Personalized learning recommendation method based on big data analysis and artificial intelligence

Abstract

The invention discloses a personalized learning recommendation method based on big data analysis and artificial intelligence. The method comprises the following steps: acquiring a key content tag and a predicted content tag of each section of video course based on a big language model; according to the static archive data and the dynamic learning condition data of the user, constructing a corresponding user basic portrait and a user learning condition portrait; according to the key content label, the user basic portrait and the user learning situation portrait, constructing a user individual learning preference prediction portrait, and further constructing a user group learning preference prediction portrait; constructing a content value model of the video course; and based on the user group learning preference prediction portrait and the content value model, generating a course recommendation list of the user individuals. According to the method, the problem of learning task deposition caused by single dependence on a resource value matching degree recommendation mechanism can be effectively avoided, the problem of low user learning efficiency caused by excessive recommended contents is avoided, and the problem of low learning interest caused by relatively repeated recommended contents for individual employees is avoided.

Description

Technical Field

[0001] The present invention belongs to the field of big data analysis and relates to the technical field of personalized recommendation of teaching video content, and more specifically to a personalized learning recommendation method based on big data analysis and artificial intelligence. Background Art

[0002] For enterprises undergoing digital transformation, it is urgent to accelerate the upgrade and transformation of the enterprise talent knowledge system through enterprise digital learning upgrades, and to serve the enterprise to efficiently respond to market changes by improving the cognitive ability and adaptability of the talent echelon to market changes, so as to achieve the goal of improving enterprise productivity and maintaining the speed of development in difficult situations. In the face of enterprise transformation and development, the efficiency of employees' learning and application of new knowledge determines the size of enterprise productivity. The need for efficient learning of enterprise talents and the need to formulate and push learning plans for thousands of people are often difficult to achieve based on a single technology to push personalized learning plans for employees due to problems such as insufficient enterprise training management and operation personnel, lack of systematic enterprise digital learning content, insufficient data on job competency knowledge system, lack of evaluation and quantitative tools for employee cognitive level, and lack of basic information on video resources.

[0003] There are currently similar methods or shortcomings of the prior art, the main ones being as follows:

[0004] 1. The technology mainly relies on the systematization of teaching video knowledge and the user's cognitive level and mastery data. By building a knowledge graph, it realizes the matching of knowledge points and course videos, and realizes the recommendation of personalized learning content. This type of teaching video is mostly subject-based courses, and the scenarios are mainly K12 subject education, college courses, or special series courses with knowledge systematization characteristics. Compared with the looseness of knowledge points in corporate digital learning courses, this type of teaching video has a higher degree of knowledge systematization in the video collection. The recommendation technology relies on the pre-constructed subject knowledge graph and the user's mastery of subject knowledge. By traversing the course knowledge graph, an adaptive learning path planning is formed. For example:

[0005] (1) "Method and system for generating personalized learning paths based on multi-course knowledge graph fusion", CN117787402B, learning path planning technology based on knowledge graph. Through automated knowledge graph fusion technology, a personalized knowledge graph of all courses in the student's personalized learning plan is constructed; the subsequent learning path is intelligently generated based on the user's current learning progress and knowledge mastery, and the personalized knowledge graph of multi-course fusion is compared.

[0006] (2) “A method, device, electronic device and storage medium for recommending vocational training courses”, CN117743695B, establishes a knowledge graph based on job competence in advance, manually annotates content labels for the course video library in advance, and establishes an association relationship between knowledge graph nodes and course video content labels; combines the mining and analysis of user behavior trajectory data and learning progress data, and calculates the weights of the content labels of the user’s courses to obtain the student’s interest and hobby portrait; then, based on the course content labels associated with the interest and hobby portrait, establishes a matching mapping relationship with the pre-generated job competence knowledge graph, thereby obtaining a knowledge graph for personalized learning of users, and based on the graph-based path planning technology, achieves personalized course recommendations for thousands of people.

[0007] 2. The technology mainly relies on rich user portrait data and learning data as the basis for content adaptive recommendation. The main application scenario of this type of technology is a relatively complete teaching system, which can track all the learning behavior data of students, including age, education level, process learning, practice, test data, and summative evaluation stage exams, learning files and other learning data, and analyze the user's learning ability, learning preferences, cognitive level and other characteristics based on rich basic data. Based on rich user portraits and weak points in knowledge points and teaching key points, personalized recommendations of teaching content can be achieved through big data analysis and mining.

[0008] For example: "An intelligent adaptive education learning method and system based on big data", CN117726485B, constructs a student learning constraint portrait by obtaining the student's family situation information, and analyzes the student's educational intention portrait data through the student's academic data in the learning process, including the course learning, practice performance and examination data of each stage and semester; generates teaching courses based on the learning constraint portrait data and learning intention data; conducts real-time evaluation of the learning process based on the real-time teaching course data, and obtains real-time evaluation data of the learning process; and generates learning content recommendations based on the real-time evaluation data of the learning process.

[0009] Therefore, how to achieve efficient completion of large-scale data when the company's digital learning content lacks systematicity, there is insufficient data on job competency knowledge systems, there is a lack of evaluation tools for employee cognitive levels, and basic content label information on video resources is missing, and to explore the potential value of data based on the lack of user learning evaluation data; to achieve dynamic learning path planning and content recommendation, to avoid the problem of accumulation of organizational must-learn tasks caused by a single reliance on resource value matching recommendation mechanisms, to avoid the problem of reduced user learning efficiency due to too much learning content recommended each time, and to avoid relatively repeated recommendations for individual employees resulting in reduced learning interest, and other issues, to effectively assist companies in upgrading digital learning, building talent echelons, and content operation management to reduce costs and increase efficiency, are issues that technical personnel in this field urgently need to address. Summary of the invention

[0010] In view of the above problems, the present invention provides a personalized learning recommendation method based on big data analysis and artificial intelligence to at least solve some of the technical problems mentioned in the above background technology.

[0011] In order to achieve the above object, the present invention adopts the following technical solution:

[0012] The personalized learning recommendation method based on big data analysis and artificial intelligence includes the following steps:

[0013] S1. Obtain key content labels and predicted content labels for each video course based on the large language model;

[0014] S2. Build corresponding user basic portraits and user learning situation portraits based on the user's static profile data and dynamic learning situation data;

[0015] S3. Constructing a user's individual learning preference prediction portrait based on the key content tags, the user's basic portrait, and the user's learning situation portrait; and constructing a user group learning preference prediction portrait based on the user's individual learning preference prediction portrait;

[0016] S4. Constructing a content value model of the video course based on the key content tags, the predicted content tags, the user learning situation portrait, the user individual learning preference predicted portrait, and the user group learning preference predicted portrait;

[0017] S5. Generate a course recommendation list for individual users based on the user group learning preference prediction portrait and the content value model.

[0018] Furthermore, the step S1 specifically includes:

[0019] Get the key field text information of each video course;

[0020] Preprocess the key field text information to obtain a basic course information table for each video course;

[0021] According to the course basic information table, the course content label of each video course is obtained based on the large language model;

[0022] Classify the knowledge domain of each video course by using a large language model to obtain a sub-knowledge domain label of each video course;

[0023] De-duplication and clustering processing are performed on the sub-knowledge domain labels by using a large language model to obtain parent knowledge domain labels;

[0024] Performing knowledge completion processing on the sub-knowledge domain label and the parent knowledge domain label through a large language model to obtain a knowledge domain classification tree corresponding to the video course;

[0025] Through the knowledge domain classification tree, the course content labels of each video course are completed and predicted to obtain the key content labels and predicted content labels of each video course.

[0026] Furthermore, the step S3 specifically includes:

[0027] S31, according to the user basic portrait, the user learning status portrait and the key content tags of each video course, obtain the user individual learning preference prediction portrait through the K-means algorithm;

[0028] S32. Divide the users into groups according to the basic user portraits to obtain a group classification data table;

[0029] S33. Construct a user group learning preference prediction portrait based on the user individual learning preference prediction portrait and the group classification data table.

[0030] Furthermore, the step S4 specifically includes:

[0031] S41, based on the key content tags, the predicted content tags, the user learning situation portrait, the user individual learning preference prediction portrait and the user group learning preference prediction portrait, obtain the user individual preference score, the user group preference score and the user learning motivation score;

[0032] S42, according to the user individual preference score, the user group preference score and the user learning motivation score, obtaining the corresponding individual preference influence factor, group preference influence factor and learning motivation influence factor;

[0033] S43, assigning weight coefficients to the individual preference influencing factor, the ethnic group preference influencing factor, and the learning motivation influencing factor;

[0034] S44. Perform weighted summation on the individual preference influencing factor, group preference influencing factor and learning motivation influencing factor after assigning weight coefficients to obtain the content value score of each course.

[0035] Furthermore, the step S41 specifically includes:

[0036] (1) Obtaining user individual preference scores:

[0037] Determine a first influencing factor according to the user's learning status portrait; the first influencing factor includes the user's learning status, learning completion percentage, cumulative learning time, cumulative learning times, whether to collect and share, and the number of times;

[0038] Determining a second influencing factor according to the predicted profile of the individual learning preference of the user; the second influencing factor includes a key content tag and a predicted content tag;

[0039] Assigning weight coefficients to the first impact factor and the second impact factor;

[0040] The first influencing factor and the second influencing factor after the weight coefficient is assigned are standardized and weighted summed to obtain the user's individual preference score;

[0041] (2) Obtaining user group preference scores:

[0042] Determine a third influencing factor based on the user learning status portrait; the third influencing factor includes the average learning completion percentage of users in the same group, the average cumulative learning time of users in the same group, the cumulative number of learning times of users in the same group, the number of collections by users in the same group, and the number of shares by users in the same group;

[0043] Determining a fourth influencing factor according to the user group learning preference prediction portrait; the fourth influencing factor includes the relevance of the key content tag and the relevance of the predicted content tag;

[0044] Assigning weight coefficients to the third impact factor and the fourth impact factor;

[0045] The third and fourth influencing factors after the weight coefficients are assigned are standardized and weighted summed to obtain the user group preference scores;

[0046] (3) Obtaining the user's learning motivation rating:

[0047] Determine the fifth influencing factor based on the company's preset operating strategy; the fifth influencing factor includes whether all employees must learn, whether the position must learn, whether new employees must learn, whether the department must learn, whether certification preparation is required, and whether it is a popular recommendation;

[0048] Assigning weight coefficients to the first impact factor and the fifth impact factor;

[0049] The first influencing factor and the fifth influencing factor after the weight coefficients are assigned are standardized and weighted summed to obtain the user's learning motivation level score.

[0050] Furthermore, the step S5 specifically includes:

[0051] Based on the content value model, all video courses on the online learning platform are traversed to obtain the content value score of each video course for the target individual user, and the video courses are sorted according to the content value score to generate a course recommendation list for the target individual user.

[0052] Furthermore, it also includes:

[0053] According to the enterprise's preset operation strategy, obtain the content that individual users must learn;

[0054] Based on the preset operational strategy priority, a corresponding must-learn task list is generated according to the must-learn content of the individual user;

[0055] Based on the enterprise's preset operating strategy, teaching plan and teaching time constraints, the priority of the required learning task list is updated.

[0056] Furthermore, it also includes:

[0057] The current dynamic learning situation data of the individual user is obtained in real time, and the user learning situation portrait is updated according to the current dynamic learning situation data; based on the updated user learning situation portrait, the course recommendation list or the required task list is updated.

[0058] Furthermore, it also includes:

[0059] According to the preset daily learning capacity restriction condition, based on the required learning task list and the course recommendation list, a daily learning to-do content list is generated.

[0060] Furthermore, it also includes:

[0061] Get user clicks, learning percentage and user reviews;

[0062] Determine a comprehensive evaluation index based on the number of user clicks, learning percentage and user evaluation;

[0063] Optimizing the weight coefficient in the content value model according to the comprehensive evaluation index;

[0064] Based on the optimized weight coefficient, the content value model is updated.

[0065] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a personalized learning recommendation method based on big data analysis and artificial intelligence, which has the following beneficial effects:

[0066] The present invention realizes content label extraction of each video course through a large language model, without the need for manual labeling, thus reducing the operating cost of manual labeling.

[0067] Under the premise that there is insufficient data on the job competency knowledge system and a lack of evaluation tools for employee cognitive levels, the present invention constructs user learning preference portraits (including user individual learning preference prediction portraits and user group learning preference prediction portraits) and content value models based on user portrait data to form a quantifiable personalized recommendation algorithm, outputting a real-time, personalized learning recommendation course list for each user, which not only meets the purpose of enterprise talent training and governance, but also meets the needs of individual personalized learning and job growth.

[0068] The present invention is based on the generalized reasoning ability of large language models and the ability to summarize and generalize long texts to achieve the division of knowledge fields of large-scale courses one by one, and based on the knowledge fields of massive courses on the platform, automatically completes and summarizes the knowledge field classification to form a hierarchical and structured knowledge field classification tree structure, which assists in providing a basis for learning path recommendations for personalized recommendation algorithms.

[0069] The present invention is based on the list of must-learn tasks for individuals obtained from the preset operation strategy of the enterprise, combined with big data analysis and training goals, and sets daily learning capacity restrictions, so as to output a daily learning to-do content list for each user in real time, thereby achieving the service purpose of "actively finding people for must-learn content" and "actively finding people for important content", truly meeting the purpose of enterprise talent development and user efficient learning. Avoid problems such as excessive recommended learning content each time, which leads to the accumulation of important content and reduced student learning interest and motivation.

[0070] The present invention realizes updating of the content value model by adjusting the weight coefficient of the content value model, so as to achieve the service purpose of "the recommended contents are what the users want".

[0071] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.

[0072] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] 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 or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0074] Figure 1 A schematic diagram of the process flow of a personalized learning recommendation method based on big data analysis and artificial intelligence provided in an embodiment of the present invention.

[0075] Figure 2 A schematic diagram of a knowledge domain classification tree structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0077] Embodiment 1:

[0078] See also Figure 1 As shown, the embodiment of the present invention discloses a personalized learning recommendation method based on big data analysis and artificial intelligence, comprising the following steps:

[0079] S1. Obtain key content labels and predicted content labels for each video course based on the large language model;

[0080] S2. Build corresponding user basic portraits and user learning situation portraits based on the user's static profile data and dynamic learning situation data;

[0081] S3. Build a prediction profile of individual user learning preferences based on key content tags, user basic profiles and user learning situation profiles; build a prediction profile of user group learning preferences based on the prediction profile of individual user learning preferences;

[0082] S4. Construct a content value model for video courses based on key content tags, predicted content tags, user learning situation portraits, user individual learning preference prediction portraits, and user group learning preference prediction portraits;

[0083] S5. Generate a course recommendation list for individual users based on the user group learning preference prediction portrait and content value model.

[0084] It should be noted that the above-mentioned marks S1-S5 are for the purpose of convenience of description and do not limit the order of implementation of the steps. Next, the above-mentioned steps are described in detail.

[0085] In the above step S1, the basic information of a large amount of video courses is analyzed, and the key content tags and predicted content tags of each video course are obtained based on the large language model; specifically, the following are included:

[0086] (1) Get the course content tags for each video course:

[0087] ① By analyzing the basic information data of teaching videos, we can obtain the unstructured key field text information of each video course, such as the course title, a small amount of course introduction, and a small amount of manually annotated course knowledge tags;

[0088] ② Preprocess the key field text information to obtain the course basic information table of each video course; specifically, the key field text information obtained above can be input into the big language model tool, and the key field text information can be cleaned and structured by the big language model to obtain a structured high-quality course basic information table, such as course title, course ID, course introduction and course label;

[0089] ③According to the course basic information table, the course content label of each video course is obtained based on the large language model; specifically, the course basic information table is imported into the large language model, and the content label is extracted once through the prompt words to obtain the preliminary content label of each video course; the invalid labels and noise labels in the preliminary content label are removed; the preliminary content labels after the removal are imported into the large language model, and the content label is extracted twice through the prompt words to obtain the structured course content label of each video course, such as the course title, course ID and course knowledge label set;

[0090] (2) Obtain the knowledge domain classification tree corresponding to all video courses:

[0091] ① Classify each video course into different knowledge domains through a large language model to obtain sub-knowledge domain labels for each video course. Specifically, sub-knowledge domain labels can be added to the structured course content labels of each video course to reflect the sub-domain classification of each video course.

[0092] ② Use the large language model to remove duplicates and cluster the sub-knowledge domain labels to obtain the parent knowledge domain labels; specifically, this can be reflected in the knowledge induction of the data table containing the sub-knowledge domain labels, and finally summarize a small number of parent knowledge domain labels (parent knowledge domain labels <10);

[0093] ③ Use the large language model to complete the knowledge of the sub-knowledge domain labels and the parent knowledge domain labels, and obtain a hierarchical and structured knowledge domain classification tree corresponding to the video course;

[0094] (3) Through the knowledge domain classification tree, the course content labels of each video course are completed and predicted to obtain the key content labels and predicted content labels of each video course.

[0095] In the above step S2, based on a small amount of static archive data and dynamic learning data of the user, a corresponding user basic profile and user learning profile are constructed; specifically, the steps include:

[0096] (1) Analyze limited user profile data to obtain key profile field data that is strongly related to user learning recommendations, such as user name, organization, department, benchmark position, position group, professional qualification and certification level, etc., to build a basic user profile.

[0097] (2) Based on limited dynamic learning data, obtain dynamic learning profiles of users in different learning activities; specific learning data include but are not limited to learned content, learning status, learning progress and test scores, etc.

[0098] In the above step S3, it specifically includes:

[0099] S31. Based on the user basic portrait, the user learning situation portrait and the key content labels of each video course, the user individual learning preference prediction portrait is obtained through the feature extraction K-means algorithm; the user individual learning preference prediction portrait includes the parent knowledge domain label, the child knowledge domain label of the user's individual learning course preference, and the key content label list data of the preferred course;

[0100] Among them, the clustering algorithm used for feature extraction is K-means, which is a clustering algorithm that is usually used to divide data points into K non-overlapping subsets (or "clusters"). In the scenario of course learning preference, the K-means algorithm can be used to group the courses learned by the user into the same cluster, and the common content labels of the courses in the same cluster are used as the content labels of the user's individual learning course preferences. See Figure 2 As shown, the content labels of the same cluster courses are the knowledge domain classification tree obtained above. Through the weight calculation of the content labels of the user's individual learning preferences, the user's individual learning preference prediction portrait under the knowledge domain classification tree can be predicted. The user's individual learning preference prediction portrait is equivalent to the knowledge domain tree structure of the user's individual learning content preference, where the terminal node in the tree structure is the key content label and the user's preference degree on the key content label;

[0101] The above K-means algorithm mainly measures the similarity between data points through the calculation formula of Euclidean distance; for two points p and q in the data set, the Euclidean distance between them can be calculated by the following formula:

[0102] d(p,q)=sqrt((p1-q1)^2+(p2-q2)^2+...+(pn-qn)^2)

[0103] Wherein, p1, p2, ..., pn and q1, q2, ..., qn represent the coordinate values ​​of point p and point q in each dimension respectively; n represents the dimension of point data. In video key frame extraction, n usually corresponds to the length of feature vector;

[0104] S32, dividing the users into groups according to the basic user portraits obtained above, and obtaining a group classification data table, such as groups with the same position, groups with the same professional qualification, etc.;

[0105] S33. Construct a user group learning preference prediction portrait based on the user individual learning preference prediction portrait and the group classification data table; the construction logic of the user group learning preference prediction portrait can refer to the construction logic of the above-mentioned user individual learning preference prediction portrait.

[0106] In the above step S4, based on the key content tags, predicted content tags, user learning situation portraits, user individual learning preference predicted portraits and user group learning preference predicted portraits obtained above, a content value model of the video course is constructed; specifically, it includes:

[0107] S41, based on the key content tags, predicted content tags, user learning situation portrait, user individual learning preference predicted portrait and user group learning preference predicted portrait obtained above, obtain the user individual preference score, user group preference score and user learning motivation score; specifically including:

[0108] (1) Obtaining user individual preference scores:

[0109] The first step is to determine the factors affecting individual user preferences:

[0110] ① Determine the first influencing factor based on the user's learning status portrait; the first influencing factor includes but is not limited to the user's learning status, learning completion percentage, cumulative learning time, cumulative learning times, whether to collect and share times;

[0111] ② Determine the second influencing factor based on the predicted profile of individual user learning preferences; the second influencing factor includes but is not limited to key content tags and predicted content tags;

[0112] The second step is to determine the weights of individual preference influencing factors and the algorithm formula to obtain the user's individual preference score I:

[0113] ① Assign weight coefficients to the first influencing factor and the second influencing factor; specifically, determine the weight coefficient of each influencing factor, and the weight coefficient can be adjusted according to actual conditions and business requirements; in the embodiment of the present invention, it is assumed that the weight coefficients are respectively \(w_1, w_2, ..., w_n\)

[0114] ② Standardize the first and second influencing factors after assigning weight coefficients to eliminate the impact of different dimensions; methods such as minimum-maximum standardization or Z-score standardization can be used. Since different influencing factors may have different dimensions and numerical ranges, directly using the original data for calculation may cause some factors to have too much influence on the results. In order to eliminate this influence, we need to standardize each influencing factor. The purpose of standardization is to convert the original data into a uniform numerical range, such as between 0 and 1, or a distribution with the same mean and standard deviation;

[0115] ③ Perform weighted summation on the standardized impact factors to obtain the user's individual preference score I; specifically, multiply each impact factor by the corresponding weight coefficient and then sum them up; the formula is as follows:

[0116] \[

[0117] I = \text{individual preference score} = w_1\times\text{standardized first influencing factor} + w_2\times\text{standardized second influencing factor} + ... + w_n\times\text{standardized nth influencing factor}

[0118] \].

[0119] (2) Obtaining user group preference scores:

[0120] The first step is to determine the factors affecting user group preferences:

[0121] ① Determine the third influencing factor based on the user's learning profile; the third influencing factor includes but is not limited to the average learning completion percentage of users in the same group (\(f_{11}\)), the average cumulative learning time of users in the same group (\(f_{12}\)), the cumulative number of learning times of users in the same group (\(f_{13}\)), the number of collections by users in the same group (\(f_{14}\)), and the number of shares by users in the same group (\(f_{15}\));

[0122] ② According to the predicted portrait of user group learning preference, determine the fourth influencing factor; the fourth influencing factor includes but is not limited to the relevance of key content tags corresponding to the learning preference courses of the same group (\(f_{21}\)) and the relevance of predicted content tags (\(f_{22}\));

[0123] The second step is to determine the weights of the factors affecting group preference and the algorithm formula to obtain the user group preference score P:

[0124] ① Assign weight coefficients to the third and fourth impact factors; assume that the weight coefficients corresponding to the third impact factor are \(w_1,w_2,\ldots,w_5\); the weight coefficients corresponding to the fourth impact factor are \(w_6,w_7\); the sum of all weight coefficients should be equal to 1, that is:

[0125] \[

[0126] w_1+w_2+\ldots+w_7=1

[0127] \]

[0128] ②Standardize the third and fourth impact factors after assigning weight coefficients

[0129] ③ Perform weighted summation on the standardized impact factors to obtain the user group preference score P; specifically:

[0130] \[

[0131] P=w_1\times\text{Normalized}(f_{11})+w_2\times\text{Normalized}(f_{12})+\ldots+w_5\times\text{Normalized}(f_{15})+w_6\times\text{Normalized}(f_{21})+w_7\times\text{Normalized}(f_{22})

[0132] \]

[0133] in:

[0134] -\(\text{Normalized}(f_{ij})\) is the value of the corresponding impact factor \(f_{ij}\) after standardized processing.

[0135] -\(w_i\) is the weight coefficient of the corresponding impact factor.

[0136] (3) Obtaining the user's learning motivation rating:

[0137] The first step is to determine the factors that influence user learning motivation:

[0138] ① The first influencing factor determined according to the user's learning profile is used as the user's learning motivation influencing factor; the first influencing factor includes but is not limited to the user's learning status (\(f_{11}\)), learning completion percentage (\(f_{12}\)), cumulative learning time (\(f_{13}\)), cumulative learning times (\(f_{14}\)), whether to collect (\(f_{15}\)) and sharing times (\(f_{16}\));

[0139] ② According to the preset operation strategy of the enterprise, determine the fifth influencing factor; the fifth influencing factor includes whether all employees must learn (\(f_{21}\)), whether the position must learn (\(f_{22}\)), whether new employees must learn (\(f_{23}\)), whether the department must learn (\(f_{24}\)), whether to prepare for certification (\(f_{25}\)) and whether it is popularly recommended (\(f_{26}\)), etc.;

[0140] The second step is to determine the weights and algorithm formulas of the factors affecting user learning motivation and obtain the user learning motivation score M:

[0141] ① Assign weight coefficients \(w_1,w_2,\ldots,w_{11}\) to the first and fifth impact factors; where \(w_1\) to \(w_6\) are the weights of the first impact factor; \(w_7\) to \(w_{11}\) are the weights of the fifth impact factor; the sum of all weight coefficients should be equal to 1, that is:

[0142] \[w_1+w_2+\ldots+w_{11}=1\]

[0143] ② Standardize the first and fifth impact factors after assigning weight coefficients

[0144] ③ Perform weighted summation on the standardized impact factors to obtain the user learning motivation score M:

[0145] \[M=\sum_{i=1}^{6}w_i\times\text{Normalized}(f_{1i})+\sum_{j=7}^{11}w_j\times\text{Normalized}(f_{2j})\]

[0146] in:

[0147] -\(\text{Normalized}(f_{ij})\) is the value of the corresponding impact factor \(f_{ij}\) after standardized processing.

[0148] -\(w_i\) is the weight coefficient of the corresponding impact factor.

[0149] S42, according to the user's individual preference score, the user's group preference score and the user's learning motivation score, obtain the corresponding individual preference influence factor (\(I\)), group preference influence factor (\(P\)) and learning motivation influence factor (\(M\));

[0150] S43. Assign weight coefficients to the individual preference influencing factors, group preference influencing factors and learning motivation influencing factors to obtain \(w_P,w_I,w_M\); the sum of all weight coefficients should be equal to 1, that is:

[0151] \[w_P+w_I+w_M=1\]

[0152] S44. Perform weighted summation on the individual preference influencing factor, group preference influencing factor and learning motivation influencing factor after assigning weight coefficients to obtain the content value score of each course:

[0153] \[C_{vs}=w_P\times P+w_I\times I+w_M\times M\].

[0154] In the above step S5, based on the predicted profile of user group learning preference and the content value model, a course recommendation list for individual users is generated; specifically:

[0155] Based on the content value model, all video courses on the online learning platform are traversed to obtain the content value score of each video course for the target user. The video courses are sorted according to the content value score to generate a course recommendation list for the target user.

[0156] Embodiment 2:

[0157] Embodiment 2 of the present invention, based on the above-mentioned embodiment 1, further includes: obtaining the individual required learning content of the user according to the preset operation strategy of the enterprise; generating a corresponding list of required learning tasks according to the individual required learning content of the user based on the preset operation strategy priority, such as required courses or learning activities for all employees, required courses or learning activities for each position, required courses or learning activities for new employees, required courses or learning tasks for departments, and required courses or learning tasks for certification preparation; updating the priority of the list of required learning tasks based on the preset operation strategy, teaching plan and teaching time constraints of the enterprise.

[0158] Embodiment three:

[0159] Embodiment 3 of the present invention, based on the above-mentioned embodiment 2, further includes: obtaining the current dynamic learning data of the individual user in real time, and updating the user learning profile according to the current dynamic learning data; based on the updated user learning profile, updating the course recommendation list obtained in embodiment 1 or the required learning task list obtained in embodiment 2; wherein:

[0160] Update the course recommendation list, including: remove completed courses, passed exams, finished exams and finished training courses; weight the content value score of courses in progress; and keep the content value score unchanged for courses not yet studied;

[0161] The list of required learning tasks will be updated, including the elimination of completed courses, passed exams, completed exams, and completed training courses.

[0162] Embodiment 4:

[0163] Embodiment 4 of the present invention, based on the above embodiment 3, further includes: determining a preset daily learning capacity restriction condition based on big data analysis and training objectives; generating a daily learning to-do content list based on a required learning task list and a course recommendation list according to the preset daily learning capacity restriction condition;

[0164] Embodiment five:

[0165] Embodiment 5 of the present invention, based on the above embodiment 4, further includes: based on the daily learning to-do content list obtained in the above embodiment 4, combined with the user's learning profile data, the user's click data on the recommended content, and the added to-do data, an evaluation model algorithm for the recommended content is constructed to update the content value model in the above embodiment 1 to adjust the course recommendation list; specifically including:

[0166] The first step is data collection: the number of user clicks (\(C_{click}\)); the learning percentage (\(C_{learn}\)); the user's evaluation of the course (\(C_{rate}\)), which refers to the behavior of adding to-do items;

[0167] The second step is to define the evaluation index: define a comprehensive evaluation index \(E\) to evaluate the quality of the recommended courses; this index can be the weighted sum of the number of clicks, learning percentage, and evaluation:

[0168] \[E=w_{click}\times C_{click}+w_{learn}\times C_{learn}+w_{rate}\times C_{rate}\]

[0169] Among them, \(w_{click},w_{learn},w_{rate}\) is the weight coefficient of the corresponding indicator and satisfies:

[0170] \[w_{click}+w_{learn}+w_{rate}=1\]

[0171] Step 3: Adjust Cvs weights: According to the comprehensive evaluation index E, adjust the weight coefficients wP, wI, wM in the content value model C_{vs}. Linear regression or other machine learning techniques can be used to find the optimal weight coefficients to maximize E.

[0172] Step 4: Algorithm optimization: Use gradient descent or other optimization algorithms to adjust the weight coefficients to maximize \(E\); cross-validation can be used to evaluate the performance of the model under different weight settings;

[0173] Step 5: Update and optimize the algorithm formula: According to the optimization results, update the calculation formula of the content value model \(C_{vs}\):

[0174] \[C_{vs}=w'_P\times P+w'_I\times I+w'_M\times M\]

[0175] Among them, \(w'_P,w'_I,w'_M\) is the optimized weight coefficient.

[0176] Step 6: Process monitoring and pre-evaluation: Implement the updated recommendation algorithm and continuously monitor its performance; regularly evaluate and adjust weight coefficients to adapt to changes in user behavior and preferences.

[0177] In summary, the embodiment of the present invention provides a personalized learning recommendation method based on big data analysis and artificial intelligence, which integrates large language model technology to extract large-scale knowledge tags from video courses; evaluates the learning preference prediction portraits of individual users and groups through machine learning technology; constructs a content value model based on data mining and big data analysis technology; evaluates the value of recommended resources through influencing factors and weight calculations; comprehensively designs a personalized recommendation algorithm model based on the enterprise's talent training management and operation strategies and teaching strategies, and adjusts the priority of recommended content; based on the personalized recommendation algorithm and user portrait data, through the schedule planning model and dynamic behavior evaluation technology, a dynamic intelligent adaptive schedule recommendation system that pushes important and fresh learning content to employees every day is realized; effectively solves the problem of insufficient manpower for extracting massive video tags; effectively avoids the difficulty of constructing a course knowledge graph due to the lack of systematization of content and insufficient job ability knowledge system data for multiple video courses, and thus cannot realize the learning path planning for thousands of people; effectively avoids the problem of lack of recommendation basis caused by the difficulty in obtaining employee cognitive level evaluation data; effectively solves the problems of too much recommended content, accumulation of key learning content of the enterprise, and repeated recommended content that affect the talent management and learning efficiency of the enterprise.

[0178] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0179] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A personalized learning recommendation method based on big data analysis and artificial intelligence, characterized in that: The steps include: S1. Obtain key content labels and predicted content labels for each video course based on the large language model; S2. Build corresponding user basic portraits and user learning situation portraits based on the user's static profile data and dynamic learning situation data; S3. Constructing a predicted profile of the user's individual learning preference based on the key content tags, the user's basic profile and the user's learning status profile; Based on the user individual learning preference prediction portrait, construct a user group learning preference prediction portrait; S4. Constructing a content value model of the video course based on the key content tags, the predicted content tags, the user learning situation portrait, the user individual learning preference predicted portrait, and the user group learning preference predicted portrait; S5. Generate a course recommendation list for individual users based on the user group learning preference prediction portrait and the content value model.

2. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 1 is characterized in that: The step S1 specifically includes: Get the key field text information of each video course; Preprocess the key field text information to obtain a basic course information table for each video course; According to the course basic information table, the course content label of each video course is obtained based on the large language model; Classify the knowledge domain of each video course by using a large language model to obtain a sub-knowledge domain label of each video course; De-duplication and clustering processing are performed on the sub-knowledge domain labels by using a large language model to obtain parent knowledge domain labels; Performing knowledge completion processing on the sub-knowledge domain label and the parent knowledge domain label through a large language model to obtain a knowledge domain classification tree corresponding to the video course; Through the knowledge domain classification tree, the course content label of each video course is completed and predicted to obtain the key content label and predicted content label of each video course.

3. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 2 is characterized in that: The step S3 specifically includes: S31, obtaining a predicted profile of individual user learning preferences through a K-means algorithm based on the user basic profile, the user learning profile, and key content tags of each video course; S32. Divide the users into groups according to the basic user portraits to obtain a group classification data table; S33. Construct a user group learning preference prediction portrait based on the user individual learning preference prediction portrait and the group classification data table.

4. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 1 is characterized in that: The step S4 specifically includes: S41, based on the key content tags, the predicted content tags, the user learning situation portrait, the user individual learning preference predicted portrait and the user group learning preference predicted portrait, obtain the user individual preference score, the user group preference score and the user learning motivation score; S42, according to the user individual preference score, the user group preference score and the user learning motivation score, obtaining the corresponding individual preference influence factor, group preference influence factor and learning motivation influence factor; S43, assigning weight coefficients to the individual preference influencing factor, the ethnic group preference influencing factor, and the learning motivation influencing factor; S44. Perform weighted summation on the individual preference influencing factor, group preference influencing factor and learning motivation influencing factor after assigning weight coefficients to obtain the content value score of each course.

5. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 4 is characterized in that: The step S41 specifically includes: (1) Obtaining user individual preference scores: Determine a first influencing factor according to the user's learning status portrait; the first influencing factor includes the user's learning status, learning completion percentage, cumulative learning time, cumulative learning times, whether to collect and share, and the number of times; Determining a second influencing factor according to the predicted profile of the individual learning preference of the user; the second influencing factor includes a key content tag and a predicted content tag; Assigning weight coefficients to the first impact factor and the second impact factor; The first influencing factor and the second influencing factor after the weight coefficient is assigned are standardized and weighted summed to obtain the user's individual preference score; (2) Obtaining user group preference scores: Determine a third influencing factor based on the user learning status portrait; the third influencing factor includes the average learning completion percentage of users in the same group, the average cumulative learning time of users in the same group, the cumulative number of learning times of users in the same group, the number of collections by users in the same group, and the number of shares by users in the same group; Determining a fourth influencing factor according to the user group learning preference prediction portrait; the fourth influencing factor includes the relevance of the key content tag and the relevance of the predicted content tag; Assigning weight coefficients to the third impact factor and the fourth impact factor; The third and fourth influencing factors after the weight coefficients are assigned are standardized and weighted summed to obtain the user group preference scores; (3) Obtaining the user's learning motivation rating: Determine the fifth influencing factor based on the company's preset operating strategy; the fifth influencing factor includes whether all employees must learn, whether the position must learn, whether new employees must learn, whether the department must learn, whether certification preparation is required, and whether it is a popular recommendation; Assigning weight coefficients to the first impact factor and the fifth impact factor; The first influencing factor and the fifth influencing factor after the weight coefficients are assigned are standardized and weighted summed to obtain the user's learning motivation level score.

6. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 1 is characterized in that: The step S5 specifically includes: Based on the content value model, all video courses on the online learning platform are traversed to obtain the content value score of each video course for the target individual user, and the video courses are sorted according to the content value score to generate a course recommendation list for the target individual user.

7. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 5 is characterized in that: Also includes: According to the enterprise's preset operation strategy, obtain the content that individual users must learn; Based on the preset operational strategy priority, a corresponding must-learn task list is generated according to the must-learn content of the individual user; Based on the enterprise's preset operating strategy, teaching plan and teaching time constraints, the priority of the required learning task list is updated.

8. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 7 is characterized in that: Also includes: Acquire the current dynamic learning data of individual users in real time, and update the learning profile of the user according to the current dynamic learning data; Based on the updated user learning profile, the course recommendation list or the required task list is updated.

9. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 8, characterized in that: Also includes: According to the preset daily learning capacity restriction condition, based on the required learning task list and the course recommendation list, a daily learning to-do content list is generated.

10. The personalized learning recommendation method based on big data analysis and artificial intelligence according to claim 9, characterized in that: Also includes: Get user clicks, learning percentage and user reviews; Determining a comprehensive evaluation index based on the number of user clicks, learning percentage, and user evaluation; Optimizing the weight coefficient in the content value model according to the comprehensive evaluation index; Based on the optimized weight coefficient, the content value model is updated.

Images