Course service personalized recommendation method based on knowledge graph

Abstract

The invention relates to the technical field of curriculum recommendation, in particular to a curriculum service personalized recommendation method based on a knowledge graph. The method comprises the following steps: acquiring course information and user interaction information on a target platform; the user interaction information comprises target user interaction information and other user interaction information; the course information is preprocessed; defining entities and relations, and extracting the entities and the relations from the preprocessed course information to establish a plurality of triple data; constructing a course knowledge graph based on the information relevance of the triple data; determining candidate recommended courses for a target user based on the user interaction information and a course knowledge graph; and respectively calculating interest values of the target user for the candidate recommended courses, and recommending the first N courses with the highest interest values to the target user. By adopting the method provided by the invention, the recommendation effect of the existing course recommendation system can be optimized, the accuracy of course recommendation can be improved, and the personalized learning requirement of the user can be met.

Description

Technical Field

[0001] This application relates to the technical field of course recommendation, and in particular to a personalized recommendation method for course services based on a knowledge graph. Background Art

[0002] In recent years, with the rapid development of technologies such as cloud computing, big data, and artificial intelligence, great changes have taken place in the field of education. Application technologies such as educational informatization, distance education, and Web 2.0 have experienced revolutionary growth, and online education has developed rapidly. The emergence of the course recommendation system has effectively alleviated the problem of "information explosion" in online education and has received increasing attention from more and more researchers as an important part of improving the quality of online education. However, the existing recommendation algorithms based on knowledge graphs ignore the connection between the user's learning state and the user's personalized needs for courses, and usually adopt a single-path recommendation method, resulting in poor recommendation effects, affecting the user's learning efficiency and being difficult to meet the personalized learning needs of different students. Therefore, it is urgent to improve the existing course recommendation methods to solve the problem of being unable to accurately locate the user's personalized learning needs. Summary of the Invention

[0003] The purpose of the embodiments of this application is to provide a personalized recommendation method for course services based on a knowledge graph to solve the problem of being unable to accurately locate the user's personalized learning needs.

[0004] To achieve the above purpose, this application provides a personalized recommendation method for course services based on a knowledge graph, and the method includes: S1. Obtain course information and user interaction information on a target platform; the user interaction information includes target user interaction information and other user interaction information; S2. Preprocess the course information; S3. Define entities and relationships, extract entities and relationships from the preprocessed course information to form several triple data; construct a course knowledge graph based on the information relevance of each triple data; S4. Determine whether the target user is in a focused learning state based on the target user interaction information; if so, determine candidate recommended courses for the target user based on the course knowledge graph using a first recommendation strategy, otherwise determine candidate recommended courses for the target user based on the course knowledge graph using a second recommendation strategy; S5. Calculate the interest values of the target user for each candidate recommended course respectively, and recommend the top N courses with the highest interest values to the target user.

[0005] In the embodiments of this application, the preprocessing of the course information includes: removing illegal data in the course information, checking for duplicate course information, and verifying the accuracy of the course information.

[0006] In the embodiments of the present application, the definition of entities and relationships includes: defining courses, chapters, test questions, knowledge points, the number of course visitors, teachers, fields, institutions, and course release times in course information as entities; defining the relationships between entities as the chapters included in a course, the knowledge points included in a chapter, the test questions included in a course, the knowledge points tested by a test question, the value of the number of course visitors, the field to which a course belongs, the release year, the teacher to whom a course belongs, and the institution to which a course belongs in course information.

[0007] In the embodiments of the present application, the target user interaction information includes historical test data that can reflect the learning progress of the target user; the first recommendation strategy includes: judging whether there are weak knowledge points of the target user in the corresponding course according to the historical test data of the target user; judging whether there are prerequisite dependent knowledge points for the weak knowledge points based on the course knowledge graph; if so, taking the course containing the prerequisite dependent knowledge points as the candidate recommended course for the target user.

[0008] In the embodiments of the present application, the target user interaction information includes first historical behavior data that can reflect the preferences of the target user, and the other user interaction information includes second historical behavior data that can reflect the preferences of other users. The second recommendation strategy includes: based on the first historical behavior data and the second historical behavior data, finding out a group of users similar to the target user in terms of preferences from other user groups; selecting courses that the target user has not browsed from the courses preferred by the group of users to form a course set; based on the course knowledge graph and the first historical behavior data, selecting courses close to the preferences of the target user from the course set to form a group of candidate recommended courses, and the remaining courses in the course set form a second group of candidate recommended courses; among them, the recommendation priority of the first group of candidate recommended courses is higher than that of the second group of candidate recommended courses.

[0009] In the embodiments of the present application, calculating the interest value of the target user for each candidate recommended course includes: obtaining the field to which the target candidate recommended course belongs based on the course knowledge graph; obtaining the average score of the target user for the courses in this field from the first historical behavior data. If there is no relevant record, a preset default constant value is used; calculating the similarity between the target user and other users; calculating the interest value of the target user for each candidate recommended course; the calculation expression of the interest value is as follows: belongs to; obtaining the target user from the first historical behavior data average score for courses in this field , if there is no relevant record, a preset default constant value is used; calculating the target user and other users similarity ; calculating the interest value of the target user for each candidate recommended course; the interest value The calculation expression is as follows: (1); in the above formula (1), represents the set of groups of users highly similar to the target user , represents other users Rating for the target candidate recommended courses indicates the average rating of the courses within the field to which the target candidate recommended courses belong, given by other users

[0010] In the embodiments of the present application, the target user interaction information includes: the total time the target user spends on the target platform within a specified time period and the learning time spent on each course; determining whether the target user is in a state of concentrated learning based on the target user interaction information includes: calculating the proportion of the learning time of the target user on each course compared to the total time respectively; determining whether there is a proportion exceeding the proportion threshold; if there is, it is determined that the target user is in a state of concentrated learning, otherwise it is determined that the target user is not in a state of concentrated learning

[0011] In the embodiments of the present application, the method further includes: generating training samples based on user interaction information and a course knowledge graph, selecting a KGAT model based on an attention mechanism, and training the KGAT model based on the attention mechanism using the training samples to obtain a course personalized recommendation model for performing steps S4 - S5

[0012] In the embodiments of the present application, the KGAT model based on an attention mechanism includes an embedding layer, an attention embedding propagation layer, and a prediction layer; the embedding layer is used to initialize the vectors of users, courses, entities, and relationships; the attention embedding propagation layer is used to update the embeddings of users and courses and determine candidate recommended courses through multi - hop neighbor aggregation; the prediction layer is used to calculate the interest value of the user in the candidate recommended courses based on the updated embeddings and generate a course recommendation list containing N courses

[0013] The personalized recommendation method for course services based on a knowledge graph provided by the present application has at least the following beneficial effects The present application utilizes the correspondence between the entities in the course knowledge graph and the courses (purchased, clicked, or browsed by the user) in the user interaction data, effectively combines the personalized preferences of different users for courses with the knowledge graph, and this combination provides additional auxiliary information for personalized recommendation, which helps to solve the problem of information sparsity. In addition, in the present application, the candidate recommended courses for the target user are determined by mining the deep connection between user interaction information and the course knowledge graph, and the interest of the target user in each candidate recommended course is quantified and predicted, and then feedback information (which is part of the user interaction information) is collected, so as to form a closed - loop, which helps to optimize the recommendation model and improve the accuracy of course recommendation

[0014] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent specific implementation part BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present application, but do not constitute a limitation to the embodiments of the present application. In the accompanying drawings: Figure 1 Schematically shown is a flowchart of a method for personalized recommendation of course services based on a knowledge graph in an embodiment. Specific Embodiments

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. It should be understood that the specific embodiments described herein are only used to illustrate and explain the embodiments of the present application and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0017] It should be noted that if there are directional indications (such as up, down, left, right, front, back, etc.) involved in the embodiments of the present application, then such directional indications are only used to explain the relative positional relationship and movement conditions between components in a specific posture (as shown in the accompanying drawings). If this specific posture changes, then such directional indications will also change accordingly.

[0018] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, then such descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In addition, the technical solutions between various embodiments can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present application. Embodiment 1

[0019] Please refer to Figure 1 , this embodiment provides a method for personalized recommendation of course services based on a knowledge graph, and the method includes: S1. Obtain course information and user interaction information on the target platform; the user interaction information includes target user interaction information and other user interaction information; Exemplarily, the Mooper dataset can be downloaded from the public dataset website. This dataset is extracted from the interaction data (user interaction information) of users participating in practical training on the Edu Coder platform and the course dataset (course information) on the platform, and information such as courses, tests, and knowledge points is modeled into a knowledge graph to construct a large-scale practice-oriented online learning dataset. Based on the above method, structured raw data can be quickly obtained, improving the data processing efficiency and the construction efficiency of the knowledge graph.

[0020] S2. Preprocess the course information; Exemplarily, save the above-downloaded user interaction information and course information as an information dataset locally, conduct a detailed analysis of the information dataset, sort out the structure of the information dataset, and check the legality of the data in the information dataset. Remove the obviously illegal data in the information dataset, check for duplicate course information. For example, some courses lack key information such as the number of visitors and release time. Replace these missing values with "Na N" and standardize the release time in the courses. During the data cleaning and preprocessing process, verify and correct the information dataset to ensure accuracy. For example, the chapter information of the course needs to be verified to ensure that the chapter belongs to this subject; verify the knowledge points of the course to ensure that all knowledge points are covered by the course. Ensuring the accuracy of the input data is an important prerequisite for ensuring the accuracy of the recommendation results.

[0021] S3. Define entities and relationships, extract entities and relationships from the preprocessed course information to form several triple data; construct a course knowledge graph based on the information relevance of each triple data; Taking the above Mooper dataset as an example, since the Mooper dataset has already stored and displayed the course information and the interaction information between users and courses in a structured manner, therefore, the knowledge extraction process can be simplified here. Specifically, entities and relationships can be defined manually and then knowledge extraction can be carried out using an automated program, thus avoiding the cumbersome extraction of knowledge from the text.

[0022] Exemplarily, when extracting entities, entities such as courses, chapters, test questions, knowledge points, the number of course visitors, teachers, fields, institutions, and course release time in the course information can be extracted. When extracting relationships, relationships such as the chapters included in the course, the knowledge points included in the chapter, the test questions included in the course, the knowledge points tested by the test questions, the value of the number of course visitors, the field to which the course belongs, the release year, the teacher to whom the course belongs, and the institution to which the course belongs in the course information are defined and extracted as (relationships between entities).

[0023] Then, identify and normalize the names of the above entities to unify the entity representations in different data sources. For example, for a course name written in traditional Chinese characters, convert it to simplified Chinese characters. Then, for different representations of the same entity, it is necessary to align their entity attributes. Some course names have both Chinese and English expressions, but they refer to the same course. Therefore, entity disambiguation needs to be performed, and these entity attributes can be unified through entity similarity calculation.

[0024] Finally, store the extracted entities and relationships in the graph database in the form of triples. Construct a course knowledge graph based on the information of the triples and select a suitable graph database system (such as Neo4j) for storage. The graph database can efficiently store and query graph-structured data, providing strong data support for subsequent recommendation algorithms. Triples are the basic units of the knowledge graph, and their form is <head entity, relationship, tail entity>. For example: <Course A, contains, Section a1>, <Course A, belongs to teacher, Teacher M>. In the above two triples, there is a common entity "Course A", that is, the information of the two triples is related. From the above two sets of triples, it can be inferred that Section a1 is explained by Teacher M.

[0025] S4. Judge whether the target user is in a state of concentrated learning based on the target user interaction information; if so, determine the candidate recommended courses for the target user using the first recommendation strategy based on the course knowledge graph, otherwise determine the candidate recommended courses for the target user using the second recommendation strategy based on the course knowledge graph; Specifically, the target user interaction information includes: the total time spent by the target user on the target platform within a specified time period and the learning time spent on each course (the above information can usually be directly obtained from the platform, and if it cannot be obtained, it can be obtained with the help of a program plugin for monitoring the system running time); the judgment of whether the target user is in a state of concentrated learning based on the target user interaction information includes: Calculate the proportion of the learning time of the target user on each course compared to the total time respectively; judge whether there is a proportion exceeding the proportion threshold; if so, determine that the target user is in a state of concentrated learning, otherwise determine that the target user is not in a state of concentrated learning.

[0026] The above total time can be set as the historical recent online time T (in minutes) on the target platform, where T = 30 - 120 min. The proportion threshold can be taken as 70% - 90%. When a user is focused on learning a certain course, they usually do not waste too much time browsing other subjects. Therefore, the courses recommended to the user at this time must be those that help the target user understand the current knowledge content. Adopting the following first recommendation strategy helps the target user deeply study the current course and improve the user's learning efficiency. If other courses in fields unrelated to the current course are recommended to the user at this time, it is likely to waste the user's time because even if the recommended courses are those the user is interested in, the user's current needs may not be dominated by interest. For example, when an exam is approaching, the user needs to quickly master the knowledge of the current course.

[0027] On the contrary, if the user does not spend a lot of time focused on the same course, it indicates that the user is not eager to master the knowledge of a certain course at present and may have some short-term interests. At this time, the user's interests can be taken as the dominant factor, and the second recommendation strategy can be adopted to recommend other types or fields of courses that the target user may be interested in, so as to broaden the user's learning and browsing scope, avoid "aesthetic fatigue", and relieve the user's boredom.

[0028] Exemplarily, in this embodiment, a course recommendation method for two paths (recommendation strategies) is given: 1) The first recommendation strategy: The target user interaction information includes historical test data that can reflect the learning progress of the target user; the first recommendation strategy includes: A1. Determine whether there are weak knowledge points of the target user in the corresponding course according to the historical test data of the target user; A2. Based on the course knowledge graph, determine whether there are prerequisite dependent knowledge points for this weak knowledge point; A3. If so, use the course containing the prerequisite dependent knowledge point as the candidate recommended course for the target user.

[0029] For example: User U is learning course C1 (including knowledge points K1, K2), and the relationship of the course knowledge graph is: U - learning - C1, C1 - contains - K1, C1 - contains - K2.

[0030] Based on the historical test data, it is detected that user U has not mastered K2 (high answering error rate), which is marked as a weak item.

[0031] According to the course knowledge graph, it is found that K2 depends on the prerequisite knowledge point K0, and course C2 contains K0.

[0032] At this time, C2 can be considered for priority recommendation to user U to help user U quickly learn knowledge point K1.

[0033] Furthermore, it is also possible to track the user's learning situation in real time and obtain user feedback. For example, when user U has learned C2, the updated relationship is: U - Mastered - K0, and the recommended list is recalculated.

[0034] 2) The second recommendation strategy: The target user interaction information includes the first historical behavior data that can reflect the preferences of the target user, and the other user interaction information includes the second historical behavior data that can reflect the preferences of other users. The second recommendation strategy includes: B1. Based on the first historical behavior data and the second historical behavior data, find a group of similar users with the same preferences as the target user from other user groups; B2. Select the courses that the target user has not browsed from the courses preferred by the group of similar users to form a course set; B3. Based on the course knowledge graph and the first historical behavior data, select the courses that are close to the preferences of the target user from the course set to form a group of candidate recommended courses, and the remaining courses in the course set form a second group of candidate recommended courses; among them, the recommended priority of the first group of candidate recommended courses is higher than that of the second group of candidate recommended courses.

[0035] In this method, by finding a group of similar users with the same preferences as the target user, it is possible to quickly screen out the courses that the target user has not browsed but may be interested in from a large amount of course resources, and ensure the accuracy of course recommendations in the case where the personal user behavior traces (such as click, browse, purchase records, etc.) are relatively sparse.

[0036] In addition, although there is a similarity between the group of similar users and the target user, there must be courses in the courses preferred by the group of similar users that are quite different from the hobbies of the target user. Therefore, to further improve the accuracy of course recommendations, in this embodiment, the courses in the course set obtained in step B2 are divided into a first group of candidate recommended courses and a second group of candidate recommended courses, and the recommended priority of the first group of candidate recommended courses is higher than that of the second group of candidate recommended courses.

[0037] S5. Calculate the interest values of the target user for each candidate recommended course respectively, and recommend the top N courses with the highest interest values to the target user.

[0038] Exemplarily, taking the course recommendation method of the above second path as an example, the calculation of the interest values of the target user for each candidate recommended course includes: Obtain the target candidate recommended course from the course knowledge graph; obtain the target user from the first historical behavior data average score for the courses in this field. If there is no relevant record, a preset default constant value is used; calculate the similarity between the target user and other users . ; Calculate the interest values of the target user for each candidate recommended course; the interest values are calculated according to the following expression: (1); In the above formula (1), represents the set of user groups similar to the target user in terms of preferences, represents other users 's rating for the target candidate recommended course , in points, represents the average rating of other users for the courses in the field to which the target candidate recommended course belongs, in points.

[0039] The above data of the target user can be obtained from the target user interaction information, and the data of other users can be obtained from other user interaction information. However, since the implementer of the course recommendation algorithm is usually the (target) platform, the platform staff can retrieve it from the background database by themselves.

[0040] By quantifying and predicting the interest of the target user in each candidate recommended course, the top N courses with the highest interest values (this value can be consistent with the number of course display positions on the login page) are recommended to the target user, and combined with the feedback information of the user (which is part of the user interaction information), a closed loop can be formed, which helps to optimize the recommendation model and improve the accuracy of course recommendation. Example 2

[0041] In this example, the method further includes: generating training samples based on user interaction information and a course knowledge graph, selecting a KGAT model based on an attention mechanism, and training the KGAT model based on the attention mechanism using the training samples to obtain a course personalized recommendation model for performing steps S4 - S5.

[0042] Specifically, the KGAT model based on the attention mechanism includes an embedding layer, an attention - embedded propagation layer, and a prediction layer; The embedding layer is used to initialize the vectors of users, courses, entities, and relationships; specifically including: 1) Feature initialization: Mapping users, courses, entities, and relationships in the course knowledge graph into low - dimensional dense vectors (embeddings). These embedding vectors serve as the basic inputs of the model for subsequent complex relationship modeling.

[0043] The embeddings of users and items are usually initialized by IDs (such as user ID, item ID); entities (such as "course", "teacher") and relationships (such as "belongs to", "release year") in the knowledge graph are also embedded into the vector space respectively. 2) Relationship modeling: Through the embedding of the knowledge graph (such as TransR, TransH, etc.), the semantic associations between entities are initially captured (for example, the potential connection between "User A likes Course B" and "Course B belongs to Field C"). The embedding layer provides the initial semantic representation for the model and is the basis for subsequent high-order relationship modeling.

[0044] The attention embedding propagation layer is used to update the embeddings of users and courses and determine candidate recommended courses through multi-hop neighbor aggregation; specifically including: 1) Information propagation: Aggregate the information of neighbor nodes layer by layer through the knowledge graph structure, and propagate the high-order relationships in the user-course interaction and the course knowledge graph (such as the multi-hop path of user → course → field → other courses). 2) Attention mechanism: Dynamically calculate the importance weights of neighbor nodes to distinguish the contributions of different neighbors to the current node (for example, users pay more attention to entities related to their current interests). During the aggregation process, the embedding of each node will be updated to the weighted sum of the neighbor node embeddings, and the weights are determined by the attention scores. 3) High-order relationship modeling: By stacking multiple attention propagation layers (such as 2-3 layers), the model can capture the semantic information of multi-hop paths (such as user → course → teacher → other courses). The attention embedding propagation layer enhances the model's understanding of the local and global graph structures, making the recommendation results more interpretable (for example, explaining the recommendation reasons by visualizing the attention weights).

[0045] The prediction layer is used to calculate the interest value of the user for the candidate recommended courses based on the updated embeddings and generate a course recommendation list containing N courses. Specifically including interaction prediction: Based on the final embedding representations of users and items, calculate the preference scores of users for items. Common methods include: Inner Product: Directly calculate the dot product of the user and item embeddings; Neural Network (such as MLP): Learn more complex interaction patterns through non-linear transformation. Knowledge enhancement: Combine the relationship path information in the knowledge graph (such as user → course → teacher) to improve the prediction accuracy.

[0046] Using this model, it is possible to simultaneously utilize explicit interaction data (such as user-course clicks) and implicit knowledge graph information (entity relationship paths), significantly improving the accuracy and interpretability of recommendations.

[0047] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0048] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0049] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that realize the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0050] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0051] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0052] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0053] Computer readable media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0054] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0055] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A personalized recommendation method for course services based on a knowledge graph, characterized in that, The method includes: S1. Obtain course information and user interaction information on the target platform; the user interaction information includes target user interaction information and other user interaction information; S2. Preprocess the course information; S3. Define entities and relationships, extract entities and relationships from the preprocessed course information to form several triple data; construct a course knowledge graph based on the information relevance of each triple data; S4. Judge whether the target user is in a state of concentrated learning based on the target user interaction information; if so, determine candidate recommended courses for the target user using the first recommendation strategy based on the course knowledge graph, otherwise determine candidate recommended courses for the target user using the second recommendation strategy based on the course knowledge graph; S5. Calculate the interest values of the target user for each candidate recommended course respectively, and recommend the top N courses with the highest interest values to the target user.

2. The personalized recommendation method for course services based on a knowledge graph according to claim 1, wherein The preprocessing of the course information includes: removing illegal data in the course information, checking for duplicate course information, and verifying the accuracy of the course information.

3. The personalized recommendation method for course services based on a knowledge graph according to claim 2, characterized in that The definition of entities and relationships includes: defining courses, chapters, test questions, knowledge points, the number of course visits, teachers, fields, institutions, and course release times in the course information as entities; defining the relationships between entities as the chapters included in the course, the knowledge points included in the chapters, the test questions included in the course, the knowledge points tested by the test questions, the value of the number of course visits, the field to which the course belongs, the release year, the teacher to whom the course belongs, and the institution to which the course belongs in the course information.

4. The personalized recommendation method for course services based on a knowledge graph according to claim 3, wherein The target user interaction information includes historical test data that can reflect the learning progress of the target user; the first recommendation strategy includes: Judging whether there are weak knowledge points of the target user in the corresponding course based on the historical test data of the target user; Judging whether there are prerequisite dependent knowledge points for the weak knowledge point based on the course knowledge graph; If so, use the course containing the prerequisite dependent knowledge point as the candidate recommended course for the target user.

5. The personalized recommendation method for curriculum services based on a knowledge graph according to claim 3, wherein The target user interaction information includes first historical behavior data that can reflect the preferences of the target user, and the other user interaction information includes second historical behavior data that can reflect the preferences of other users. The second recommendation strategy includes: Based on the first historical behavior data and the second historical behavior data, find a group of users similar to the target user's preferences from other user groups; Select courses that the target user has not browsed from the courses preferred by the group of users similar to the target user to form a course set; Based on the course knowledge graph and the first historical behavior data, select courses from the course set that are close to the target user's preferences to form a group of candidate recommended courses, and the remaining courses in the course set form a second group of candidate recommended courses; Among them, the recommendation priority of the first group of candidate recommended courses is higher than that of the second group of candidate recommended courses.

6. The personalized recommendation method for curriculum services based on a knowledge graph according to claim 5, wherein The calculation of the interest values of the target user for each candidate recommended course includes: Obtaining target candidate recommended courses based on a curriculum knowledge graph Field of belonging; Obtain the target user from the first historical behavior data The average score of the courses in this field , if there is no relevant record, then use a preset default constant value; Calculate the target user and other users similarity ; Calculate the interest value of the target user for each candidate recommended course; the calculation expression of the interest value is as follows: （1） In the above formula (1), represents the set of user groups similar to the target user in terms of preferences, represents other users 's ratings for the target candidate recommended course , represents the average rating of courses within the field to which the target candidate recommended course belongs by other users. ​ 7. The personalized recommendation method for course services based on a knowledge graph according to claim 1, wherein The target user interaction information includes: the total time spent by the target user on the target platform within a specified time period and the learning time spent on each course; the judgment of whether the target user is in a state of concentrated learning based on the target user interaction information includes: Calculate the proportion of the learning time of the target user in each course compared to the total time respectively; Determine whether there is a proportion exceeding the proportion threshold; If there is, determine that the target user is in a state of concentrated learning, otherwise determine that the target user is not in a state of concentrated learning.

8. The personalized recommendation method for course services based on a knowledge graph according to claim 1, characterized in that The method further includes: generating training samples based on user interaction information and a course knowledge graph, selecting a KGAT model based on an attention mechanism, and training the KGAT model based on the attention mechanism using the training samples to obtain a course personalized recommendation model for performing steps S4 to S5.

9. The personalized recommendation method for course services based on a knowledge graph according to claim 8, wherein, The KGAT model based on the attention mechanism includes an embedding layer, an attention embedding propagation layer, and a prediction layer; the embedding layer is used to initialize the vectors of users, courses, entities, and relationships; The attention embedding propagation layer is used to update the embeddings of users and courses and determine candidate recommended courses through multi-hop neighbor aggregation; the prediction layer is used to calculate the interest value of the user for the candidate recommended courses based on the updated embeddings and generate a course recommendation list containing N courses.

Images