A learning path generation method, device, equipment, medium and program product for senior members

Abstract

The application relates to the technical field of online learning, and particularly relates to a learning path generation method, device and equipment for senior members, a medium and a program product. The method comprises the following steps: firstly, in response to a question input by a senior member, extracting a question feature according to a corresponding subject type of the question; secondly, inputting the feature into a knowledge point matching model, obtaining a core knowledge point by combining a preset knowledge point description-feature keyword mapping library; thirdly, obtaining a mastery degree rating by a multi-dimensional weighting algorithm according to historical learning data of the senior member on the core knowledge point; fourthly, screening similar knowledge points from a preset knowledge point graph based on the core knowledge point, and obtaining an initial screening question list by combining a knowledge point ID-question ID list mapping relationship; and finally, secondarily screening the list and preferentially returning a question corresponding to a weak knowledge point, so that a precise and personalized learning path can be generated, and the learning efficiency of the senior member is improved.

Description

Technical Field

[0001] This application relates to the field of online learning technology, and more specifically, to a method, apparatus, device, medium, and program product for generating learning paths for premium members. Background Technology

[0002] Current learning path generation solutions for advanced members in the K-12 education sector have significant shortcomings. Most systems can only provide single answers based on user-input questions, failing to connect related knowledge points, resulting in fragmented question recommendations that struggle to form a systematic learning path. Furthermore, existing solutions do not accurately tailor recommendations based on members' knowledge mastery, often leading to a disconnect between recommended questions and members' weaknesses. Additionally, traditional knowledge point matching relies heavily on simple keyword comparisons, neglecting subject characteristics and logical connections between knowledge points, resulting in low matching accuracy and inefficient learning. This fails to meet the personalized learning needs of "searching one question to learn a whole category of questions," and thus fails to satisfy advanced members' demands for precise and systematic learning. Summary of the Invention

[0003] In response, this application provides a method, apparatus, device, medium, and program product for generating learning paths for advanced members, to at least partially solve the aforementioned technical problems.

[0004] This application provides a method for generating learning paths for premium members, including the following steps: In response to receiving a question input by a senior member, the system extracts the corresponding question features based on the question type, where the question type corresponds to different subject types. The extracted question features are input into the knowledge point matching model, which obtains the core knowledge points corresponding to the question based on a pre-set knowledge point description-feature keyword mapping library. Based on the senior members' historical learning data on the core knowledge points, a multi-dimensional weighted algorithm is used to obtain a rating of the senior members' mastery of the core knowledge points. Based on the core knowledge points, similar knowledge points are selected from the preset knowledge point graph. Based on the core knowledge points and similar knowledge points, a preliminary list of questions is obtained through the preset knowledge point ID-question ID list mapping relationship. The initial screening list of questions is then further filtered, and questions corresponding to knowledge points with a weak mastery level are selected as the preferred ones.

[0005] In another aspect, this application also provides a learning path generation device for premium members, comprising: The question feature extraction module is used to extract corresponding question features based on the question type in response to receiving a question input from a senior member. The question type corresponds to different subject types. The core knowledge point acquisition module is used to input the extracted question features into the knowledge point matching model. The knowledge point matching model obtains the core knowledge points corresponding to the question based on the preset knowledge point description-feature keyword mapping library. The rating module is used to obtain a rating of the senior member's mastery of the core knowledge points based on the senior member's historical learning data on the core knowledge points through a multi-dimensional weighted algorithm. The initial screening module is used to screen similar knowledge points from a preset knowledge point map based on the core knowledge points. Based on the core knowledge points and similar knowledge points, a preliminary screening question list is obtained through a preset knowledge point ID-question ID list mapping relationship. The secondary screening module is used to perform a secondary screening of the initial screening question list, and to preferentially return questions corresponding to knowledge points whose mastery level is rated as weak.

[0006] This application also provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the learning path generation method for advanced members as described above.

[0007] In another aspect, this application provides a computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor to implement the learning path generation method for advanced members as described above.

[0008] Another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the learning path generation method for advanced members as described above.

[0009] This application generates learning paths for premium members, accurately matching subject-specific questions with core knowledge points. It uses a multi-dimensional weighted algorithm based on historical learning data to determine the level of knowledge mastery, and then filters questions based on a knowledge point graph and ID mapping relationship, prioritizing questions corresponding to weak knowledge points. This solves the problems of fragmented recommendations and disconnect from weak areas in traditional solutions, while improving the accuracy of knowledge point matching. It achieves a systematic learning approach of "searching for one question and learning a whole category of questions," providing personalized and precise learning paths for premium members in the K-12 field, effectively improving learning efficiency and meeting their needs for systematic learning. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of a learning path generation method for advanced members provided in an embodiment of this application.

[0012] Figure 2 This is a schematic diagram of the title feature matching and extraction process provided in the embodiments of this application.

[0013] Figure 3 This is a schematic diagram of a learning path generation device for advanced members provided in an embodiment of this application.

[0014] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0016] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, or product comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, apparatus, or products.

[0017] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0018] Please seeFigure 1 As shown, Figure 1 This is a flowchart illustrating a learning path generation method for premium members disclosed in an embodiment of this application. Figure 1 As shown, the learning path generation method for premium members may include the following operations: S101, in response to receiving a question input by a senior member, extract the corresponding question features based on the question type, wherein the question type corresponds to different subject types; S102, input the extracted question features into the knowledge point matching model. The knowledge point matching model obtains the core knowledge points corresponding to the question based on the preset knowledge point description-feature keyword mapping library. S103, Based on the senior member's historical learning data on the core knowledge points, a multi-dimensional weighted algorithm is used to obtain a rating of the senior member's mastery of the core knowledge points. S104, Based on the core knowledge points, select similar knowledge points from the preset knowledge point map, and based on the core knowledge points and similar knowledge points, obtain a preliminary list of questions by using the preset knowledge point ID-question ID list mapping relationship. S105, perform a second screening on the initial screening question list, and preferentially return the questions corresponding to the knowledge points whose mastery level is rated as weak.

[0019] In one embodiment, unstructured questions (text input or image search) entered by members are transformed into structured, subject-specific feature data. Specifically, the member-input questions are first preprocessed, which, for example, includes: Typically, premium members can input questions using either text input (manually entering the question text) or image search input (taking a picture of the question), which is then processed to obtain the standard question text.

[0020] For text input processing, for example, it directly receives text entered by members and uses regular expressions to filter redundant symbols in the question stem, such as "

Question

[0021] For the processing of camera search input, for example, after a member uploads a question image, the OCR (Optical Character Recognition) engine is called to convert the image into text. Exemplarily, the OCR engine loads a recognition model dedicated to K12 disciplines (optimizing the recognition accuracy of formulas, symbols, ancient poems and essays), and respectively recognizes the formula symbols in science questions, the ancient poems and essays, and special punctuation in liberal arts questions, such as the "。", ";" in classical Chinese, and the "’s" in English. The liberal arts dedicated character library is loaded to correct the recognition errors of classical Chinese words such as "之乎者也". For example, when "乎" is recognized as "呼", the ancient poem corpus is automatically matched for correction.

[0022] Next, for different types of questions and the characteristics of different disciplines, corresponding features are extracted. In one embodiment, based on the characteristics of the structured and standardized K12 discipline knowledge, the question features are extracted by means of keyword matching.

[0023] Specifically, based on a preset discipline keyword library, the content matching the keyword library is quickly found from the question text, and then these matching contents are mapped to the three dimensions of "entity, problem-solving / answering goal, limiting condition / examination point", and finally structured question features are formed.

[0024] At the same time, in advance, according to disciplines and feature dimensions, the corresponding discipline keyword template libraries are sorted out. For example, the entity keyword template of mathematics includes square, side length, π, and the problem-solving goal keyword template includes finding the area, solving equations, and proving; the examination point keyword template of history includes significance, reason, and time.

[0025] When extracting features, the question text is used to compare with the template library to find the overlapping keywords, and then these keywords are classified into the corresponding feature dimensions. For example, "find XX" in the template library belongs to the "problem-solving goal" dimension. If the question contains "find the area of a rectangle", then "find the area of a rectangle" is the matched "problem-solving goal" feature.

[0026] As an example, take the math problem: Given that the side length of a square is 5 cm, find its area, and the history problem: Briefly describe the historical significance of the May 4th Movement in 1919 as an example. As Figure 2 shown, the complete process of question feature matching and extraction specifically includes: S201, based on the discipline keyword template library, determine the discipline involved in the question. Specifically, first look at the discipline-specific keywords in the question. For example, a math problem contains "square, side length" (math entity keywords), and a history problem contains "1919, May 4th Movement" (history-specific keywords) to quickly determine the discipline; based on the determined discipline, call the corresponding keyword template library of the discipline. Math calls the math keyword template library, and history calls the history keyword template library.

[0027] S202 involves scanning the question text sentence by sentence and matching keywords in the template library. Specifically, for matching math questions, for example, scanning "given that the side length of a square is 5cm" will match "square" and "side length 5cm" (keyword templates belonging to the mathematical "entity" dimension). Scanning "find its area" matches "find area" (a keyword template belonging to the "problem-solving objective" dimension in mathematics); There are no other restrictions, such as "side length > 0", so there are no matching keywords in the "limiting conditions" dimension.

[0028] For historical question matching, for example, scanning "May Fourth Movement in 1919": matching "1919" and "May Fourth Movement" (keyword templates belonging to the historical "entity" dimension); Scanning "Briefly describe the historical significance" matches "Briefly describe the significance" (a keyword template belonging to the "answer objective" dimension of history), and "historical significance" matches the keyword template of the "examination point" dimension (examining the "significance of the event").

[0029] S203, categorize the matched keywords into corresponding feature dimensions. Specifically, the keywords matched in step 202 are organized according to three dimensions: "entity, problem-solving objective / answering objective, and limiting conditions / examination points," forming the final core features. For example, Math problem extraction results: Entity (subject-specific element): Square, 5cm on each side; Problem objective: Find the area; Limitations: None.

[0030] History question extraction results: Entities (subject-specific elements): 1919, the May Fourth Movement; Answer objective: Briefly describe the (action) + historical significance (content); Examination point: The historical significance of the May Fourth Movement In one embodiment, the construction process of the preset knowledge point graph includes: based on the textbook versions and chapter-section structure of K12 subjects, breaking down knowledge into multi-level knowledge points with granularity ranging from large to small, labeling each knowledge point with a unique knowledge point ID, and constructing a knowledge point graph in the format of "parent node-child node-relationship type" through preset logical relationships between knowledge points; wherein the logical relationships include prerequisite dependencies, sibling associations, and extensions.

[0031] Specifically, the construction of the knowledge point map is based on the textbooks of various K-12 subjects, ensuring that the knowledge system is fully adapted to the teaching syllabus and students' learning progress. The specific operation is as follows: It organizes all core subjects in the K-12 stage, such as mathematics, Chinese, English, physics, chemistry, biology, history, geography, and politics, and covers the mainstream textbook versions for each subject, such as the People's Education Press version, Beijing Normal University Press version, and Jiangsu Education Press version for mathematics; and the Ministry of Education's compiled version for Chinese, to ensure that senior members in different regions and using different textbooks can match the corresponding knowledge system. The book-chapter-section structure of each textbook is used as the initial hierarchical framework of the graph, such as Junior High School Mathematics - Grade 9, Volume 1 - Chapter 22 Quadratic Functions - Section 1 Graph of Quadratic Functions.

[0032] In this embodiment, the textbook content is broken down into five levels of structured knowledge points according to the logic of granularity from large to small, ensuring that the granularity of knowledge is neither redundant nor incomplete. Each level of knowledge point corresponds to a clear teaching scope. For example, the specific hierarchical division is shown in Table 1 below: Table 1

[0033] In this way, it is ensured that each lower-level knowledge point is completely subordinate to the higher-level knowledge point. For example, "solving the vertex coordinates of a quadratic function with parameters" belongs to "solving the vertex coordinates of a quadratic function graph". Furthermore, there is no overlap or omission of knowledge points at the same level. For example, "the general form of a quadratic function" and "the vertex form of a quadratic function" are core knowledge points at the same level, covering all the content of "quadratic function expressions".

[0034] Assign a unique knowledge point ID to each decomposed knowledge point. For example, the ID format can adopt a structured rule of subject code-grade-textbook volume-chapter-section-level-serial number, ensuring that the subject, grade, textbook position, and level of the knowledge point can be located through the ID. For example: The ID for the general form of a quadratic function (y=ax²+bx+c) can be K12_MATH_9_U1_22_1_4_01 The meanings of each part are as follows: K12 (K12 education field) _MATH (subject: mathematics) _9 (grade: ninth grade) _U1 (textbook volume: first volume) _22 (chapter: chapter 22) _1 (section: section 1) _4 (level: level 4 core knowledge points) _01 (serial number: the first knowledge point under this level).

[0035] In this embodiment, based on the inherent logic of K12 knowledge, such as learning order and related applications, a "parent node-child node-relationship type" association is established between knowledge points, where the relationship type includes three types of relationships: prerequisite dependency, sibling association, and extension.

[0036] Among them, the prerequisite dependency relationship is used to define the learning order of learning before learning. If you must learn knowledge point A before you can learn knowledge point B, then A is the parent node of B and B is the child node of A. The relationship type is prerequisite dependency. Among them, for the same-level association, for example, the definition of knowledge linkage at the same level and with strong association, such as the same-level knowledge points under the same superior knowledge point, or two core knowledge points under the same section, if there is a relationship of mutual explanation and collaborative application, then they are parent nodes / child nodes of each other (without strict order, bidirectional association is possible), and the relationship type is same-level association, which supports the cross-knowledge point application of comprehensive questions.

[0037] Among them, for the extended relationship, for example, defining the basic → advanced knowledge deepening, if knowledge point B is an advanced application or extension of knowledge point A (you can learn B after mastering A, and you can learn B through other paths even if you don't master A, but mastering A can improve the efficiency of understanding B), then A is the parent node of B and B is the child node of A. The relationship type is extended, supporting advanced members with extra learning capacity for in-depth learning.

[0038] Optionally, information such as knowledge point ID, parent node ID, child node ID, relationship type, knowledge point name, and textbook location can be stored in the Neo4j graph database (which supports efficient node and relationship queries) to form a parsable structured graph. Neo4j is a high-performance NoSQL graph database that stores structured data on the network instead of in tables.

[0039] In one embodiment, based on the knowledge point graph and question resources, all questions are structured and their core elements are extracted. The core elements include at least the question stem text, solution steps, answer, knowledge points tested, difficulty level, and question type. The tested knowledge points are associated with one or more knowledge point IDs in the knowledge point graph, and a mapping relationship between knowledge point IDs and question IDs is established based on the tested knowledge points.

[0040] In this embodiment, based on the structured processing of knowledge point graphs and question resources, a large number of unstructured questions, such as image search questions and text questions, are transformed into searchable, associative, and analyzable structured data. At the same time, by examining the association between knowledge points and knowledge point IDs, a mapping relationship between knowledge point IDs and question IDs is constructed.

[0041] Specifically, for question resources, the core elements are extracted in the following order: question text → solution steps → answer → knowledge points tested → difficulty level → question type. The extraction logic varies slightly for different subjects, but all are based on the knowledge point map as the correlation standard.

[0042] For example, the text of the question stem can be extracted directly from the text. For science questions, the formula format is retained, and for humanities questions, the text structure is retained. For example, the "reading material and questions" are separated in Chinese reading comprehension questions, and the "material description and questions" are separated in history questions. For problem-solving steps, for example, for science problems (mathematics / physics / chemistry), extract the step-by-step solution process and mark the knowledge points corresponding to each step; for humanities problems (Chinese / history / politics), extract the answer framework. For the answers, for example, for science questions, the objective answers are directly extracted, such as "the vertex coordinates are (-1, -1)" and "the buoyancy is 5N". For humanities questions, the "key points of the reference answer" are extracted, such as the key points of the answer to the history question "the significance of the May Fourth Movement": "the proletariat stepped onto the historical stage and promoted the spread of Marxism", to ensure that the answer is consistent with the problem-solving steps / answering ideas.

[0043] In this embodiment, the knowledge points being examined correspond to the examination points in the aforementioned question features. The association between the examination points and the knowledge point names in the knowledge graph is established through keyword matching. Specifically, the core keywords of the examination points are extracted from the question features and compared with the name keywords of the knowledge points in the knowledge point graph to find the knowledge point with the highest degree of overlap, thereby establishing a connection.

[0044] Specifically, keyword overlap can be calculated, and the matching degree can be calculated by the number of overlapping keywords and keyword weights. Higher weights are assigned to core subject keywords. A successful match is determined when the number of overlaps is ≥1 and the weighted matching degree is ≥0.6. After a successful match, a mapping relationship is established between the knowledge point IDs and the question ID lists.

[0045] As for the difficulty level, it can be determined by the number of knowledge points; the more knowledge points, the higher the difficulty.

[0046] Optionally, the mapping relationship between knowledge point IDs and question IDs is stored in the STORED high-performance key-value storage system. The key is the knowledge point ID, and the value is the list of question IDs (with question difficulty levels). STORED supports O(1) retrieval efficiency, ensuring that the question list can be quickly retrieved in high-concurrency scenarios. STORED, short for Bitalostored, is a high-performance key-value storage system with performance exceeding that of traditional Redis. The internal storage engine adopts structures such as Bithash and Bitalostree, effectively solving the read / write amplification problem of traditional LSM-Tree.

[0047] In one embodiment, based on subject teaching and research experience and historical answer data statistics, a knowledge point description-feature keyword mapping library is established, forming a four-dimensional mapping relationship of knowledge point ID-feature type-feature keyword-keyword weight; wherein, the knowledge point ID is consistent with the knowledge point ID in the knowledge graph, the feature type corresponds to the extracted question features, the feature keyword is the feature keyword associated with the knowledge point, and the keyword weight represents the contribution of the keyword to the matching of the knowledge point.

[0048] In this embodiment, taking the solution of the vertex coordinates of a quadratic function as an example, the four-dimensional mapping relationship corresponds to the three dimensions of the problem characteristics, as exemplarily shown in Table 2 below. Table 2

[0049] The knowledge point ID is completely consistent with the ID in the knowledge point graph; the feature type corresponds to the three dimensions of the question features (entity, problem-solving objective, and limiting conditions), with science subjects often using "XX entity" or "XX operation objective," and humanities subjects often using "XX event" or "XX answer direction"; the feature keywords are common expressions of the knowledge point in the question, such as finding the vertex coordinates being a typical problem-solving objective keyword for finding the vertex coordinates of a quadratic function; the keyword weight indicates the importance of the keyword in matching the knowledge point (the higher the weight, the more likely it is to belong to this knowledge point when the keyword appears), such as "y=ax²+bx+c" being the core formula of a quadratic function with a weight of 0.92.

[0050] In one embodiment, the knowledge point matching model is based on an optimized Bayesian algorithm. Specifically, for the input question feature set F, the likelihood probability of each feature in the feature set F is adjusted based on the preset weights of the feature types and the keyword weights in the knowledge point description-feature keyword mapping library. This includes: calculating the original likelihood probability of each feature, multiplying the original likelihood probability by the feature type weight and the corresponding keyword weight to obtain the optimized likelihood probability of each feature, and calculating the optimized likelihood probability of the question feature set F based on the optimized likelihood probability of each feature.

[0051] In this embodiment, the formula for the Bayesian algorithm is:

[0052] in, The i-th knowledge point in the knowledge point map, such as "solving the coordinates of the vertex of a quadratic function", ID: K12_MATH_9_U1_22_1_01; F is the set of question features extracted from the knowledge point map. Given that the question possesses characteristic F, this question pertains to the knowledge point. The posterior probability (the final output metric of the model; the higher the probability, the stronger the match). , indicating knowledge points The likelihood probability of feature set F appearing in the corresponding question (calculated statistically using the knowledge point description-feature keyword mapping library). , indicating knowledge points Prior probability in all questions (i.e., the proportion of questions corresponding to this knowledge point to the total number of questions, used to balance the matching bias between high-frequency and low-frequency knowledge points). , representing the marginal probability of feature set F appearing in the question (a fixed value for all knowledge points, negligible in calculation, only requiring comparison). (size).

[0053] In this embodiment, considering that the question feature set F contains multiple sub-features, to reduce computational complexity, based on the assumption of Naive Bayes, the sub-features are conditionally independent, that is:

[0054] in, is a single feature in the feature set F.

[0055] In one embodiment, feature type weights are introduced. and keyword weight Afterwards, a single feature The likelihood probability is adjusted to:

[0056] Accordingly, the likelihood probability of the feature set F is adjusted as follows:

[0057] Where m is the number of feature types. Let be the number of keywords in the t-th feature class.

[0058] Specifically, taking the problem feature set F, such as {F1: entity y=ax²+bx+c, F2: problem objective: finding vertex coordinates}, as an example, the specific steps are as follows: First, analyze the frequency with which this feature appears in historical data and pertains to a specific knowledge point: For feature F1 (y=ax²+bx+c), for example, in the past 1000 questions containing this feature, 900 of them belong to the problem of finding the vertex coordinates of a quadratic function, and the original likelihood probability = 900 / 1000 = 0.9; For feature F2 (finding vertex coordinates), for example, out of the past 800 questions containing this feature, 720 belong to this knowledge point, and the original likelihood probability = 720 / 800 = 0.9.

[0059] Next, two weights are introduced for adjustment (weight values ​​range from 0 to 1, with larger values ​​indicating greater importance): For feature type weights, for example, corresponding to the three dimensions of question features (entity, problem-solving objective, and limiting conditions), the importance of teaching and research pre-planning is as follows: if the problem-solving objective best reflects the core of the examination, the weight is set to 0.9; and if the entity is second, the weight is set to 0.8. For keyword weights, for example, from the knowledge point description-feature keyword mapping library, such as y=ax²+bx+c, the keyword weight for this knowledge point is 0.92, and the vertex coordinates are 0.88.

[0060] Finally, the optimized likelihood probabilities of all features are multiplied together to obtain the probability that the entire feature set belongs to the knowledge point. For example, the optimized likelihood probability of F is 0.662 × 0.713 ≈ 0.472 (that is, about 47.2% probability that it belongs to the knowledge point).

[0061] Repeat the above calculation for all possible knowledge points, and finally select the knowledge point with the highest optimized likelihood probability of feature set F as the matching result.

[0062] As an example, if the overall probability of solving for the vertex coordinates of a quadratic function is 47.2%, while the probability of solving for other knowledge points, such as the properties of the graph of a quadratic function, is only 20%, then the question is determined to be about solving for the vertex coordinates of a quadratic function.

[0063] Optionally, before adjusting the likelihood probability of each feature in the feature set F, the likelihood probability of each feature is smoothed using Laplace to obtain the smoothed likelihood probability of each feature, and the smoothed likelihood probability of each feature is adjusted to obtain the optimized likelihood probability of each feature.

[0064] In some possible implementations, in the basic Bayesian algorithm, if a feature f is at a knowledge point It has never appeared in history questions (i.e.) ), which will lead to It was directly ruled out. The matching probability is particularly evident when dealing with new question types and rare keywords. To address this issue, this invention employs Laplace Smoothing, with the specific formula as follows: For a single feature The likelihood probability, after smoothing, is calculated as follows:

[0065] in, For knowledge points The corresponding questions contain features. The number of times; For knowledge points The corresponding total number of questions; Smoothing coefficient (empirical value, ranging from 0.1 to 1; in this example, it is set to 0.5 to balance the smoothing effect with the true probability). Let t be the total number of keywords for the t-th feature in the mapping library.

[0066] In one embodiment, for S102, the extracted question features, such as entity: y=ax²+bx+c, and problem-solving objective: to find vertex coordinates, are input into the knowledge point matching model. The model will then perform a preliminary comparison with the feature keywords of all knowledge points in the knowledge point description-feature keyword mapping library, and filter out candidate knowledge points with overlapping keywords for preliminary comparison to obtain candidate knowledge points. Next, for each candidate knowledge point, the model uses an optimized Bayesian algorithm to calculate the overall matching probability of the question features belonging to the candidate knowledge point. The calculation method is the same as in the above embodiment and will not be repeated here. Finally, by comparing the overall matching probabilities of all candidate knowledge points, the model selects the 1-2 knowledge points with the highest probabilities, which are the core knowledge points corresponding to the question.

[0067] In one embodiment, for S103, based on all the member's historical learning records for core knowledge points, such as solving the vertex coordinates of a quadratic function, for example, data in three dimensions are extracted: The accuracy rate includes the total number of questions on this knowledge point that have been answered in the past 30 days, for example, 20 questions, and the number of questions answered correctly, for example, 12 questions. The accuracy rate is calculated as 12 / 20 = 60%. Answering time includes the average time spent on questions about that knowledge point, such as an average of 4 minutes per question, and then comparing it with the average time spent by members of the same grade, such as 3 minutes per question, to determine whether it takes too long; The types of mistakes include whether they are "conceptual errors" (e.g., confusing formulas) or "calculation errors" (e.g., using the wrong calculation symbols). If conceptual errors account for a high percentage of mistakes, such as 80% of the mistakes being conceptual errors, it indicates a weak foundation.

[0068] Next, different weights are assigned to the three dimensions, with a total weight of 100%, distributed according to importance. The score for each dimension is then calculated, and finally, the scores are summed to obtain the overall score.

[0069] Finally, the pre-defined rules for matching scores and ratings directly match the results, for example, 0-40 points: Not mastered (accuracy rate <50%, with many conceptual errors); 41-60 points: Weakness (accuracy rate 50%-60%, or taking too long); 61-85 points: Proficient (accuracy rate 60%-85%, most mistakes are calculation errors); 86-100 points: Proficient (accuracy rate > 85%, short time, few wrong answers).

[0070] For S104, after identifying the core knowledge points, find similar knowledge points in the knowledge graph of the aforementioned embodiment, and then use the knowledge point ID-question ID mapping relationship of the aforementioned embodiment to find the corresponding questions, quickly generating a preliminary question list. This is accomplished in two exemplary steps: First, based on the pre-defined knowledge point map, identify similar knowledge points that have a specific relationship with the core knowledge points, including: Prerequisite knowledge points, such as the core knowledge point being the graph of a quadratic function, and solving a quadratic equation in one variable in the graph is a prerequisite knowledge point, should be added to the list of similar knowledge points first. Related knowledge points at the same level, such as solving the vertex coordinates of a quadratic function and the core knowledge point of the graph of a quadratic function, are related at the same level and should also be added to the list; Expanding on existing knowledge points, such as the application of quadratic functions in practical problems, is also an expansion and should be included.

[0071] This results in a list containing core knowledge points and similar knowledge points, such as [“Graph of a quadratic function”, “Solving a quadratic equation in one variable”, “Solving for the coordinates of the vertex of a quadratic function”, “Application of quadratic functions in practical problems”].

[0072] Next, based on the identified list of knowledge points, the mapping relationship between knowledge point ID and question ID is searched. For example, if the knowledge point ID for the quadratic function graph is K12_MATH_9_U1_22_1_4_03, the corresponding question ID list is found in the mapping relationship, such as [Q20240101, Q20240103]. Perform this operation on each knowledge point ID in the list, merging all found question IDs together and removing duplicate question IDs; Finally, a preliminary list of questions is obtained, such as [Q20240101, Q20240103, Q20240205, Q20240307]. These questions test core knowledge points or similar knowledge points, and will be further filtered and recommended to senior members.

[0073] In one embodiment, for S104, when performing a second screening of the initial question list, the core is to prioritize recommending questions corresponding to knowledge points where the member's mastery level is rated as weak, ensuring that the learning path is precisely targeted at the weak points. For example, the specific process is as follows: Extract the IDs of weak knowledge points. For example, from the members' mastery level rating results, find all the knowledge point IDs rated as weak, such as the ID for "solving the vertex coordinates of a quadratic function": K12_MATH_9_U1_22_1_4_03. The question-knowledge point association matching, for example, iterates through each question in the initial screening question list, finds the corresponding knowledge point ID (there may be multiple) for each question through the mapping relationship between knowledge point ID and question ID list, and determines whether it contains weak knowledge point IDs.

[0074] Based on the weakness ranking questions obtained from the aforementioned embodiments, specifically, ranking weights are first set. For example, the weight of questions that directly correspond to weak knowledge points is the highest (100%); the weight of questions that correspond to weak knowledge points that are prerequisites or extensions of weak knowledge points is the second highest (80%); and the weight of questions that correspond to weak knowledge points that are related to the same level as weak knowledge points is the lowest (60%).

[0075] Next, the weak link score for each question is calculated based on the weight. For example, if a question corresponds to a weak knowledge point ID, it gets 100 points. If the prerequisite knowledge points for the corresponding weak knowledge points are mentioned, such as the quadratic equation being a prerequisite for the quadratic function, 80 points will be awarded; and so on, with scores assigned according to weight.

[0076] Finally, the results are filtered and returned. For example, high-scoring questions are prioritized, such as questions with scores ≥80 (questions that directly correspond to weak or strongly related knowledge points) to form a candidate list; medium-related questions are added, such as questions with scores 60 or higher that are related to the same level if the number of questions in the candidate list is insufficient (e.g., <10 questions) to ensure the number of recommendations; and the questions are sorted from highest to lowest score, such as returning a sorted list of questions to ensure that members prioritize practicing the questions that can best improve their weak points.

[0077] Therefore, based on the above embodiments, the transformation from "single question search" to "precise recommendation of similar questions" and then to "closed loop of knowledge point mastery" has been realized, truly achieving the goal of "searching for one question and learning a type of question", solving the pain points of fragmented recommendations and unrelated learning in traditional membership systems, and improving the personalized learning efficiency of K12 users.

[0078] Figure 3 This is a schematic diagram of a learning path generation device for premium members disclosed in an embodiment of this application. Figure 3 As shown, the device includes: The question feature extraction module 301 is used to extract corresponding question features based on the question type in response to receiving a question input from a senior member, wherein the question type corresponds to different subject types. The core knowledge point acquisition module 302 is used to input the extracted question features into the knowledge point matching model. The knowledge point matching model obtains the core knowledge points corresponding to the question based on the preset knowledge point description-feature keyword mapping library. The rating module 303 is used to obtain a rating of the senior member’s mastery of the core knowledge points based on the senior member’s historical learning data on the core knowledge points through a multi-dimensional weighted algorithm. The initial screening module 304 is used to screen similar knowledge points from a preset knowledge point map based on the core knowledge points. Based on the core knowledge points and similar knowledge points, a preliminary screening question list is obtained through a preset knowledge point ID-question ID list mapping relationship. The secondary screening module 305 is used to perform secondary screening on the initial screening question list, and preferentially return the questions corresponding to the knowledge points whose mastery level is rated as weak.

[0079] Specific limitations regarding the learning path generation device for advanced members can be found in the limitations of the learning path generation method for advanced members mentioned above, and will not be repeated here. Each module in the aforementioned learning path generation device for advanced members can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the electronic device in hardware format or independently of it, or stored in the memory of the electronic device in software format, so that the processor can call the corresponding operations of each module.

[0080] It should be noted that, in order to highlight the innovative aspects of this application, this embodiment does not include modules that are not closely related to solving the technical problems proposed in this application, but this does not mean that there are no other modules in this embodiment.

[0081] like Figure 4 As shown, the electronic device 4 provided in this application may include a memory 12, a processor 13 and a bus, and may also include a computer program stored in the memory 12 and executable on the processor 13, such as a learning path generation program for senior members.

[0082] The memory 12 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 12 can be an internal storage unit of the electronic device 4, such as a portable hard drive of the electronic device 4. In other embodiments, the memory 12 can be an external storage device of the electronic device 4, such as a plug-in portable hard drive, smart memory card, secure digital card, flash memory card, etc., equipped on the electronic device 4. Furthermore, the memory 12 can include both internal and external storage units of the electronic device 4. The memory 12 can be used not only to store application software and various types of data installed on the electronic device 4, such as code generated for learning paths for advanced members, but also to temporarily store data that has been output or will be output.

[0083] In some embodiments, the processor 13 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 13 is the control core of the electronic device 4, connecting various components of the entire electronic device 4 via various interfaces and lines. It executes programs or modules stored in the memory 12 (e.g., a learning path generation program for advanced members) and calls data stored in the memory 12 to perform various functions of the electronic device 4 and process data.

[0084] The processor 13 executes the operating system of the electronic device 4 and various installed applications. The processor 13 executes the applications to implement the steps in the above-described method for generating learning paths for advanced members.

[0085] For example, the computer program may be divided into one or more modules, which are stored in the memory 12 and executed by the processor 13 to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device 4.

[0086] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The software functional module stored in the storage medium includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute some functions of the learning path generation method for advanced members described in the various embodiments of this application.

[0087] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the learning path generation method for advanced members.

[0088] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for generating learning paths for premium members, characterized in that, Includes the following steps: In response to receiving a question input by a senior member, the system extracts the corresponding question features based on the question type, where the question type corresponds to different subject types. The extracted question features are input into the knowledge point matching model, which obtains the core knowledge points corresponding to the question based on a pre-set knowledge point description-feature keyword mapping library. Based on the senior members' historical learning data on the core knowledge points, a multi-dimensional weighted algorithm is used to obtain a rating of the senior members' mastery of the core knowledge points. Based on the core knowledge points, similar knowledge points are selected from the preset knowledge point graph. Based on the core knowledge points and similar knowledge points, a preliminary list of questions is obtained through the preset knowledge point ID-question ID list mapping relationship. The initial screening list of questions is then further filtered, and questions corresponding to knowledge points with a weak mastery level are selected as the preferred ones.

2. The learning path generation method for premium members according to claim 1, characterized in that, The construction process of the preset knowledge point graph includes: Based on the textbook versions and chapter-section structure of various K12 subjects, knowledge is broken down into multi-level knowledge points with granularity ranging from large to small. Each knowledge point is labeled with a unique knowledge point ID. Through the pre-set logical relationship between knowledge points, multiple knowledge points are constructed into a knowledge point graph in the format of "parent node-child node-relationship type". The logical relationships mentioned above include prerequisite dependencies, sibling associations, and extensions.

3. The method for generating a learning path for premium members according to claim 2, characterized in that, It also includes, Based on the knowledge point map and question resources, all questions are structured and their core elements are extracted. The core elements include at least the question stem, solution steps, answer, knowledge points tested, and difficulty level. The knowledge points being examined are associated with one or more knowledge point IDs in the knowledge point graph, and a mapping relationship between knowledge point IDs and question ID lists is established based on the knowledge points being examined.

4. The method for generating a learning path for premium members according to claim 3, characterized in that, Based on subject teaching and research experience and historical answer data statistics, a knowledge point description-feature keyword mapping library was established, forming a four-dimensional mapping relationship of knowledge point ID-feature type-feature keyword-keyword weight; Wherein, the knowledge point ID is consistent with the knowledge point ID in the knowledge graph, the feature type corresponds to the extracted question features, the feature keyword is the feature keyword associated with the knowledge point, and the keyword weight represents the contribution of the keyword to the matching of the knowledge point.

5. The learning path generation method for premium members according to claim 4, characterized in that, The knowledge point matching model is based on an optimized Bayesian algorithm. Specifically, for an input question feature set F, the likelihood probability of each feature in the feature set F is adjusted based on the preset weights of the feature types and the keyword weights in the knowledge point description-feature keyword mapping library. This includes: calculating the original likelihood probability of each feature, multiplying the original likelihood probability by the feature type weight and the corresponding keyword weight to obtain the optimized likelihood probability of each feature, and calculating the optimized likelihood probability of the question feature set F based on the optimized likelihood probability of each feature.

6. The method for generating a learning path for premium members according to claim 5, characterized in that, It also includes, Before adjusting the likelihood probability of each feature in the feature set F, the likelihood probability of each feature is smoothed using Laplace smoothing to obtain the smoothed likelihood probability of each feature. The smoothed likelihood probability of each feature is then adjusted to obtain the optimized likelihood probability of each feature.

7. A learning path generation device for premium members, characterized in that, include: The question feature extraction module is used to extract corresponding question features based on the question type in response to receiving a question input from a senior member. The question type corresponds to different subject types. The core knowledge point acquisition module is used to input the extracted question features into the knowledge point matching model. The knowledge point matching model obtains the core knowledge points corresponding to the question based on the preset knowledge point description-feature keyword mapping library. The rating module is used to obtain a rating of the senior member's mastery of the core knowledge points based on the senior member's historical learning data on the core knowledge points through a multi-dimensional weighted algorithm. The initial screening module is used to screen similar knowledge points from a preset knowledge point map based on the core knowledge points. Based on the core knowledge points and similar knowledge points, a preliminary screening question list is obtained through a preset knowledge point ID-question ID list mapping relationship. The secondary screening module is used to perform a secondary screening of the initial screening question list, and preferentially return the questions corresponding to the knowledge points whose mastery level is rated as weak.

8. An electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the learning path generation method for advanced members as described in any one of claims 1-6.

9. A computer-readable medium having stored thereon computer program instructions that can be executed by a processor to implement the learning path generation method for advanced members as described in any one of claims 1-6.

10. A computer program product comprising a computer program that, when executed by a processor, implements the learning path generation method for premium members as described in any one of claims 1-6.

Images