1.3.1. Log Analysis and Information-seeking Behavior in the Context of the Digital Library

Libraries have traditionally performed user Transaction Log Analysis (TLA) on Online Public Access Catalogs (OPACs) for user research since the 1980s. The Online Computer Library Center (OCLC) studied log data from 1981 to 1983 by analyzing transactions and in-depth interviews. A TLA or online monitoring analysis of the interaction time and content is generated by the user from the system terminal, which is collected automatically (Rice & Borgman, 1983). Since the emergence of the Internet in 1993, TLA has been redefined as the analysis of research interactions that are electronically recorded between an online search system and users who search for information on that system (Peters, 1993). Many attempts have been made to implement log analyses in user studies. Transaction log data provide a method for data collection while logging analysis servers as a method of data research. A TLA is defined as an online system and user study method that analyzes the system’s performance and user behavior (Jansen et al., 2009).

On the other hand, online data monitoring, known as a type of TLA, is one of several ways to capture the detailed process of how users search for information at a reasonable price. TLA is a data collection technique rather than a research design. It can be used in empirical or field research either by itself or alongside other data collection methods (Borgman et al., 1996). TLA was reported as an appropriate technique for evaluating a search process because it is unobtrusive and provides a qualitative analysis of a quantitative model (Penniman & Dominick, 1980; Rice & Borgman, 1983). Log data are related to the qualitative evaluation of digital libraries and the qualitative evaluation process of interaction with services from a digital library (Koch et al., 2004). Transaction log data is also used as an evaluation tool for improving the user interface or system design. Further, it even supports qualitative evaluations of information behavior paths in an information retrieval system (Borgman, 1986; Rice & Borgman, 1983). As the information use behavior can be analyzed through log analysis, the log analysis method is argued to be an appropriate method for research in the web search field, which involves diverse, heterogeneous user classes (Park & Lee, 2007). Accordingly, studies that analyze search behaviors based on user log data have been conducted (Choi et al., 2018; Choo et al., 2000; Park, 2011; Park & Lee, 2007, 2013).

The berrypicking model as the process of acquiring information through web browsing was proposed by Bates (1989). The users may change their search query, but their searches may shift toward a new direction depending on the search results, or they may change information in each stage of their search process. Users will move from one search information source to another through various methods, such as footnotes, journal citations, reference literature, or local searches, or they may change the domain, and even when they are searching for a different information source. Agosti et al. (2012) analyzed the ten years of research publications about web service and digital library logs. This research showed that web search engine log analyses addressed matters related to representing, saving, organizing, and approaching important information items composing webpages. In the last ten years, this field of research was divided into three parts: analyzing the details of queries made by users from a search engine, how users interact with the search engine, and how the search engine organizes the results. A new log trend pertains to mobile devices, social networks, personalization, and multimedia service technology related to user information-seeking behaviors or system use (Agosti et al., 2012). A log analysis for the digital library is done by analyzing the access log statistics of the library user community to evaluate the digital library and user community (Bollen & Luce, 2002).

1.3.2. Log Analysis of Digital Library in South Korea

Many public libraries in South Korea have also utilized log data to analyze user information behaviors. For example, a study was conducted on the behavior of archive users based on weblog analyses (Lee & Yim, 2015). Another study proposed solutions for strengthening connections between websites on the National Archives of Korea and further developing the search service through analyzing search queries from users and the paths through which they end up on websites from the archives (Jin & Rieh, 2018). Finally, another study analyzed the status of library culture programs based on user participation log data through a case from the National Library of Korea in Sejong (Choi et al., 2018).

Various studies have been conducted through analysis of the NDSL. First, user log files were used to analyze the information needs and use behavior of users of the NDSL (Yoo, 2002). Search log keywords from the NDSL were analyzed, and the automatic category information of the document selected by the user and the categories between words were analyzed (Lee et al., 2012). Finally, an analysis was performed on the correlation between the categorized results of search queries and the categories of accessed documents. One study proposed a user login service policy after conducting a log analysis to promote use of the NDSL (Kim et al., 2013). Another study analyzed the characteristics of queries in one year’s worth of log data from the NDSL. In this study, 70% of the queries consisted of one or two words, and 27.29% of the users spent one minute or less on the site (Park & Lee, 2013). However, the session length of users on the NDSL was longer than the time spent on general commercial search services. Moreover, log data from the NDSL reflected time passing over three years. The queries and browsing requests of long-term users were analyzed, and it was found that while the simplicity of queries and shortness of usage duration remained the same as time passed, the research subject seemed to change with the passage of time. Search queries with a high frequency of use were analyzed, and the information use behavior was analyzed (Park & Lee, 2016). Since log analysis for libraries or information services in Korea primarily constitutes research regarding statistics on the number of uses or search terms, studies are needed to analyze log data from various aspects to understand users’ information behavior. Therefore, different from the previous studies of log data of the NDSL, this study focused on revealing users’ information behavior rather than search terms. Therefore, the utilization of log data could be expanded.

3.2.1. Information Seeking Path Analysis: Service-main-class and Content-main-class

Usage frequencies were studied to determine the relationship between variables in the information-seeking path ‘Service-main-class’ and ‘Content-main-class.’ Accordingly, a research hypothesis was formulated, and cross-tabulation analysis was performed using the χ² Test to determine a group difference or correlation between the attributes of ‘Service-main-class’ and ‘Content-main-class.’

In order to determine the statistical significance of ‘Service-main-class’ and ‘Content-main-class,’ the log data for ‘Service-main-class’ of January 12 was tested. The items ‘search,’ ‘detailed view,’ and ‘full-text view’ were observed to be the most frequently used nominal variables. On the other hand, the more frequently used items in ‘Content-main-class’ were ‘Journal,’ ‘Paper,’ ‘Patent,’ ‘Report’ and Researcher.’ Based on this sample data, a cross-tabulation analysis was conducted to test the independence of ‘Service-main-class’ and ‘Content-main-class.’ The results of the analysis are presented in Table 5. Furthermore, for the test of significance between ‘Service-main-class,’ and ‘Content-main-class,’ a cross-tabulation analysis was performed, and the χ² value was observed to be statistically significant at the 0.05 level.

As presented in Table 5, ‘Detailed view’ was most frequently accessed in ‘Service-main-class,’ followed by ‘Full-text view’ and ‘Search.’ In ‘Content-main-class,’ ‘paper’ was the most intensely used item, which is also observed to be the most widely used item in ‘Search’ and ‘Full-text view.’ This demonstrates that most NDSL services are related to theses and dissertations. This finding can also be confirmed by the log data analysis of the information-seeking paths. The item ‘Paper,’ which pertains to the content available in the NDSL services, is utilized with various services. Therefore, it is necessary to develop strategies to create more diversified services that use such contents (theses and dissertations), or to identify any negative user experience in existing services to improve the contents or services from the users’ perspective.

3.2.2. Information Seeking Path Analysis: Service-sub-class and Content-sub-class

This analysis investigates and identifies users’ information-seeking paths in the log data’s ‘Service-sub-class’ and ‘Content-sub-class’ shown in the log data. Lower in the hierarchy, located in the lower layer of the corresponding main classification, the sub-classifications of services or contents can be reached by taking a path through the ‘Service-main-class’ or ‘Content-main-class.’ To determine if there is a group difference or correlation between the attributes of ‘Service-sub-class’ and ‘Content-sub-class,’ a research hypothesis was formulated and then a χ² test, i.e., a cross-tabulation analysis was performed.

Using the log data of January 12, 2018 as the sample data, a cross-tabulation analysis was performed to test for the group independence of ‘Service-sub-class’ and ‘Content-sub-class.’ The results are presented in Table 6. The results of the cross-tabulation analysis, which was conducted to test the significance of ‘Service-sub-class’ and ‘Content-sub-class,’ showed that the χ² value was observed to be statistically significant at the 0.05 level.

For ‘Content-sub-class,’ ‘Detailed view’ was the most frequently used in 3,541 out of 9,073 cases. Users frequently used ‘Detailed view’ contents, such as ‘Domestic paper,’ ‘International paper,’ and ‘Theses and dissertations.’ ‘Detailed view’ is a high-frequency service because it allows users to view abstracts and bibliographies or information about authors and references, as well as the full-text version of texts. A close look at the frequency of ‘Service-sub-class’ and ‘Content-sub-class’ for each information-seeking characteristic of a path reveals that ‘all papers’ of ‘Content-sub-class’ made through ‘Quick search’ in ‘Service-sub-class’ was the most frequently used item.

In this path, a brief list of all papers is displayed as the results for ‘Quick search,’ making it suitable for users who prefer performing a quick search. Another ‘Service-sub-class’ with a high usage frequency of ‘Domestic paper’ was ‘PDF full-text view.’ This path indicates the users’ tendency to download a full-text paper after quickly searching and retrieving information with search terms that suit their information needs. In the information-seeking path for ‘Service-sub-class’ and ‘Content-sub-class,’ ‘Domestic paper’ through ‘Detailed view’ was also popular among users. ‘Theses and dissertations’ through ‘Link to other institutions’ for full-text were further observed to be popular among users. User information-seeking paths to access the contents available in other institutions can be observed from ‘link to other institutions for full-text,’ which is a popular item. Therefore, the process of satisfying user information needs could be identified along the path by analyzing the log data.

3.3.1. Demographic Statistics of Identified Users

The demographic characteristics of users’ subjects and preferred information types, identified through the contextual analysis of log data are demonstrated in Table 7. The number of male users was 2.75 times larger than the number of female users among the 30 users selected for the analysis. As for the users’ academic fields, seven users (23%) pursued electrical engineering, four users (13%) were in bioengineering, three users (10%) were in computer engineering, and three users (10%) were in medicine. The 30 logged-in users had 607 pageviews altogether. The 30 identified users and their log data were analyzed to assess whether they demonstrated individually unique information-seeking behavior in their circumstances and/or contextual information needs. The preferred types of information were 246 cases (40.54%) for ‘Theses,’ 305 cases (50.25%) for ‘Report,’ and 31 cases (5.11%) for ‘Patent.’ The type of information NDSL users preferred was analyzed by reflecting the cases where an individual requested more than one type of information. The frequency of ‘Patent’ or ‘Journal’ was also not high. On the other hand, 305 cases (50.25%) of ‘Report’ indicate that ‘Report’ was the most frequently used. The result showed that the most frequent use of content by identified users was ‘Report.’ The identified users had a higher preference for ‘Report’ than for ‘Paper.’ This shows a different aspect from the overall usage frequency statistics of NDSL in terms of log pattern analysis.

3.3.2. Visualization of Log Context Analysis

This study visualized user information-seeking behavior using log data, since log data was an electronic code and showed the user information-seeking behavior that the user interacted with while accessing NDSL. From the perspective of log analysis, log context analysis included user information and log pattern analysis. It would be desirable to analyze the user’s information-seeking behavior based on the model, improve the systems, and provide the preferred information to all users. Since most users accessed the NDSL without logging in, it will be possible to provide various services only after analyzing personal information and researching the use of log information of users who have accessed the system after logging in. Only a few steps are currently presented, but adding various elements will make it possible to create subdivided service conditions that satisfy users.

3.3.3. Creating Visualization Diagram of Log Data

The visualization diagram was applied to User 27’s cases, as shown in Fig. 1. This figure visualized the flow of User 27’s behavior as time passed when they accessed the NDSL and used the services and content that leave log data. The x-axis shows pageviews, and the flow of time is seen from left to right. The y-axis shows services and content as one set and is displayed as arbitrary numbers. For example, User 27’s first pageview in Fig. 1 is labeled 33. Label 33 here means that the user checked the ‘Detailed view’ and ‘Report,’ as shown in the center-right of Fig. 1. The user’s second pageview was 40, which shows that the content report was viewed through ‘Full-text view.’ This figure visualizes how User 27 accessed services and content through 21 pageviews.

Fig. 1

Visualization diagram of log context analysis: example of User 27.

Microsoft Excel (Microsoft, Redmond, WA, USA) was used to visualize the flow of time, content, and services for the visualization diagram. Given the nominal scales of log data, services and content were bundled as shown in Table 8, and an arbitrary number was attached to each pair to create an Excel cross table that visualizes this data. Combinations of seven Service-main-classes and seven Content-main-classes (7×7) resulted in 49 types of service and content pairs, each of which was given an arbitrary number from 1 to 49. Fig. 1 was created based on these categories.

The columns in Table 8 are the commonly used service types, and the rows at the top of the cross table are the list of user content types. The number in each cell shows that the service on the left and the content on the right were used. The service users’ patterns of using NDSL can be seen by looking at the category to the left of the number to understand each user’s combination of services and content. The content can be seen in the content category from the rows above. For example, the service and content that the user accesses can be understood by checking the number on the line graph in Fig. 1, and then finding the number it corresponds to. For example, number 49 means that the user searched to look up a ‘paper’; the number 42 means that the user looked up a ‘paper’ through ‘full-text view.’ This information is indicated as numbers on the diagram, showing which service or content was used from the relevant pageview.

3.3.4. Application of Visualization Diagram Targeting Identified User on NDSL

The visualization diagram of log context analysis was used for 30 cases as shown in Fig. 2. The lines in the line graph represent the information-seeking behavior of 30 users; these 30 users had different use patterns, and each of their unique information-seeking behavior was visualized through a line. This is consistent with the fact that the information-seeking behaviors of users, which are regarded as the most critical factor in the user study model, are complex according to the situation, context, and social environment and are dynamic according to the context and process. Therefore, the users’ information-seeking behaviors may be understood better through a user model rather than a technical description. The x-axis that represents time in the search formula indicates users’ information-seeking behaviors over time. The y-axis represents a pair of services and content, and uses an arbitrary number based on the user’s type of service and content the user availed. For example, User 3 viewed two ‘Papers’ through Search’ before ending their search. User 2 only looked at ‘Paper’ using services and content. This shows constant changes in information-seeking behaviors in the line regarding individuality, circumstances, and context, as pointed out in various user studies.

Fig. 2

Visualization diagram of context analysis: 30 users.

In the future, a method that can group these use behaviors while emphasizing each user’s individuality will be sought. With respect to content, users could be grouped into users who prefer ‘Papers,’ users who prefer ‘Reports,’ and so on. In terms of services, although there are various types of users, such as those who only use ‘Search’ and ‘Full-text view,’ users who typically use ‘search,’ ‘Detailed view,’ or ‘Full-text view,’ and users who use customized services like ‘Save’ after logging in, if users can be grouped, it would be best to establish a way to provide services based on these groupings. If services can be provided based on understanding the information behavior of users, it will become easier to help users fulfill their information needs.