KR102375755B1 - System and method for recommendation of courses based on course similarity and computer program for the same - Google Patents

Abstract

The present invention relates to a system for recommending a subject based on subject similarity, which comprises: a first data analysis unit for analysing text data associated to one or more subjects to generate lecture content information for each of the subjects; a second data analysis unit for analyzing attendance history data of a student for each of the subjects to generate student characteristic information on each of the subjects; a similarity determining unit for generating similarity information on each of the subjects using the lecture content information and the student characteristic information; and a recommended subject determining unit for generating subject recommendation information on a user on the basis of subject-of-interest information received from the user and the similarity information. Accordingly, when the system for recommending a subject is used, a subject can be recommended at high accuracy as compared to that of the conventional invention by determining the similarity between subjects in consideration of a lecture content level and student characteristics at the same time. Also, a wide variety of subjects meeting the interest of each student can be recommended.

Description

SYSTEM AND METHOD FOR RECOMMENDATION OF COURSES BASED ON COURSE SIMILARITY AND COMPUTER PROGRAM FOR THE SAME

Embodiments relate to a subject recommendation system and method based on subject similarity, and a computer program therefor. More specifically, embodiments relate to a subject recommendation model capable of determining a recommended subject based on subject similarity calculated by considering lecture content and student characteristics together and a technology for providing the same.

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

Discussion on the introduction of AI is also actively taking place in the education field, and the Korean government is supposing and promoting personalized education using AI technology as the development direction of education. In addition, as a related prior art, Korean Patent Application Laid-Open No. 10-2021-0026234 discloses an apparatus and method for providing an AI-based user-customized math education service.

However, in conventional education-related AI technologies, including those disclosed in Korean Patent Application Laid-Open No. 10-2021-0026234, there is no insight in analyzing unstructured data such as keywords of research papers, syllabus of liberal arts lectures, and learning goals through AI. Moreover, there is no disclosure of a method for deriving similarity between subjects using such unstructured data or providing it visually.

Laid-Open Patent Publication No. 10-2021-0026234

According to one aspect of the present invention, unstructured data can be used to derive similar subjects, the degree of similarity between subjects can be determined in consideration of the dimension of lecture content and student characteristics, and similar subjects can be visualized and provided in the form of subject guidance. It is possible to provide a subject recommendation system and method based on subject similarity and a computer program for the same.

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

According to an aspect of the present invention, a subject recommendation system based on subject similarity includes: a first data analyzer configured to generate lecture content information for each of the one or more subjects by analyzing text data related to one or more subjects; a second data analysis unit configured to generate student characteristic information for each of the one or more subjects by analyzing the attendance history data of the students for each subject; a similarity determining unit configured to generate similarity information for each subject using the lecture content information and the student characteristic information; and a recommended subject determining unit configured to generate subject recommendation information for the user based on the interest subject information and the similarity information received from the user.

The subject recommendation system according to an embodiment further includes a preprocessor configured to extract the text data from at least one of a syllabus of the subject, a learning goal of the subject, a curriculum for each department, and a published thesis keyword for each academic field.

In one embodiment, the first data analysis unit learns the lecture contents of the subject by word frequency-inverse document frequency-based analysis of the text data, and the lecture contents based on the learning result of the lecture contents and generate information.

In an embodiment, the second data analysis unit is configured to generate attendance rate data for each department based on the attendance history data of a plurality of students, and generate the student characteristic information using the attendance ratio data for each department of the students for each subject. more composed.

In an embodiment, the similarity determining unit is further configured to generate the similarity information for each subject by using a cosine similarity between matrices respectively corresponding to the lecture content information and the learner characteristic information.

In an embodiment, the interest subject information includes information on at least one of the user's attendance history and desired subjects. In this case, the recommended subject determination unit determines the preferred subject of the user using at least one of the course history and desired subject information, and determines the recommended subject based on the degree of similarity to the preferred subject to obtain the subject recommendation information. further configured to create

In an embodiment, the recommended subject determining unit provides the subject recommendation information to the user by using a subject map defined by a plurality of dimensions corresponding to the characteristics of each subject and in which subjects are located closer to each other as the degree of similarity between subjects increases. further configured to do so.

According to an aspect of the present invention, a method for recommending a subject based on subject similarity includes: generating, by a subject recommendation system, text data related to one or more subjects to generate lecture content information for each of the one or more subjects; generating, by the subject recommendation system, information on student characteristics for each of the one or more subjects by analyzing attendance history data of students for each subject; generating, by the subject recommendation system, similarity information for each subject using the lecture content information and the student characteristic information; and generating, by the subject recommendation system, subject recommendation information for the user based on the subject information of interest and the similarity information received from the user.

The subject recommendation method according to an embodiment further includes the step of extracting, by the subject recommendation system, the text data from at least one of a syllabus of the subject, a learning goal of the subject, a curriculum for each department, and a published thesis keyword for each academic field do.

In an embodiment, the generating of the lecture content information may include: learning, by the subject recommendation system, the lecture contents of the subject through word frequency-inverse document frequency-based analysis of the text data; and generating, by the subject recommendation system, the lecture content information based on a learning result of the lecture content.

In an embodiment, the generating of the student characteristic information may include: generating, by the subject recommendation system, data on the attendance rate for each department based on the attendance history data of a plurality of students; and generating, by the subject recommendation system, the student characteristic information by using the data on the attendance rate by department of the students for each subject.

In an embodiment, the generating of the similarity information includes calculating, by the subject recommendation system, a cosine similarity between matrices respectively corresponding to the lecture content information and the learner characteristic information.

In an embodiment, the interest subject information includes information on at least one of the user's attendance history and desired subjects. In this case, the generating of the course recommendation information may include: determining, by the course recommendation system, the user's preferred course using information on at least one of the course history and desired courses; and determining, by the subject recommendation system, a recommended subject based on the degree of similarity to the preferred subject.

In the subject recommendation method according to an embodiment, the subject recommendation system is defined by a plurality of dimensions corresponding to the characteristics of each subject, and the subject recommendation information using a subject map in which subjects are located closer to each other as the degree of similarity between subjects increases The method further includes providing to the user.

The computer program according to an aspect of the present invention is combined with hardware to execute the subject recommendation method based on subject similarity according to the above-described embodiments, and may be stored in a computer-readable recording medium.

According to the subject recommendation system and method based on subject similarity according to the embodiments, a model using only one of the prior art, for example, lecture content or student characteristics, by determining the similarity between subjects by considering the lecture content dimension and student characteristics at the same time Subjects can be recommended with much higher accuracy compared to the previous technology, and by recommending a wide variety of subjects that meet the interests of each student, there is an advantage that better results can be derived in terms of the scope and diversity of recommendations compared to the prior art.

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

1 is a schematic block diagram of a subject recommendation system based on subject similarity according to an exemplary embodiment.
2 is a flowchart illustrating each step of a method for recommending a subject based on subject similarity according to an exemplary embodiment.
3 is a conceptual diagram illustrating a process of generating lecture content information by a subject recommendation method based on subject similarity according to an exemplary embodiment.
4 is a conceptual diagram illustrating a process of generating student characteristic information by a subject recommendation method based on subject similarity according to an exemplary embodiment.
5 to 7 are graphs illustrating an effect of a subject recommendation method based on subject similarity according to an exemplary embodiment in comparison with the prior art.
8A and 8B are conceptual diagrams illustrating examples of subject guidance provided by a subject recommendation method based on subject similarity according to an exemplary embodiment.

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

1 is a schematic block diagram of a subject recommendation system based on subject similarity according to an exemplary embodiment.

Referring to FIG. 1 , the subject recommendation system 3 based on subject similarity includes a user device 1 of a student, an educational institution server 2 and/or an academic information providing server 4 through a wired and/or wireless network. It may be configured to operate while communicating. A communication method through a wired and/or wireless network may include any communication method capable of networking an object and an object, and is not limited to wired communication, wireless communication, 3G, 4G, or other methods.

For example, wired and/or wireless networks 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, VoIP (Voice) over Internet 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 ultrasonic-based communication to refer to a communication network by one or more communication methods selected from the group consisting of However, the present invention is not limited thereto.

In one embodiment, the subject recommendation system 3 includes a first data analysis unit 31 , a second data analysis unit 32 , a similarity determining unit 33 , and a recommended subject determining unit 35 . In an embodiment, the subject recommendation system 3 may further include a preprocessor 30 . Also, in one embodiment, the subject recommendation system 3 may further include a database (DB) 34 .

The systems, devices, and servers described herein may be wholly hardware, or may have aspects that are partly hardware and partly software. For example, the system, apparatus, and server of the present specification, and each unit included therein, may collectively refer to hardware and related software for processing and/or sending and receiving data in a specific format and content in an electronic communication manner. there is. As used herein, 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. In addition, 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 subject recommendation system 3 according to the present embodiment is not necessarily intended to refer to separate devices physically separated from each other. That is, the pre-processing unit 30, the first data analysis unit 31, the second data analysis unit 32, the similarity determining unit 33, the DB 34, and the recommended subject determining unit 35 of FIG. 1 are The hardware constituting the subject recommendation system 3 is functionally classified according to the operation performed by the corresponding hardware, and each part does not necessarily have to be provided independently of each other. Of course, depending on the embodiment, at least one of the respective parts of the subject recommendation system 3 may be implemented as separate devices physically separated from each other.

The subject recommendation system 3 generates lecture content information and student characteristic information by analyzing text data related to one or more subjects corresponding to the recommendation target and attendance history data of a plurality of students, and determines the similarity between subjects based on this. is configured to In addition, the subject recommendation system 3 receives subject information of a user who wants to receive a subject recommendation from the user device 1 , and generates subject recommendation information based on the subject of interest information and the above-described similarity information to generate subject recommendation information from the user device 1 . ) can be provided.

To enable the user to transmit subject information of interest and receive subject recommendation information, the subject recommendation system 3 communicates with an application (or an app) running on the user device 1 to perform the function of the application. It can perform the function of an application service server that enables it. Alternatively, in another embodiment, the subject recommendation system 3 may be implemented in the form of a web server that provides a web page accessible by a web browser running on the user device 1 .

In the embodiment shown in Fig. 1, the user device 1 is shown in the form of a notebook computer. However, this is an example, and the type of the user device 1 is not limited to those shown in the drawings. For example, a user who wants to receive a course recommendation is 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). A function provided by the subject recommendation system 3 may be used by using an arbitrary computing device such as a set-top box for etc.

The preprocessor 30 is a part for extracting text data, which is a basis for calculating the degree of similarity between subjects. The preprocessor 30 may extract text data from syllabus and/or learning goals of a subject, curriculum for each department, and the like. For the above operation, the preprocessor 30 may be configured to collect information from the server 2 of an educational institution such as a university that provides lectures of each subject. In addition, the preprocessor 30 may extract text data from published thesis data for each academic field in order to determine a keyword of an academic field corresponding to a subject in generating lecture content information. To this end, the preprocessor 30 may be configured to communicate with the academic information providing server 4 that provides research thesis data.

In addition, the pre-processing unit 30 may be configured to extract characteristics of course students through pre-processing of attendance history data of students for each subject. Furthermore, the preprocessor 30 may be further configured to extract preferred subject information through preprocessing of the subject information of interest transmitted from the user device 1 by the user who wants to receive a subject recommendation.

The first data analysis unit 31 may generate lecture content information for each subject by analyzing text data related to one or more subjects. In one embodiment, the first data analysis unit 31 learns the lecture content of the subject by analyzing the text data based on Term Frequency-Inverse Document Frequency (TF-IDF), Lecture content information can be generated based on the learning result of the content. A detailed operation of the first data analysis unit 31 will be described later in detail with reference to FIG. 3 .

The second data analysis unit 32 may generate student characteristic information for each subject by analyzing the attendance history data of the students for each subject. For example, the second data analysis unit 32 may generate attendance rate data for each department based on the attendance history data of a plurality of students, and generate student characteristic information using the attendance ratio data for each department of the students for each subject. there is. A detailed operation of the second data analyzer 32 will be described later with reference to FIG. 4 .

The similarity determining unit 33 may generate similarity information for each subject using the lecture content information generated by the first data analysis unit 31 and the student characteristic information generated by the second data analysis unit 32 . there is. For example, the similarity determining unit 33 uses the cosine similarity between the lecture content matrix in which the academic area data for each subject is embedded and the student characteristic matrix in which the attendance ratio by department of students who have taken the course is embedded. Star similarity can be determined. A detailed process for determining the degree of similarity for each subject will be described later in detail with reference to FIG. 2 .

In an embodiment, the lecture content information generated by the first data analysis unit 31 , the student characteristic information generated by the second data analysis unit 32 , and/or the similarity determination unit 33 generated by the The similarity information may be stored in the DB 34 of the subject recommendation system 3 .

The recommended subject determining unit 35 may generate subject recommendation information for the target user by using the subject information of interest received from the user device 1 and the subject-specific similarity information generated by the similarity determining unit 33 . there is. For example, the recommended subject determining unit 35 may determine the user's preferred subject based on the recommended subject user's attendance history and/or desired course, and may determine the recommended subject based on the degree of similarity to the preferred subject.

The recommended subject determination unit 35 may provide the generated subject recommendation information to the user device 1 . In this case, the subject recommendation information may be visualized in the form of a subject map indicating the degree of similarity between subjects. A subject map is defined by a plurality of dimensions corresponding to the characteristics of each subject, and the higher the similarity between subjects, the closer to each other the subjects may be, in the form of a flat or three-dimensional map. In this case, the UMAP (Uniform Manifold Approximation and Projection) algorithm may be used in order to reduce the academic areas and department characteristics for each subject to a limited number of dimensions (eg, two-dimensional). The form of subject guidance will be described later in detail with reference to FIGS. 8A and 8B.

2 is a flowchart illustrating each step of a method for recommending a subject based on subject similarity according to an exemplary embodiment. The subject recommendation method according to the embodiments may be performed using the subject recommendation system according to the embodiments. Hereinafter, for convenience of explanation, a subject recommendation method according to the present embodiment will be described with reference to FIGS. 1 and 2 .

Embodiments of the present invention are implemented using a subject-based recommendation model in which a subject user who wants to receive a subject recommendation recommends a lecture similar to an interested lecture. More specifically, the embodiments calculate lecture content information and learner characteristic information for each subject by embedding data for each subject in a characteristic matrix at the lecture content dimension and student characteristic dimension, and use this to calculate the degree of similarity between subjects. can Thereafter, subject recommendation information may be provided to the user in a manner of recommending a subject having a high degree of similarity to the user's preferred subject.

Specifically, the first data analysis unit 31 of the subject recommendation system 3 may generate an academic field keyword matrix for each subject by performing keyword analysis on research thesis data (S11). Also, the first data analysis unit 31 may generate a subject information keyword matrix through morphological analysis of subject information data such as syllabus for each subject ( S12 ). Next, the first data analysis unit 31 may generate lecture content information by embedding each subject in the lecture content matrix using the academic field keyword matrix and the subject information keyword matrix (S13).

For the above operation by the first data analysis unit 31 , the preprocessor 30 of the subject recommendation system 3 may extract keywords for each academic area from the research thesis data in the form of text data. In addition, the preprocessor 30 may extract text data corresponding to a keyword through morphological analysis from subject information such as syllabus or learning goals of each subject, and curriculum data for each department. In addition, the TF-IDF value may be used to reflect the importance of each keyword in each matrix generated by the first data analysis unit 31 .

3 is a conceptual diagram of a process of generating lecture content information by a subject recommendation method based on subject similarity according to an exemplary embodiment.

Referring to FIG. 3 , research thesis data may be used to embed the academic domain characteristics of each subject corresponding to the recommended subjects. For example, each thesis of the research thesis data usually includes about 5 to 7 keywords, and is usually assigned to about 1 to 3 academic fields according to the characteristics of the submitted journal. In an embodiment, the research thesis data may be obtained from the academic information providing server 4 (FIG. 1).

In this embodiment, the frequency of occurrence of keywords included in the thesis is counted according to the academic field in which the thesis is included, and the document-word matrix 301 of the keyword in the academic field is constructed based on the aggregated data. Each row of the document-word matrix 301 corresponds to a keyword, and each column corresponds to an academic area. In an embodiment, the above-described document-word matrix 301 may be configured by excluding keywords that appear less than a certain number of times by summing up all academic areas and/or academic areas in which the recorded keyword is less than a certain number of times. However, this is only an example, and in another embodiment, the document-word matrix 301 may be configured to include all academic areas and all keywords.

In this case, each component of the document-word matrix 301 may include a TF-IDF value for each combination of a keyword of a research thesis and an academic field. The TF-IDF value is a numerical value indicating how important a keyword is within a specific academic field, considering the overall frequency of occurrence of the keyword and the frequency of occurrence within a specific academic field. It is used to properly control the importance within the field. In an embodiment, the TF-IDF value may be normalized so that the TF-IDF value for each keyword-scientific field combination is between 0 and 1.

Meanwhile, subject information may be used to embed characteristics corresponding to contents of each subject in recommended subjects. In this case, the subject information indicates the contents learned in the lectures of each subject. For example, the more similar the contents covered in the two lectures are academically, the more similar the two subjects can be considered to each other. In this embodiment, the syllabus and/or learning goals input by the instructor in order to analyze the learning content of the subject, curriculum information provided by the educational institution server, etc. may be used as subject information. In an embodiment, syllabus, learning goals, curriculum information, and the like may be obtained from the educational institution server 2 ( FIG. 1 ).

In this embodiment, first, subject information keywords can be extracted through morphological analysis of subject information data, which is unstructured data. At this time, in one embodiment, for example, words not closely related to the educational content of each lecture, such as [final, paper, submission, .... study, teaching, necessary], may be excluded from keywords by classifying them as preset stopwords. The matrix 302 of keywords extracted from the subject information data may be converted into a keyword-subject matrix 303 through transpose, and each row of the keyword-subject matrix 303 corresponds to each subject and each column corresponds to each keyword.

In one embodiment, similar to the academic field keyword matrix 301, the TF-IDF value for each keyword-subject pair is calculated in consideration of how many times a specific keyword appeared in one subject (TF) and how many subjects it appeared in (DF). It can also be calculated, and a higher TF-IDF value means that a specific keyword may be an important keyword representing the subject. Also, in an embodiment, the TF-IDF value for each subject of each subject information keyword may be normalized to a value between 0 and 1.

Next, the lecture content matrix 310 may be generated using the academic field keyword matrix 301 and the subject information keyword matrix 303 . At this time, each row of the lecture content matrix 310 corresponds to each subject, and each column corresponds to each academic area. In addition, each component of the lecture content matrix 310 may be calculated as an average value obtained by applying a TF-IDF weight to the academic domain characteristic of the subject information keyword. In other words, the academic domain characteristics of keywords that are important for each subject are reflected more strongly when calculating the academic domain characteristics of the subject.

For example, in academic field m, the TF-IDF value of keyword i is represented by F in Equation 1 below, the TF-IDF value of keyword i in subject information j is represented by C in Equation 2 below, and subject information j is When the number of keywords is expressed as N _j , the TF-IDF value F ′ in which the weight of subject information j belonging to the academic field m is reflected can be calculated as in Equation 3 below.

[Equation 1]

[Equation 2]

[Equation 3]

Referring back to FIGS. 1 and 2 , the second data analysis unit 32 of the subject recommendation system 3 generates information about the characteristics of students for each subject in order to reflect the characteristics of the students who have listened to the lecture for each subject. can To this end, the second data analysis unit 32 may use the attendance history data of a plurality of students. First, the second data analysis unit 32 calculates the ratio of the credits completed in a specific department among the total credits completed by the students for each student in the course history data, thereby calculating a course history characteristic matrix for each student (S14) . To this end, the pre-processing unit 30 may extract information such as a student number, subject, and opened department from the course history data in the form of text data.

Next, the second data analysis unit 32 may calculate the student characteristics for each subject by using the attendance history characteristic matrix of the students who took the corresponding course from the attendance history data for one or more subjects to be recommended. There is (S15). For example, the average of the attendance history characteristic matrix of students who have taken the same course may be used as the student characteristic information of the corresponding course.

4 is a conceptual diagram illustrating a process of generating student characteristic information by a subject recommendation method based on subject similarity according to an exemplary embodiment.

Referring to FIG. 4 , in the present embodiment, the attendance history data is used to calculate the TF-IDF value for each subject based on the attendance history of the students. From the course history data 400 including identification information (eg, student number) of each student and information on the courses each student has taken (eg, courses and departments), a subject ratio matrix 401 may be calculated. Each row of the subject ratio matrix 401 corresponds to a student, and each column corresponds to a department in which a subject is opened. That is, the value of each component of the subject ratio matrix 401 represents the ratio of the total number of credits acquired in the department corresponding to the corresponding column among the total credits acquired by the students corresponding to the corresponding row.

Also, it is possible to extract the student matrix 402 for each subject from the attendance history data. Each row of the learner matrix 402 for each subject corresponds to a respective subject, and each column of the learner matrix 402 corresponds to a respective learner. By using the calculated subject ratio matrix 401 and the student matrix 402 for each subject, it is possible to obtain a student characteristic matrix 410 in which the characteristics of the attendance rate by department of the students who have taken each course are embedded. That is, each row of the student characteristic matrix 401 represents a respective subject, and each column represents an average attendance ratio for each open department of students who have taken the corresponding subject.

Referring back to FIGS. 1 and 2 , the similarity determining unit 33 of the subject recommendation system 3 includes the lecture content matrix 310 ( FIG. 3 ) and the student characteristic matrix 410 ( FIG. 4 ) generated as described above. can be used to calculate the degree of similarity between subjects (S16). In an embodiment, the similarity information between subjects may be calculated by cosine similarity based on each of the lecture content matrix and the student characteristic matrix. For example, if lecture content matrices (or student characteristic matrices) of two different subjects are A and B, respectively, the cosine similarity between the two matrices can be calculated as in Equation 4 below.

[Equation 4]

Based on the cosine similarity calculated as in Equation 4, similarity information between subjects based on lecture content information and information on similarity between subjects based on student characteristic information may be generated, respectively.

Next, the recommended subject determination unit 35 of the subject recommendation system 3 may receive information on the subject of interest of the target user in order to determine a recommended subject suitable for the target user who wants to receive the subject recommendation (S17). At this time, the subject information of interest may include a desired course designated by the user, and/or may include the user's existing attendance record.

Next, the recommended subject determination unit 35 may determine the user's preferred subject based on the subject information of interest, and determine a subject having a high similarity with the preferred subject in the similarity information as a recommended subject with a preset number of subjects (S18) ). If the subject information of interest includes the user's desired course, the desired course itself corresponds to the preferred course. can

In one embodiment, in order to select preferred subjects from the existing course history, subjects other than subjects that the user has necessarily completed (major subjects or compulsory liberal arts subjects, etc.) may be determined as preferred subjects. This is to exclude subjects that the user must take for graduation, and select subjects that are taken only according to the user's preference.

The recommended subject determining unit 35 may provide subject recommendation information including the determined recommended subject to the user device 1 ( S19 ). For example, the recommended subject determination unit 35 may determine n subjects having a higher similarity as recommended subjects based on the similarity to the user's preferred subjects (n is an arbitrary natural number). As the value of n is large, a wider range of recommendations is made, and as the value of n is smaller, a recommendation of a detailed range is made. Therefore, the value of n is also referred to as a recommendation range in this specification.

In an embodiment, the recommended subject determination unit 35 may determine the final recommended subject by reflecting a weight in consideration of (i) the number of selections, (ii) a desired course order, and (iii) a similarity percentile. At this time, the number of selections is to reflect the weight of the top n recommended courses based on the degree of similarity to the number of preferred courses. In addition, the desired course order is to reflect the preferred course order written by the user as a weight for the preferred subject similar to the recommended subject. In addition, the similarity percentile is to reflect the value obtained by converting the degree of similarity with the preferred subject to the percentile as a weight.

In one embodiment, the recommended subject determination unit 35 is configured to, when a recommended subject user does not register a desired course and does not have an existing course record of the user, based on the preferred subject of other students having the same characteristics as the corresponding user. You can also decide which subjects to recommend. In this case, other students may refer to students having the same major as the recommended target user, but is not limited thereto. Also, the preferred course of other students may refer to a desired course entered by other students or a course determined based on the existing course histories of other students.

In order to verify the performance of the subject recommendation model according to the embodiments of the present invention described above, the present inventors compared the subjects recommended by the recommendation model with the actual course registration results of students. For this purpose, data from 9,275 students who applied for courses in the second semester of 2020 among students who applied for three or more courses by the first semester of 2020 were used. Based on the data up to the first semester of 2020, a list of recommended subjects for each student was calculated and compared with the subjects selected by the students in the second semester of 2020. In this specification, for the sake of brevity of expression, the subjects that students have applied for and the subjects registered in the desired subjects are expressed as elective subjects. The average number of subjects selected by the students was 2.94, and the standard deviation was 1.86.

For performance verification, precision rate and recall rate, which are widely used for recommendation model performance verification, were used. Precision was calculated as [precision = (number of successful subject recommendations) / (number of recommended subjects)] through the number of recommended successes compared to the number of recommendations did In addition, the recall rate was calculated as [recall rate = (number of successful subject recommendations) / (number of elective subjects)] through the ratio of the number of successful subject recommendations to the number of elective subjects.

5 to 7 are graphs illustrating an effect of a subject recommendation method based on subject similarity according to an exemplary embodiment in comparison with the prior art.

5 shows a comparison of the recommendation accuracy by the subject recommendation method according to an embodiment with the prior art through a recall rate and a precision curve, and FIGS. 5 (a) to (f) are 3, 5, The case where 10, 30, 50, and 100 is used is shown. In contrast to this embodiment, the virtual prior art was divided into a model using only “student characteristics” and a model using only “lecture content” according to the data utilized. As the recommendation range increased, a recommendation result reflecting the student's overall interest was calculated, and as the recommendation range was narrow, the recommendation result based on the student's specific interest subject was calculated.

As shown in FIG. 5 , the subject recommendation method according to the present embodiment performed better than a model using only one of “student characteristics” and “lecture content”. In particular, when the recommendation ranges were 3, 5, and 10, the precision and recall of the subject recommendation method according to this embodiment were excellent, and it was confirmed that it was preferable for analysis and service application. Also, a trade-off between precision and recall was confirmed according to the number of recommended subjects. In other words, the more subjects were recommended, the higher the probability that there was a subject recommended by the model among the subjects selected by the student, but the probability that the student actually selected the recommended subject was low. Based on this, the present invention may be implemented by defaulting to recommending 21 subjects in order to recommend many subjects to students. .

6 shows a comparison of recommendation accuracy by the subject recommendation method according to an embodiment with the prior art in terms of coverage and Shannon entropy, and FIG. 6 (a) shows the coverage for each model. , (b) of FIG. 6 shows the variety of recommendations for each model.

In FIG. 6 , the coverage was calculated as the ratio of subjects recommended by the recommendation model among 281 subjects, and the diversity recommended by each model means how many different subjects are recommended by each student, and was calculated using the Shannon entropy formula. As shown, the subject recommendation method according to the embodiment of the present invention recommended more subjects than a model using only one of “student characteristics” or “lecture content”, and has the widest subject coverage It can be seen that compared to the model of

7 is a comparison of the recommendation result by the subject recommendation method according to an embodiment and the subject recommendation result based on popular subjects by department. b) represents the coverage and Shannon entropy.

In the result shown in FIG. 7 , the subject recommendation method according to the embodiment of the present invention shows the result of applying the recommendation range 3 to the subject recommendation method. In addition, popular subjects by department were calculated based on the results of recommending the subjects that were taken the most in the department until the first semester of 2020 among the courses opened in the second semester of 2020. As shown, the popular subject-based recommendation showed better recommendation performance in terms of precision and recall, but in terms of coverage of recommended subjects and diversity of recommendations, the subject recommendation method according to the embodiment of the present invention is wider and wider than the popular subject-based recommendation model. It was found to produce various recommendation results.

8A and 8B are conceptual diagrams illustrating examples of subject guidance provided by a subject recommendation method based on subject similarity according to an exemplary embodiment.

The subject map shown in FIGS. 8A and 8B shows the course content matrix and subject information matrix for each subject. McInnes, L., Healy, J., and Melville, J. co-authored paper "Umap: Uniform manifold approximation and projection for dimension reduction" (arXiv preprint arXiv: 1802.03426, 2018) shows the results visualized after being reduced to two dimensions using the UMAP algorithm disclosed in (arXiv preprint arXiv: 1802.03426, 2018). This means that it is different.

8A is a diagram showing that nodes have different hatchings according to departments opened in each subject on the subject map, and subjects opened in the same department are located adjacent to each other, so that subjects according to an embodiment of the present invention It can be seen that the similarity calculation between subjects by the recommendation method is accurate. In FIG. 8A , each node is distinguished through hatching on the black-and-white map. However, when the subject map is provided in color, the opened department may be distinguished through the color of each node.

In addition, FIG. 8B shows subjects with strong characteristics in a specific academic area by using the academic area characteristic data, and subjects with high accounting characteristics such as investment economic analysis, technology management and strategy, entrepreneurship for engineering students, and corporate law. is shown brightly. However, the same method of visualization can be applied to other academic fields. In addition, in another embodiment, by classifying the brightness of the nodes based on the subject information matrix rather than the academic area, it is possible to configure the subject map to brightly display subjects with a high ratio of students taking classes in a specific department. . In this way, when the recommended subject information is provided in the form of a reduced-dimensional subject map using the UMAP algorithm, it is possible to provide information so that the global structure and topological structure can be effectively captured even in low dimensions. There is an advantage.

As described above, according to the embodiments of the present invention, after extracting keywords by performing morphological analysis on syllabus and learning goals for each subject, the weight for each subject-keyword can be calculated through the TF-IDF value, and the calculated The lecture content characteristic vector of each subject content can be generated by using the subject-keyword-based weight and the research paper keyword-based glossary. In addition, a student characteristic vector may be generated based on the ratio of students enrolled in each subject to each department. The degree of similarity between subjects is calculated through the cosine similarity based on the lecture content characteristic vector and student characteristic vector for each subject, and subjects with a high degree of similarity to the subjects of interest for each student can be recommended as recommended subjects. Furthermore, subject recommendation information can be visually provided using the subject map generated by applying the UMAP algorithm to the academic domain characteristic vector for each subject.

According to the embodiments of the present invention, it is possible to calculate a recommendation result consistent with the actual student's course selection, and maintain the accuracy of the recommendation model, compared to recommending a course in consideration of only one piece of information among lecture content or student characteristics. However, it has the advantage of obtaining better results in terms of the scope and diversity of recommendations. Through this, when using the embodiments of the present invention, it is possible to implement a customized service for nurturing creative and convergence-type future talents by recommending a wide variety of subjects that meet the interests of each student, which is a recommendation of a second major or career path, micro Alternatively, there is an advantage that can be extended to nano-degree related services.

Meanwhile, the operation by the method for recommending a subject based on subject similarity according to the embodiments described above may be at least partially implemented as a computer program and recorded in a computer-readable recording medium. A computer-readable recording medium in which a program for implementing an operation by a subject recommendation method based on subject similarity according to embodiments is recorded and includes all types of recording devices in which computer-readable data is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. In addition, the computer-readable recording medium may be distributed in a network-connected computer system, and the computer-readable code may be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present embodiment may be easily understood by those skilled in the art to which the present embodiment belongs.

Although the present invention described above 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 variations of the embodiments are possible therefrom. However, such modifications should be considered to be within the technical protection scope of the present invention. Accordingly, the true technical protection scope of the present invention should be defined by the technical spirit of the appended claims.

Claims (15)

As a subject recommendation system based on subject similarity,
a first data analysis unit configured to generate lecture content information for each of the one or more subjects by analyzing text data related to the one or more subjects;
a second data analysis unit configured to generate student characteristic information for each of the one or more subjects by analyzing the attendance history data of the students for each subject;
a similarity determining unit configured to generate similarity information for each subject using the lecture content information and the student characteristic information; and
a recommended subject determination unit configured to generate subject recommendation information for the user based on the subject information of interest received from the user and the similarity information;
Further comprising a preprocessing unit that extracts keywords for each academic area from the subject information including the syllabus of the subject, the learning goal of the subject, the curriculum for each department, and the research thesis for each academic field, and pre-processes it in the text data format,
The subject information of interest includes at least one of the user's attendance history and desired subjects,
The recommended subject determination unit determines the user's preferred subject by using at least one information among the user's attendance history and desired subjects, and determines the recommended subject based on the degree of similarity to the preferred subject to obtain the subject recommendation information. further configured to create,
The recommended subject determining unit is further configured to select a preferred subject by excluding the subject that the user obligatedly took from the user's attendance record when the user's preferred subject is determined based on only the user's attendance record,
The recommended subject determination unit determines a recommended subject based on the preferred subjects of different students having the same characteristics as the user, when the user's attendance history and desired subjects are not entered in the interest subject information;
The preferred subjects of different students who have the same characteristics as the above users are,
A course recommendation system based on subject similarity that includes one or more of the desired courses entered by different students and one or more courses determined based on the previous course histories of different students.
delete According to claim 1,
The first data analysis unit is further configured to learn the lecture contents of the subject by word frequency-inverse document frequency-based analysis of the text data, and to generate the lecture content information based on the learning result for the lecture contents Subject recommendation system based on the constructed subject similarity.
According to claim 1,
The second data analysis unit is further configured to generate attendance rate data for each department based on the attendance history data of a plurality of students, and generate the student characteristic information using the department-specific attendance ratio data of the students for each subject. Based on the subject recommendation system.
According to claim 1,
The subject recommendation system based on subject similarity is further configured to generate the similarity information for each subject by using the cosine similarity between matrices corresponding to the lecture content information and the student characteristic information, respectively.
delete According to claim 1,
The subject similarity determination unit is further configured to provide the subject recommendation information to the user using a subject map defined by a plurality of dimensions corresponding to the characteristics of each subject and in which subjects are located closer to each other as the degree of similarity between subjects is higher. Based on the subject recommendation system.
generating lecture content information for each of the one or more subjects by analyzing text data related to the one or more subjects by the subject recommendation system;
generating, by the subject recommendation system, information on student characteristics for each of the one or more subjects by analyzing attendance history data of students for each subject;
generating, by the subject recommendation system, similarity information for each subject using the lecture content information and the student characteristic information; and
generating, by the subject recommendation system, subject recommendation information for the user based on the subject information of interest and the similarity information received from the user;
The subject recommendation system further comprises the step of extracting, by the subject recommendation system, keywords for each academic area from subject information including the syllabus of the subject, the learning goal of the subject, the curriculum for each department, and the research thesis for each academic field, and pre-processing it in the text data format, and ,
The subject information of interest includes at least one of the user's attendance history and desired subjects,
The step of generating the subject recommendation information includes:
determining, by the subject recommendation system, the preferred subject of the user by using information on at least one of the course history and desired subjects; and
determining, by the subject recommendation system, a recommended subject based on the degree of similarity to the preferred subject;
The step of determining the user's preferred subject includes:
If the user's preferred subject is determined based on only the user's attendance record, the user's compulsory subject is excluded from the user's attendance record to determine the preferred subject;
The step of determining the recommended subject is,
When the user's course history and desired course are not entered in the subject information of interest, a recommended subject is determined based on the preferred subjects of different students having the same characteristics as the user;
The preferred subjects of different students with the same characteristics as the above users are,
A course recommendation method based on subject similarity, including at least one of the desired courses entered by different students and one or more courses determined based on the previous course histories of different students.
delete 9. The method of claim 8,
The step of generating the lecture content information includes:
learning, by the subject recommendation system, the lecture contents of the subject through word frequency-inverse document frequency-based analysis of the text data; and
and generating, by the subject recommendation system, the lecture content information based on a learning result of the lecture content.
9. The method of claim 8,
The step of generating the student characteristic information includes:
generating, by the subject recommendation system, data on the attendance rate for each department based on the attendance history data of a plurality of students; and
and generating, by the subject recommendation system, the student characteristic information by using the data on the attendance rate by department of students for each subject.
9. The method of claim 8,
The generating of the similarity information includes, by the subject recommendation system, calculating a cosine similarity between matrices corresponding to the lecture content information and the student characteristic information, respectively.
delete 9. The method of claim 8,
The subject recommendation system further includes providing the subject recommendation information to the user by using a subject map defined by a plurality of dimensions corresponding to the characteristics of each subject and in which subjects are located closer to each other as the similarity between subjects is higher. Subject recommendation method based on subject similarity.
A computer program stored in a computer-readable recording medium in combination with hardware to execute the subject recommendation method based on subject similarity according to any one of claims 8, 10 to 12 and 14.

Images