Course teaching resource personality recommendation method based on AI big data

Abstract

The invention discloses a curriculum teaching resource personalized recommendation method based on AI big data, and relates to the technical field of curriculum recommendation, and the method comprises the following steps: S1, defining a plurality of features for each curriculum teaching resource, and carrying out the assignment of each feature through expert evaluation, automatic evaluation or learning path analysis, enabling each course teaching resource to obtain a feature set with a plurality of features and corresponding feature values; s2, on the basis of the mutual relation between the curriculum teaching resource features, setting correlation values between each feature in the feature set and other features; and S3, non-privacy learning behavior data of the student is collected through the corresponding learning platform, and the learning behavior data is authorized non-privacy data. By combining the learning behavior data of the students and the dynamic correction of the course features, the accuracy and practicability of the personalized recommendation system are improved, and the recommendation process can more flexibly adapt to the change requirements of the students.

Description

Technical Field

[0001] The present invention relates to the field of course recommendation technology, and specifically to a method for personalized recommendation of course teaching resources based on AI big data. Background Art

[0002] With the continuous development of educational technology, personalized learning recommendation systems have become an important tool for improving teaching quality and learning efficiency. In recent years, with the widespread application of big data and artificial intelligence technologies, more and more educational platforms have adopted recommendation algorithms based on student behavior data. These algorithms aim to provide customized learning resources by analyzing dynamic data such as students' learning paths, interests, preferences, and learning progress. These personalized recommendation systems can recommend the most appropriate course content based on students' needs and behavioral changes, thereby helping students achieve a better learning experience and results.

[0003] Existing personalized course recommendation methods typically rely on students' historical learning behavior and course content characteristics, using traditional techniques such as collaborative filtering and content-based recommendation to provide recommended courses. These methods have achieved good results in practical applications and can recommend appropriate courses based on students' historical behavior and academic performance data.

[0004] After searching, a Chinese patent (publication number: CN117194716A) discloses a personalized recommendation system for educational robot courses based on big data. This patent obtains video interest and tag preference and preference tags by analyzing the user's viewing behavior within a preset period; obtains similar preference users based on the difference in preference tags between users, and then obtains suspected preference tags that the user has not been exposed to; by analyzing the tag preference and contact frequency in the suspected preference tag video, combined with the preference differences, the user's potential preference for the suspected preference tag is obtained; the potential preference tags are screened out, and based on the preference tags, potential course videos of interest are recommended to the user; the user's potential preferences are continuously updated to make personalized recommendations.

[0005] Existing technologies still have certain limitations when dealing with the cross-influence between student emotional fluctuations, dynamic learning progress, and course characteristics. They are unable to fully and dynamically adjust recommended content to meet students' personalized learning needs. Therefore, how to dynamically adjust course recommendations based on the relationship between students' real-time behavioral data, learning emotions, learning progress, and course characteristics to achieve more accurate personalized recommendations has become an important research direction in the field of current educational recommendation systems. To this end, this application proposes a personalized recommendation method for course teaching resources based on AI big data. Summary of the Invention

[0006] The purpose of the present invention is to provide a personalized recommendation method for course teaching resources based on AI big data to solve the problems mentioned in the above background technology.

[0007] The present invention can be implemented through the following technical solutions: A personalized recommendation method for course teaching resources based on AI big data, the method comprising the following steps: S1. Define multiple features for each course teaching resource and assign values to each feature, and each course teaching resource thus obtains a feature set with multiple features and corresponding feature values; Each feature covers different dimensions of the course teaching resource and provides basic data for subsequent recommendations, which can be obtained through expert evaluation, automated evaluation or learning path analysis; Among them, expert evaluation: The difficulty of a course can be determined through the evaluation of domain experts; experts assign a difficulty score to the course according to factors such as the complexity of the course content, the required background knowledge, learning objectives, etc.; Automated evaluation (based on course content analysis): Use natural language processing (NLP) technology to analyze course materials (such as video scripts, textbooks, lecture notes, etc.); Learning path analysis: Based on the historical learning data of students, if the time required for students to complete the course is long and the failure rate is high, it is automatically inferred that the course is of high difficulty; The features of the course teaching resources include but are not limited to the following categories: Course difficulty: Represents the difficulty of the course content, expressed using a digital scale; Course duration: Represents the learning duration of the course content, with the unit of minutes; the course duration can be determined by the video playback duration, the reading duration of documents, or the average time to complete assignments; Learning objective: Represents the learning objective or expected learning outcome of the course, expressed using a standardized digital label; the value range of this feature is from 0 to 1, indicating the target level of the course; Course type: Represents the presentation form or type of the course; the value range of this feature is from 0 to 1, and different course types are represented by numbers; Course popularity: Calculated based on historical data such as student evaluations, learning participation, and scores, with a value range of 0 to 1; the higher the value, the more popular the course; Target learning group: Represents the learning group suitable for the course, with a value range of 0 to 1, and the larger the value, the higher the level of the suitable group; After all the feature values are assigned, each course teaching resource will generate a corresponding feature set, which contains all the features of the course and their corresponding feature values; The feature set of each course teaching resource represents the multi-dimensional attributes of the course and is the basic input data in the subsequent personalized recommendation system; S2. Based on the mutual relationships between the characteristics of course teaching resources, correlation values are set for each characteristic in the characteristic set with the remaining characteristics. The correlation value between characteristics indicates that when the characteristic value of a certain characteristic changes, the characteristic values of other related characteristics will also be adjusted accordingly; The realization of characteristic values is to set the association rules between characteristics by using data analysis techniques or expert knowledge, analyze the relationships between the characteristics of course teaching resources based on historical data, and set weight correlation values for each pair of related characteristics (such as course duration and course difficulty); S3. Through the corresponding learning platform, collect the learning behavior data of students, and the learning behavior data is authorized non-private data. The non-private data specifically includes: Course access data: The access frequency and duration of students to various course contents (such as the number of times a student accesses the videos, reads articles, or participates in exercises of a certain course per week); Course participation: The frequency of students' participation in in-course interactions, such as the number of views in the course discussion area, the number of question answers, quiz participation, etc.; Learning time: The learning time spent by students on various courses, such as the learning duration invested in a certain course per week; S4. Process and model the collected learning behavior data of students to construct the behavior portrait of the corresponding students. The specific steps of learning behavior data processing include: Data cleaning: Remove invalid data; Data standardization: Standardize different behavior data to a unified scale; And the modeling of learning behavior data is as follows: Based on the learning behavior data of students, establish a student portrait covering information such as their learning progress, interest preferences, and course participation; use clustering algorithms to classify students, and customize personalized recommendation strategies according to the learning needs of different groups; use regression analysis algorithms to evaluate the relationship between students' grades and their learning behaviors to predict students' preferences for course characteristics; S5. Dynamically correct the characteristic values of the corresponding characteristics in the course teaching resources according to the learning behavior data of students; S6. Based on the characteristic values of each characteristic in the corrected characteristic set, dynamically update the course teaching resources, and use AI big data technology to generate a personalized course recommendation list for students, which specifically includes: Recommendation model training: Use collaborative filtering, matrix factorization, or deep neural network models to train the recommendation system, learn the personalized needs of students, and generate a recommendation list; Recommendation strategy optimization: Continuously optimize the recommendation model according to the real-time feedback of students (such as the completion of courses, course ratings, etc.) to ensure that the recommended content meets the latest needs of students; S7. A dynamically updated curriculum teaching resource recommendation list according to the learning progress, interest preferences, and emotional changes of students, specifically including: When the learning progress or interest of a student changes, the system will recalculate the recommendation list based on the new data and give priority to recommending courses that meet the current needs; The interaction data of students during the learning process (such as participation in the discussion area, assignment feedback) will be continuously fed back to the recommendation system to adjust the subsequent recommended course content; Among them, the method for obtaining the learning progress is as follows: The database system stores the progress data of students and updates the system status through real-time data synchronization technology. After each chapter or learning unit in each curriculum teaching resource on the learning platform is completed, the learning progress of students is recorded, and the LMS and learning analysis system are used to track the progress of students in real time; Whenever a student completes a task or module, the system will automatically update the learning progress of the student and record what the student has completed; The method for obtaining the participation degree of students is to participate in discussions, answer questions, watch videos, etc.; The method for obtaining interest preferences is: through interest preference modeling or real-time detection of interest changes; Real-time detection of interest changes uses a machine learning model to analyze the behavior data of students, identify interest changes, perform time series analysis on the behavior data of students, and detect the changing trend of interest in real time; The method for obtaining emotional changes is: sentiment analysis based on a dictionary: By matching sentiment words in the text through a sentiment dictionary, calculate the sentiment polarity (positive or negative); When the student's mood is positive, the recommendation system tends to recommend courses with higher challenges to stimulate the enthusiasm and spirit of challenge of the student; When the student's mood is negative, the recommendation system will give priority to recommending courses with lower difficulty and stronger support to reduce the learning pressure of the student and increase their confidence in learning.

[0008] A further technical improvement of the present invention lies in: In step S2, the method for setting the association value between each feature and the remaining features includes: For the association value between feature i and feature j, determine the association value between them ; The association value ranges between [0, 1]; where 0 means irrelevant and 1 means completely relevant; And the calculation formula of the association value is ; In the formula, is the covariance between feature i and feature j, representing the linear relationship between the two; for example, the linear relationship between "course difficulty" and "course duration". By collecting the learning records of multiple students, if the course difficulty increases and the course duration also increases accordingly, then there is a positive correlation between these two features and the covariance is positive; if the course duration decreases as the difficulty increases, the covariance is negative; and are the variances between feature i and feature j respectively, representing the variation range of each feature.

[0009] A further technical improvement of the present invention lies in that each feature in the feature set is set with a weight , and the calculation rule between each feature is as follows: When the eigenvalue of a certain feature changes, its weight must be greater than the weight of the feature to be calculated for the change to spread; Features with small weights will not affect the changes of other features and will only be affected by the changes of other features; Features with the same weight affect each other; And the correlation between each feature in the feature set is represented by the correlation matrix A: ; where, the on the diagonal is 1, indicating that each feature is completely correlated with itself; When the eigenvalue of feature changes, only when the weight of feature is greater than the weight of the feature to be calculated , can the change amount of feature be calculated through the correlation value using the formula: ; In the formula, is the indicator function, which takes the value of 1 if and only if the weight of feature is greater than the weight of feature , indicating that feature can affect feature ; is the change amount of feature ; And features with weights greater than the feature to be calculated will spread the influence through the correlation value, while features with smaller weights will not affect other features; Features with the same weight will affect each other and propagate the influence through their correlation values.

[0010] A further technical improvement of the present invention lies in that: when the feature values of multiple features change simultaneously, the feature change amount of the target feature is affected by multiple features, and the following operations are performed: Z1. Determine which features have an impact on the target feature through the correlation value and weight between features, and establish an influence feature set for the target feature ; Specifically, it is obtained through calculation and judgment using the correlation matrix A and weight ; Z2. When calculating, preferentially select the feature with the largest weight, and use the change amount of this feature to adjust the feature value of the target feature Z3. If the feature values of multiple features change simultaneously, and their influence propagation is calculated through the correlation value , then the final feature change amount of the target feature is determined by the feature with the largest weight among the multiple features that change simultaneously, that is, select the feature with the largest weight in the influence feature set; The calculation formula for the feature change amount is: ; In the formula, represents the correlation value between the feature with the largest weight and the target feature is the change amount of the feature with the largest weight

[0011] A further technical improvement of the present invention lies in that: if the changes of multiple features with the same weight have an impact on the target feature , then select the feature with the largest weight and the largest influence on the change amount from the change amounts of the multiple features as the determinant of the change amount of the target feature.

[0012] A further technical improvement of the present invention lies in that: in step S5, the method for dynamically correcting the feature value of a feature is: Q1. Establish an update rule for each feature in the feature set of course teaching resources; Q2. Map the collected learning behavior data to each feature in the course teaching resources to form a one-to-one correspondence, and match the mapping result with a preset mapping table to obtain the correction amount for the corresponding feature; Q3. For each feature in the course teaching resources , based on the mapping result in step Q2, through the formula , obtain the new feature value of the feature ; In the formula, is the original feature value of the feature ; is the correction coefficient, which is calculated through experiments or historical data; is the correction amount of the feature , which is obtained from the result of mapping the learning behavior data and the course teaching resources.

[0013] A further technical improvement of the present invention lies in: in step S6, the method for generating the course recommendation list includes: Y1. Represent the feature set of each course teaching resource as , where is the feature value of the i-th feature; Convert the student behavior data into a vector to generate a behavior feature set T, , where is the feature value of the student on a certain specific i behavior; Y2. Calculate the similarity between the feature set C of each course teaching resource and the behavior feature set T of the student behavior data; Y3. Based on the similarity in Y2, select the top N course teaching resources for sorting.

[0014] Compared with the prior art, the present invention has the following beneficial effects: By combining the learning behavior data of students and the dynamic correction of course features, the present invention not only improves the accuracy and practicality of the personalized recommendation system, but also enables the recommendation process to more flexibly adapt to the changing needs of students; compared with the prior art, the present invention can dynamically adjust the course features according to factors such as the progress, grades, and emotional fluctuations of students during the learning process, making the recommended course content more in line with the real-time needs of students and improving the real-time response ability and accuracy of the recommendation system; And the present invention can optimize the recommendation result by real-time detecting the emotional state of students. Through the recognition and adaptation of students' emotions, the recommendation system can automatically adjust the recommended content according to the emotional state of students, which can help students maintain a positive learning attitude, enhance their learning motivation, and thus improve the learning effect; In summary, the proposed solution can consider student behavior data in more dimensions in personalized recommendation, combine the interrelationships between course features, ensure the efficiency and personalization of the recommendation results, and significantly improve the learning experience of students. The ability to dynamically adjust course features and recommended content enables the system to better adapt to the needs of students at different learning stages, providing a more accurate and flexible way to recommend learning resources for educational platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings.

[0016] Figure 1 is a flowchart of the present invention; Figure 2 is a process mind map of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, with reference to the accompanying drawings and preferred embodiments, describe in detail the specific embodiments, structures, features, and effects of the present invention.

[0018] Please refer to Figure 1-2 as shown, the present invention provides a personalized recommendation method for course teaching resources based on AI big data, and the method includes the following steps: S1. Define multiple features for each course teaching resource and assign values to each feature, so that each course teaching resource obtains a feature set with multiple features and corresponding feature values; Each feature covers different dimensions of the course teaching resource and provides basic data for subsequent recommendations, which can be obtained through expert evaluation, automated evaluation, or learning path analysis; Among them, expert evaluation: The difficulty of a course can be determined through the evaluation of domain experts. The expert assigns a difficulty score to the course according to factors such as the complexity of the course content, the required background knowledge, and the learning objectives; Automated evaluation (based on course content analysis): Use natural language processing (NLP) technology to analyze course materials (such as video scripts, textbooks, lecture notes, etc.). For example, by analyzing features such as the complexity of terms, the number of formulas, and the code density in the course, calculate the difficulty score of the course; Learning path analysis: According to the historical learning data of students, if the time required for students to complete the course is long and the failure rate is high, it is automatically inferred that the course has a high difficulty; The features of the course teaching resource include but are not limited to the following categories: Course Difficulty: Represents the difficulty of the course content, expressed using a numerical scale; the value range of this feature is between 0 and 1, where 0 indicates easy and 1 indicates difficult; the difficulty is determined by factors such as the complexity of the course content, the required background knowledge, and the course objectives; The assignment rules for course difficulty can be: If the course is an introductory course, assign a value of 0.2; if it is an intermediate course, assign a value of 0.5; if it is an advanced course, assign a value of 0.8; if it is an expert-level course, assign a value of 1; Course Duration: Represents the learning duration of the course content, with the unit of minutes; the course duration can be determined by the video playback duration, the duration of reading documents, or the average time to complete assignments; The assignment rules for course duration can be: The course duration value can be set as the total learning time of the course, such as 30 minutes, 60 minutes, 90 minutes, etc.; Learning Objectives: Represents the learning objectives or expected learning outcomes of the course, expressed using standardized digital tags; the value range of this feature is between 0 and 1, indicating the target level of the course; The assignment rules for learning objectives can be: Assign a value of 0.2 for the learning objectives of basic courses, 0.5 for advanced courses, and 0.8 for in-depth courses; Course Type: Represents the presentation form or type of the course, such as video, article, interactive exercise, etc.; the value range of this feature is between 0 and 1, using numbers to represent different course types; The assignment rules for course type can be: Assign a value of 0.8 for video courses, 0.5 for article textbooks, 0.2 for exercise questions, and 1 for mixed courses; Course Popularity: Calculated based on historical data such as student evaluations, learning participation, and ratings, with a value range of 0 to 1; the higher the value, the more popular the course; The assignment rules for course popularity can be: According to the course ratings and student participation, if the course rating exceeds 4.5, assign a popularity value of 1; if the rating is between 3.5 and 4.5, assign a popularity value of 0.7; if the rating is below 3.5, assign a value of 0.3; Target Learning Group: Represents the learning group suitable for the course, with a value range of 0 to 1, and the larger the value, the higher the level of the suitable group; The assignment rules for the target learning group can be: Assign a value of 0.2 for beginner courses, 0.5 for intermediate learner courses, and 0.8 for advanced learner courses; After all feature values are assigned, each course teaching resource will generate a corresponding feature set, which contains all the features of the course and their corresponding feature values. The feature set of each course teaching resource represents the multi-dimensional attributes of the course and is the basic input data in the subsequent personalized recommendation system; S2. Based on the mutual relationships between the characteristics of course teaching resources, correlation values are set for each characteristic in the characteristic set with the remaining characteristics. For example, there may be a certain positive correlation between the difficulty of a course and the learning objectives (courses with higher difficulty usually contain more complex learning objectives). The correlation value between characteristics indicates that when the characteristic value of a certain characteristic changes, the characteristic values of other related characteristics will also be adjusted accordingly; The realization of characteristic values is to set the association rules between characteristics by using data analysis techniques or expert knowledge, analyze the relationships between the characteristics of course teaching resources based on historical data, and set weight correlation values for each pair of related characteristics (such as course duration and course difficulty); The method of setting correlation values for each characteristic with the remaining characteristics includes: For the correlation value between characteristic i and characteristic j, determine the correlation value between them ; The correlation value ranges from [0, 1]; where 0 indicates no relevance and 1 indicates full correlation; And the calculation formula for the correlation value is ; In the formula, Cov(i, j) is the covariance between characteristic i and characteristic j, indicating the linear relationship between the two; for example, the linear relationship between "course difficulty" and "course duration", by collecting the learning records of multiple students, if the course difficulty increases and the course duration also increases accordingly, then there is a positive correlation between these two characteristics and the covariance is positive; if the course duration decreases as the difficulty increases, the covariance is negative; and are the variances between characteristic i and characteristic j respectively, indicating the variation ranges of their respective characteristics; the "course difficulty" and "course duration" of course teaching resources will each have their own variation ranges. For example, for a group of course teaching resources, the possible value range of "course difficulty" may be from 1 to 3, and the variation range of duration may be from 30 minutes to 180 minutes; For "course difficulty", calculate the variance of this characteristic. Assume that the difficulty values of multiple courses are 1, 2, and 3 respectively, and calculate their variance; for "course duration", calculate its variance by collecting the duration data of courses (such as 30 minutes, 60 minutes, 120 minutes, etc.); Suppose it is calculated that the covariance between the "course difficulty" and "course duration" of the course is 0.3, the variance of the difficulty characteristic is 0.5, and the variance of the duration characteristic is 1.0. Then through the formula: The calculated correlation value between "course difficulty" and "course duration" is 0.424; Each characteristic in the characteristic set is set with a weight , each feature The calculation rule between them is: When the feature value of a certain feature changes, the change will be propagated only if its weight is greater than the weight of the feature to be calculated; Features with small weights will not affect the changes of other features, but only accept the influence of changes in other features; Features with the same weight influence each other; And each feature in the feature set The correlation between them is represented by the correlation matrix A: ; Among them, the on the diagonal is 1, indicating that each feature is completely correlated with itself; When the feature changes in feature value, only when the weight of the feature is greater than the weight of the feature to be calculated can the change amount of the feature be calculated through the correlation value The change formula of the feature is: ; ; In the formula, is an indicator function. When and only when the weight of the feature is greater than the weight of the feature it takes the value of 1, indicating that the feature can affect the feature ; ; is the change amount of the feature ; Moreover, features with weights greater than the feature to be calculated will propagate the influence through the correlation value, while features with smaller weights will not affect other features; Features with the same weight will influence each other and propagate the influence through their correlation values; When the feature values of multiple features change simultaneously, the feature change amount of the target feature is affected by multiple features, and the following operations are performed: Z1. Determine which features affect the target feature through the correlation value between features and the weight . For the target feature , establish its set of influencing features ; Specifically, it is obtained through calculation and judgment by means of the correlation matrix A and the weights ; Z2. When calculating, the feature with the largest weight is preferentially selected, and the change amount of this feature is used to adjust the eigenvalue of the target feature ; Z3. If the eigenvalue changes of multiple features occur simultaneously, and their influence propagation is calculated through the correlation value , then the target feature The final change amount of the feature is determined by the feature with the largest weight among the multiple features that occur simultaneously, that is, the feature with the largest weight is selected from the set of influencing features ; ; The calculation formula for the feature change amount is: ; In the formula, represents the correlation value between the feature with the largest weight and the target feature ; is the change amount of the feature with the largest weight ; Moreover, if the changes of multiple features with the same weight affect the target feature , then from the change amounts of the multiple features, the feature with the largest weight and the largest change amount influence is selected as the determining factor for determining the change amount of the target feature ; S3. Through the corresponding learning platform, collect the learning behavior data of students, and the learning behavior data is authorized non-private data. The non-private data specifically includes: Course access data: The access frequency and duration of students to various course contents (such as the number of times a student accesses a video, reads an article, or participates in practice questions of a certain course per week); Course participation: The frequency of students' participation in in-course interactions, such as the number of times of browsing the course discussion area, answering questions, participating in quizzes, etc.; Learning time: The learning time spent by students on various courses, such as the learning duration invested in a certain course per week; The specific learning behavior data is interconnected with the learning platform through the API interface to obtain the public learning behavior data of students, and these learning behavior data include text data such as students' comments, discussion contents, course feedback, learning logs, etc. in the platform learning platform; And after obtaining the learning behavior data, perform text preprocessing on it, including: Noise removal: Remove stop words (such as "de", "shi", etc.), special symbols, meaningless characters, etc. from the text; Word segmentation and part-of-speech tagging: Use NLP tools (such as spaCy, NLTK) to segment the text and extract meaningful words; Sentiment vocabulary construction: Utilize sentiment lexicons (such as SentiWordNet, sentiment dictionary) to enhance the representation ability of sentiment vocabulary to improve the accuracy of sentiment analysis; S4. Process and model the collected student learning behavior data to construct a behavior portrait of the corresponding student. The specific steps for processing the student learning behavior data include: Data cleaning: Remove invalid data (such as learning records of students who quit midway or duplicate access data); Data standardization: Standardize different behavior data to a unified scale (such as converting learning duration to hours and learning progress to a percentage); And the modeling of student learning behavior data is as follows: Based on the student's learning behavior data, establish a student portrait covering information such as their learning progress, interest preferences, and course participation; Use clustering algorithms (such as K-means) to classify students and customize personalized recommendation strategies according to the learning needs of different groups; Use regression analysis algorithms to evaluate the relationship between the student's grades and their learning behavior (such as learning duration, participation, etc.) to predict the student's preference for course features; S5. Dynamically modify the feature values of the corresponding features in the course teaching resources according to the student's learning behavior data. For example: Course duration adjustment: If a student spends a large amount of time on a certain course module, the system may infer that the course is more difficult for this student, so the course difficulty feature value of this course will be increased; Interest adjustment: If a student frequently accesses a certain type of course (for example, video courses), the system will increase the feature value of this course type, thereby recommending more similar types of courses; And when the feature value of a certain feature in the course teaching resources changes, other features associated with this feature will be adjusted based on the preset association values; For example, when the course difficulty feature value changes, the system will automatically adjust the feature values of features such as learning objectives and course duration according to the association rules; In step S5, the method for dynamically modifying the feature values of features is as follows: Q1. Establish update rules for each feature in the feature set of the course teaching resources; For example, for the feature "course difficulty" in the course, if the student's performance in this course is poor (for example, the average test scores of multiple students are low), then the feature value of the feature "course difficulty" will be increased based on the preset correction coefficient; Conversely, the feature value will be decreased by the correction coefficient; And the eigenvalue correction method for each feature is a phased correction. For example, by presetting the interval of the correction time, within the time interval period, only one modification is made; Q2. Map the collected learning behavior data to each feature in the course teaching resources to form a one-to-one correspondence, and match the mapping result with a preset mapping table to obtain the correction amount for the corresponding feature; The mapping table provides a basis for subsequent feature correction by defining the relationship between student behavior data features and course features. Each entry in the mapping table represents the degree of association between a certain student behavior and a certain course feature, and usually uses the correction amount to represent the impact of this behavior on the course feature; For example, if a certain student shows strong interest and completes a large number of assignments in the course teaching resources, the system will consider that the student has a high adaptability to the "course difficulty" feature of this course teaching resource, and accordingly correct the "course difficulty" feature value of this course teaching resource; Q3. For each feature in the course teaching resources , based on the mapping result in step Q2, through the formula , obtain the new eigenvalue of feature ; In the formula, is the original eigenvalue of feature ; is the correction coefficient, which is calculated through experiments or historical data; is the correction amount of feature , which is obtained from the result of mapping the learning behavior data and the course teaching resources; S6. Dynamically update the course teaching resources based on the eigenvalues of each feature in the corrected feature set, and use AI big data technology to generate a personalized course recommendation list for students, which specifically includes: Recommendation model training: Use collaborative filtering, matrix factorization or deep neural network models to train the recommendation system, learn the personalized needs of students, and generate a recommendation list; Recommendation strategy optimization: Continuously optimize the recommendation model according to the real-time feedback of students (such as the completion of courses, course ratings, etc.) to ensure that the recommended content meets the latest needs of students; Specifically, a recommendation list is generated through collaborative filtering or content-based recommendation methods; Collaborative filtering: By analyzing the similarity between different students, recommend courses based on historical behavior and rating data. Collaborative filtering can be divided into user-based collaborative filtering and item-based collaborative filtering; Content-based recommendation: By calculating the similarity between the features of courses (such as difficulty, duration, learning objectives) and the student's interest profile, suitable courses for the student are recommended; At the same time, it can also be through feature vectorization and recommendation result generation. The specific process is as follows: Convert the student profile and the features of course teaching resources into vector forms through feature vectorization, and then use algorithms such as cosine similarity and Euclidean distance to calculate the similarity between the student profile and the course feature vectors to obtain the recommended courses; And based on the feature vectors, the system generates a personalized course recommendation list according to factors such as the student's interests and learning progress. The recommendation results are sorted according to relevance to recommend the courses that best meet the student's needs; Finally, transmit the recommendation results to the student's learning platform (such as a mobile APP or a web application) through an interface for the student to view; The method for generating a course recommendation list includes: Y1. Represent the feature sets of each course teaching resource as where is the feature value of the i-th feature; Convert the student behavior data into vector form to generate a behavior feature set T, where is the feature value of the student on a specific i behavior; For example, learning duration vectorization: Normalize the duration data of the student on each course module and learning resource. Specifically, convert the duration data into a value between [0,1] through min-max standardization; Participation vectorization: The student's participation in the course can be represented by their activity frequency. For example, the number of times the student participates in activities such as submitting homework, taking quizzes, and participating in course discussions; The number of participation times of the student in each learning task can be counted to form the participation vector of each course; In this embodiment, assume that the student participated in 5 discussions, submitted 3 homework assignments, and took 2 quizzes in a week. Then the participation vector can be represented as: P=(5,3,2). Subsequently, normalize the participation vector to obtain a vector between [0,1]; Grade data vectorization: Standardize the grades of each student in different courses so that they are in the same dimension; Click behavior vectorization: The student's click behavior in the course can be represented by the type and quantity of the course resources they access; For example, the student clicked on 10 videos, 5 articles, 2 homework tasks, etc. within a certain period of time; Take each behavior category (video, article, homework, etc.) as a dimension, calculate the click times of the student on each category and standardize: Suppose a student clicks on 10 videos, 5 articles, and 2 assignment tasks in a course, and the vector representation is: O = (10, 5, 2). Then, normalize this behavior vector so that the values in each dimension are within a unified range; Y2. Calculate the similarity between the feature set C of each course teaching resource and the behavior feature set T of the student behavior data; In this embodiment, cosine similarity is used for calculation, and the final result range is [0, 1], where 1 represents complete similarity and 0 represents complete dissimilarity; Y3. Based on the similarity in Y2, select the top N course teaching resources for sorting; S7. Dynamically update the recommended list of course teaching resources according to the student's learning progress, interest preferences, and emotional changes, specifically including: When the student's learning progress or interest changes, the system will recalculate the recommended list based on the new data and give priority to recommending courses that meet the current needs; The interaction data of the student during the learning process (such as participation in the discussion area, assignment feedback) will be continuously fed back to the recommendation system to adjust the subsequent recommended course content.

[0019] Among them, the method for obtaining the learning progress is: the database system saves the student's progress data and updates the system status through real-time data synchronization technology. After each chapter or learning unit in each course teaching resource on the learning platform is completed, the student's learning progress is recorded. The commonly used data includes: the chapters or modules completed by the student, test scores, and assignment scores; And use the LMS and learning analysis system to track the student's progress in real time; Whenever the student completes a task or module, the system will automatically update the student's learning progress and record what the student has completed; The student's participation (such as participating in discussions, answering questions, watching videos, etc.); The method for obtaining interest preferences is: through interest preference modeling or real-time detection of interest changes; Among them, interest preference modeling uses tools such as Python or Apache Spark to process the student behavior data for interest preference modeling and analysis. For example, the student's interest preferences are modeled through click behavior on learning content, learning duration, assignment scores, etc.; based on the modules frequently participated in by the student during the learning process, the course topics concerned, and the viewing time, etc., to infer their interests; Real-time detection of interest changes uses machine learning models (such as decision trees, random forests, K-means clustering, etc.) to analyze the student behavior data to identify interest changes; Perform time series analysis on the student's behavior data to detect the changing trend of interest in real time; For example, if a student shows a high level of engagement (such as viewing duration, rating, etc.) with a certain type of course over a period of time, the system will increase the recommendation weight for courses of that category; The method for obtaining emotional changes is as follows: sentiment analysis based on a dictionary: match emotional words in the text through an emotional dictionary and calculate the emotional polarity (positive or negative); When the student's emotion is positive, the recommendation system tends to recommend courses with higher challenges to stimulate the student's enthusiasm and spirit of challenge; When the student's emotion is negative, the recommendation system will give priority to recommending courses with lower difficulty and stronger support to reduce the student's learning pressure and enhance their confidence in learning.

[0020] The above are only the preferred embodiments of the present invention and do not impose any formal limitations on the present invention. Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or equivalent changes and modifications within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A personalized recommendation method for course teaching resources based on AI big data, characterized in that, It includes the following steps: S1. Define multiple features for each course teaching resource, and assign values to each feature through expert evaluation, automated evaluation or learning path analysis, so that each course teaching resource obtains a feature set with multiple features and corresponding feature values; S2. Based on the mutual relationship between the features of the course teaching resources, set association values between each feature in the feature set and the remaining features; S3. Through the corresponding learning platform, collect non-private learning behavior data of students; S4. Process and model the collected learning behavior data of students to construct a behavior portrait of the corresponding students; S5. Dynamically correct the feature values of the corresponding features in the course teaching resources according to the learning behavior data of students; S6. Based on the feature values of each feature in the corrected feature set, dynamically update the course teaching resources, and generate a personalized course recommendation list for students based on the updated course teaching resources; S7. Dynamically update the course teaching resource recommendation list according to the learning progress, interest preferences and emotional changes of students.

2. The personalized recommendation method for course teaching resources based on AI big data according to claim 1, wherein In step S2, for feature i and feature j, through the formula: Determine the associated value therebetween ; Correlation value ranges from [0, 1]; and, is the covariance between feature i and feature j, representing the linear relationship between the two; and are the variances between feature i and feature j respectively, representing the range of variation of their respective features.

3. The personalized recommendation method for course teaching resources based on AI big data according to claim 2, wherein, Each feature in the feature set is set with a weight , and each feature The calculation rule between them is as follows: When the feature value of a certain feature changes, its weight greater than the weight of the feature to be calculated will spread the change; Features with small weights will not affect the changes of other features, but only accept the influence of changes in other features; Features with the same weight affect each other.

4. The personalized recommendation method for course teaching resources based on AI big data according to claim 3, wherein Each feature in the feature set The correlation between them is represented by the correlation matrix A: ; among them, the on the diagonal is 1, indicating that each feature is completely related to itself; When the eigenvalue of the feature changes, the formula adopted is: ； wherein, is an indicator function, and takes the value of 1 if and only if the weight of feature is greater than the weight of feature , indicating that feature can affect feature ; ​​ as a feature amount of change 5. The personalized recommendation method for course teaching resources based on AI big data according to claim 4, wherein When the eigenvalue of multiple features changes simultaneously, the target feature feature change amount is affected by multiple features and the following operations are performed: Z1. Determine which features have an impact on the target feature through the correlation value and weight between features, and establish its set of influencing features { , , for the target feature . For the target feature , establish its set of influencing features { , ,...,[[]] }; Z2. When calculating, preferentially select the feature with the largest weight, and use the change amount of this feature to adjust the eigenvalue of the target feature ; Z3. If the eigenvalue changes of multiple features occur simultaneously, and the influence propagation between them is calculated through associated values then select the feature with the maximum weight from the set of influencing features ; Feature change amount The calculation formula is as follows: ; In the formula, represents the maximum weight feature and the association value between the target feature ; is the maximum weight feature of the change amount.

6. The personalized recommendation method for course teaching resources based on AI big data according to claim 5, characterized in that, If multiple feature changes with the same weight affect the target feature then, from the change amounts of the multiple features, select the feature with the largest weight and the greatest impact on the change amount as the determinant of the change amount of the target feature.

7. The personalized recommendation method for course teaching resources based on AI big data according to claim 1, characterized in that, In step S5, the method for dynamically correcting the feature value of a feature is: Q1. Establish an update rule for each feature in the feature set of the course teaching resource; Q2. Map the collected learning behavior data to each feature in the course teaching resource to form a one-to-one correspondence, and match the mapping result with a preset mapping table to obtain the correction amount of the corresponding feature; Q3. For each feature in the course teaching resources , based on the mapping result in step Q2, through the formula , obtain the new feature value of the feature ; ; In the formula, is the feature 's original feature value; is the correction coefficient; is the feature correction amount of.

8. The personalized recommendation method for course teaching resources based on AI big data according to claim 1, wherein In step S7, the method for generating the course recommendation list includes: Y1. Represent the feature set of each course teaching resource as , where is the eigenvalue of the i-th feature; Convert the student behavior data into vectors to generate a behavior feature set T, , where is the feature value of the student on a specific behavior i; Y2. Calculate the similarity between the feature set C of each course teaching resource and the behavior feature set T of the student behavior data; Y3. Based on the similarity in Y2, select the top N course teaching resources for sorting.

Images