Personalized teaching recommendation system for higher vocational students

Abstract

The invention discloses a personalized teaching recommendation system for higher vocational students. The system belongs to the technical field of education informatization and artificial intelligence, and comprises a student portrait construction module, a teaching resource label management module, an intelligent recommendation engine module, a recommendation display and reason generation module, a teacher management and intervention module, and a learning process monitoring and feedback module. All the modules are mutually connected to form a recommendation-feedback-optimization closed-loop recommendation structure; according to the method, the recommendation accuracy and the post matching degree are improved; cold start, behavior-driven and semantic-driven multi-strategy fusion is supported, and recommendation accuracy and adaptability are improved; recommendation result interpretability and teacher cooperation capability are realized; dynamic evolution of portraits and self-adaption of recommendation strategies are realized; the system can be widely applied to various vocational education scenes. According to the invention, an intelligent recommendation system containing multi-module cooperation is constructed, and a complete closed loop of accurate layering of students, adaptive distribution of resources and interpretability of recommendation results is realized.

Description

Technical Field

[0001] The present invention belongs to the technical fields of educational informatization and artificial intelligence, and relates to a personalized teaching recommendation system for higher vocational students; in particular, it relates to a personalized teaching recommendation system based on student portraits and multi-label resource management, which is applicable to the construction of intelligent teaching platforms at the higher vocational education stage. Background Art

[0002] In the existing higher vocational education, teaching resources are generally uniformly distributed in units of classes, lacking in-depth analysis and adaptation to individual differences, post goals, and learning styles, resulting in insufficient student learning enthusiasm and uneven effects. At present, although some intelligent teaching platforms support the push of teaching resources and the collection of learning data, they generally lack effective student portrait modeling, post-oriented ability label systems, and dynamic learning path planning mechanisms, and it is difficult to meet the personalized teaching needs of "one size fits one person" in vocational education. Summary of the Invention

[0003] In view of the above problems, the purpose of the present invention is to provide a personalized teaching recommendation system for higher vocational students; it focuses on the technical path of "student portrait × teaching resource label × multi-strategy fusion recommendation", aiming to solve the problems of mismatch between resources and students' abilities and post goals, and rigid recommendation mechanisms in existing teaching systems.

[0004] The technical solution of the present invention is as follows: A personalized teaching recommendation system for higher vocational students described in the present invention; the overall architecture of this system includes six core functional modules: a student portrait construction module, a teaching resource label management module, an intelligent recommendation engine module, a recommendation display and reason generation module, a teacher management and intervention module, and a learning process monitoring and feedback module. Each module is interconnected through data streams and control logics to form a closed-loop recommendation structure of recommendation - feedback - optimization.

[0005] Further, the student portrait construction module: collects data such as students' course grades, practical training performance, post intentions, and learning preferences, and forms a quantifiable and evolvable student portrait through structured label modeling for driving personalized recommendation decisions;

[0006] The teaching resource label management module: assigns multi-dimensional labels to teaching resources such as courses, question banks, and simulation training, including knowledge points, post suitability, certificate goals, teaching forms, and learning stages, to achieve semantic organization and post-oriented management of teaching resources;

[0007] The intelligent recommendation engine module: The system integrates three algorithm strategies of tag matching recommendation, collaborative filtering recommendation, and semantic parsing recommendation, and designs a fusion weight configuration mechanism to dynamically adjust the strategy weights according to dimensions such as "cold start level", "behavior density", and "semantic preference" in the student profile, realizing multi-strategy fusion recommendation and dynamic recommendation strategy optimization;

[0008] Based on the student profile and resource tags, by integrating three strategies:

[0009] (1). Tag matching;

[0010] (2). Collaborative filtering (User-Based);

[0011] (3). Semantic recommendation;

[0012] The recommendation engine sorts according to the total score and screens the final recommendation set, automatically generating personalized learning recommendation content; attaching "recommendation reasons" to each recommended content, such as: "Based on your relatively weak foundation, recommend 'Civil Aviation Freight Basics' as the starting course;";

[0013] The recommendation display and reason generation module: The system not only outputs the course list, but also attaches recommendation reasons, path stage suggestions, and job adaptation annotations, forming a trinity of "recommendation - explanation - path" interpretable output, enhancing the credibility of recommendations and the visibility of learning goals;

[0014] The teacher management and intervention module: Teachers can view student profile clustering, individual recommendation paths, resource adaptability scores, learning progress, and data reports through the background, supporting manual intervention methods such as "assigning tasks by group / by tag", "blocking recommended resources", and "pushing new courses", forming a teacher-system collaborative promotion mechanism;

[0015] The learning process monitoring and feedback module: The system continuously collects students' behavioral data (learning completion rate, homework performance, test results, video viewing rate, etc.), analyzes the acceptance of recommendations and the effectiveness of resources, updates the status of student profile tags, dynamically adjusts the recommendation strategy parameters, and dynamically corrects the next round of recommendation strategies, forming a recommendation closed-loop; Each module of the system is connected through data streams to realize a complete closed-loop process of "student profile - intelligent recommendation - recommendation display - behavioral feedback - profile update", ensuring the personalization, dynamics, and job adaptability of teaching resource recommendations.

[0016] Specifically, 1. Student profile construction module:

[0017] This module is used to construct a multi-dimensional, structured, and dynamically evolvable student profile; The profile tag system = a structured tag set used to describe aspects such as students' characteristics, status, goals, and preferences, supporting functions such as personalized recommendation, learning path planning, and dynamic behavior analysis;

[0018] The system uniformly classifies student portrait tags into five categories: ability mastery, learning preference, career goal, status control, and recommendation feedback; each tag can be sourced from initial input, student behavior, recommendation feedback, or semantic expression, featuring unified structure, clear source, updatability, and matchability; the above tag system supports the tag matching path, strategy weight adjustment, and recommendation interpretability output module in the recommendation engine, and has the ability of dynamic evolution;

[0019] (1) Classification of students' dynamic behaviors

[0020] [[ID=⑧]]The system sets a total of five categories of student behavior types, and each type of behavior contains several behavior indicators; the behavior classification and indicators are as follows:

[0021]

[0022]

[0023] (2) Student portrait tag system:

[0024] The system divides student portrait tags into five categories, and each category of tags has clear data sources, construction methods, and evolution logics, supporting both the initial recommendation in the cold start stage and dynamic evolution during the continuous accumulation of student behavior;

[0025] ① Ability mastery tags (T1):

[0026] Tag examples: Weak - ULD loading, Well - mastered - cargo station process, Need to strengthen - order filling skills;

[0027] Tag sources: Test scores (C1 - C3), Assignment completion rate (B1), Score stability (B2), Self - assessment questionnaire;

[0028] Construction method: Trigger tags according to the concentration of wrong questions, low average score, self - assessment items, etc., and the initial score defaults to 0.6;

[0029] Evolution method: Automatically update the tag weights for each new test / assignment. If the correct rate increases, the tag weakens; if wrong questions are concentrated, the weight increases;

[0030] ② Learning preference tags (T2):

[0031] Tag examples: Prefer short videos, Prefer task - driven courses, Resist graphic resources;

[0032] Tag sources: Resource click - through rate (D1), Completion rate (D2), Video completion rate (A1), Enrollment questionnaire preference items;

[0033] Construction method: The initial questionnaire can specify the preference type, and preference labels are automatically generated when the click ratio in system behavior data is greater than the set threshold;

[0034] Evolution method: Conduct a sliding window analysis of behavior data (last 7 days / 30 days), and the weights of preference labels increase and decrease dynamically; if the preference is no longer obvious, the weight is automatically reduced;

[0035] ③. Occupational goal type labels (T3):

[0036] Label examples: Preparing to take a freight certificate exam, target position - aircraft materials warehouse, paying attention to security inspection practical operations;

[0037] Label sources: Enrollment target questionnaire, semantic input keywords (E3), manual annotation by teachers;

[0038] Construction method: Target labels are automatically generated after keywords or position keywords hit the system label library, and teachers can also add them manually on the teacher side;

[0039] Evolution method: The semantic frequency weight increases, and if not mentioned for a long time, it fades. The system supports a "goal confirmation mechanism" to lock this label;

[0040] ④. Status control type labels (T4):

[0041] Label examples: Cold start users, high behavior density, recent increase in bounce rate;

[0042] Label sources: Learning behavior density (A2), homework / quiz frequency (B1 / C1), recommended click-through rate (D1), bounce behavior analysis;

[0043] Construction method: The platform system initializes the score, and when the behavior density is lower than the set threshold, the "cold start" label is automatically added;

[0044] Evolution method: The status score is updated once a day. High-active behaviors can clear the cold start label, and the active label is generated by judging the sliding window density;

[0045] ⑤. Recommendation feedback type labels (T5):

[0046] Label examples: High recommendation satisfaction, decreasing recommendation acceptance;

[0047] Label sources: Recommendation completion rate (D2), feedback score (D3), recommended bounce rate;

[0048] Construction method: The system records the recommendation response behavior and statistically analyzes the trend of click-through rate / completion rate;

[0049] Evolution method: By default, it is updated once after 10 recommended behaviors in the sliding window; the label is used to assist in adjusting the recommendation frequency and content types;

[0050] The above five types of tags are uniformly saved in a structured format, and the fields include: tag ID, tag name, tag category, weight value ([0-1]), source path, whether it can be updated, update strategy (such as "add 0.1 to the weight for each quiz error, subtract 0.05 for correct answers"), update timestamp, and regularly adjust the scoring value according to the evolution of students' behaviors; this tag system runs through the whole process of the recommendation system's portrait - strategy matching - weight fusion, and is the key foundation for the personalized recommendation accuracy and interpretability of this system;

[0051] The system collects students' raw data and behavioral data, and generates an initial tag set T_u through the portrait feature extraction algorithm; each tag contains fields such as tag ID, tag name, tag category, weight value (W(u,t), [0-1]), source path, whether it can be updated, update strategy, update timestamp, etc., and records the portrait timestamp and data credibility level;

[0052] (3) Tag evolution mechanism:

[0053] The system has a portrait evolution mechanism, and the weight of each tag is dynamically adjusted according to students' behavioral performances, supporting the addition, weakening, and elimination of tags;

[0054] ① Tag addition mechanism;

[0055] When certain tag-related features first appear in students' behavioral data, or when a certain keyword frequently mentioned in the semantic input hits the tag library, the system will automatically add the corresponding tag for this student;

[0056] For example: high error rate of a certain knowledge point in students' quizzes → generate the tag "weak - ULD loading";

[0057] Frequent appearance of keywords such as "freight certificate" and "CAC certification" in semantic input → generate the tag "target certification - CAC";

[0058] The initial weight of the newly added tag is usually set to 0.5, and the system can be configured according to historical experience;

[0059] Tag sources: quiz scores (C1 - C3), homework completion rate (B1), score stability (B2), self-assessment questionnaire;

[0060] Construction method: trigger tags according to the concentration of wrong questions, low average score, self-assessment items, etc., and the initial score is default 0.6;

[0061] Evolution method: automatically update the tag weight for each new quiz / homework, the tag weakens when the correct rate improves, and the weight increases when wrong questions are concentrated;

[0062] ② Tag weakening mechanism:

[0063] When a student's learning performance in a certain label direction continues to improve, the system will automatically reduce the weight of this label based on the trend of behavioral indicators; for example:

[0064] The average score of the student in the last 3 tests is higher than 90% → the label weight decreases by 0.1 before each round of recommendation;

[0065] The click share of video resources drops below 30% → the weight of the learning preference label "prefers short videos" decreases;

[0066] The behavioral trend is analyzed according to a sliding window (such as 7-day or 30-day behavior);

[0067] ③. Label elimination mechanism:

[0068] When a certain label has no behavior hits in N consecutive learning cycles and the current weight is already lower than the threshold (such as 0.1), the system will execute the label elimination process;

[0069] The label is set to the "frozen" state and no longer participates in the recommendation ranking;

[0070] It can be set to logical deletion or archived retention, and the system supports configuration of whether to automatically clear;

[0071] Ensure that the output of the recommendation system is no longer affected by expired labels;

[0072] ④. Evolution update logic and execution method:

[0073] The system performs portrait evolution update regularly every day, and can also perform quick evolution before each round of recommendation; all label evolutions are coordinated by the "label weight management module" and completed by the "behavioral indicator analysis module"; the weight update range is limited to [0,1], and the system supports the "minimum weight protection" mechanism to prevent misdeletion; the system supports configuration of policy parameters (such as rise / fall thresholds, cooling periods, etc.);

[0074] This mechanism realizes the continuous evolution of the portrait and the automatic update of the label system, ensuring that the system's recommendation strategy always responds to the dynamic changes of students, and improving the accuracy, timeliness and intelligence level of recommendations.

[0075] 2. Teaching resource label management module

[0076] This module is responsible for structuring label modeling of teaching resources such as courses, training, and question banks; the system supports three methods: manual annotation, teacher upload and synchronous annotation, and content recognition assisted annotation, which are used to realize the structured label annotation, classification, weight setting and update of teaching resources; this module supports the label path matching strategy (α path) between the student portrait and resource labels, and is a key part of the personalized recommendation engine;

[0077] (1). Label type and source method:

[0078] Learning resource tags are classified into six categories according to their semantic functions:

[0079] Knowledge point tags: such as "ULD loading process" and "cargo station operation specifications", which are manually marked by teachers or extracted by the system through teaching text NLP;

[0080] Ability tags: such as "order printing ability" and "English reading and writing ability", which are parsed from the job ability map or configured by teachers;

[0081] Teaching stage tags: such as "Civil Aviation Cargo - Basic Chapter" and "Exam Simulation - Advanced Course", which are automatically extracted from the course structure;

[0082] Media form tags: such as "video courses", "task packages", and "interactive simulations", which are generated by the system automatically parsing resource metadata;

[0083] Exam - oriented tags: such as "CAC exam courses" and "Junior Freight Agent Certificate", which are generated according to the corresponding exam syllabus of the course or marked by teachers;

[0084] Teaching objective tags: such as "master ULD operation" and "understand the security inspection process", which are generated from the teaching plan objective structure or set by teachers;

[0085] (2) Label structure and binding mechanism:

[0086] Each teaching resource can be bound to multiple tags. Each tag supports the following fields: tag ID, tag name, tag type, tag weight (importance), source method, bound resource ID, and update time. The relationship between tags and resources is many - to - many, and the system supports binding maintenance through the management interface or automated interfaces;

[0087] (3) Tag weight setting and update method:

[0088] The degree of association between a tag and a course is represented by the weight value W(t, i), ranging from [0.3, 1.0]: the main tag is 1.0, the secondary tag is 0.7, and the attached tag is 0.4. The weight can be set by teachers, and the system also supports automatically evaluating the weight through keyword frequency and teaching key point analysis. When the resource content changes, the system supports manually or automatically adjusting the bound tags and their weights;

[0089] The learning resource tag management module can be linked with the course structure system and the teacher management interface, supporting tag import, intelligent tagging, and feedback correction of recommended tags, to build a complete resource semantic structure system.

[0090] 3. Intelligent recommendation engine module

[0091] 3.1 Three - strategy integration:

[0092] (1) Label matching recommendation algorithm: The system calculates the recommendation score based on the matching degree between the student portrait label set \(T_u\) and the resource label set \(T_i\):

[0093]

[0094] In the formula, \(W(u,t)\) represents the weight of label \(t\) in the student portrait, and \(W(t,i)\) represents the contribution degree of label \(t\) to teaching resource \(i\); the higher the score, the more the resource meets the student's current learning needs, serving as the basis for sorting the α path;

[0095] (2) Collaborative filtering recommendation algorithm: When the system accumulates student behavior data, it adopts the user-based collaborative filtering algorithm to make prediction recommendations based on the similarity between students and resource preferences, and predicts the score

[0096]

[0097] In the formula, \(sim(u,v)\) represents the similarity between students \(u\) and \(v\), and \(r_{v,i}\) is the score or preference of similar students for the course; this method can enhance the diversity and accuracy of recommendations;

[0098] (3) Semantic recommendation assistance module: The system integrates a natural language processing module (such as based on the deepseek model) to parse the intent of the student's natural language input;

[0099] For example, when the student inputs "I want to obtain a freight certificate", the system identifies the keywords "obtain a certificate" and "freight", converts them into the label "certificate = freight", and then calls the label matching module for resource recommendation and outputs the "recommendation reason" to explain the recommendation logic;

[0100] The recommendation fusion strategy in this system adopts three parallel strategies: label matching, collaborative filtering, and semantic understanding; the outputs of the above three sub-modules are respectively normalized and fused according to the strategy weights α, β, and γ:

[0101] \(S_{total}(i)=\alpha\cdot S_{label}(i)+\beta\cdot S_{cf}(i)+\gamma\cdot S_{deepseek}(i)\)

[0102] α: The weight of the label matching strategy;

[0103] β: The weight of the collaborative filtering strategy;

[0104] γ: The weight of the semantic recommendation strategy (automatically supplemented);

[0105] Among them, \(\alpha+\beta+\gamma = 1\), and the multi-strategy recommendation score is calculated by setting the fusion weight coefficients α, β, and γ; the recommendation engine sorts according to the total score and screens the final recommendation set; the system supports the explanation of the fusion recommendation source and weight annotation to improve the controllability and interpretability of the recommendation;

[0106] 3.2. Definition Framework of Behavioral Indicators:

[0107] In the system initialization stage, the recommendation engine performs multi-strategy fusion recommendation operations using the preset initial weight template TemplateA (α = 0.5, β = 0.3, γ = 0.2); this setting is applicable to the recommendation scenario where user portrait data is dominant, collaborative behavior data is scarce, and semantic input is not yet sufficient during the cold start stage;

[0108] After the system runs, the fusion weight coefficients α, β, and γ will be automatically adjusted according to the students' behaviors; typical behavioral indicators include: video completion rate (A1), homework score stability (B2), quiz coverage (C1), recommended click-through rate (D1), semantic recommendation hit rate (E2), etc.;

[0109] All indicators are normalized and participate in the calculation logic of the weight dynamic fusion strategy to achieve intelligent evolution control of the recommendation path under multi-dimensional behavioral feedback;

[0110] The calculation of the fusion recommendation weight adopts a behavioral indicator-driven strategy, as follows:

[0111] (1) Calculation of the weight α of the label matching strategy:

[0112] α = 0.2×(1 - A1)+0.2×(1 - B2)+0.2×(1 - D1)+0.2×(1 - D3)+0.2×(1 - C2)

[0113] When the following behaviors are "poor" (small values), α↑:

[0114] Low video completion rate A1 → indicating insufficient learning input;

[0115] Poor homework stability B2 → indicating weak mastery and the need for label backup;

[0116] Low recommended click-through rate D1 and low feedback score D3 → indicating that the current recommendation is inaccurate and relying on the portrait is more stable;

[0117] Low average quiz score C2 → the system needs to first give basic resource recommendations based on labels;

[0118] The higher α is, the less sufficient the students' current behavioral data is, and the more the system trusts the portrait to recommend "basic adaptation courses";

[0119] (2) Calculation of the weight β of the collaborative filtering strategy:

[0120] β = 0.25×B1+0.25×C1+0.2×D2+0.15×C3+0.15×A2

[0121] When the following behaviors are "positive", β↑:

[0122] High completion rate B1 of homework → indicating activity and stability;

[0123] High coverage C1 of quizzes → wide mastery and a basis of labels;

[0124] High completion rate D2 of recommendations → indicating that the recommended resources are truly accepted;

[0125] Low concentration C3 of wrong questions → no obvious weak points in knowledge points;

[0126] High density A2 of learning behaviors → overall activity, and collaborative calculation has reference value;

[0127] The higher β is, the richer the student behavior data is, and the system can find collaborative recommendations of "similar behaviors and consistent preferences" from similar students;

[0128] (3) Calculation of the weight γ of the semantic recommendation strategy:

[0129] γ = min[1 - α - β, 0.5×(E1×E2 + E3)]

[0130] γ comes from the remaining weight but is limited by semantic activity; Semantic activity indicators:

[0131] E1: ↑Semantic input frequency → students actively ask questions

[0132] E2: ↑Semantic hit rate → GPT recommendations hit the student's intention

[0133] E3: ↑Keyword coverage rate → high degree of matching between the input and the system labels;

[0134] The higher γ is, the more students strongly use semantic dialogue as an interaction method, and the system should use GPT more to understand semantic intentions and recommend resources;

[0135] Finally, the system normalizes and adjusts α, β, and γ to ensure that the sum of the three weights is 1, and dynamically maps them to the corresponding policy branches in the recommendation engine for the input of the weight dynamic adjustment model, so as to achieve the adaptive fusion control at the recommendation strategy level and improve the recommendation accuracy and interpretability of the system in different portrait states;

[0136] 3.3 Summary of the recommendation process:

[0137] After the student completes the learning task, the behavior collection module records the behavior data;

[0138] The system updates the student portrait labels;

[0139] The weight calculation module calls the latest indicators to calculate α, β, and γ;

[0140] The recommendation engine integrates three types of strategies to output recommended resources;

[0141] The recommendation display module displays personalized recommendations and attaches reasons for the recommendations;

[0142] Teachers can intervene in the recommendation results through the platform and view the basis.

[0143] 4. Recommendation Display and Reason Generation Module

[0144] This module is responsible for presenting the recommendation results to students and teachers in an interactive form; the system adopts a three-stage output method of "course card + recommendation path + recommendation reason"; each recommended course is accompanied by recommendation explanations such as the source of recommendation (tag matching, collaborative filtering, semantic recommendation), matching weight, and job suitability tips, enhancing the credibility of the recommendation and the visibility of learning objectives;

[0145] The recommendation reason generation engine integrates three types of traceability links: tag recommendation ("You are identified as a hands-on type, matching practical training courses"), collaborative behavior reference ("90% of the students with a similar portrait to yours have selected this course"), and semantic understanding explanation ("You mentioned that your exam goal is freight transportation, so this special training is recommended"); the system supports version recording of recommendation results and backtracking of recommendation history, facilitating learning trajectory analysis.

[0146] 5. Teacher Intervention Support Module

[0147] This module provides a teacher background interface for viewing student portraits, recommendation lists, and path planning situations, and supports manual intervention in the recommendation plan; teachers can perform operations such as setting recommendation shielding (excluding courses), weight fine-tuning (strengthening the priority of a certain type of course), and manual push (mandatory resource injection) for individual students or groups;

[0148] The intervention logic is embedded in the recommendation engine fusion layer, and weight modes of "system first" or "teacher first" can be set to form a teacher-system collaborative recommendation framework; at the same time, a task distribution mechanism is supported, and teachers can screen student groups based on tag conditions and assign hierarchical tasks to meet personalized teaching under large-class teaching.

[0149] 6. Learning Process Monitoring and Feedback Module

[0150] The system collects students' course learning behavior data in real time (clicks, start rates, completion rates, quiz scores, video dragging behaviors, etc.), and generates a "recommendation acceptance score" for each recommended course; the feedback engine compares the recommendation reasons with the actual learning behaviors, performs difference analysis, and dynamically adjusts the portrait tags;

[0151] The update methods include: weight fine-tuning (e.g., "video preference" rises from 0.6 to 0.8), label status update (mastered, weakly mastered, not mastered), label addition (automatically generating new labels based on new behaviors), etc.; all updates are written into the portrait log, supporting teachers to view and trace the data; through the cycle of feedback - adjustment - recommendation, the system realizes self-adaptive optimization and personalized evolution;

[0152] By continuously collecting students' learning behavior data (such as video viewing rate, homework submission situation, stage test scores, etc.), the system can correct students' portrait labels in real time, dynamically adjust the recommendation strategy parameters, and realize portrait update and closed-loop optimization driven by learning behavior.

[0153] The beneficial effects of the present invention are as follows: 1. Improve the recommendation accuracy and job matching degree; 2. Support the integration of multiple strategies such as cold start, behavior-driven and semantic-driven, and improve the recommendation accuracy and adaptability; 3. Have the interpretability of recommendation results and the ability to collaborate with teachers; 4. Realize the dynamic evolution of the portrait and the self-adaptation of the recommendation strategy; 5. The system has good scalability and adaptability and can be widely applied to various vocational education scenarios; the present invention has good scalability, practicability and engineering implementation value; the present invention focuses on the core path of "student portrait × resource label × multi-strategy fusion recommendation", constructs an intelligent recommendation system including multi-module collaboration, and realizes a complete closed-loop of accurate student stratification, resource adaptation distribution and interpretability of recommendation results. Brief Description of the Drawings

[0154] Figure 1 It is the structure diagram of the personalized teaching recommendation system for higher vocational students in the present invention;

[0155] Figure 2 It is the class diagram of the personalized teaching recommendation system in the present invention;

[0156] Figure 3 It is the data flow diagram of the personalized teaching recommendation system in the present invention;

[0157] Figure 4 It is the flowchart of multi-strategy fusion recommendation in the present invention;

[0158] Figure 5 It is the closed-loop timing diagram of the influence of student portrait label update on the recommendation engine in the present invention. Detailed Embodiment

[0159] The following further elaborates on the specific technical solutions of the present invention with specific examples.

[0160] The present invention further verifies its technical effects through teaching practice cases; in the course of "Civil Aviation Cargo Transportation" in a certain higher vocational college, 80 students majoring in "Civil Aviation Transportation Service" in the 2023 grade are selected and divided into an experimental group and a control group of 40 people each for a comparative test on the effectiveness of the recommendation system.

[0161] The experimental group adopted the personalized recommendation system described in the present invention, and the control group adopted the traditional unified resource push method; the test period was three weeks, and the weight configuration of the initial recommendation strategy integration of the system was: tag matching α = 0.5, collaborative filtering β = 0.3, semantic recommendation γ = 0.2, and the recommendation update period was once a week.

[0162] The test process is as follows:

[0163] (1), In the first week, the system generated the initial recommendation and recorded the student behavior data (clicks, views, assignments, tests, etc.);

[0164] (2), In the second and third weeks, the system updated the portrait tags and reconstructed the recommendation path according to the behavior feedback, forming a recommendation - feedback - evolution closed loop;

[0165] (3), Teachers in the experimental group could perform recommendation intervention operations on some students;

[0166] The comparison of some behavior results is as follows:

[0167] Click - through rate of recommended resources: 82% in the experimental group, 61% in the control group, an increase of 21%;

[0168] Completion rate of video courses: 74% in the experimental group, 53% in the control group, an increase of 21%;

[0169] Average score of unit tests: 80.6 in the experimental group, 72.1 in the control group, an increase of 11.8%;

[0170] Recommendation satisfaction score (on a 5 - point scale): 4.3 in the experimental group, 3.1 in the control group, an increase of 38.7%;

[0171] The effective triggering rate of tag evolution was 82%, and the average number of portrait coverage dimensions increased from 3.2 to 5.6 items.

[0172] Comparison of technical effects by role:

[0173]

[0174] The above results show that by integrating the recommendation strategy with the dynamic portrait optimization mechanism, the present invention significantly improves the adaptability of the recommended content and the enthusiasm of learning behaviors. The recommendation results are interpretable, and the teacher intervention mechanism ensures the path controllability. The overall recommendation system shows good technical implementation effects and promotion application values in the higher vocational teaching scenario.

[0175] The following is a specific combination with the actual scenario of the system in the teaching of the higher vocational civil aviation transportation service major;

[0176] I. Initialization of user portrait

[0177] Student Zhang, a sophomore majoring in civil aviation transportation services, is currently studying the course "Civil Aviation Cargo Transportation"; after logging in to the platform for the first time, he completed the enrollment guidance questionnaire; based on the student status information, questionnaire answers and teacher annotations, the system generated the following initial portrait tags:

[0178] Ability mastery category: Weak - ULD loading and handling specifications (weight 0.8);

[0179] Learning preference category: Prefers short video courses (weight 0.7);

[0180] Career goal category: Target position - cargo station operation post, preparing to obtain the civil aviation freight forwarder qualification certificate (weight 0.9);

[0181] Status category label: Cold start user (initial mark);

[0182] The system identifies it as a cold start state and automatically calls the initial weights of the fusion strategy: α = 0.5, β = 0.3, γ = 0.2;

[0183] II. Execution of the first round of recommendations

[0184] Student's semantic input: "I want to master ULD loading skills and prepare for the freight certificate exam;" Based on the three - path score fusion of α (tag matching), β (collaborative filtering), and γ (semantic understanding), the system generates the following recommended list:

[0185]

[0186]

[0187] III. Recording of behavior feedback

[0188] Zhang clicks on the recommended course 1 "ULD Loading and Handling Specifications", completes the entire video content, and finishes the in - class exercises and course feedback; the system records the following behaviors:

[0189] Index Number Index Name Value Description A1 Video Completion Rate 96％ Indicates a high level of learning investment in this resource D1 Recommended Click Behavior Yes Click on Recommended Course 1 D2 Recommended Task Completion Rate Yes Complete relevant quizzes and exercises

[0190] IV. Label evolution and strategy weight update

[0191] Based on the behavior feedback and evolution rules, the system automatically executes label evolution and weight strategy switching:

[0192] The weight of the label "Prefers short video courses" ↑ to 0.8 (A1 = 96%);

[0193] The label "Weak in ULD loading" remains unchanged temporarily (the quiz module has not been completed);

[0194] The status of the label "Cold start user" is lifted and switched to the "Activated" status;

[0195] The fusion weights are updated as: α = 0.6, β = 0.25, γ = 0.15.

[0196] V. Second-round recommendation generation

[0197] The system generates the recommended resource sequence again, reflecting the changes after the portrait update and strategy adjustment:

[0198]

[0199]

[0200] So far, the system has completed the complete adaptive closed-loop process of "recommendation - learning - feedback - update - re-recommendation".

[0201] VI. Explanation of technical effects

[0202] Compare the recommendation performance before and after:

[0203] The click-through success rate has increased from 33% to 66%;

[0204] The recommendation satisfaction level has increased from unrated to 5 stars;

[0205] The average total recommendation score has increased from 0.84 to 0.91;

[0206] The number of tag hits has increased from 2 to 3.

[0207] This specific implementation process shows that the system can automatically generate a recommendation path based on the initial portrait of students, realize portrait evolution through behavior feedback, and dynamically adjust the fusion strategy to achieve a closed-loop process of higher vocational course recommendation that combines ability priority, interest adaptation, and goal orientation, with high implementability, intelligence, and promotion value.

Claims

1. A personalized teaching recommendation system for higher vocational students, characterized in that, It includes a student portrait construction module, a teaching resource label management module, an intelligent recommendation engine module, a recommendation display and reason generation module, a teacher management and intervention module, and a learning process monitoring and feedback module; each module is interconnected through data streams and control logic to form a closed-loop recommendation structure of recommendation - feedback - optimization.

2. The personalized teaching recommendation system for higher vocational students according to claim 1, wherein The student portrait construction module collects data on students' course grades, practical training performance, career intentions, and learning preferences, and forms a multi-dimensional, structured, and dynamically evolvable student portrait through structured label modeling. The student portrait labels are uniformly divided into five categories: ability mastery, learning preference, career goal, status control, and recommendation feedback. Each label comes from initial input, student behavior, recommendation feedback, or semantic expression. Specifically, it includes: (1) students' dynamic behaviors, (2) the student portrait label system, and (3) the label evolution mechanism.

3. The personalized teaching recommendation system for higher vocational students according to claim 2, wherein The student portrait label system refers to dividing the student portrait labels into five categories, and each category of label has clear data sources, construction methods, and evolution logics. Specifically, it includes: ① The ability mastery label, i.e., T1. ② The learning preference label, i.e., T2. ③ The career goal label, i.e., T3. ④ The status control label, i.e., T4. ⑤ The recommendation feedback label, i.e., T5. The above five categories of labels are uniformly saved in a structured format, and the fields include: label ID, label name, label category, weight value, source path, whether it can be updated, update strategy, update timestamp, and the score value is dynamically adjusted regularly according to the evolution of student behavior. Collect the original data and behavior data of students, and generate an initial label set T_u through the portrait feature extraction algorithm; each label contains fields such as label ID, label name, label category, weight value, source path, whether it can be updated, update strategy, update timestamp, and records the portrait timestamp and data credibility level.

4. An individualized teaching recommendation system for higher vocational students according to claim 2, characterized in that, The label evolution mechanism refers to: the system has a portrait evolution mechanism, and the weight of each label is dynamically adjusted according to the performance of student behavior, supporting label addition, weakening, and elimination. Specifically, it includes: ① The label addition mechanism. ② The label weakening mechanism. ③ The label elimination mechanism. ④ The evolution update logic and execution method. This mechanism realizes the continuous evolution of the portrait and the automatic update of the label system, ensuring that the system's recommendation strategy always responds to the individual dynamic changes of students, and improving the accuracy, timeliness, and intelligence level of recommendations.

5. The personalized teaching recommendation system for higher vocational students according to claim 1, wherein The teaching resource label management module is responsible for structuring label modeling of teaching resources such as courses, practical training, and question banks, including knowledge points, job suitability, certificate goals, teaching forms, and learning stages, to achieve semantic organization and job-oriented management of teaching resources. This module supports the label path matching strategy between the student portrait and resource labels and is an integral part of the personalized recommendation engine; specifically, it includes: (1) label types and source methods, (2) label structures and binding mechanisms, and (3) label weight setting and update methods.

6. The personalized teaching recommendation system for higher vocational students according to claim 1, wherein The intelligent recommendation engine module integrates three algorithm strategies: tag matching recommendation, collaborative filtering recommendation, and semantic parsing recommendation, and designs a fusion weight configuration mechanism to dynamically adjust the strategy weights according to the cold start level, behavior density, and semantic preference dimensions in the student profile, so as to achieve multi-strategy fusion recommendation and dynamic recommendation strategy optimization; The recommendation engine sorts according to the total score and filters the final recommendation set, and automatically generates personalized learning recommendation content.

7. An individualized teaching recommendation system for higher vocational students according to claim 6, characterized in that, Specifically: First, the integration of three strategies: (1) Tag matching recommendation algorithm: The system calculates the recommendation score according to the matching degree between the student profile tag set T_u and the resource tag set T_i: In the formula, W(u,t) represents the weight of tag t in the student profile, and W(t,i) represents the contribution degree of tag t to teaching resource i; the higher the score, the more the resource meets the current learning needs of the student, and it serves as the basis for sorting the α path; (2) Collaborative filtering recommendation algorithm: When the system accumulates student behavior data, it adopts the user-based collaborative filtering algorithm to predict and recommend through the similarity between students and resource preferences, and predicts the score: In the formula, sim(u,v) represents the similarity between student u and v, and r_v,i is the rating or preference of similar students for the course; (3) Semantic recommendation auxiliary module: The system integrates a natural language processing module to parse the intent of the student's natural language input; Second, the behavior index definition framework: In the system initialization stage, the recommendation engine performs multi-strategy fusion recommendation operations using the preset initial weight template TemplateA; This setting is applicable to the recommendation scenario where the user profile data is mainly in the cold start stage, the collaborative behavior data is less, and the semantic input is not sufficient; Third, the summary of the recommendation process.

8. The personalized teaching recommendation system for higher vocational students according to claim 1, characterized in that, The recommendation display and reason generation module is responsible for presenting the recommendation results to students and teachers in an interactive form; it adopts a three-section output method of "course card + recommendation path + recommendation reason"; each recommended course is accompanied by the recommendation source, matching weight, and job adaptation tips to enhance the recommendation credibility and learning and semantic understanding explanations.

9. The personalized teaching recommendation system for higher vocational students according to claim 1, characterized in that The teacher management and intervention module provides a teacher background interface, through which the teacher can view the student profile clustering, individual recommendation path, resource suitability score, learning progress, and data report in the background, forming a teacher-system collaborative co-recommendation mechanism.

10. A personalized teaching recommendation system for higher vocational students according to claim 1, characterized in that, The learning process monitoring and feedback module collects the student's behavior data in real time, analyzes the recommendation acceptance and resource effect, updates the status of the student profile tags, dynamically adjusts the recommendation strategy parameters, and dynamically corrects the next round of recommendation strategies, forming a recommendation closed-loop.

Images