KR20230018677A - System for recommending implementation of unborrowed book - Google Patents

Abstract

The present invention relates to an unborrowed book recommending system, and more specifically, to an unborrowed book recommending system, which considers diversity of books, rather than recommending books based on popular books, to provide customized services. The unborrowed book recommending system includes: a data storage unit for storing collection list information, loan history information, member information, and detailed book information of a book loan system in which a plurality of books and members are registered; a data set building unit which builds an integrated data set for the information; a text mining unit; a topic clustering unit which classifies the plurality of books into a plurality of clusters and assigns a unique topic to each cluster; a popular category extraction unit; a TF-IDF vectorization unit; a cosine similarity calculating unit; and an unborrowed book recommendation unit which recommends the unborrowed books based on an average value of a cosine similarity value.

Description

Unloaned book recommendation system {SYSTEM FOR RECOMMENDING IMPLEMENTATION OF UNBORROWED BOOK}

The present invention relates to a system for recommending un-loaned books, and more particularly to a system for recommending un-loaned books that is personalized based on the diversity of books, rather than recommending books based on popular books.

While the roles and functions of domestic public libraries are diversifying, internally, various problems are emerging due to biased book lending. In addition, recently, a book recommendation system focusing on popular books has been introduced in public libraries, but due to this, the diversity of books that users can access is limited.

In addition, the role and function of the existing public library is expanding due to the recent development of IT technology and the increase in various demands of public library users due to the 4th industrial revolution. This goes beyond the function of borrowing books, which is the role of the existing public library, and has a close relationship with the lives of residents as activities such as cultural centers, exhibitions, movie viewing, and local communities are performed.

Although the importance of public libraries is increasing due to the quantitative growth and diversity of roles of public libraries, there are problems such as internally biased book lending. is appearing In order to solve this problem, public libraries regularly recommend books using curators and establish policies to promote use by the person in charge, but there are many limitations.

In order to solve this problem, various previous studies have been conducted to implement a personalized book recommendation system, and most of the previous studies have implemented using collaborative filtering algorithms, content-based filtering algorithms, and hybrid algorithms. This has the advantage of implementing a scientific recommendation system based on machine learning by using individual book loan history data, but has limitations in recommending various books by recommending books with a high frequency of loaning.

The present invention relates to a system for recommending unleavened books, and is intended to provide a system for recommending books that are not loaned out based on popular books but are customized for individuals considering the diversity of books.

The technical problems to be achieved by the present invention are not limited to the above-mentioned technical problems, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below. You will be able to.

The non-lending book recommendation system of the present invention,

A data storage unit for storing collection list information, loan history information, member information, and detailed book information of a book rental system in which a plurality of books and members are registered;

a data set construction unit for constructing an integrated data set for the collection list information, the loan history information, the member information, and the detailed book information;

a text mining unit that performs text mining processing on text data from among the integrated data set;

The plurality of books are classified into a plurality of clusters by topic modeling based on spherical K-means clustering and latent dirichlet allocation (LDA) using the data processed by text mining in the text mining unit as an input value, respectively. a topic clustering unit that assigns a unique topic to clusters of;

a popular category extracting unit which classifies the plurality of books into a list of loaned books and a list of un loaned books based on the integrated data set, and extracts a top unique topic from the list of loaned books as a popular category;

a TF-IDF vectorization unit calculating TF-IDF values for each of the plurality of books for feature vectorization based on term frequency-inverse document frequency (TF-IDF) for the plurality of books;

a cosine similarity calculation unit for calculating a cosine similarity value using the TF-IDF values of the books in the uncirculated book list and the TF-IDF values of the books in the loaned book list; and

and an unleased book recommendation unit for recommending unleased books based on an average value of cosine similarity values for a plurality of books in the unleased book list for each book in the unleased book list.

In the uncirculated book recommendation system of the present invention, the collection list information includes book name, author, publisher, publication year, ISBN (international standard book number), set ISBN, additional code, subject classification number, number of books, number of loans, and book registration. It includes at least one information of a date, and the loan history information includes at least one or more of a user number, a book number, a Korean decimal classification (KDC), a book name, an author, a publisher, a loan date, a return date, and a media type for each loan case. The member information includes at least one of user number, member registration date, date of birth, gender, postal code, book rental number and overdue number, and the book detailed information includes book name, KDC, publication date, author , a publisher, a book image URL, and a book introduction may include at least one or more information.

In the unlendered book recommendation system of the present invention, the data set building unit integrates book introductions of books with the same book name for the detailed book information, and deletes detailed book information for books with blank book introductions. A member information organizing unit combining the loan history data and the member information based on a user number (User key), and generating an age group variable, which is categorical data, from the date of birth, which is discrete data; A loan information data set generating unit that generates a loan information data set by merging the loan history information and the member information, and an integrated data set construction unit that integrates the loan information data set and the detailed book information to build the integrated data set. It may include wealth.

In the non-loaned book recommendation system of the present invention, the text mining unit includes a text cleansing unit that removes numbers or special characters from text data from the detailed book information, and a noun that extracts nouns from the text data from the detailed book information. An extraction unit, a stopword removal unit that removes stopwords from text data of the detailed book information, and the text data processed through the text cleansing unit, the noun extraction unit, and the stopword removal unit are separated into morpheme units, It may include a tokenization unit that performs tokenization.

In the system for recommending books not yet loaned according to the present invention, text data among the detailed book information may be a book name or a book introduction.

In the unleavened book recommendation system of the present invention, the topic clustering unit may perform spherical K-means clustering by utilizing the names of books included in the plurality of books and nouns extracted from book introductions through word tokenization.

In the non-loaned book recommendation system of the present invention, the popular category extraction unit classifies an integrated data set into a plurality of member types based on the member information, and independently extracts a popular category for each of the plurality of member types. it could be

In the system for recommending books not yet loaned according to the present invention, the plurality of member types may be by gender or age group.

In the uncirculated book recommendation system of the present invention, the popular category extraction unit extracts a plurality of popular categories, the TF-IDF value is calculated based on the popular category as a calculation unit, and the TF-IDF vectorization unit extracts a plurality of popular categories, The IDF value may be calculated by Equation 1 below.

[Equation 1]

W _i,j is the TF-IDF value for word i in book j, tf _i,j is the frequency count of word i in the analysis domain specified in book j, df _i is the number of books containing word i, and N is the number of books belonging to the calculation unit.

In the unlisted book recommendation system of the present invention, the cosine similarity calculation unit may calculate the cosine similarity by Equation 2 below.

[Equation 2]

cos(θ) is the cosine similarity value between the book of the uncirculated book list and the book of the loaned book list, A _i is the TF-IDF value for word i of the book of the un loaned book list, and B _i is the above It is the TF-IDF value for word i of the book in the borrowed book list, and n is the number of types of word i.

The non-borrowed book recommendation system of the present invention extracts popular categories by gender and age group using spherical K-means clustering and LDA-based topic modeling based on the book loan history of public libraries, and then extracts popular categories based on content By implementing a system that recommends non-lending books with similar characteristics to popular books in the category through one-to-many correspondence in filtering, it contributes to users learning various knowledge and ideologies, and also contributes to mitigating biased book lending and activating the use of books. can contribute

The system for recommending un-borrowed books of the present invention implements a content-based filtering book recommendation system that recommends un-borrowed books by reflecting the preferences of the user group based on the lending history and member information of the public library. As a problem that occurs in collaborative filtering, the cold start problem of not being able to recommend any item to a new user for whom no existing data has been accumulated is solved by the present invention's uncirculated book recommendation system through cluster analysis and LDA technique in topic modeling. It can be solved by extracting popular books in popular categories by group and then recommending similar books that are not loaned through cosine similarity calculation.

1 is a block diagram showing an unexploited book system of the present invention.
2 is a block diagram showing a data set building unit.
3 may be a UI screen output from the unbooked book system of the present invention.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. In this process, the size or shape of the components shown in the drawings may be exaggerated for clarity and convenience of explanation. In addition, terms specifically defined in consideration of the configuration and operation of the present invention may vary according to the intentions or customs of users and operators. Definitions of these terms should be made based on the content throughout this specification.

Hereinafter, “unit” or “unit” in the description may be implemented by hardware and software combination or hardware alone.

Hereinafter, referring to FIGS. 1 to 3, the unbooked book recommendation system 100 of the present invention will be described in detail.

As shown in FIG. 1, the unextracted book recommendation system 100 of the present invention,

a data storage unit 110 storing collection list information, loan history information, member information, and detailed book information of the book rental system 11 in which a plurality of books and members are registered;

a data set construction unit 120 for constructing an integrated data set for the collection list information, the loan history information, the member information, and the detailed book information;

a text mining unit 130 that performs text mining processing on text data among the integrated data set;

The plurality of books are classified into a plurality of clusters by topic modeling based on spherical K-means clustering and latent dirichlet allocation (LDA) using the text mining processed data in the text mining unit 130 as an input value. a topic clustering unit 140 that assigns a unique topic to each cluster;

a popular category extracting unit (150) for classifying the plurality of books into a list of loaned books and a list of non-loaned books based on the integrated data set, and extracting, as a popular category, a top unique topic with respect to the list of loaned books;

a TF-IDF vectorization unit 160 for calculating TF-IDF values for each of the plurality of books for feature vectorization based on term frequency-inverse document frequency (TF-IDF) for the plurality of books;

a cosine similarity calculation unit (170) for calculating a cosine similarity value using the TF-IDF values of the books in the uncirculated book list and the TF-IDF values of the books in the loaned book list; and

It may include an unleased book recommendation unit 180 for recommending unleased books based on an average value of cosine similarity values for a plurality of books in the unleased book list for each book in the unleased book list.

The book rental system 11 in which a plurality of books and members are registered may be a public library. Alternatively, the book rental system 11 may be an e-book service system operated as a network.

The data storage unit 110 may be a data storage medium as a database.

The collection list information may be information about books held in the book rental system 11 such as a public library. The collection list information may include at least one of book name, author, publisher, publication year, ISBN (international standard book number), set ISBN, additional code, subject classification number, number of books, number of loans, and book registration date. there is. The ISBN may be a book number (international standard book number) issued to distribute books to bookstores. The set ISBN is an ISBN assigned to books in a set, and if there are books in the same set, all of them may be assigned the same set ISBN. The additional sign may be a sign designated for classifying the reader target, publication type, and content of the book.

The loan history information may be information on loans generated in the book rental system 11 such as a public library. The loan history information may include at least one of a user number, a book number, a Korean decimal classification (KDC), a book name, an author, a publisher, a loan date, a return date, and a media type for each loan case. KDC may be a book classification key according to the Korean decimal classification method.

Member information may be personal attribute information such as age and address of members who use the book rental system 11 such as a public library. More specifically, the member information may include at least one of user number, member registration date, date of birth, gender, postal code, book rental number, and overdue number.

The detailed book information may be information on properties such as book names, book introduction contents, publication year, etc. of books stored or registered in the book management system 11 such as a public library. Specifically, the detailed book information may include at least one of book name, KDC, publication date, author, publisher, book image URL, and book introduction. The detailed book information may be data collected through Open API provided by Library Information Naru.

In the data storage unit 110, the book detailed information collected as semi-structured data in the form of XML or JSON may be converted into structured data through Python and stored.

Table 1 below is a table showing detailed information of collection list information, loan history information, member information, and detailed book information.

information data details Collection catalog information Title, Author, Publisher, Year of publication, ISBN, Set ISBN, Additional code, Subject classification number, Number of books, Number of loans, Registration date and etc Loan history information User key, Book key, KDC, Title, Author, Publisher, Loan date, Return date, Media type and etc Profile User key, Registration date, Date of birth, Gender, Zipcode, Number of loans, Number of overdue and etc Book details Title, KDC, Date of publication, Author, Publisher, Book Image URL, Book Outline and etc

As shown in FIG. 2, the data set building unit 120,

A book introduction organizing unit 122 for integrating book introductions of books with the same book name for the detailed book information and deleting detailed book information for books with blank book introductions;

A member information organizing unit 121 that combines the loan history data and the member information based on a user number (User key) and generates an age group variable, which is categorical data, from the date of birth, which is discrete data;

a loan information data set generating unit 123 generating a loan information data set by merging the collection list information, the loan history information, and the member information;

It may include an integrated data set building unit that builds the integrated data set by integrating the loan information data set and the detailed book information.

If a book introduction is blank in the detailed book information or a book with the same book name exists, the book introduction organizing unit 122 may remove the blank row through R and integrate the book introduction of the book with the same book name. R is a programming language specialized in statistical calculations and data analysis.

The member information organizing unit 121 may combine loan history information and member information based on a user number (User key) to recommend books based on popular categories by gender or age group. The member information organizing unit 121 may remove data other than the user number, registration date, date of birth, and gender required for book recommendation. User key may be a unique identification code or ID assigned to each user.

The member information organizing unit 121 may remove member information having a date of birth after the set date of birth based on the set date of birth in order to remove outliers and missing values. For example, by setting the date of birth to December 31, 2019, member information about members whose birth year is after 2020 and missing member information can be collectively deleted.

The member information organizing unit 121 may convert the discrete type of date of birth data into a categorical type to generate an age group variable. For example, as an age variable, a categorical variable may be created with minors (19 years old or younger), 20's, 30's, and the like.

The loan information data set generator 123 may merge the collection list information, the loan history information, and the member information based on book names, and add ISBN information to the loan information data set.

When integrating the loan information data set and the detailed book information, the integrated data set building unit may use an ISBN as a standard. If the ISBN of the loan information data set is 10 digits, it can be converted to 13 digits and then merged based on the detailed book information and ISBN.

At this time, if the ISBN of the loan information data set is an outlier or missing value, the SequenceMatcher library calculates the book name similarity between the book name of the loan information data set and the book name of the detailed book information, and the book name similarity exceeds 90% In this case, the loan information data set and the detailed book information may be integrated. SequenceMatcher is a Python standard library that provides a function to quantify the mutual similarity of two strings. Data can be deleted if the book name similarity is less than 90%.

The integrated data set building unit may build the integrated data set by integrating the book introduction of the detailed book information with the loan information data set.

The text mining unit 130,

A text cleansing unit for removing numbers or special characters from text data of the detailed book information;

A noun extraction unit extracting nouns from text data among the detailed book information;

a stopword removal unit for removing stopwords from text data of the detailed book information;

It may include a tokenization unit that separates Korean into morpheme units and tokenizes the text data processed through the text cleansing unit, the noun extraction unit, and the stopword removal unit.

The tokenization unit may perform a tokenization operation of separating Hangul into morpheme units by utilizing the KoNLPy library and complementary stopword dictionary, and may extract only words whose parts of speech are nouns.

The complementary stopword dictionary is created in the tokenization unit, and it may be prepared because there is a limitation in applying only the stopword dictionary provided by the KoNLPy library to extract only meaningful words. Specifically, the complementary stopword dictionary can be created as follows.

After identifying keywords in individual topics by using LDA-based topic modeling, among keywords, words unnecessary for book category classification are selected as stopwords, entered into a text file, and then saved. Then, the stopwords entered in the text file are removed from all words, and the keywords in the topic are identified by using topic modeling again. If new stopwords are additionally found, the stopwords are added to the text file. Repeat this process until no more stopwords are identified.

Table 2 below is a table showing functions of the data set building unit 120 and the text mining unit 130.

composition function Data set building unit removal of missing values and outliers, conversion of data type, addition of required data, calculation of similarities between strings, sequential combination of 4 materials text mining unit text cleansing, tokenization, noun extraction, stopwords removal(286 cases)

Text data among the detailed book information may be a book name or a book introduction.

The topic clustering unit 140 may perform spherical K-means clustering using nouns extracted through word tokenization from book names and book introductions included in the plurality of books.

The topic clustering unit 140 classifies all collected books into clusters using spherical K-means clustering and LDA-based topic modeling for actual book classification, and each cluster means Topics can be defined.

Spherical K-means clustering is a machine learning technique suitable for cluster analysis of high-dimensional data such as document sets. LDA (Latent Dirichlet Allocation) is a probabilistic topic model technique that describes which topics exist in each document for a given document.

For example, all books can be classified into 20 clusters by performing spherical K-means clustering using the book names and nouns extracted by word tokenization in book introductions. If the number of topics is too large, there are difficulties in interpretation, so 20 topics may be appropriate.

The topic clustering unit 140 may assign a unique topic individually contained in each cluster by utilizing LDA-based topic modeling for a plurality of clusters.

For example, unique topics include literature/travel, social issues, psychology/politics, religion/philosophy, health/cooking, diplomacy/world politics, youth education, hobbies/life/IT, foreign languages, fashion/design, medicine/world novels. , self-development/entrepreneurship, fiction, world history/history, culture/art, education, prose/poetry, science, comics/entertainment, and Korean essays.

The topic clustering unit 140 may assign a unique topic to each of a plurality of books and add it to a data set.

The popular category extraction unit 150 may extract a popular category based on member information. The popular category extraction unit 150 may extract a popular category from among unique topics generated by the topic clustering unit 140 based on the number of book rentals among member information. The popular category extraction unit 150 may extract a plurality of popular categories.

The popular category extracting unit 150 may create a plurality of member types and individually extract a popular category for each of the plurality of member types. The popular category extraction unit 150 may classify an integrated data set into a plurality of member types based on the member information, and independently extract a popular category for each of the plurality of member types. The plurality of member types may be by gender or age group. For example, gender is divided into two types, male and female, and age group is divided into 6 types: 10 or younger, 20s, 30s, 40s, 50s, 60s or older for each gender, and combined into a total of 12 member types. We can classify the data set, and for this, we can extract the top 3 popular categories for each member type. Table 3 below is a table showing examples of popular categories extracted for each member type.

gender age group Popular category 1st 2nd place 3rd place male 0-19 Youth education Comic/Entertainment Science 20-29 Self-improvement/Start-up Literature/Travel Hobby/ Life Style/ IT 30 to 39 Self-improvement/Start-up Literature/Travel Hobby/ Life Style/ IT 40 to 49 Self-improvement/Start-up Youth education Comic/Entertainment 50-59 Literature/Travel Self-improvement/Start-up Hobby/ Life Style/ IT 60~ Literature/Travel Self-improvement/Start-up World history/ History female 0-19 Youth education Comic/Entertainment Science 20-29 Literature/Travel Self-improvement/Start-up Novel 30 to 39 Youth education Literature/Travel Comic/Entertainment 40 to 49 Literature/Travel Youth education Comic/Entertainment 50-59 Literature/Travel Self-improvement/Start-up Novel 60~ Literature/Travel Comic/Entertainment Youth education

The extracted popularity category may be a calculation unit in calculating a TF-IDF value described later. That is, the TF-IDF value may be calculated with a popular category as a calculation unit. Specifically, the TF-IDF value may be calculated for each group by grouping books by popular category. For example, if youth education and hobbies/life/IT are extracted as popular categories, and 30 books belong to youth education and 20 books belong to hobby/life/IT, then 30 books for youth education are counted as a unit of calculation. TF-IDF values for each of the 30 books can be calculated, and TF-IDF values for each of the 20 books for hobbies, life, and IT can be calculated using 20 books as a calculation unit. When popular categories are extracted for each member type, each popular category for each member type may be a calculation unit. For example, in the case of Table 3, 3 popular categories were extracted for every 12 member types, and therefore, the number of calculation units may be 36.

The TF-IDF vectorization unit 160 may calculate the TF-IDF value by Equation 1 below.

[Equation 1]

W _i,j is the TF-IDF value for word i in book j, tf _i,j is the frequency count of word i in the analysis domain specified in book j, df _i is the number of books containing word i, and N is the number of books belonging to the calculation unit.

Word i may be words described in the analysis area of books belonging to the calculation unit. The word i is a noun and may have stop words removed. For example, if the words a, b, c, and d are described in the analysis area of one book and the words b, c, e, and f are described in the analysis area of another book of the same computational unit, the words i may include a, b, c, d, e, f. The analysis area may be the entire book or may be designated as a part of the book name or book introduction. Within the computational unit, if the number of types of word i is a total of n, the books corresponding to the computational unit can be calculated as vector values having n components, and the component values in the vector values of the books can be TF-IDF values there is. That is, if the total number of types of word i within a computational unit is n, each book corresponding to the computational unit may have n TF-IDF values. Therefore, if N books are included in one calculation unit and n types of words are listed in the analysis areas of N books, the TF-IDF value in one calculation unit (one popular category) is N×n dogs can be produced.

Feature vectorization is a digitization method that converts text into features with vector values. Among them, the term frequency-inverse document frequency (TF-IDF) method calculates the importance of each word in a specific document in a document set consisting of multiple documents. . TF-IDF has the advantage of better reflecting the characteristics of each word by not only assigning additional points to the higher the frequency of words in a specific document, but also giving a penalty to words that appear frequently in multiple documents.

The cosine similarity calculation unit 170 may calculate the cosine similarity by Equation 2 below.

[Equation 2]

cos(θ) is the cosine similarity value between the book of the uncirculated book list and the book of the loaned book list, A _i is the TF-IDF value for word i of the book of the un loaned book list, and B _i is the above It is the TF-IDF value for word i of the book in the borrowed book list, and n is the number of types of word i.

The cosine similarity means a numerical value of similarity between vectors calculated through a cosine value corresponding to an angle between two vectors. The angle formed by the two vectors can have a value of 1 when the angle is 0°, 0 when the angle is 90°, and -1 when the angle is 180°.

Cosine similarity has the advantage of being suitable for measuring the similarity between vectors in a multi-dimensional space due to its dimension-independent characteristics.

The popular category extraction unit 150 may extract a plurality of popular categories. Specifically, a plurality of popular categories may be extracted for each member type.

The unleavened book recommendation unit 180 may rank the books in the unlendered book list based on an average value of the cosine similarity value for each of the plurality of popular categories. That is, the average of the cosine similarities between each unleased book and all borrowed books may be calculated, and the recommended rank for each unleased book may be calculated for each popular category by arranging them in descending order.

Table 4 below is a table showing an example of recommending uncirculated books calculated based on the average value of cosine similarity.

As shown in FIG. 3 , the system 100 for recommending unexamined books of the present invention may recommend unextracted books. As shown in the screen shown in FIG. 3 , the unlendered book recommendation unit 180 may send unlended book recommendation information to the user terminal 13 .

Embodiments according to the present invention have been described above, but these are merely examples, and those skilled in the art will understand that various modifications and embodiments of equivalent range are possible therefrom. Therefore, the true technical protection scope of the present invention should be defined by the following claims.

11...Book lending system
13 ... user terminal
100...unloaned book recommendation system
110 ... data storage unit
120 ... data set building unit
121...member information organization department
122... Book Introduction Organizing Department
123 ... loan information data set generation unit
124 ... integrated data set generation unit
130 ... text mining unit
140 ... topic clustering unit
150...Popular Category Extraction Unit
160...TF-IDF vectorization unit
170 ... cosine similarity calculation unit
180... unchecked book recommendation unit

Claims (10)

A data storage unit for storing collection list information, loan history information, member information, and detailed book information of a book rental system in which a plurality of books and members are registered;
a data set construction unit for constructing an integrated data set for the collection list information, the loan history information, the member information, and the detailed book information;
a text mining unit that performs text mining processing on text data from among the integrated data set;
The plurality of books are classified into a plurality of clusters by topic modeling based on spherical K-means clustering and latent dirichlet allocation (LDA) using the data processed by text mining in the text mining unit as an input value, respectively. a topic clustering unit that assigns a unique topic to clusters of;
a popular category extracting unit which classifies the plurality of books into a list of loaned books and a list of un loaned books based on the integrated data set, and extracts a top unique topic from the list of loaned books as a popular category;
a TF-IDF vectorization unit calculating TF-IDF values for each of the plurality of books for feature vectorization based on term frequency-inverse document frequency (TF-IDF) for the plurality of books;
a cosine similarity calculation unit for calculating a cosine similarity value using the TF-IDF values of the books in the uncirculated book list and the TF-IDF values of the books in the loaned book list; and
and an unleased book recommendation unit for recommending an unleased book based on an average value of cosine similarity values for a plurality of books in the borrowed book list for each book in the unleased book list. According to claim 1,
The collection list information includes at least one of book name, author, publisher, year of publication, international standard book number (ISBN), set ISBN, additional code, subject classification number, number of books, number of loans, and book registration date,
The loan history information includes at least one information of a user number, book number, KDC (Korean decimal classification), book name, author, publisher, loan date, return date, and media type for each loan case,
The member information includes at least one of user number, member registration date, date of birth, gender, postal code, book rental number and overdue number,
The book detailed information includes at least one of book name, KDC, publication date, author, publisher, book image URL, and book introduction. According to claim 2,
The data set building unit,
A book introduction organizing unit for integrating book introductions of books with the same book name for the detailed book information and deleting detailed book information for books with blank book introductions;
A member information organizing unit for combining the loan history data and the member information based on a user number (User key) and generating an age group variable as categorical data from the date of birth as discrete data;
a loan information data set generating unit generating a loan information data set by merging the collection list information, the loan history information, and the member information;
and an integrated data set building unit that builds the integrated data set by integrating the loan information data set and the detailed book information. According to claim 1,
The text mining unit
A text cleansing unit for removing numbers or special characters from text data of the detailed book information;
A noun extraction unit extracting nouns from text data among the detailed book information;
a stopword removal unit for removing stopwords from text data of the detailed book information;
and a tokenization unit that separates Korean into morpheme units and tokenizes the text data processed through the text cleansing unit, the noun extraction unit, and the stopword removal unit. According to claim 4,
The text data of the detailed book information is the book name or book introduction. According to claim 1,
The topic clustering unit,
An unloaned book recommendation system that performs spherical K-means clustering by using nouns extracted by word tokenization from book names and book introductions included in the plurality of books. According to claim 1,
The popular category extraction unit,
Based on the member information, the integrated data set is classified into a plurality of member types,
An unloaned book recommendation system for independently extracting a popular category for each of the plurality of member types. According to claim 7,
The plurality of member types are by gender or age group. According to claim 1,
The popular category extracting unit extracts a plurality of popular categories;
The TF-IDF value is calculated based on the popular category as a calculation unit.
The TF-IDF vectorization unit,
Unlendered book recommendation system that calculates the TF-IDF value by Equation 1 below:
[Equation 1]

W _i,j is the TF-IDF value for word i in book j, tf _i,j is the frequency count of word i in the analysis domain specified in book j, df _i is the number of books containing word i, and N is the number of books belonging to the calculation unit. According to claim 1,
The cosine similarity calculation unit,
Unleavened book recommendation system that calculates the cosine similarity by Equation 2 below:
[Equation 2]

cos(θ) is the cosine similarity value between the book of the uncirculated book list and the book of the loaned book list, A _i is the TF-IDF value for word i of the book of the un loaned book list, and B _i is the above It is the TF-IDF value for word i of the book in the borrowed book list, and n is the number of types of word i.

Images