Online First

Table 1 shows the bibliographic metadata acquired from CrossRef. The bibliographic metadata is composed of journal, journal_issue, journal_article, title, and citation list. The journal includes journal title, the abbreviation of the journal title, ISSN, DOI, and resources. The journal_issue includes issue information, DOI, and date of publication. The article includes article title, author, the first page, and the last page of the article.

Therefore, the schema is designed to focus on journal, journal_issue, article, reference information, and co-authors to configure the bibliography system in this study. In particular, because of the large amount of articles, bibliographic information, and co-authors, the system is designed in consideration of variation.

As shown in Table 2, it is essential that the journal schema uses journal title, the abbreviation of the journal title, ISSN, and DOI (Digital Object Identifier). In particular, because abbreviations of the journal titles are generally used in bibliographic data, it is essential to have them for future journal matching. Because authors generally use journal title abbreviations to write their bibliography, they must be managed together. Journals can be found more accurately with ISSN or DOI, rather than with a journal title, because a journal title can have variations resulting from errors or journal title changes.

As shown in Table 3, the article schema information is composed of article title, year and month of publication, the first page of the relevant article, and DOI. It also has the journal key value for the relevant article and the key value of the issue information, in order to determine the relevant journal and issue information quickly. A key value is assigned within the architecture to enable journals, issues, and bibliographic information to be found easily, focusing on articles, as described above.

Table 4 shows the architecture of bibliographic data schema. That is, it is composed of bibliographic data values of the relevant article, for example, bibliography number, journal title, ISSN, author, and the first page of the article. If there is a doi value in the bibliographic data provided by CrossRef, other field values are provided blank. Therefore, if there is a doi value only to provide the bibliography service, it is necessary to find the relevant article to obtain information values including journal, article title, and year of publication and then to enter data values. It is also necessary to have an article key value for linking the relevant article to the bibliography list.

The recent data boom contributes to a great number of systems (150) for supporting NoSQL (Not Only SQL). For example, exemplary systems include the column-oriented storage system, Hadoop/HBase and Cassandra, the document-oriented storage system, MongoDB and CouchDB, and the tuple-oriented storage system, DynamoDB and Redis. The aforementioned NoSQL systems have their own weaknesses and strengths. Therefore, it is very important to select an ideal system for the intended purpose.

In this study, MongoDB, a document-oriented storage system, is used for the bibliographic data, to manage data in document-wise manner while handling large amounts. MongoDB provides data sharding, called auto-sharding. Data sharding is essential to handle large amounts of bibliographic data. Sharding adds a device for addressing increased load and data without an impact on applications, in the case of frequent writing or if more disk space is required. MongoDB supports the replica-set to recover stored data safely.

Fig. 1. 
Architecture of sharding system

In Figure 1, the bibliography system is composed of 3 Mongod servers, 3 Config servers, and one MongoS server. The MongoDB developers recommend adding the replica-set server in order to implement stable system services. First, the cited analyzer reads article data to examine DOI values. If there are DOI values, it uses the DOI identifier to search for citation data, and then to test cited data matching. It then stores only the matching information in the Cited DB.

Table 5 shows the architecture of DB collection in the bibliography system. Therefore, the system is composed of the JOURNAL collection with the information for journals, the ISSUE collection with the issue information for the journals, the ARTICLE collection with article information, the CITATION collection with bibliography information of the articles, and the CITED collection with cited article information.

In this study, ID keys are essential for connecting and identifying collections. For example, it is essential that an article has more journal and issue information, and connection information. It is designed to have connection information for citing articles, and connection information for cited articles. Therefore, Figure 2 shows the relation between citing and cited. That is, the citing information means the reference data cited by an article, and the cited information means the articles which cite the relevant article.

In Figure 2, the cited collection has both citing information and cited information. This capability is used to provide navigation between authors or articles in the design.

Fig. 2. 
Relation between article citing and cited

The constructed CrossRef bibliographic data includes 22,000 journals, 28 million pieces of article collection data, 266 million pieces of citing collection data, 28 million pieces of cited collection data, and other analysis-type data, which makes 300 million data in total.

An analysis of the bibliographic data provided by CrossRef reveals that an article has approximately 10 bibliographic entries on the average. In this study, the system is designed to do auto-sharding for the collections of large articles, citing and cited, as shown in Table 6. General MySQL and data loading took 7 days and 9 hours, but MongoDB data loading took 2 days and 2 hours. The system designed as described above implements faster data loading than general RDBMS, and data search speed is also faster.

Figure 2 shows how to analyze cited bibliography information. First, the cited analyzer reads article data to examine DOI values. If there are DOI values, it uses the DOI identifier to search for citation data and then to test cited data matching to store only the matching information in the Cited DB.

Fig. 3. 
Cited analyzer flow chart

Second, if there is no DOI value, it uses data including journal ISSN, the year of issue, and the first page of the article to carry out cited data matching, and stores only the matching information in the cited DB. Stored information includes journal key, article key, cited journal key, and cited article key.

In this study, it is possible to create a journal report to calculate the Impact factor by using the cited mapping information. The system is also enabled to create networks between journals and article authors by using the cited information, and also to create a map between cited and citing articles. Therefore, the cited information is the most important factor to use the bibliographic data.

Fig. 4. 
Bibliography system user interface

In this study, 300 million pieces of CrossRef bibliographic data are used to build a NoSQL-based system as shown in Figure 4, and to develop an algorithm for analyzing data quickly. A method of various citation services is studied by using bibliography citing and cited information.

Figure 5 shows the greatest strength of using bibliographic data by using cited bibliographic data to reflect on the search rank algorithm. The result is shown in the higher rank of cited information for the searched articles by users. However, because articles of higher recognition in the past have more cited information, it is necessary to consider the latest years to reflect them on the search rank algorithm.

Fig. 5. 
Search rank algorithm

Figure 5 shows the visualized ranks for cited information and years with the axis X for document weight and the axis Y for years, to design a system which enables users to select search results in consideration of weight and years.

When a user selects View for the searched article, the citing bibliographic information of the relevant article is provided, and the information of articles cited by other authors for the relevant article is also viewed, as shown in Figure 6. The trend of relevant cited article is viewed by showing the cited information for each year. Therefore, it is easy to examine how many times the relevant article has recently been cited, and was cited in the past.

Fig. 6. 
Trend of cited and citing/cited information

With respect to the above article, it is shown that it continued to be cited in the past, but the level of citing the article has recently dropped.

Figure 7 shows navigation through article authors. All articles have citing information. However, they may not have cited information. Therefore, when a user searches for articles, they can be visualized in network nodes focusing on citing and cited information.

In this study, it is assumed that “an author more frequently cited has better recognition in the relevant field” in order to examine an author’s impact. This service is used to examine the relationship between authors or the impact of the relevant author.

Figure 7 shows the result of citing and cited navigation among authors for the exemplary articles searched with “health care”. The words in red are cited information and those in blue are citing information, with respect to implemented navigation through citing and cited authors.

For real-time search, the schema is configured to keep connection information focusing on articles, the ID key values of citing and cited collections, shown in Table 5. The user interface is designed so that users can control the number of citing and cited articles in real time.

Fig. 7. 
Network among authors by using cited information

Most users want to find articles and journals they want with a few keywords. Therefore, the most ideal journals can be recommended by using user’s keywords to know which journal publishes articles focusing on them and to use the cited index of the relevant journal.

Figure 8 shows how to group and visualize the search results focusing on journals of higher cited indexes and which journals publish articles related to the relevant keywords by using the sample keywords “Health Care”. The result is that the Health Affairs (in blue) has recently published many articles for the relevant key word, and has a higher citation index.

In this study, the user interface is designed to allow sampling of search results to be controlled between highest 1 to 100%. The 10 highest journals are grouped to visualize the result as shown in Figure 8. The result shows analyzed result for each year with respect to Health Affairs, American Journal of Public Health, and Community Health.

Fig. 8. 
Finding a relevant journal

The citation analysis service is the same as the citation analysis report in that they suggest bibliometric analysis and citation indexes for analyzed articles, but different from each other depending on whether the analysis is carried out in real time, or in a particular time range. That is, the citation analysis service is for bibliometric analysis of academic literature searched by using keywords. The citation analysis report is for major academic journals published in a given period, bibliometric analysis reports and indexes in the unit of institutions and countries. In this respect, Web of Science and SCOPUS correspond to the citation analysis service. JCR and SCIMAGO correspond to the citation analysis report. The most important service for using the bibliographic data is the citation analysis service and report.

Figure 9 shows how to calculate the Impact factor by using the cited frequency for the last 2 years. It is enabled to calculate the Immediacy index and also the Impact factor for the time period from minimum one to 5 years.

In this study, the Impact factor for journals for each year is obtained for 22,000 journals by establishing the 300 million bibliographic data. Therefore, more bibliographic data would enable more accurate Impact factors to be obtained.

Fig. 9. 
Impact factor for journals

In this study, the cited information is used to make a network among article authors and a network among journals. Figure 10 shows the results of using the cited information to make a journal map. This uses the citation index information to visualize the network relationship among journals. Therefore, the relation information among journals is known.

Fig. 10. 
Relationship among journals by using cited information

For example, the allergy journals have information interconnected with expert opinions on therapeutic patents, hematologic disorder journals, blood journals, AIDS, and the like.

In this way, users can know which journals are interconnected for studies by using the citation information among journals. It is also possible to classify topics of the above journals to analyze interconnected studies for which subject areas interact.