Author:  Qijia Liu^1,a,*

 Affiliation:

^1,a,*College of Education, Capital Normal University Beijing, 100048, China

 All authors’ email:

^aEmail: 18610408037@163.com

 All authors’ bio:

Qijia Liu, female, from Beijing, China, born in 2004, an undergraduate student, majoring in Educational Technology

Abstract: With the vigorous development of online education, learning resources have grown exponentially, and the problem of “learning resource overload” has become increasingly prominent, leading learners to face dilemmas such as “choice paralysis”. To solve this problem, this paper takes college students as the research object, designs and implements a personalized learning resource recommendation system based on the collaborative filtering algorithm. The system adopts a three-tier architecture with separation of front and back ends. The front end uses Vue and Nuxt frameworks, and the back end is developed with Spring Boot and Spring Cloud frameworks. In the design of the recommendation algorithm, first, a student profile containing static data and dynamic behavior data is constructed; then, the traditional collaborative filtering algorithm is improved by introducing resource weight factors and rating difference factors into the Pearson similarity calculation, and integrating the K-means clustering algorithm to optimize the recommendation process. Experimental results show that the improved recommendation algorithm has significantly improved in indicators such as accuracy, recall, and F1 value. Compared with the traditional UserCF algorithm, the final UserCF-KMeans algorithm has an accuracy increase of about 4.51%, a recall increase of about 14.33%, and an F1 value increase of about 10.22%. The system can provide learners with personalized learning resources that are more in line with their learning needs and preferences, and provides a reference for solving the problem of personalized learning in educational informatization.

Keywords: Personalized recommendation system; UserCF-KMeans; K-means; Pearson similarity

1. Introduction

In recent years, with the in-depth integration of information technology and the education field, the online education industry has shown an explosive growth trend [1-3]. The rapid emergence of Massive Open Online Courses (MOOCs), professional learning platforms, and various digital teaching resource libraries has led to an exponential expansion in the number of online learning resources, covering fields from traditional subject knowledge to vocational skills and interest development[4,5]. This inclusiveness of resources has broken through the limitations of time and space, providing learners with unprecedented opportunities for autonomous learning, theoretically greatly improving the convenience and flexibility of knowledge acquisition [6].

However, the rapid expansion of resources has also triggered the phenomenon of “learning resource overload”, that is, the cognitive dilemma where the number and types of resources faced by learners far exceed their ability to effectively screen, process, and absorb [7-9]. Studies have shown that when the supply of resources exceeds the individual’s information processing threshold, learners often fall into a state of “choice paralysis” and need to spend a lot of redundant time to locate content that meets their cognitive needs, learning goals, and knowledge foundation [10,11]. This inefficient resource screening process not only directly reduces learning efficiency but also may cause negative effects such as cognitive load overload and reduced learning motivation, ultimately leading to a significant decline in the quality of learning experience [12].

To sum up, how to overcome the problem of information overload and enable learners to find the resources they want among a large amount of data, so that learners can have a good learning experience, has become an urgent problem to be solved in current online learning systems. For this reason, Robert Armstrong and others first proposed the concept of a personalized recommendation system at the American Association for Artificial Intelligence. Early research on personalized recommendation systems mainly focused on the teacher collaboration level, constructing an associated recommendation model by mining information of teachers with similar courses [13]; with the rise of MOOC platforms, scholars began to turn to the learner dimension. Bousbahi et al. built a static recommendation system based on learner portraits [14], and Li et al. further introduced user behavior pattern analysis to solve the new user cold start problem [15]. The research paradigm has gradually developed towards dynamization and hybridization. Salehi et al. constructed a dynamic multiple preference model by fusing collaborative filtering and genetic algorithms, which significantly improved the accuracy and diversity of recommendations [16]. At the same time, scholars such as Hudak expanded the theoretical basis of learner modeling from an interdisciplinary perspective, incorporating cognitive characteristics from fields such as psychology and education into the label system [17]. This evolution path presents three significant characteristics: shifting from teacher-centered to learner-centered, deepening from static recommendation to dynamic modeling, and moving from a single algorithm to hybrid intelligent optimization [18-23].

Based on this, this paper addresses the problem of learners’ “information disorientation” caused by overload of learning resources in the field of online education. It takes college students as the main research object, designs and implements a personalized learning resource recommendation system. The research establishes a student portrait model by collecting learners’ static attributes and dynamic behavior data; at the same time, it innovatively introduces resource weight factors and rating difference factors into the traditional Pearson similarity calculation, and combines the K-means clustering algorithm to optimize the user collaborative filtering recommendation process. This research provides learners with personalized learning resources that are more in line with their learning needs and preferences, and provides a reference for solving the problem of personalized learning in educational informatization.

2. System Scheme Design

2.1 Overall System Scheme Design

The system follows the principle of layered design in the overall design, adopts the method of separating the front and back ends for system development. The front end is quickly built using Vue and Nuxt frameworks [24,25], and the back end is developed using Spring Boot framework [26,27], supplemented by Spring Cloud [28] to ensure good maintainability and scalability of the system. According to the idea of B/S architecture and the specific functional requirements of the system, a three-tier architecture model is adopted. The overall architecture design of the system is shown in Figure 1.

Figure 1 Three-tier Architecture Design Framework Diagram of the System

The main function of the UI display layer is to provide users with an interactive Web interface through which users can input relevant information and display course information received from the server. Users access through browsers, which mainly include the system homepage display, recommended resources interface, learning resource details page, user personal center interface, etc.

The main function of the BLL business logic layer is to be responsible for data transmission and processing. This module plays a key core role. It interacts with the instructions from the user layer, executes relevant businesses specifically, and finally transfers the data needed by users from the data access layer to the user layer, playing a connecting role. To explain the main functions of this layer in more detail, it can be divided into: data processing module related to the data access layer, user portrait module and recommendation algorithm module related to recommendation, UI interaction module related to the front-end interface, and background management module and comment collection module related to users.

As the bottom design framework of the entire system, the DAL data access layer mainly includes all operations related to the database. The BLL business logic layer decomposes specific businesses into add, delete, check, and modify operations on the data in the database, and distributes the obtained data results to the UI display layer, so that users can intuitively obtain relevant data resources. The data resources involved in the online course learning system include user personal information, behavior information, and course resource libraries.

Figure 2 Structure Design Diagram of System Functional Modules

Figure 2 shows the specific functional module diagram. In this system, the main functional modules can be subdivided into the front-end student learning module and the back-end management module. The student learning module includes students’ interaction with learning resources, students’ personal information, lecturers’ detailed information, etc.; the back-end module is mainly the administrator module, including user registration and management, lecturer management, course review and classification, etc.

The basic information module includes user registration, login, and personal center. Users register via mobile phone numbers, and the system verifies the input information to ensure the uniqueness of the account. The personal center mainly records the relevant content filled in by students during registration, a brief introduction to students, learning records, course purchase records, etc. Students can view their information, delete relevant browsing and learning footprints, and modify personal introductions.

The course module is the core module of the system, mainly including personalized course recommendation, course details viewing, and course search functions. The personalized recommendation module is a key part of the entire system. It needs to profile users, improve existing recommendation algorithms, implement learning resource recommendation algorithms based on user portraits, and in addition, tag users according to their personal attributes and historical behavior data, and continuously update user portraits to recommend learning resources that meet their personal preferences and actual needs; the course details module mainly displays details of learning resources, including course introduction, resource explanation, chapter arrangement, etc.; the course search module mainly provides users with retrieval functions. When users have clear learning goals, they can find the learning resources they want through keyword retrieval.

The user behavior module mainly includes browsing, commenting, and collecting functions, which can reflect the user’s preference for a certain learning resource to a certain extent. The system can recommend interesting learning resources for them based on this information.

The background management module mainly has user management functions and course management functions. The user management function mainly includes permission management, student information management, administrator information management, and lecturer information management; the course management function mainly includes course information records, course classification management, and course evaluation systems.

2.2 System Database Design

According to the system requirements, this paper determines the system entities and their attributes, clarifies the relationships between various entities, constructs an entity-relationship model according to the relationships between entities, then draws an Entity-Relationship Diagram (E-R Diagram) [29,30], and designs the data table fields of each entity. The main entity relationships of the system are shown in Figure 3.

Figure 3 E-R Diagram of Online Learning Resource Recommendation System

The main entities in the system are shown in Figure 3 This paper sorts out each entity in the system and their relationships with the administrator as the center, including nine entities such as students, teachers, teaching resources, and administrators. Among them, administrators can add, delete, check, and modify all students, teachers, and teaching resources; for students and courses, there is a many-to-many topological structure. A student can learn multiple interesting learning resources, and similarly, each learning resource may be learned by multiple people; for lecturers and courses, the topological relationship is one-to-many, and a teacher can teach multiple courses in this field. In addition, course classification, course chapters, course sections, and course introductions are all related to courses, and comments are published by students to describe relevant courses. The course recommendation module recommends personalized learning resources that meet students’ learning characteristics and needs through specific recommendation algorithms based on their professional characteristics and behavioral characteristics in the system.

3. Design and Implementation of Learning Resource Recommendation Algorithm

3.1 Design of Recommendation Algorithm Scheme

3.1.1 Overall Design of Recommendation Algorithm Scheme

Learning resources have higher complexity, strong pertinence, and professionalism. Therefore, traditional recommendation algorithms have poor recommendation effects and weak pertinence on learning resources. However, a significant feature of learning resources is that the clustering effect is obvious, and the user groups of the same type of resources are highly concentrated. Therefore, this paper introduces user portraits to solve this problem, combining user portraits with online learning resource recommendation [31,32], making the final recommendation results more in line with learners’ learning needs and interests. The overall design scheme of the recommendation algorithm is shown in Figure 4.

Figure 4 Flowchart of Overall Design Scheme of Recommendation Algorithm

This paper first constructs a portrait for users to obtain a labeled representation of the user, and then recommends learning resources with relevant labels to the user. Considering the clustering effect and relevance of learning resources, this paper selects collaborative filtering recommendation algorithms and clustering algorithms, supplemented by students’ basic information and professional characteristics. Then, together with the system’s popular courses, the initial recommendation candidate set is generated. Finally, the final recommendation list is determined by filtering according to the student’s user portrait.

3.1.2 Evaluation Indicators of Recommendation Algorithm

To quantify the prediction accuracy of recommended learning resources, two indicators, MAE (Mean Absolute Error) and RMSE (Root Mean Square Error), are first used for evaluation [26,31].

In the test set,  represents the learner,  represents the learning resource, and  is a set containing all students and learning resources. If  represents the real rating of student  on learning resource , and  represents the rating of student  on learning resource  predicted by the recommendation system, then  can be used to describe the error between the predicted rating and the real rating, which can be calculated by formula (1).

(1)

MAE depicts the prediction performance of the system by calculating the average absolute value of the difference between the real value and the predicted value. We can judge the accuracy of the recommendation according to the size of the deviation. The smaller the value, the higher the accuracy of the algorithm, and vice versa.

The mathematical formula of RMSE can be expressed as formula (2).

(2)

RMSE represents the accuracy of the algorithm by calculating the two-norm between the real value and the predicted value. Like MAE, the smaller the value, the smaller the deviation and the better the effect.

Since this paper is based on a Top-N recommendation system, the recommendation results are the top N learning resources with the highest ratings selected from all learning resources and recommended to students. Therefore, this paper further uses evaluation indicators based on Top-N recommendations to evaluate the algorithm [12,23].

The online learning resource recommendation system of this recommendation algorithm can be evaluated by the following three indicators. First, Precision, which represents the proportion of the number of course resources that users are truly interested in to the number of resources recommended in Top-N, is defined as formula (3).

(3)

Among them,  is the list of learning resources recommended to students according to their behavior, and  is the set of all learning resources that students are truly interested in.

Second, Recall, which represents the proportion of learning resources that students are interested in in the recommendation list recommended to students to all learning resources that students are interested in in the dataset, is defined as formula (4).

(4)

For Top-N recommendation systems, precision and recall can reflect the accuracy of the recommendation algorithm to a certain extent, but sometimes there may be contradictions between these two indicators. When there are many learning resources that users are interested in, but the system’s recommendation list contains only one accurate recommendation, the precision is 100%, but the recall is very low. To solve this special case, we introduce a comprehensive evaluation indicator F to weighted and harmonize P and R, as shown in formula (5).

(5)

Among them, the value range of  is [-1,1]. When , it is the common F1 indicator, also called F1-Score.The larger the value of , the more accurate algorithm is.

3.2 Implementation and Improvement of Collaborative Filtering Algorithm

In the education field, due to the significant clustering characteristics of students in the same major in the selection of learning resources, that is, they tend to use similar learning resources, this paper selects the user-based collaborative filtering recommendation algorithm as the basic framework and improves and optimizes it.

The specific improvement paths are as follows: First, optimize the similarity calculation model. On the basis of the Pearson correlation coefficient, introduce the weight factor of students’ public learning resources and user rating difference characteristics, and construct a collaborative filtering similarity calculation method that integrates multiple factors, in which different influencing factors are given different weights to improve calculation accuracy; second, introduce the user portrait mechanism, and realize the comprehensive utilization of student information by quantifying the matching degree between learning resources and student characteristics, thereby enhancing the reliability of recommendation results [31,32]; third, integrate the clustering algorithm into the collaborative filtering framework [33-35], which not only improves the accuracy and quality of recommendations but also effectively reduces the computational complexity and significantly optimizes the operating efficiency of the algorithm.

3.2.1 Construction of Student Portrait

From the perspective of the learner data model, it can be divided into static data types and behavioral data models. Among them, the static data model mainly includes students’ basic learning conditions and personal preferences; the behavioral data model mainly includes various operation information of students on learning resources in the learning system.

The static data model mainly includes students’ individual characteristics. This paper determines learners’ individual characteristics with students’ basic attributes, course selection information, learning styles, and static hobbies as labels, and conducts user profiling based on students’ static data. The static data label system model is shown in Table 1.

Table 1 Student Static Data Label System

Dynamic characteristics refer to some implicit characteristics that gradually appear with the occurrence of learners’ learning behaviors, such as learners’ cognitive status, main learning content, mastery of learning resources, and evaluation information of some learning resources. Dynamic data will change with learners’ learning behaviors and be continuously updated. Therefore, dynamic data is more important for building learner models. This paper mainly constructs the model of learners’ dynamic data through the data collection layer, data analysis layer, and presentation layer, as shown in Figure 5.

Figure 5 Student Dynamic Data Label System

Next, the dynamic interest preference characteristics and dynamic cognitive level characteristics are labeled. Each student’s learning process is a constantly changing dynamic process. Various operations of students in the learning process will generate different behavioral information, which will reflect students’ dynamic interest preference characteristics. The labeling process is as follows [31].

Step 1: Construct a student-teaching resource matrix  according to students’ learning behaviors, where  represents the behavior weight of the m-th learner  on the n-th learning resource. Each value is the superposition of different behaviors of learners on the learning resource. If  , it means that learner  has no behavior on learning resource .

(6)

Step 2: Construct a learning resource-learning label matrix  to characterize the characteristics of learning resources.

(7)

Each row represents a learning resource , and each column represents a learning label . The element  in matrix  indicates whether the i-th learning resource  has the j-th learning label.  is a binary matrix composed of 0 and 1 , where 1 means the resource has the label, and 0 means it does not.

Step 3: Construct a student-learning label matrix  according to the student-learning resource matrix  and the learning resource-learning label matrix , which can be expressed as:

(8)

is used to represent the dynamic interest preference matrix of students. The element  in matrix  represents the cumulative behavior weight of the i-th learner  on the j-th learning label. The calculation formula is:

(9)

Through the above method, the behavior weight of learners on each learning label can be obtained, thereby confirming the user portrait and its vectorized representation.

3.2.2 Implementation of Collaborative Filtering Algorithm Based on Student Portrait

This section mainly introduces the user-based collaborative filtering algorithm and adds student portraits to improve the accuracy of recommendations. The specific steps are as follows.

First, construct a student-learning resource matrix . We assume that the set of students is  and the set of learning resources is . It can be seen that the number of students is m and the number of learning resources is n. The rating matrix  is shown in Table 2.

Table 2 Student-Learning Resource Rating Matrix

Among them,  represents the rating of student i on teaching resource j. When , it means that student i has no behavior on learning resource j. This paper obtains the rating matrix through the historical information of students in the learning system, including watching time, browsing records, evaluation systems, and course collections. The specific rating standards are shown in Table 3.

Table 3 Student-Learning Resource Rating Matrix

From Table 3, the rating of student i on learning resource j can be calculated using formula (10).

(10)

Then, this paper calculates user similarity. The traditional Pearson correlation coefficient can directly and effectively calculate the correlation between two users, but it does not fully consider the impact of the number of course resources commonly rated by students on similarity calculation. Therefore, this paper considers the weight of common learning resources between two students, which is calculated using formula (11).

(11)

In formula (11),  represents the number of all learning resources that both student u and student v have rated, and  represents the number of all learning resources that student u and student v have rated respectively. The similarity between two students is proportional to the number of common learning resources, but  does not consider the impact of differences in rating standards between students. Therefore, on the basis of learning resource weights, this paper considers the impact of rating differences, and the formula is defined as (12).

(12)

In formula (12),  represents the intersection of learning resources of student u and student v, that is, common learning resources,  represents the number of elements in the set ,  represents the rating of student u on learning resource  i, and  represents the average rating of student u on all learning resources. The same applies to student v and will not be repeated. To highlight the impact of students’ rating standards on similarity, it is necessary to consider the difference in the average rating of all students, not just the intersection of learning resources.

This paper forms the final student similarity calculation method by integrating Pearson similarity, learning resource weight, and rating difference, and the calculation formula is as (13).

(13)

Using this similarity calculation method, the k nearest neighbors with the highest similarity to student u are obtained. Then, based on the ratings of these k similar students on learning resource i, the rating of student u on learning resource i is predicted, and the top N course resources with high predicted ratings are recommended to students. The specific calculation formula is as (14).

(14)

In formula (14),  represents the predicted rating of student u on learning resource i,  epresents the average rating of students on resource i,  is the set of the top  students with the highest similarity to student u,  is the similarity between student u and student v, and  represents the rating of student u on learning resource i.

3.2.3 Collaborative Filtering Recommendation Integrating K-means Algorithm

The collaborative filtering algorithm integrating the K-means algorithm takes the K-means clustering result as input, divides students into clusters with high similarity, and only calculates student similarity within the cluster, avoiding traversal of the entire dataset, thereby improving the accuracy and efficiency of the collaborative filtering algorithm [33-35].

The input of the algorithm is the dataset  for clustering and the number of clusters N , and the output is a set divided into N clusters. The specific implementation steps are as follows.

Step 1: Randomly select N students from the set  as initial cluster centers. Calculate the distance from each student object in the set  to the initial cluster center , and select an appropriate cluster for the student according to the distance.

Step 2: Calculate the sum of squared errors to evaluate the clustering performance. Assume that the initial dataset is , which contains M student objects. The clustered sample set is , which is divided into N clusters. Update the center point of the cluster to the mean of all elements , and the calculation formula is as (15). The two-norm of the distance from all elements in the cluster to the center point is recorded as the sum of squared errors SSE, and the calculation formula is (16). The smaller the value of SSE, the smaller the error and the better the clustering effect.

(15)

(16)

Step 3: Update the cluster centers using the gradient descent method based on the sum of squared errors. Repeat this step until the error is less than the specified value or the maximum number of iterations is reached, then stop updating, and finally obtain a set divided into N clusters.

Step 4: Take the initially obtained set of student-learning resources containing N clusters as the input of the collaborative filtering recommendation algorithm, find other people similar to the student within the cluster, take the learning resource ratings of similar students as input, and calculate the predicted rating of the student on the learning resource through formula (14). Recommend the top learning resources to the target student.

4. Research Results

To verify the effectiveness of the improved student similarity calculation method proposed in this paper, this paper calculates evaluation indicators such as RMSE (Root Mean Square Error), MAE (Mean Absolute Error), accuracy, recall, and F1 value under different similarity calculation methods, and analyzes their changing trends with the number of neighbors k. The experimental results are shown in Figures 6 to 9.

Figure 6 Curves of RMSE with k under Different Similarities

It can be seen from Figures 6 and 7 that with the increase of the number of neighbors k, both RMSE and MAE show a gradually decreasing trend. This indicates that under different similarity calculation methods, the recommendation accuracy gradually improves with the increase of the number of neighbors. However, under the same k value, the RMSE and MAE values corresponding to the improved Pearson similarity calculation method in this paper are significantly lower than those of the traditional Pearson similarity, cosine similarity, and Jaccard similarity. This shows that after integrating the learning resource weight factor and rating difference factor into the Pearson similarity, this paper can more accurately measure the similarity between students, thereby improving the recommendation accuracy and effectively reducing the recommendation error.

Figure 7 Curves of MAE with k under Different Similarities

Further, it can be seen from Figures 8 to 10 that under different similarity calculation methods, the accuracy, recall, and F1 value all gradually increase with the increase of the number of neighbors k. This is consistent with the changing trends of RMSE and MAE, further verifying the conclusion that the recommendation accuracy improves with the increase of the number of neighbors. Under the same k value, the accuracy, recall, and F1 value corresponding to the improved Pearson similarity calculation method in this paper are significantly higher than those of the other three similarity calculation methods. This indicates that the improved similarity calculation method in this paper can more comprehensively consider the similarity between students, not only improving the recommendation accuracy but also improving the recommendation recall and comprehensive performance (F1 value), and has a better application effect in the recommendation system.

Figure 8 Curves of Precision with k under Different Similarities

Figure 9 Curves of Recall with k under Different Similarities

Figure 10 Curves of F1 Value with k under Different Similarities

To determine the optimal number of clustering centers N, this paper conducts experiments under different N values and calculates the change of RMSE with the number of neighbors. The experimental results are shown in Figure 11.

Figure 11 Changes of RMSE with the Number of Neighbors under Different Numbers of Clustering Centers

It can be seen from the figure that when N = 6 or N = 8, the RMSE value is relatively small, and with the increase of the number of neighbors, the downward trend of RMSE is relatively gentle. This indicates that when N = 6 or N = 8, the clustered student-learning resource rating matrix can better reflect the similarity between students, thereby improving the recommendation accuracy. When N is too small (such as N = 2 or N = 4), although the similarity between students in the clustered clusters is high, the number of clusters is small, which cannot fully explore the diversity and personalized characteristics among students, leading to certain limitations in recommendation accuracy. When N is too large (such as N = 10), although the number of clusters is large, which can divide student groups more carefully, the similarity between students in the clusters is relatively reduced, increasing computational complexity, and the improvement effect on recommendation accuracy is limited. Therefore, considering both recommendation accuracy and computational efficiency, this paper selects N = 8 as the optimal number of clustering centers.

To comprehensively evaluate the effect of the proposed recommendation algorithm improvement measures in this paper, this paper calculates evaluation indicators such as RMSE (Root Mean Square Error), MAE (Mean Absolute Error), accuracy, recall, and F1 value under the improved algorithms at different stages, and analyzes their changing trends. These include the traditional user-based collaborative filtering algorithm (User-based CF), the collaborative filtering algorithm after improving the student similarity calculation method (UserCF-Improved), the collaborative filtering algorithm after further adding user portraits (UserCF-Profile), and finally the collaborative filtering algorithm integrating the K-means clustering algorithm (UserCF-KMeans).

Table 4 Performance Evaluation of Recommendation Algorithms

It can be seen from Table 4 that in terms of the three evaluation indicators of accuracy, recall, and F1 value, the performance has been significantly improved with the gradual improvement of the recommendation algorithm. It shows that improving the student similarity calculation method, introducing user portraits, and integrating the K-means clustering algorithm can effectively improve the efficiency and accuracy of the algorithm, reduce the amount of calculation, and further improve the recommendation quality. Compared with the traditional UserCF algorithm, the final UserCF-KMeans algorithm has increased by about 4.51% in accuracy, 14.33% in recall, and 10.22% in F1 value, with significant performance improvement.

To sum up, the experimental results show that with the gradual addition of improvement measures, the performance of the recommendation algorithm has been significantly improved in various evaluation indicators. The method proposed in this paper has good accuracy and F1 value, can better meet students’ learning needs, and provides an effective recommendation method for online learning resource recommendation systems.

5. Conclusion

This study addresses the problem of learners’ “information disorientation” caused by overload of learning resources in the field of online education. It takes college students as the research object, builds a learning resource library around high-frequency electronic circuit-related knowledge points, and designs and implements a personalized learning resource recommendation system based on the collaborative filtering algorithm.

The research constructs a comprehensive student portrait model by collecting learners’ static attributes and dynamic behavior data, and innovatively introduces resource weight factors and rating difference factors into the traditional Pearson similarity calculation, combined with the K-means clustering algorithm to optimize the user collaborative filtering recommendation process. The system is implemented based on a three-tier architecture model using a front-end and back-end separation development method, ensuring good maintainability and scalability. Experimental results show that the improved algorithm is superior to the traditional collaborative filtering algorithm in evaluation indicators such as MAE, RMSE, accuracy, recall, and F1 value. Among them, the algorithm integrated with K-means clustering has the most significant performance improvement, with accuracy, recall, and F1 value increasing by about 4.51%, 14.33%, and 10.22% respectively compared with the traditional algorithm.

In conclusion, the personalized learning resource recommendation system proposed in this study can effectively solve the problem of learning resource overload, provide learners with learning resources more in line with their needs and preferences, provide a useful reference for solving the problem of personalized learning in educational informatization, and also provide a practical basis for the application and optimization of related recommendation algorithms in the education field.

References

• Wang H, Fu W. Personalized learning resource recommendation method based on dynamic collaborative filtering[J]. Mobile Networks and Applications, 2021, 26(1): 473-487.
• Zhang Q. Construction of personalized learning platform based on collaborative filtering algorithm[J]. Wireless Communications and Mobile Computing, 2022, 2022(1): 5878344.
• Hidayat A F, Suwawi D D J, Laksitowening K A. Learning content recommendations on personalized learning environment using collaborative filtering method[C]//2020 8th international conference on information and communication technology (ICoICT). IEEE, 2020: 1-6.
• Jin W. User interest modeling and collaborative filtering algorithms application in English personalized learning resource recommendation: W. Jin[J]. Soft Computing, 2023: 1-14.
• Hu X, Wang Y, Chen Q B, et al. Research on personalized learning based on collaborative filtering method[C]//Journal of Physics: Conference Series. IOP Publishing, 2021, 1757(1): 012050.
• Alanya-Beltran J. Personalized learning recommendation system in e-learning platforms using collaborative filtering and machine learning[C]//2024 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI). IEEE, 2024: 1-5.
• Chatti M A, Dakova S, Thüs H, et al. Tag-based collaborative filtering recommendation in personal learning environments[J]. IEEE Transactions on learning technologies, 2013, 6(4): 337-349.
• Zhong M, Ding R. Design of a personalized recommendation system for learning resources based on collaborative filtering[J]. International Journal of Circuits, Systems and Signal Processing, 2022, 16(1): 122-31.
• Segal A, Katzir Z, Gal K, et al. Edurank: A collaborative filtering approach to personalization in e-learning[C]//Educational Data Mining 2014. 2014.
• Harpale A S, Yang Y. Personalized active learning for collaborative filtering[C]//Proceedings of the 31st annual international ACM SIGIR conference on Research and development in information retrieval. 2008: 91-98.
• Bourkoukou O, El Bachari E. E-learning personalization based on collaborative filtering and learner’s preference[J]. Journal of Engineering Science and Technology, 2016, 11(11): 1565-1581.
• Senthil Kumaran V, Latha R. Towards personal learning environment by enhancing adaptive access to digital library using ontology-supported collaborative filtering[J]. Library Hi Tech, 2023, 41(6): 1658-1675.
• Garcia E, Romero C, Ventura S. A collaborative educational association rule mining tool[J]. Internet High Educ, 2011, 14(2): 77 – 88.
• Bousbahi F, Chorfi H. MOOC – Rec: A Case Based Recommender System for MOOCs[J]. Procedia – Social and Behavioral Sciences, 2015, 195: 1813 – 1822.
• Li Y, Li H. MOOC – FRS: A new fusion recommender system for MOOCs[C]. IEEE advanced information technology electronic and automation control conference, 2017, 1481 – 1488.
• Salehi M, Kamalabadi IN, Ghoushchi MBG. An effective recommendation framework for personal learning environments using a learner preference tree and a GA[J]. IEEE Trans Learn Technol, 2013, 6(4): 350 – 363.
• Hudrk G M, Kihn P. Labeling: Pedagogy and Politics[M]. New York, USA: Routledge, 2014.
• Gm D, Goudar R H, Kulkarni A A, et al. A digital recommendation system for personalized learning to enhance online education: A review[J]. IEEE Access, 2024, 12: 34019-34041.
• Ilidio P, Gharahighehi A, Nakano F K, et al. Personalized Learning in K-12 Education: Exploring Weak-Labels for a Random Forest-based Collaborative Filtering Approach[C]//ALL@ AIED. 2024: 38-48.
• Wang H, Wei Z. Research on Personalized Learning Route Model Based on Improved Collaborative Filtering Algorithm[C]//2021 2nd International Conference on Big Data & Artificial Intelligence & Software Engineering (ICBASE). IEEE, 2021: 120-123.
• Cheng Y. Research on Personalized Learning Recommendation Model Based on User Collaborative Filtering Algorithm[C]//2024 3rd International Conference on Robotics, Artificial Intelligence and Intelligent Control (RAIIC). IEEE, 2024: 389-392.
• Sun B, Tan H, Yang D, et al. The system of personalized learning resource recommendation and experimental teaching based on collaborative filtering[C]//2022 IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, 2022: 862-866.
• Ngo H, Vo K, Nguyen T. Personalized Learning Path Recommendations: Fusing Knowledge Graph Embedding, Sequence Mining, and Collaborative Filtering[C]//2024 IEEE International Conference on Big Data (BigData). IEEE, 2024: 8145-8153.
• Wu L. Analysis and design of personalized learning resources recommendation system based on collaborative filtering algorithm[C]//2023 IEEE 12th International conference on communication systems and network technologies (CSNT). IEEE, 2023: 585-590.
• Liu D. Research on Personalized Resource Recommendation Based on User Profile and Collaborative Filtering Algorithm[J]. Advances in Applied Sociology, 2023, 13(11): 841-849.
• Salehi M, Nakhai Kamalabadi I, Ghaznavi Ghoushchi M B. Personalized recommendation of learning material using sequential pattern mining and attribute based collaborative filtering[J]. Education and Information Technologies, 2014, 19(4): 713-735.
• Raj N S, Renumol V G. A systematic literature review on adaptive content recommenders in personalized learning environments from 2015 to 2020[J]. Journal of Computers in Education, 2022, 9(1): 113-148.
• Ali S M, Ghani I, Latiff M S A. Interaction-based collaborative recommendation: a personalized learning environment (PLE) perspective[J]. KSII Transactions on Internet and Information Systems (TIIS), 2015, 9(1): 446-465.
• Li J, Ye Z. Course recommendations in online education based on collaborative filtering recommendation algorithm[J]. Complexity, 2020, 2020(1): 6619249.
• Lin Y, Wu J, Lin M. Online learning platform based on AI adaptive learning and collaborative filtering algorithm[J]. Journal of Computational Methods in Sciences and Engineering, 2025: 14727978251361823.
• Zhang Z. A method of recommending physical education network course resources based on collaborative filtering technology[J]. Scientific Programming, 2021, 2021(1): 9531111.
• Lin Y, Wu J, Lin M. Online learning platform based on AI adaptive learning and collaborative filtering algorithm[J]. Journal of Computational Methods in Sciences and Engineering, 2025: 14727978251361823.
• Xu M, Kong L. Design and Implementation of Personalized Learning Resource Recommendation System based on Collaborative Filtering Algorithm[C]//2025 3rd International Conference on Data Science and Information System (ICDSIS). IEEE, 2025: 1-6.
• Tahir S, Hafeez Y, Abbas M A, et al. Smart learning objects retrieval for E-Learning with contextual recommendation based on collaborative filtering[J]. Education and Information Technologies, 2022, 27(6): 8631-8668.
• Shi Y, Yang X. A personalized matching system for management teaching resources based on collaborative filtering algorithm[J]. International Journal of Emerging Technologies in Learning (iJET), 2020, 15(13): 207-220.