple selection difficulty score falls into the step difficulty range, determine the multiple selection difficulty difference based on the multiple selection difficulty score, the preceding difficulty score, and the path difficulty score; Step 210: When the difference in difficulty between multiple selections is not greater than a preset buffer threshold, update the learning path based on the multiple selection node and the preceding node; The learning path planning method also includes: Step 211: Determine the number of insertions based on the preceding node and the selected node; Step 212: If the number of insertions is greater than 1, determine the difficulty variance based on the adjacent difficulty difference; Step 213: Determine the number of learning steps based on the learning path; Step 214: Determine the learning efficiency by combining the difficulty variance and the number of learning points, and determine the coverage quantity based on the necessary nodes and learning path; Step 215: Determine the number of nodes based on the necessary nodes; Step 216: Calculate the quotient of the coverage quantity and the number of nodes, and define it as the coverage rate; Step 217: Determine the path score by combining the learning efficiency and coverage, and select a learning path based on the path score.

2. The knowledge graph-based personalized learning method according to claim 1, characterized in that, It also includes a question sheet generation method, which includes: Step 300: Retrieve the custom question set and read the number of favorites, pass rate, and freshness from the custom question set; Step 301: Determine the question score based on the number of collections, pass rate, and freshness. Step 302: If the score of the question sheet is greater than the preset high-quality threshold, determine the proportion of question sheets according to the passing coefficient; Step 303: Generate a personalized question sheet based on the question sheet ratio, the custom question sheet, and the preset basic question sheet.

3. The knowledge graph-based personalized learning method according to claim 2, characterized in that, The question paper generation method also includes: Step 304: Collect the question sheet results based on the personalized question sheet; Step 305: Determine the answer nodes and their corresponding node accuracy rates based on the results of the question sheet; Step 306: If the accuracy rate of the node is lower than the preset lag threshold, determine the proportion of question types in response to the accuracy rate of the node; Step 307: Update the personalized question sheet according to the proportion of the question types, and determine the answer sequence based on the question sheet results; Step 308: Determine the sequence accuracy based on the answer sequence; Step 309: If the sequence accuracy is lower than the preset threshold for adding points, determine the insertion node by combining the answer sequence and the preset knowledge graph; Step 310: Update the personalized question sheet according to the inserted node.

4. The knowledge graph-based personalized learning method according to claim 3, characterized in that, It also includes an occlusion recognition method, which includes: Step 400: If the node accuracy is lower than a preset lag threshold and / or the sequence accuracy is lower than a preset increment threshold, determine the error location in response to the question sheet results and the personalized question sheet; Step 401: Determine the error frequency based on the error location; Step 402: When the error frequency is greater than the preset repetition threshold, retrieve the error text of the question by combining the error location and the personalized question sheet; Step 403: Determine the overlapping area based on the erroneous text; Step 404: Update the personalized question sheet in response to the overlapping area.

5. The knowledge graph-based personalized learning method according to claim 4, characterized in that, The occlusion recognition method further includes: Step 405: When the error frequency is greater than the preset repetition threshold, collect the answer image; Step 406: Identify the answer duration from the answer image based on the error location, and determine the answer threshold based on the node accuracy and sequence accuracy; Step 407: If the answering time is lower than the answering threshold, retrieve the correct text of the question based on the error location and the personalized question sheet; Step 408: Determine the combined region based on the correct text; Step 409: Determine the occlusion area by combining the combined area and the overlapping area; Step 410: Update the personalized question sheet in response to the occluded area.

6. The knowledge graph-based personalized learning method according to claim 5, characterized in that, The occlusion recognition method further includes: Step 411: If the answering time is lower than the answering threshold, identify constraint features from the correct text; Step 412: Determine the question stem region based on the constraint features, and identify the correct and incorrect answers from the correct text; Step 413: When the incorrect response coincides with the overlapping area, the correct response is defined as a response feature; Step 414: Update the combined region by combining the question stem region and the response features.

7. A personalized learning system based on knowledge graphs, characterized in that, include: The data acquisition module is used to collect identity information, learning objectives, question results, and answer images. A memory for storing the program of the knowledge graph-based personalized learning method as described in any one of claims 1 to 6; The processor is the unit of memory that allows programs to be loaded and executed by the processor.

Images