Learning resource pushing method based on artificial intelligence

Abstract

The invention belongs to the field of course recommendation, and provides an artificial intelligence-based learning resource pushing method, which comprises the following steps that: a learning system obtains the age of a user, and determines a first brain development degree evaluation result according to the age of the user; inputting the first brain development degree evaluation result into a first model, wherein the first model recommends a first learning resource; inputting the first brain development degree evaluation result and a first learning resource into a second model, wherein the second model generates a cognitive competence test question according to the age of the user and the first learning resource; inserting cognitive competence test questions in the teaching process of the first learning resource; sending the user age and cognitive competence test question test result to a third model, and outputting a brain development degree evaluation correction result of the user by the third model; and sending a brain development degree evaluation and correction result to the first model, and recommending a second learning resource by the first model according to the brain development degree evaluation and correction result.

Description

Technical Field

[0001] The present invention belongs to the field of course recommendation, and in particular relates to a learning resource push method based on artificial intelligence. Background Art

[0002] Traditional learning resource recommendation systems usually push resources based on users' interest preferences or historical learning records, but this method ignores the actual situation of users' cognitive ability differences and brain development levels to a certain extent, making it difficult to achieve accurate adaptation to the learning needs of different users.

[0003] Existing studies have shown that the degree of development of the human brain varies significantly at different ages, and these differences directly affect learners' cognitive abilities, information processing speeds, and learning outcomes. Therefore, incorporating the user's brain development level into the core parameters for learning resource recommendations can better match the user's actual learning needs, thereby improving learning efficiency. However, the assessment of brain development involves multiple complex cognitive ability dimensions (such as memory, logical reasoning ability, language comprehension ability, etc.), and the assessment results may deviate from the user's actual cognitive level. How to dynamically correct and optimize the recommendation strategy remains an important technical challenge. Summary of the invention

[0004] In order to solve the problems in the prior art, the present invention provides a learning resource push method based on artificial intelligence, comprising the following steps:

[0005] Step S10, the learning system obtains the user's age, and the learning system determines a first brain development degree assessment result according to the user's age;

[0006] Step S20, the learning system inputs the first brain development level assessment result into a first model, and the first model recommends a first learning resource;

[0007] Step S30, the learning system inputs the first brain development assessment result and the first learning resource into a second model, and the second model generates cognitive ability test questions according to the user age and the first learning resource;

[0008] Step S40, the learning system inserts the cognitive ability test questions into the teaching process of the first learning resource;

[0009] Step S50, the learning system sends the user's age and the cognitive ability test results to the third model, and the third model outputs the user's brain development level assessment correction result;

[0010] Step S60: The learning system sends the revised result of the brain development level assessment to the first model, and the first model recommends a second learning resource based on the revised result of the brain development level assessment.

[0011] Furthermore, the first brain development level assessment result is an average development level reference value obtained by analyzing statistical data according to age distribution.

[0012] Furthermore, the first model is a rule-based model or a collaborative filtering model or a neural network model.

[0013] Furthermore, the collaborative filtering model generates learning resource recommendations corresponding to the user evaluation results by mining the learning resource preference data of users whose development level deviations are within a preset value.

[0014] Furthermore, the second model is a rule-based adaptive test question generation model or a deep learning-based dynamic evaluation model.

[0015] Furthermore, the cognitive ability test questions are of the same subject type as the first learning resource.

[0016] Furthermore, the learning system dynamically selects a suitable teaching node as a question insertion point based on monitoring the user's learning status of the first learning resource.

[0017] Furthermore, the third model extracts features from the received test results of the user's age and cognitive ability test questions, and normalizes the test results using a preset feature weight rule.

[0018] Furthermore, the third model compares the user's test results with statistical data of users of the same age group based on a preset brain development assessment rule library, and calculates the deviation between the user's current ability and the average brain development level.

[0019] Furthermore, the first model analyzes and corrects the score weights of each capability indicator in the result, and gives priority to recommending learning resources that can improve the user's weak capability areas.

[0020] The present invention provides a learning resource push method based on artificial intelligence. By introducing the user's brain development assessment, combined with cognitive ability test results and dynamic correction mechanism, it solves the problems of insufficient assessment accuracy, weak resource recommendation targeting, and disconnection between assessment and recommendation in the prior art. Its beneficial effects are mainly reflected in the following aspects:

[0021] The present invention preliminarily evaluates the user's cognitive ability by combining multi-dimensional indicators of user age and brain development level, and dynamically corrects the evaluation model using the cognitive ability test results to ensure that the evaluation results can accurately reflect the user's actual cognitive level and avoid the deviation problem caused by single-dimensional evaluation of traditional methods.

[0022] The present invention seamlessly inserts cognitive ability test questions combined with learning resource content into the teaching process, achieving a close integration of learning and assessment, which not only ensures the user's learning experience, but also can capture the user's cognitive ability changes in real time, and further optimize the assessment and recommendation strategies.

[0023] The present invention dynamically adjusts the recommendation strategy of learning resources according to the corrected brain development level assessment results, gives priority to matching learning resources in the user's weak areas, while taking into account the expansion of strong abilities, ensuring that the difficulty and content of the learning resources are highly matched with the user's current cognitive level and learning needs, and effectively improving learning outcomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0026] Below, the invention is preferably described in conjunction with the accompanying drawings and specific implementation methods.

[0027] This embodiment solves the above problem through the following steps:

[0028] In one embodiment, reference Figure 1 The present invention provides a learning resource push method based on artificial intelligence, which involves a technical solution for analyzing user characteristics through an intelligent model and accurately matching personalized learning content to meet the different cognitive levels and learning needs of users.

[0029] Specifically, the method comprises the following steps:

[0030] Step S10: The learning system obtains the user's age, and the learning system determines a first brain development degree assessment result according to the user's age.

[0031] The evaluation result is an average development level reference value obtained by analyzing large-scale statistical data by age distribution. Specifically, the first brain development level evaluation result is determined by a pre-built model, which is based on statistical data of brain cognitive abilities of users of different age groups, including the distribution range of core indicators such as attention, memory, language comprehension ability and executive function, for quantitative evaluation. However, due to differences between individuals, the first brain development level evaluation result is a general reference value at the age level, and the specific development level of the actual individual may deviate from the statistical mean. The evaluation result provides a preliminary basis for the subsequent push of learning resources and dynamic adjustments.

[0032] Optionally, the brain development results can be mapped according to the user's age. For example, when the user is 10 years old, the corresponding memory score is 80 / 100 and the logic score is 60 / 100. Similarly, for other age groups, the corresponding average brain development scores are determined based on statistical data.

[0033] Step S20: The learning system inputs the first brain development level assessment result into a first model, and the first model recommends a first learning resource.

[0034] The first model may adopt a rule-based model, determine a preset learning resource set through the first brain development level assessment result, and recommend the preset learning resource set to the user.

[0035] Exemplarily, a set of learning resources corresponding to different developmental levels is set in the database, such as a preset number of learning resources corresponding to a logical reasoning ability score of 60 / 100-70 / 100. When the user's first brain development level assessment result meets the set rules, the corresponding learning resources are recommended.

[0036] The first model can adopt a collaborative filtering model to generate learning resource recommendations corresponding to the user's evaluation results by mining the learning resource preference data of users with the same or similar developmental levels; or adopt a content-based recommendation model to combine specific cognitive indicators (such as attention level, language ability, etc.) in the user's first brain development level assessment results with the matching degree of content characteristics of learning resources, and output the first learning resource that best suits the user.

[0037] Furthermore, the collaborative filtering model is implemented in the following manner:

[0038] The feature vector of each user is expanded to the following form:

[0039] f u =[a u ,memory u ,logic u ,language u,p u ]

[0040] in:

[0041] f u A feature vector representing the user;

[0042] a u Indicates the user's age;

[0043] memory u Indicates the user's memory score;

[0044] logic u Indicates the user's logical reasoning ability score;

[0045] language u Indicates the user's language comprehension ability score;

[0046] p u Represents the user preference score, which can be counted from the user's historical behavior.

[0047] The feature vector of each learning resource is defined as:

[0048] f r =[d r ,memory r ,logic r ,language r ]

[0049] in,

[0050] f r A feature vector representing learning resources;

[0051] d r Indicates the difficulty of the resource;

[0052] memory r Indicates the degree of demand for memory capacity by resources;

[0053] logic r Indicates the degree to which the resource requires reasoning ability;

[0054] language r Indicates the degree to which resources require language comprehension capabilities.

[0055] Based on the user feature vector and resource feature vector, define the matching function between users and resources:

[0056]

[0057] in:

[0058] wi is a weight parameter used to balance the importance of different cognitive dimensions (such as memory ability, logical ability, etc.).

[0059] f u,i and f r,i are the i-th dimension in the user and resource feature vectors respectively.

[0060] The matching function is calculated for the user and the resource to obtain an m×n user-resource interaction matrix M. The element M in the user-resource interaction matrix i,j Indicates the matching degree of the i-th user to the j-th resource.

[0061] Construct user set U = {u1,u2…u m}, where m represents the number of users, u m represents the mth user.

[0062] Construct a learning resource set R = {r1, r2…r n}, where n represents the number of resources, r n Indicates the nth resource.

[0063] For any user u, determine the set of users N(u) whose similarity is greater than a preset threshold, and determine the degree of coordination between user u and resource i based on the behavior of these users.

[0064]

[0065] Among them, sim(u,v) represents the similarity between users u and v, which can be calculated using Pearson correlation coefficient, cosine similarity, etc. v,i Indicates the matching degree between user v and resource i.

[0066] The traditional collaborative filtering model only makes recommendations based on the user's historical behavior and rating data, ignoring the differences in users' cognitive abilities. The improved model introduces the user's brain development assessment results (such as memory ability, logical reasoning ability, language comprehension ability, etc.), which can more accurately capture the user's learning needs. Exemplarily, when the user's first brain development assessment result shows that his memory ability score is high (for example, a score of 85 / 100), but his logical reasoning ability score is low (for example, a score of 40 / 100), the first model compares other user data with similar brain development assessment results in the large-scale learning database, and extracts learning resources that are commonly used by such users and have significant effects, such as interesting math puzzle resources and spatial logic training games for improving logical reasoning ability.

[0067] In addition, the first model can adopt a neural network model, in which the input layer receives the first brain development level assessment result, the hidden layer processes the input data through a multidimensional weight matrix constructed through training data, and the output layer generates a list of learning resources suitable for the user's current cognitive level.

[0068] Specifically, the first model structure is as follows:

[0069] Input Layer:

[0070] The input layer receives the user’s brain development assessment results and related features, including the following information:

[0071] User age: indicates the basic age stage of users.

[0072] Memory ability score: Quantify the evaluation value of the user's memory ability.

[0073] Logical reasoning ability score: The score of the user's logical ability.

[0074] Language comprehension score: A score of the user's language ability.

[0075] User preference characteristics: For example, user preferences for certain subjects or resource types (such as mathematics, language learning, etc.).

[0076] These features constitute the user's feature vector, which serves as the input of the model.

[0077] Hidden layer:

[0078] The hidden layer performs multi-layer nonlinear processing on the input user features:

[0079] The first hidden layer: performs preliminary processing on the basic features of user input and extracts important feature relationships.

[0080] For example, based on the combination of memory ability and logical ability, the user's potential needs in different learning areas are extracted.

[0081] The second hidden layer: combines resource tag characteristics and user features to learn more complex interaction relationships.

[0082] For example, matching users with low logic ability scores with resource logic training requirements can discover potential resource recommendation priorities.

[0083] The third hidden layer: reduces the dimension of features and generates matching scores between users and resources to provide support for the output layer.

[0084] Output layer:

[0085] The output layer generates a probability distribution for learning resource recommendations, and the recommendation probability of each learning resource indicates the possibility that the user is suitable for the resource. The model outputs a recommended list of learning resources and sorts them by priority. Exemplarily, the input layer receives the results of the first brain development assessment, including specific indicators such as an attention score of 75 / 100 and a language comprehension ability of 90 / 100; the hidden layer processes the input features through a weight matrix generated by the training data, and outputs the learning resources that best match these indicators, such as difficult reading comprehension exercises, interactive learning animations that require high concentration, and memory-based word memory challenge tasks.

[0086] Through the above method, the first model can recommend the first learning resource that matches the user's cognitive ability based on the user's specific brain development characteristics, thereby ensuring the personalization and targeting of the learning content.

[0087] The first learning resource refers to personalized learning tutorials recommended based on the user's brain development assessment results. These tutorials are presented in the form of specific courses and cover learning content of different difficulty levels and themes. The first learning resource can include mathematics courses, language courses, science courses, professional skills courses, etc., aiming to meet the user's current cognitive ability and learning needs, and promote the maximization of learning effects through highly adaptable content design.

[0088] For example, user characteristics: Age: 10 years old Brain development assessment: weak logical reasoning ability, high memory ability score

[0089] Learning goal: Improve logical reasoning ability

[0090] Recommended first learning resources:

[0091] Course type: elementary mathematics course. Step S30, the learning system inputs the first brain development assessment result and the first learning resource into a second model, and the second model generates cognitive ability test questions according to the first brain development assessment result and the first learning resource.

[0092] In this step, the cognitive ability test questions are used to measure whether the first brain development level assessment results match the user's actual abilities; the test questions are designed to be combined with the first learning resource content and integrated into the learning process to achieve dynamic assessment of the user's cognitive abilities while the user is learning; the second model includes a rule-based adaptive test question generation model or a deep learning-based dynamic assessment model.

[0093] The second model can be a rule-based adaptive test question generation model, which generates targeted cognitive ability test questions based on the cognitive ability score range (such as memory, attention, logical reasoning ability, etc.) in the first brain development level assessment result and the content characteristics of the first learning resource. Through the accuracy rate, reaction time and answering mode of the user's completion of the test questions, measure whether the user's actual cognitive ability is consistent with the assessment results. By combining the user's brain development assessment results, test questions that assess the user's cognitive ability are generated to verify the matching degree between the learning resources and the user's ability. According to the content of the first learning resource, test questions that are naturally embedded in the learning process are designed to avoid user-perceived obvious evaluation behaviors.

[0094] For example, if the first brain development assessment result shows that the user's logical reasoning ability score is 60 / 100, and the first resource is a mathematics class, the test questions generated by the model may be "find an increasing pattern in a series of numbers" or "determine whether two mathematical statements contradict each other". By analyzing the user's answer performance, it is verified whether his logical reasoning ability meets the assessment expectations.

[0095] The second model can be a dynamic evaluation model based on deep learning. The model compares the user's real-time answer data (such as accuracy, reaction speed, error pattern, etc.) with the pre-trained ability evaluation network, and dynamically generates subsequent test questions to more accurately verify the user's actual ability.

[0096] In one implementation, the dynamic evaluation model may include a generator and a discriminator. In the dynamic evaluation model, the generator is used to generate personalized cognitive ability test questions, and the discriminator is used to determine whether the generated test questions meet the user's ability characteristics and learning resource requirements, thereby achieving high-quality test question generation.

[0097] Generator:

[0098] The generator is a neural network that takes the following input data:

[0099] User features (embedding vectors).

[0100] Learn resource features (embedding vectors).

[0101] Random noise vector (used to introduce test question diversity).

[0102] The generator maps the input to the test question feature space through a multi-layer neural network, including question content, question type, difficulty, etc. For example:

[0103] Input: [user features + learning resource features + noise].

[0104] Output: [Question content: "Calculate the next number in the following sequence"; Question type: Multiple choice; Target ability: Logic; Difficulty: Medium].

[0105] The output characteristics of the generator are decoded by the decoding module to generate specific test questions for users to answer.

[0106] Discriminator:

[0107] The discriminator is a binary classification neural network that evaluates whether the generated test questions meet the requirements of the user's ability level and the content of the learning resource. Its input includes:

[0108] Characteristics of the generated test questions (such as question type, difficulty, and target ability).

[0109] User’s real characteristics (such as brain development assessment results).

[0110] Learning resource characteristics (such as knowledge point category and content difficulty).

[0111] The discriminator outputs a binary probability, indicating whether the generated test question meets the requirements, which can be determined by statistical features or manually:

[0112] 0: The question is not suitable for the user or learning resource.

[0113] 1: The test questions meet the requirements. When the user's first brain development assessment result shows that the language comprehension ability is 80 / 100, and the first learning resource is language and literature, the model will initially generate test questions of language and literature, such as "read a short passage and answer questions". If the user's accuracy rate is significantly higher than expected, the model will dynamically adjust to generate more complex test questions (such as "analyze the cause and effect relationship between sentences in a given paragraph") to further verify the accuracy of the assessment results.

[0114] The cognitive ability test questions generated by the second model are highly integrated with the first learning resource and have the same subject type. The learning system completes the assessment of the user's cognitive ability while providing teaching content, ensuring the continuity of the learning process and the authenticity of the test question assessment, and providing efficient support for subsequent learning resource recommendations and ability assessment corrections.

[0115] Step S40: the learning system inserts the cognitive ability test questions into the teaching process of the first learning resource.

[0116] Specifically, the learning system dynamically selects appropriate teaching nodes as test question insertion points based on monitoring the user's learning status of the first learning resource (including learning time, interactive operations, completion progress, etc.). For example, when the first learning resource is mathematical operation skill training, the learning system inserts a logical reasoning question after the user completes a certain number of basic operation tasks to evaluate the user's thinking coherence and comprehension ability.

[0117] Furthermore, the learning system can adjust the presentation of test questions according to the characteristics of cognitive ability test questions (such as test difficulty, answering time, and assessment objectives). For example, for test questions that take a long time, the learning system can choose to insert test questions at key stages or intervals of the teaching module; for simple instant assessment test questions, the learning system can present them in real time in the form of pop-up windows when the user is learning key knowledge points, so as to capture the user's cognitive ability performance in a timely manner during the teaching process.

[0118] Through the above insertion strategy, the learning system can achieve seamless integration of cognitive ability test questions and teaching content, dynamically evaluate the user's cognitive level while ensuring the user's learning experience, and provide data support for the optimization and push of subsequent learning resources.

[0119] In step S50, the learning system sends the user's age and the cognitive ability test results to the third model, and the third model outputs the user's brain development assessment correction result. The correction result is used to optimize the push of subsequent learning resources. The third model achieves comprehensive assessment and correction of the user's cognitive ability by integrating the rule model and the deep learning model.

[0120] The third model extracts features from the received test results of the user's age and cognitive ability, including indicators such as the user's answer accuracy, reaction time, error distribution, and answer logic.

[0121] For users of different age groups, the model normalizes the test results through preset feature weight rules to make the test results of different users comparable.

[0122] The third model compares the user's test results with the statistical data of users of the same age group based on the preset brain development assessment rule library, and calculates the deviation between the user's current ability and the average brain development level. For example, if a 10-year-old user's memory ability score is significantly higher than the average level, but the attention score is lower, the rule model quantifies the deviation as an assessment correction factor. In this case, the memory ability score can be increased, while the attention score can be lowered.

[0123] Step S60: The learning system sends the revised result of the brain development level assessment to the first model, and the first model recommends a second learning resource based on the revised result of the brain development level assessment.

[0124] The first model receives the revised results of the brain development assessment, which include the user's specific scores and deviation information in multiple cognitive ability indicators (such as memory ability, attention, logical reasoning ability, language comprehension ability, etc.).

[0125] The model analyzes and modifies the score weights of each capability indicator in the result, giving priority to recommending learning resources that can improve the user's weak capability areas, while retaining learning content that can further improve the user's strong capabilities.

[0126] The first model screens the resources in the learning resource database, combines resource tags (such as knowledge point classification, difficulty level, interaction form, etc.) with the cognitive ability requirements in the evaluation and correction results, and selects the learning resources that best suit the user's current ability level.

[0127] For example, when the user's logical reasoning ability correction result shows a score of 65 / 100 and a language comprehension ability of 90 / 100, the model will preferentially match interactive learning resources containing logical reasoning training, while adding an appropriate amount of language comprehension challenge content.

[0128] Furthermore, the first model dynamically adjusts the recommendation strategy through historical learning data and current evaluation correction results, including the gradual progression of resource difficulty, the diversification of learning task types, and the focus of resource allocation.

[0129] For example, for users who show weak logical ability and whose learning goal is in the field of mathematics, the model prioritizes recommending elementary mathematical problems related to logical reasoning, such as "guessing the pattern of number series" or "simple logical relationship diagram", while appropriately adding interesting interactive content to enhance learning interest.

[0130] The prior art mentioned in the above background technology section and specific embodiment section of the present invention can be used as part of the present invention to understand the meaning of some technical features or parameters.

Claims

1. A learning resource push method based on artificial intelligence, characterized in that: The method comprises the following steps: Step S10, the learning system obtains the user's age, and the learning system determines a first brain development degree assessment result according to the user's age; Step S20, the learning system inputs the first brain development level assessment result into a first model, and the first model recommends a first learning resource; Step S30, the learning system inputs the first brain development assessment result and the first learning resource into a second model, and the second model generates cognitive ability test questions according to the first brain development assessment result and the first learning resource; Step S40, the learning system inserts the cognitive ability test questions into the teaching process of the first learning resource; Step S50, the learning system sends the user's age and the cognitive ability test results to the third model, and the third model outputs the user's brain development level assessment correction result; Step S60: The learning system sends the revised result of the brain development level assessment to the first model, and the first model recommends a second learning resource based on the revised result of the brain development level assessment.

2. The method for pushing learning resources based on artificial intelligence according to claim 1, characterized in that: The first brain development level assessment result is an average development level reference value obtained by analyzing statistical data according to age distribution.

3. The method for pushing learning resources based on artificial intelligence according to claim 1, characterized in that: The first model is a rule-based model, a collaborative filtering model, or a neural network model.

4. The method for pushing learning resources based on artificial intelligence according to claim 3 is characterized in that: The collaborative filtering model mines the learning resource preference data of users whose development level deviation is within a preset value, and generates learning resource recommendations corresponding to the user evaluation results.

5. The method for pushing learning resources based on artificial intelligence according to claim 1, characterized in that: The second model is a rule-based adaptive test question generation model or a deep learning-based dynamic evaluation model.

6. The method for pushing learning resources based on artificial intelligence according to claim 1, characterized in that: The cognitive ability test questions are of the same subject type as the first learning resource.

7. The method for pushing learning resources based on artificial intelligence according to claim 1, characterized in that: The learning system dynamically selects a suitable teaching node as a question insertion point based on monitoring the user's learning status of the first learning resource.

8. The method for pushing learning resources based on artificial intelligence according to claim 1, characterized in that: The third model extracts features from the received test results of the user's age and cognitive ability test questions, and normalizes the test results using a preset feature weight rule.

9. The method for pushing learning resources based on artificial intelligence according to claim 1, characterized in that: The third model is based on a preset brain development assessment rule library, compares the user's test results with statistical data of users of the same age group, and calculates the deviation between the user's current ability and the average brain development level.

10. The method for pushing learning resources based on artificial intelligence according to claim 1, characterized in that: The first model analyzes and corrects the score weights of each capability indicator in the result, and gives priority to recommending learning resources that can improve the user's weak capability areas.

Images