KR102557827B1 - System and method for recommending related courses based on graph data and recording medium for performing the same - Google Patents

Abstract

The present invention relates to a graph data-based related subject recommendation system and method, and a recording medium for performing the same.
A system for recommending related subjects based on graph data according to the present invention includes a pre-processing unit for generating a course registration data set by pre-processing course registration history data during a set period; a data construction unit that builds network data between subjects connected in a precedent relationship based on a point in time when one class is enrolled by using the generated course registration data set; an embedding unit that embeds the constructed inter-subject network data into an academic area dimension using an inductive learning-based Graph SAGE algorithm; and a subject recommendation unit for recommending subjects similar to the subject input from the user by using the characteristics of the node for each subject according to the embedding result.

Description

Graph data-based related subject recommendation system and method, and recording medium for performing it

The present invention relates to a graph data-based related subject recommendation system and method, and a recording medium for performing the same, and more particularly, a graph data-based related subject recommendation system and method for recommending related subjects to students using graph data. , it relates to a recording medium for performing this.

Recently, many experts predict that artificial intelligence (AI) technology will bring about revolutionary changes in our society. In particular, due to COVID-19, a large amount of data is accumulating as digitalization in each field of society is rapidly progressing, and the introduction and spread of AI technology is accelerating.

In the field of education, there are active discussions on the introduction of AI, and recently, as social demand for student-tailored education services has increased, the importance of personalized recommendations using AI technology is increasing.

However, in the conventional education-related AI technology, insight is not reached in analyzing unstructured data such as keywords of research papers, syllabuses and learning objectives of liberal arts lectures through AI, and similarity between subjects using such unstructured data. A method of deriving or providing it visually is not disclosed at all.

In general, a certain pattern appears in the behavior of students taking one course and taking the next course due to the department's curriculum or the continuity of their interests. This relationship between subjects can be used as useful information in the process of students selecting subjects and developing their interests academically. Until now, however, students' access to this information has been limited.

Therefore, it is necessary to develop a system that provides information on subjects related to subjects of interest and recommends similar subjects from other departments.

The background technology of the present invention is disclosed in Republic of Korea Patent Publication No. 10-1754723 (2017. 07. 06. notice).

A technical problem to be achieved by the present invention is to provide a graph data-based related subject recommendation system and method for recommending related subjects to students using graph data, and a recording medium for performing the same.

A system for recommending related subjects based on graph data according to an embodiment of the present invention to achieve this technical problem includes a pre-processing unit for generating a course registration data set by pre-processing course registration history data during a set period; a data construction unit that builds network data between subjects connected in a precedent relationship based on a point in time when one class is enrolled by using the generated course registration data set; an embedding unit that embeds the constructed inter-subject network data into an academic area dimension using an inductive learning-based Graph SAGE algorithm; and a subject recommendation unit for recommending subjects similar to the subject input from the user by using the characteristics of the node for each subject according to the embedding result.

At this time, the data construction unit may construct network data between subjects by calculating relation data between subjects through subjects connected in a precedent relationship based on the time point of one lecture with reference to the course registration data set.

In addition, the embedding unit randomly extracts neighboring nodes for calculating the feature vector of the target node, selects a neighboring node directly connected to the target node and n connections of the neighboring node to the neighboring node, and selects a characteristic vector of the selected neighboring node. It is possible to perform the node representation learning of the Graph Sage Algorithm in the order of updating the characteristics of the target node by integrating and embedding the characteristics of the node for each subject into a new dimension by performing the expression learning of the target node using an inductive learning method.

In addition, by using the network data between the built courses, each time a student takes a prerequisite course and then takes a later course, the connection between the prerequisite course and the later course is increased, so that the first and subsequent courses are taken simultaneously. a prior/post enrollment conditional probability calculation unit that calculates prior/post enrollment conditional probability data through a value obtained by dividing the frequency by the frequency of taking prerequisite courses; a concurrent course enrollment network building unit that constructs concurrent course enrollment network data in which two courses taken simultaneously in one semester are connected using the generated course registration data set; and a placement table generating unit generating a placement table for providing a list of previous, concurrent, and subsequent courses for a specific subject using the calculated prior/post-enrollment conditional probability data.

In addition, when a specific subject is taken, the arrangement table generating unit calculates a list of previous courses in the order of highest probability of taking the previous course, and calculates a list of concurrent courses in order of highest probability of taking the same course. , it is possible to calculate and provide a list of courses to be taken in the order of the most likely courses to be taken thereafter.

In addition, a method for recommending related subjects based on graph data according to another embodiment of the present invention includes generating a course registration data set by pre-processing course registration history data during a set period; constructing network data between subjects connected in a precedence relationship based on a point in time of attending one lecture by using the generated course registration data set; embedding the constructed inter-subject network data in the dimension of an academic domain using an inductive learning-based Graph SAGE algorithm; and recommending subjects similar to the subject input from the user by using characteristics of nodes for each subject according to the embedding result.

At this time, the step of constructing the network data between subjects is to construct the network data between subjects by calculating the relationship data between subjects through the subjects connected in a precedent relationship based on the time of taking one lecture with reference to the course registration data set. can

In addition, the step of embedding randomly extracts neighboring nodes for calculating the feature vector of the target node, selects a neighboring node directly connected to the target node and n connections of the neighboring node to the neighboring node, and selects the selected neighboring node. The node representation learning of the Graph Sage Algorithm can be performed in the order of updating the characteristics of the target node by integrating the feature vectors, performing the expression learning of the target node using an inductive learning method, and embedding the characteristics of each subject node into a new dimension. .

In addition, after the step of constructing the network data, each time a student takes a prerequisite course and then takes a subsequent course by using the network data between the constructed courses, the connection between the prerequisite course and the subsequent course is established. calculating conditional probability data of prior and subsequent enrollment through a value obtained by dividing the frequency of taking prerequisite courses by the frequency of concurrent enrollment of prior and subsequent courses; constructing concurrent course enrollment network data in which two courses taken simultaneously in one semester are connected using the generated course registration data set; and generating a placement table for providing a list of previous, concurrent, and subsequent courses for a specific subject by using the calculated prior/post-enrollment conditional probability data.

In addition, in the step of generating the arrangement table, when a specific course is taken, a list of previous courses taken in the order of the highest probability of taking the previous course is calculated, and a list of concurrent courses taken in order of the highest probability of taking the same course. It is possible to calculate and provide a list of subsequent courses in the order of the highest probability to be taken later.

In addition, a computer program for performing the graph data-based related subject recommendation method may be recorded in a computer-readable recording medium according to another embodiment of the present invention.

As described above, according to the present invention, by recommending related subjects to students using graph data, it is possible to recommend a curriculum considering actual students' course attendance patterns, thereby improving student satisfaction.

In addition, according to the present invention, there is an advantage in enabling more accurate subject embedding and recommendation even in a situation where text data to be analyzed is relatively insufficient by considering related subjects together compared to the existing method using only text data.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a system configuration diagram illustrating a system for recommending related subjects based on graph data according to an embodiment of the present invention.
2 is a flowchart illustrating an operation flow of a method for recommending related subjects based on graph data according to an embodiment of the present invention.
3 is a diagram exemplarily illustrating a curriculum recommendation process for each department according to an embodiment of the present invention.
FIG. 4 is a flowchart illustrating a detailed operation flow for step S60 of FIG. 2 .
FIG. 5 is a diagram for explaining the process of FIG. 4 .
6 is a diagram illustratively illustrating a result of node embedding for each subject according to an embodiment of the present invention compared with the conventional one.
7 is a diagram exemplarily illustrating results of recommending related subjects according to an embodiment of the present invention.
8 is a diagram illustratively illustrating similar subject recommendation results from other departments according to an embodiment of the present invention.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings. In this process, the thickness of lines or the size of components shown in the drawings may be exaggerated for clarity and convenience of description.

In addition, terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of a user or operator. Therefore, definitions of these terms will have to be made based on the content throughout this specification.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

First, a related subject recommendation system based on graph data according to an embodiment of the present invention will be described with reference to FIG. 1 .

1 is a system configuration diagram illustrating a system for recommending related subjects based on graph data according to an embodiment of the present invention.

Referring to FIG. 1, the graph data-based related subject recommendation system 3 communicates with a user terminal 1 possessed by a student, an educational institution server 2, or an academic information providing server 4 through a wired or wireless network. It can be configured to operate while A communication method through a wired or wireless network may include all communication methods that can be networked between objects and objects, and are not limited to wired communication, wireless communication, 3G, 4G, or other methods.

For example, a wired or wireless network may include Local Area Network (LAN), Metropolitan Area Network (MAN), Global System for Mobile Network (GSM), Enhanced Data GSM Environment (EDGE), High Speed Downlink Packet Access (HSDPA), W -CDMA (Wideband Code Division Multiple Access), CDMA (Code Division Multiple Access), TDMA (Time Division Multiple Access), Bluetooth, Zigbee, Wi-Fi, Voice over Internet (VoIP) Protocol), LTE Advanced, IEEE802.16m, WirelessMAN-Advanced, HSPA+, 3GPP Long Term Evolution (LTE), Mobile WiMAX (IEEE 802.16e), UMB (formerly EV-DO Rev. C), Flash-OFDM, iBurst and MBWA (IEEE 802.20) systems, HIPERMAN, Beam-Division Multiple Access (BDMA), Wi-MAX (World Interoperability for Microwave Access), and ultrasound communication may refer to a communication network by one or more communication methods selected from the group consisting of , but is not limited thereto.

In an embodiment of the present invention, the graph data-based related subject recommendation system 3 includes a pre-processing unit 31, a data construction unit 32, a first-post enrollment conditional probability calculation unit 33, and a concurrent course enrollment network construction unit 34 , a placement table generator 35, an embedding unit 36, an input unit 37, and a subject recommendation unit 38.

The systems, devices and servers described herein may be entirely hardware, or may have aspects that are part hardware and part software. For example, the systems, devices, and servers of the present specification, and each unit included therein, may collectively refer to hardware and software related thereto for processing or exchanging data in a specific form and content or by electronic communication. In this specification, terms such as "unit", "module", "device", "terminal", "server" or "system" are intended to refer to a combination of hardware and software driven by the hardware. For example, the hardware may be a data processing device including a CPU or other processor. Also, software driven by hardware may refer to a running process, an object, an executable file, a thread of execution, a program, and the like.

In addition, each element constituting the related subject recommendation system 3 according to the present embodiment is not necessarily intended to refer to a separate device that is physically separated from each other. That is, the pre-processing unit 31 of FIG. 1, the data construction unit 32, the prior-post enrollment conditional probability calculation unit 33, the concurrent course enrollment network establishment unit 34, the placement table generation unit 35, and the embedding unit 36 , The input unit 37 and the subject recommendation unit 38 functionally divide the hardware constituting the related subject recommendation system 3 according to the operation performed by the corresponding hardware, and each unit is provided independently of each other. It's not what it should be. Of course, depending on the embodiment, one or more of the parts of the related subject recommendation system 3 may be implemented as a separate device that is physically separated from each other.

The related subject recommendation system 3 may recommend subjects similar to the input subject or similar subjects in other departments to the user who inputs the subject. The related subject recommendation system 3 communicates with an application (or app) running on the user terminal 1 so that the user can input and transmit a subject and receive corresponding information, thereby performing the function of the application. It can perform the function of an application service server that enables Alternatively, in another embodiment, the related subject recommendation system 3 may be implemented in the form of a web server providing a web page accessible by a web browser running on the user terminal 1 .

In the embodiment shown in FIG. 1, the user terminal 1 is shown in the form of a notebook computer. However, this is just an example, and the type of user terminal 1 is not limited to that shown in the drawings. For example, a user who wants to be recommended for a major may use a mobile communication terminal such as a smartphone, a personal computer, a notebook computer, a personal digital assistant (PDA), a tablet, and an Internet Protocol Television (IPTV) The function provided by the related subject recommendation system 3 may be used by using an arbitrary computing device such as a set-top box.

The pre-processing unit 31 creates a course registration data set by pre-processing the course registration history data during the set period.

In detail, the pre-processing unit 31 extracts a number of course registration history data generated by a number of students for the past N years in order to construct graph data between subjects. Among these, only courses with 2 or more credits are maintained among the remaining courses, excluding course history data related to courses that are not suitable for the course recommendation service or that require responses. In addition, it is desirable to reflect only the number of first-time courses per student in the analysis and exclude re-takes from the analysis. Through this, a course registration data set for final analysis is created.

The data builder 32 builds network data between subjects connected in a precedence relationship based on the time point of one lecture by using the course registration data set generated through the preprocessor 31.

At this time, the data construction unit 32 references the course registration data set and constructs network data between subjects by calculating relation data between subjects through subjects connected in a precedent relationship based on the point in time of one lecture.

In detail, the data construction unit 32 refers to the course registration data set preprocessed through the preprocessing unit 31 and builds network data between subjects connected in a precedent relationship based on the time of class enrollment.

For example, if a student took a course called "Basics of Psychology I" in the first semester of the freshman year and then took a course called "Personality Psychology" in the second semester of the freshman year, then "Basics of Psychology I" and "Personality Psychology" would be Establish network data between subjects by linking each other with each other in the relationship between prerequisite courses and post course courses. At this time, links are not created between courses applied for in the same semester.

The prerequisite course enrollment conditional probability calculation unit 33 uses the inter-course network data constructed in the data construction unit 32 to determine whether or not a student takes a prerequisite course and then takes a subsequent course whenever a student takes a prerequisite course. By increasing the connection of courses taken, the conditional probability data of prior and subsequent enrollment is calculated by dividing the frequency of taking prerequisite courses from the frequency of concurrently taking prior and subsequent courses.

In detail, the conditional probability calculation unit 33 of taking a prerequisite course (target), when a prerequisite course (source) is taken based on the network data between subjects constructed in the data building unit 32, is a conditional probability of taking a later course (target) Calculate (p_source_target).

That is, each time a student takes a prerequisite course and then takes a later course, the connection (n_link) between the prerequisite course and the subsequent course is increased by one time. Then, the conditional probability of prior and subsequent enrollment is calculated by dividing the frequency of taking prerequisite courses by the frequency of concurrent enrollment of prior and subsequent courses. Among them, it is desirable to remove the first and second subject pairs whose number of connections is less than n times. The graph data between subjects calculated in this way are sorted in the order of high conditional probability to calculate the conditional probability data of the first and second classes.

For example, assuming that the prerequisite course is A and the subsequent course is B, the conditional probability data for prior and subsequent enrollment is calculated as (the number of courses B taken after taking A) / (the total number of other courses taken after taking A) yield

The concurrent course enrollment network construction unit 34 uses the course registration data set generated through the preprocessing unit 31 to construct concurrent course enrollment network data in which two courses taken simultaneously in one semester are connected.

That is, the simultaneous course enrollment network construction unit 34 calculates concurrent course graph data in a manner similar to the method of calculating the first and subsequent course networks. In this case, the two courses included in the simultaneous enrollment network means that students take the corresponding courses simultaneously in one semester.

The placement table generator 35 creates a placement table for providing a list of previous, concurrent, and subsequent courses for a specific subject using the prior/post-enrollment conditional probability data calculated by the prior/post-enrollment conditional probability calculation unit 33.

At this time, when a specific subject is taken, the arrangement table generator 35 calculates a list of previous courses taken in the order of highest probability of taking a specific course, and lists concurrent courses in order of highest probability of taking the same course. is calculated, and a list of subsequent courses is calculated and provided in the order of the most likely courses to be taken later.

The embedding unit 36 embeds the inter-subject network data constructed in the data construction unit 32 at the level of the academic domain by using the Graph SAGE algorithm of the inductive learning method.

At this time, the embedding unit 36 randomly extracts neighboring nodes for calculating the feature vector of the target node, selects a neighboring node directly connected to the target node and n connections of the neighboring node to the neighboring node, and selects the selected neighboring node. The node representation learning of the Graph Sage Algorithm can be performed in the order of updating the characteristics of the target node by integrating the feature vectors of , performing expression learning of the target node using an inductive learning method, and embedding the characteristics of each subject node into a new dimension. there is.

That is, the related subject recommendation system 3 according to an embodiment of the present invention uses a representation learning technique capable of generating a vector representation of each subject (node) by reflecting not only subject characteristics but also relationship data between subjects. used In particular, by using the graph sage algorithm, efficient learning is possible even when there is a lot of node characteristic information.

The subject recommendation unit 38 recommends subjects similar to the subject input from the user through the input unit 370 by using the characteristics of nodes for each subject according to the embedding result of the embedding unit 36 .

At this time, the subject recommendation unit 38 may recommend a subject having the highest cosine similarity according to Equation 1 based on the subject A and the feature vector B selected by the student.

Hereinafter, a method for recommending related subjects based on graph data according to an embodiment of the present invention will be described with reference to FIGS. 2 to 8 .

FIG. 2 is a flow chart illustrating an operation flow of a method for recommending related subjects based on graph data according to an embodiment of the present invention. Referring to this flow chart, specific operations of the present invention will be described.

According to an embodiment of the present invention, first, the pre-processing unit 31 pre-processes course registration history data for a set period to create a course registration data set (S10).

At this time, in step S10, students extract a number of course registration history data generated by a number of students over the past N years to construct graph data between subjects.

In addition, the data construction unit 32 builds network data between subjects connected in a precedent relationship based on the time point of one lecture by using the course registration data set generated in step S10 (S20).

At this time, in step S20, with reference to the course registration data set in step S10, data on the relationship between subjects is calculated through subjects connected in a sequential relationship based on the point in time of taking one lecture, and network data between subjects is constructed.

In addition, the prerequisite course enrollment conditional probability calculation unit 33 uses the inter-course network data constructed in step S20 to determine the number of prerequisite courses and subsequent courses each time a student takes a prerequisite course and then takes a subsequent course. By increasing the connection, conditional probability data for prior and subsequent enrollment is calculated through a value obtained by dividing the frequency of taking prerequisite courses from the frequency of concurrent enrollment of prior and subsequent courses (S30).

In addition, the concurrent course enrollment network construction unit 34 constructs concurrent course enrollment network data in which two courses simultaneously taken in one semester are connected using the course registration data set generated in step S10 (S40).

Then, the placement table generator 35 creates a placement table for providing a list of previous, concurrent, and subsequent courses for a specific subject using the conditional probability data calculated in step S30 (S50).

That is, in step S50, when a specific course is taken, a list of previous courses taken in the order of the highest probability of taking the previous course is calculated, and a list of concurrent courses is calculated in order of the highest probability of taking the same course. A list of subsequent courses is calculated and provided in the order of the most likely courses to be taken.

In addition, in step S50, the placement table generator 35 constructs previous, concurrent, and subsequent course enrollment graph data between subjects by using the course registration data set generated in step S10 and the probability data calculated in step S30. At this time, it is possible to generate a placement table in which up to 21 subjects with high probability of taking previous, concurrent, and subsequent courses are calculated for each subject.

3 is a diagram exemplarily illustrating a curriculum recommendation process for each department according to an embodiment of the present invention.

As shown in FIG. 3 , the placement table generator 35 may recommend a curriculum for each department using the generated placement table. This information can be used to search for advanced courses to be taken after taking a specific course according to students' interests, or to search for courses to be taken before taking the advanced course they want to take.

In addition, the embedding unit 36 embeds the inter-subject network data constructed in step S20 into the dimension of the academic domain using the Graph SAGE algorithm of an inductive learning method (S60).

At this time, step S60 randomly extracts neighboring nodes for calculating the characteristic vector of the target node, selects a neighboring node directly connected to the target node and n connections of the neighboring node to the neighboring node, and selects the characteristic vector of the selected neighboring node. , update the characteristics of the target node, and perform node representation learning of the Graph Sage Algorithm in the order of embedding the characteristics of each subject node into a new dimension by performing the expression learning of the target node using an inductive learning method.

That is, the related subject recommendation system 3 according to an embodiment of the present invention uses a representation learning technique capable of generating a vector representation of each subject (node) by reflecting not only subject characteristics but also relationship data between subjects. used In particular, by using the graph sage algorithm, efficient learning is possible even when there is a lot of node characteristic information.

In detail, the graph sage algorithm learns the representations of nodes in an heuristic way. That is, even if a new subject is added, it is not necessary to recalculate the expressions of all subjects included in the network between all subjects.

That is, when a new subject is added, information on the precedence relationship cannot be obtained from the existing student's course history data, so a representation of the subject cannot be generated using a transformative learning method. Therefore, in an embodiment of the present invention, it is preferable to apply a graph sage algorithm of an inductive learning method capable of generating an expression for a newly established subject.

At this time, since the graph sage algorithm uses a sampling technique, efficient expression learning is possible even when the feature vector of a node is large. To implement the graph sage algorithm, you can use the graph sage functions of the Stellargraph framework.

In addition, subject node expression learning is performed in the following way.

Even when the feature vector of the node is large, the lecture content for each subject of the network between subjects is used to generate the subject expression by using the strength of the Graph Sage algorithm, which can efficiently learn the expression.

In addition, to vectorize the contents of lectures by subject, we use the academic area embedding method, which is a natural language analysis technique developed by ourselves. Through the academic area embedding method, the lecture contents of all subjects are expressed as 152-dimensional academic area characteristics.

Each dimension is normalized using the maximum-minimum method to perform node expression learning for each subject that will follow.

In an embodiment of the present invention, the graph data of the prior relationship between courses has a network structure with directionality between prerequisite courses and subsequent courses. It is desirable to use a direct graph sage algorithm to reflect this relationship in expression learning. Learning the expression of a node using graph data is a type of transformative learning, but in the graph sage algorithm according to an embodiment of the present invention, it is performed like inductive learning by using a sampling technique.

Specifically, n neighboring nodes and non-neighboring nodes that are actually connected to the target node to learn the expression are randomly selected. The algorithm learns through a process of predicting whether a corresponding node and a randomly selected node are connected to each other by using the feature vector of the input node.

At this time, in order to utilize both the characteristics of the surrounding nodes and the information that can be obtained from the graph structure, the characteristics of the surrounding nodes are used together to calculate the characteristic vector of the target node.

FIG. 4 is a flowchart illustrating a detailed operation flow for step S60 in FIG. 2, and FIG. 5 is a diagram for explaining the process of FIG.

As in FIGS. 4 and 5, first, the embedding unit 36 randomly selects neighboring nodes to be used to calculate the feature vector of the target node (S61).

Next, the embedding unit 360 selects a neighbor node directly connected to the target node and n connections from the neighbor node to the neighbor node, integrates the feature vectors of the selected neighbor nodes, and updates the target node's characteristics to update the target node's characteristics. A process of integrating node information is performed (S62).

In detail, the graph sage algorithm makes the calculation process efficient even if the size of the network is large by making the sampling process random. At this time, it is possible to select up to how many connections to select and how many connections to select using hyperparameters. In the embodiment of the present invention, five connections are extracted for each node, but this is only an example for convenience of explanation, and it is not desirable to limit the number of extractions.

Select 5 connections from a neighbor node directly connected to the target node to a neighbor node of that neighbor node. At this time, the number of nodes that can be selected for each step can be directly set as a parameter. 10 neighboring nodes in the first step and 50 neighboring nodes in the second step are selected, and when the neighboring nodes are selected, the properties of the target node are updated by integrating the feature vectors of the selected neighboring nodes. At this time, there are Mean, Max, Attentional, etc. as a method of integrating the information of neighboring nodes. In the embodiment of the present invention, the attentional method, which has been shown to have the best performance in previous studies, may be used.

Finally, the embedding unit 360 performs a node expression generation process for performing expression learning of a target node in an inductive learning method (S63).

In detail, while learning the prediction task described above, the expression of the target note is learned. 90% of all nodes are classified as training data and 10% as training data, and the progress of learning can be checked using a binary classification accuracy indicator. The edge weight is set as the conditional probability of enrollment between subjects, the batch is 100, and the epoch is repeated 3 times. Through this process, the characteristics of the subject node are embedded into a new 50-dimensional space.

When a subject is input from the user through the input unit 37 (S70), the subject recommendation unit 38 recommends similar subjects to the subject input by the user using the characteristics of the node for each subject according to the embedding result of step S60 ( S80).

6 is a diagram illustratively illustrating a result of node embedding for each subject according to an embodiment of the present invention compared with the conventional one.

FIG. 6(a) is an example showing a conventional embedding result, and FIG. 6(b) is an example showing a node embedding result for each subject according to an embodiment of the present invention.

That is, in an embodiment of the present invention, low-dimensional embedding of a specific subject is performed using a graph sage algorithm. At this time, not only the characteristics of the subject, but also the characteristics of the related subject and the relationship with the related subject are all considered.

To check the performance of these embeddings, qualitative evaluation based on domain knowledge for each major is required. In the embodiment of the present invention, a dimensionality reduction and visualization technique using UMAP was used to check approximate embedding performance. Cosine distance was used as a metric, n_neighbors was set to 100, and min_dist was set to 1 to prevent nodes from overlapping after visualization.

As can be seen by comparing the result using only the existing subject characteristics of FIG. 6 (a) and the result of dimension reduction and visualization of the embedding using the Graph Sage algorithm (b), it is clear that subject nodes are at a glance when using the Graph Sage algorithm. You can check.

7 is a diagram illustratively illustrating results of recommending related subjects according to an embodiment of the present invention.

7 is an example of a screen actually provided based on graph data between subjects. When a user selects a course, you can see that the courses with the highest probability of taking the course before, after, and concurrently are displayed in order.

8 is a diagram illustratively illustrating similar subject recommendation results from other departments according to an embodiment of the present invention.

As shown in FIG. 8 , subjects in other departments similar to subjects input by the user may be recommended.

Such a graph data-based related subject recommendation system and method may be implemented as an application or implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer readable recording medium may include program instructions, data files, data structures, etc. alone or in combination.

Program instructions recorded on the computer-readable recording medium may be those specially designed and configured for the present invention, or those known and usable to those skilled in the art of computer software.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like.

Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter or the like as well as machine language codes such as those produced by a compiler. The hardware device may be configured to act as one or more software modules to perform processing according to the present invention.

As described above, the graph data-based related subject recommendation system and method according to an embodiment of the present invention recommends related subjects to students using graph data, thereby recommending a curriculum that considers actual student attendance patterns. satisfaction can be improved.

In addition, according to the embodiment of the present invention, there is an advantage in enabling more accurate subject embedding and recommendation even in a situation where text data to be analyzed is relatively insufficient by considering related subjects together compared to the existing method using only text data.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. will be. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the claims below.

1: User terminal
3: Related subject recommendation system
31: pre-processing unit
32: data construction unit
33: Priority and subsequent enrollment conditional probability calculation unit
34: Concurrent course network construction department
35: batch table generation unit
36: embedding unit
37: input unit
38: subject recommendation department

Claims (11)

a pre-processing unit that pre-processes course registration history data during the set period to create a course registration data set;
a data construction unit that builds network data between subjects connected in a precedent relationship based on a point in time when one class is enrolled by using the generated course registration data set;
an embedding unit that embeds the constructed inter-subject network data into an academic area dimension using an inductive learning-based Graph SAGE algorithm; and
A subject recommendation unit recommending subjects similar to subjects input from a user by using characteristics of nodes for each subject according to the embedding result;
Using the network data between the built courses, each time a student takes a prerequisite course and then takes a later course, the connection between the prerequisite course and the later course is increased to increase the frequency of taking the first and subsequent courses at the same time. a prior/post-enrollment conditional probability calculation unit that calculates prior/post-enrollment conditional probability data through a value obtained by dividing the frequency of taking prerequisite courses;
a concurrent course enrollment network building unit that constructs concurrent course enrollment network data in which two courses taken simultaneously in one semester are connected using the generated course registration data set; and
The system for recommending related subjects based on graph data further comprising a placement table generating unit for generating a placement table for providing a list of previous, concurrent, and subsequent courses for a specific subject using the calculated prior and subsequent course enrollment conditional probability data. According to claim 1,
The data building unit,
A graph data-based related subject recommendation system that constructs network data between subjects by calculating relationship data between subjects through subjects connected in a precedent relationship based on the time of one course enrollment by referring to the course registration data set. According to claim 1,
The embedding part,
A neighbor node for calculating the feature vector of the target node is randomly extracted, a neighbor node directly connected to the target node and n connections from the neighbor node to the neighbor node are selected, and the feature vectors of the selected neighbor node are integrated to select the target node. A graph data-based related subject recommendation system that performs node expression learning of the Graph Sage Algorithm in the order of updating the characteristics of and embedding the characteristics of each subject node into a new dimension by performing the expression learning of the target node using an inductive learning method. delete According to claim 1,
The arrangement table creation unit,
If a specific course is taken, the previous course list is calculated in the order of the highest probability of taking the previous course, and the concurrent course list is calculated in the order of the highest probability of taking the same course. A graph data-based related subject recommendation system that calculates and provides a list of courses taken in the order of the highest subject. In the related subject recommendation method performed by the graph data-based related subject recommendation system,
generating a course registration data set by pre-processing course registration history data during a set period;
constructing network data between subjects connected in a precedence relationship based on a point in time of attending one lecture by using the generated course registration data set;
embedding the constructed inter-subject network data in the dimension of an academic domain using an inductive learning-based Graph SAGE algorithm; and
recommending a subject similar to the subject input from the user by using characteristics of nodes for each subject according to the embedding result;
After the step of building the network data,
Using the network data between the built courses, each time a student takes a prerequisite course and then takes a later course, the connection between the prerequisite course and the later course is increased to increase the frequency of taking the first and subsequent courses at the same time. calculating pre-enrollment conditional probability data through a value obtained by dividing the frequency of taking prerequisite courses;
constructing concurrent course enrollment network data in which two courses taken simultaneously in one semester are connected using the generated course registration data set; and
The related subject recommendation method further comprising generating a placement table for providing a list of previous, concurrent, and subsequent courses for a specific subject using the calculated prior and subsequent enrollment conditional probability data. According to claim 6,
The step of building the network data between subjects,
A method for recommending related subjects that constructs network data between subjects by calculating relationship data between subjects through subjects connected in a precedent relationship based on the time of enrollment of one course by referring to the course registration data set. According to claim 6,
In the embedding step,
A neighbor node for calculating the feature vector of the target node is randomly extracted, a neighbor node directly connected to the target node and n connections from the neighbor node to the neighbor node are selected, and the feature vectors of the selected neighbor node are integrated to select the target node. A related subject recommendation method that performs node expression learning of the Graph Sage Algorithm in the order of updating the characteristics of and embedding the characteristics of each subject node into a new dimension by performing the expression learning of the target node using an inductive learning method. delete According to claim 6,
The step of creating the arrangement table,
If a specific course is taken, the previous course list is calculated in the order of the highest probability of taking the previous course, and the concurrent course list is calculated in the order of the highest probability of taking the same course. A method of recommending related courses that calculates and provides a list of subsequent courses in the order of the highest. A computer-readable recording medium on which a computer program for performing the graph data-based related subject recommendation method according to claim 6 is recorded.

Images