3.1. Library Transformation: Traditional Libraries Heading into the Artificial Intelligence Era

Universities often organize their libraries to offer open information resources for users. Traditional libraries are physical institutions that serve as a repository of information resources. They are spaces where users can access print resources such as books, journals, newspapers, magazines, and other materials for educational, recreational, research, and other purposes. These libraries often have several types of spaces: quiet reading areas, individual learning spaces, and collaboration and discussion spaces. There are librarians who assist users in retrieving and utilizing the available information resources and systems (Ashikuzzaman, 2023).

Recent developments in technology have had a high impact on the library, as many users are promptly adapting to digital libraries. This is because these resources are flexibly accessible via the Internet. They are also known as online resources. User behaviors have changed, resulting in an increased number of digital reading materials and digital media. However, university libraries are still considered vital learning spaces. According to Asif and Singh (2020), traditional libraries have been turned into digital libraries due to technical developments. Libraries have a wide variety of e-resources, e-services, and other functions, as well as distributing information in e-formats to meet the information needs of modern library users.

Over the past decade, libraries have developed their activities and affordance with the use of technology. For example, augmented reality and virtual reality, emerging technologies that utilize virtual objects to enhance reality, have been used to improve the library user experience and engagement. Several types of software, applications, and systems have been developed to support the operation of the library and user education (Oyelude, 2017; Yu et al., 2023). Recently emerged generative artificial intelligence tools, such as ChatGPT, have also been integrated into the library system. According to Ehrenpreis and DeLooper (2022), using an AI Chatbot on library websites to answer questions helps retrieve or recommend information sources like a librarian.

3.2. Reading Behaviors

Various unexpected changes have occurred due to the great impact of the COVID-19 pandemic. These changes have affected many sectors, mostly in social, economic, and education domains. Consequently, people were strongly urged to stay home, and even work and study are transformed into online modes (MacKenzie, 2020). Libraries had to be physically closed. To some extent, their services were provided by online systems. At the same time, digital content became the sole option for these libraries’ users (Ćirić & Ćirić, 2021). As a result, user behavior has changed with technology that supports easier access to information, and users can access it anywhere, anytime, through the Internet. As a consequence, the rapid ongoing transition of reading from print to screen is clearly visible. This creates challenges to the use of physical books. At the same time, it creates the opportunity for digital books and reading materials due to an increasing number of digital reading devices, such as computers and laptops, tablets, smartphones, and other forms (Horstmann & Brase, 2016).

The paradigm of reading, in particular for users, is increasingly screen-based rather than paperbound. As Internet use has developed increasingly, more reading information resources have transformed into digital format, gaining more readers through this platform. In contrast, paper-based reading also has its particular benefits, such as time efficiency (Delgado et al., 2018), eye-friendliness (Tosun, 2014), comprehensibility (Duran, 2013; Jabr, 2013), and simplicity (Jabr, 2013). Regardless of the fact that some people refuse to use paper due to the increasing campaign of wasting less paper (Parlindungan et al., 2021), many people still find paper their preferred reading material. Additionally, some research has argued that digital environments may not always be best suited to fostering deep comprehension and learning. For example, Mangen et al. (2013) found that students who read texts in a printed format scored significantly better on the reading comprehension test than students who read the texts digitally. Similarly, Delgado et al. (2018) suggested that providing students with printed texts, despite the appeal of computerized study environments, might be an effective way to improve comprehension outcomes. Therefore, the role of libraries, which contain a large number of paperbound materials, are essential for education.

However, given the unexpected attack of COVID-19, users’ reading behaviors and library usages are expected to change. For instance, Ćirić and Ćirić (2021) found that the average reading time of participants from Serbia increased by approximately 130% during the lockdown. Baharuddin and Mohamad Rosman (2020) further found that information quality and responsiveness are two important factors that predicted potential library e-services utilization during COVID-19. However, these studies focused on the reading behaviors of digital text by using data gathered from a survey. There is a lack of research utilizing recorded trace data to examine the factors or behaviors of library users during the post-COVID-19 era. Hence, this present research aims to investigate how the behaviors of library users have changed from pre- to post-COVID-19.

3.3. Trace Data Analysis

Interaction with the available information systems offered by the library generates a large volume of data. This data is often used to understand the behavior of users in various fields. For example, trace data is recognized as a highly insightful source of information in education (Gašević et al., 2015). It has been reported as a more reliable source as compared to the use of self-report instruments such as questionnaires, interviews, or think-aloud protocols (Srivastava et al., 2022; Zhou & Winne, 2012). Trace data has also been used in business to understand customer behaviors (Kröckel & Bodendorf, 2012). Similarly, it was used to explore the performance conformation of the hospital process (Gao et al., 2020). As presented thus far, trace data records insightful information about users’ interactions that could be used to strengthen the understanding of user interactions with the system.

To understand the library user and analyze activity relevant to the library, trace data could be used. However, trace data records a large amount of data; a simple count might not offer actionable insights to improve an existing activity and its relevant activities. ML as well as data mining algorithms are often used to unveil the insight hidden within the trace data. ML and data mining algorithms are mathematics-based algorithms that can be grouped under the umbrella of AI. It can be programmed to order the computer to automatically retrieve, recognize, identify, classify, or predict some information from the given data set.

One of the ML algorithms used to mine trace data is process mining. Process mining enables automatically creating a process model from event logs without prior knowedge of the process and analyzing the performance of processes, such as identifying bottlenecks, delays, or resource inefficiencies. Originally, this technique was used to analyze and improve business processes by extracting information from event logs and trace data generated by information systems (Janssenswillen et al., 2019). Recent literature also applies such methods to understand the activity flow of the library process. For instance, Kouzari and Stamelos (2018) applied process mining to examine the differences in operational practices across libraries using the same integrated library system. They found that even though two organisation use the same system, there were differences in terms of practices, activities, sequences, and approaches followed in daily tasks. Such findings could be used to improve existing practices. Similarly, Dzihni et al. (2019) used process mining to examine the bottleneck in the helpdesk process of a university’s academic information system. They were able to indentify two activities in the helpdesk process that required improvement, which were activities during input, and seeing the ticket or request.

Despite the usefulness of process mining, to the best of our knowledge there are a very small numbers of research studies using process mining to examine the processes and activities of the library, especially library users’ behaviors that might have changed due to COVID-19. Therefore, this research aims to explore the changes in library users’ behaviors while using the library resources by comparing pre and post-COVID-19. The behaviors of library users were automatically recorded when they interacted with the system. Therefore, two dimensions of library usage were explored, including library physical access and facilities reservation, and online behaviors.

4.1. Library Context

This study focused on library user visitation and utilization of the resources provided by the university library. The secondary data from the system’s recorded interactions were used. The library focused on in this study is one of Thailand’s southernmost university libraries. The university has several campuses with their own libraries. The library offers services to several types of users, including academics, researchers, graduates, undergraduates, university employees, and external university users, including adults and children. There are eight main faculties offered on the campus. Library users can physically visit the library during working hours from 8:30 to 20:30 daily. During examination weeks the library offers 24-hour access. When users pass through the entry gate, the automated system will read the ID and start counting the entry.

The university’s library offers a variety of resources. This includes paperbound and digital-based reading materials, audio and video materials, computers, working spaces, meeting rooms, discussion rooms, and electronic devices (Prince of Songkla University Pattani Campus, n.d.). Some of the resources are made available via online systems. Some resources are physically available at the library. This study focused on items that were made available to physically borrow at the university, such as paper-based books, meeting rooms, and facilities.

However, borrowing items from the library must be done through an electronic system. In general, graduates and external users can borrow these resources for seven days. They can renew the borrowed resources by themselves via the online system. All borrowed items need to be physically returned to the library by a specific time or else the user will be fined. Undergraduate students, university employees, and academicians are allowed to borrow items for 30 days and renew them through the online system. In 2018, the renewing function was limited to one time per item. However, in 2022 multiple renewals were allowed. An e-mail reminder will be automatically sent to users if they approach the due date.

4.2. Data Analysis

A large volume of data is recorded via the library system. The collected data in 2018 and 2022 were used to understand the behaviors of library users during pre- and post-COVID-19. The data were integrated from several resources to analyze the user’s digital footprint left behind when interacting with the library systems. Fig. 1 presents the relevant sequence of the use of library resources.

Fig. 1

Sequence diagram illustrating sequences and supported systems relevant to this study (own data processing).

Physical entry to the library was recorded by using the university’s self-developed system, called the ASSO system (Prince of Songkla University Pattani Campus, Pattani, Thailand). Only anonymized IDs and the number of times users performed specific actions, such as accessing the library or checking out items, were used in this study. The reservations for meeting rooms, collaborative working spaces, computers, and other facilities were recorded via the room booking system. Only anonymized IDs and a number of reservation counts were used. Borrowing and returning the items are operated via the university’s self-developed system, called ALIST OPAC (Prince of Songkla University Pattani Campus, Pattani, Thailand). Log data recording the user’s interaction with the system was used. Based on the exploration, the main learning events recorded by the system included check-in items, check-out items, reminder e-mails, and renewal items. Also, some minor events were recorded, such as inserting, updating, deleting the remark, holding, canceling the hold item, and reporting the lost item. A brief description and its corresponding event are presented in Table 1.

To prevent any identification of library users, only the anonymized ID, corresponding department, type of user, events, borrowed items, and timestamp were used to understand the behaviors of the users.

The university library recorded a huge amount of data; hence, to understand the users’ behavior extensively, the ML algorithm was used to analyze the data. In this study, process mining was used to formulate the sequence of activities of library users. Before applying process mining, the trace and its sequences were explored by the bupaverse R package (Janssenswillen et al., 2019). The trace explorer function enables us to visualize sequences within the coverage ranges, most frequently observed sequences, and infrequent sequences. Then, the bupar R library (Janssenswillen et al., 2019), one of the most frequently used R libraries, was used to explore the activities and sequence of activities.

The sequence of activities was illustrated by using a directly-follows graph (DFG). DFG is the simplest and most commonly used method to formulate the process based on the occurring event (Van Der Aalst, 2012). It simply requires an activity instance, sometimes referred to as case, activity, and timestamp. Some other variables are optional when computing the diagram. To address the research aim, the sequence of activities performed by library users was formulated for each iteration of the borrowed item by each user. Hence, the case variable referred to any related events performed by a user from borrowing until returning the specific item. The activity referred to the event recorded in the system as presented in Table 1. The timestamp refers to the specific time when the event occurred.

As a result of using DFG via bupaR R library, the sequence of activities was illustrated. It is composed of nodes, numbers, and flows (arrows). Each node represents each activity, as presented in Table 1. The flows of activities were illustrated by using the arrows to represent the direction from the source activity to the next activity. Hence, the behaviors derived from user interactions prior to COVID-19 and after COVID-19 can be examined.

5.1. Accessing and Facilities Reservation

Exploration into library usage before and after COVID-19 revealed a decreased number of library users. Prior to the global pandemic, 9,674 library users physically visited the library in 2018. The majority of them were undergraduates (5,314, 54.93%), academics and researchers (3,394, 35.08%), external university users (397, 4.10%), graduates (328, 3.39%), and university employees (242, 2.50%). A total number of 192,920 counts were recorded for physical entry to the library. The average number of visits was 11.33 times. However, in 2022 after the global pandemic, the number of library users who physically visited the library decreased to 7,511. Similarly to the pre-COVID-19 period, most of them were undergraduates (4,442, 59.14%), academics and researchers (2,164, 28.81%), external university users (654, 8.71%), graduates (189, 2.52%), and university employees (62, 0.82%). The visitation counts came to 94,768 times. The average number of visits was 7.57 times. The distribution of visitation counts showed similar patterns, as presented in Fig. 2.

Fig. 2

Number of times library users physically visited the library (own data processing).

Even though the number of visits decreased, the number of meeting rooms and co-working space reservations increased. That is, in 2018, 675 reservations were recorded in the database. In 2022, the number of reservations increased to 1,552.

When looking into online behaviors (refer to Fig. 3), it was revealed that most library users were undergraduate students enrolled in the faculty of education, humanities and social sciences, and political sciences. In 2022, there was a very large increase in library users from the faculty of political sciences, the faculty of Islamic sciences, and those from functional units as well as from other campuses.

Fig. 3

Number of library users (own data processing).

5.2. User Behaviors When Interacting with the Online Library System

Fig. 4 presents the sixteenth most frequently occurring sequences of interaction exhibited by library users. These sequences covered 13% of all sequences. As shown in Fig. 4, the most frequently observed sequence was checking out items from the library and returning the items. Other patterns that frequently occurred were checking out items from the library, followed by the reminder e-mail regarding the due date, and then returning the item to the library after receiving the reminder e-mail. When looking at the whole diagram, it is clear that e-mail reminders were the most frequently occurring event when interacting with the ALIST system, as presented in Fig. 5. The node in Fig. 5 represents the type of interaction with the ALIST system. The darker shade of node color represents the higher records of such interactions. The percentage in the node indicates the proportion of interaction against all recorded interactions. The numbers located inside the node show the actual number of recorded interactions. The arrow indicates the flow of the activity from the one interaction and its following interaction. The stronger and thicker the line signifies the higher frequency of occurrence and vice versa.

Fig. 4

Sixteenth most frequently occurring activity in 2018 (own data processing).

Fig. 5

Overall activity sequence of user interaction with the online library system in 2018 (own data processing).

In 2018 (refer to Fig. 5), most library users began their activity by borrowing items (start → check-out item) from the library. Note the beginning of interactions by returning the borrowed items (start → check-in item) to the library, as this action might be a result of continued activity from the previous year. The proportion of interactions between checking-in (check-in item → check-out item) and checking-out items (check-out item → check-in item) from the library is almost equal (22.31% and 22.80%, respectively). This indicates that the user will normally borrow the items from the library and return them after some time. However, the strongest, darkest shade of nodes is the reminder e-mail (accounting for 41.86% of all recorded interactions), showing a strong direction to check-in (reminder e-mail → check-in item) and check-out (reminder e-mail → check-out item) of items from the library. This means that most of the time, library users receive a reminder e-mail regarding their borrowing, returning, and due date of the items.

Similar to 2018, in 2022 the most frequently occurring activities were checking out items from the library and returning the items to the library. This sequence accounted for 9.61% of all detected sequences. Also, library users often borrowed multiple items at a time, as indicated by consecutive check-outs followed by separate check-ins. The main difference observed was that in 2022, the number of e-mail reminders to return books was much lower than in 2018. The interaction of renewed items was also observed in the 16 most frequently observed sequences (Fig. 6).

Fig. 6

Sixteenth most frequently occurring activity in 2022, covering 31% of sequences (own data processing).

Fig. 7 presents the overall sequence of activities based on library users’ interactions when borrowing items from the library in 2022. Note that the number of interactions was slightly lower than in 2018. This is due to the smaller number of library users in 2022 as compared to 2018. Hence, the percentage of proportion is better for representing user interactions with the online library system. Similar to 2018, most library users began their interaction with the system by checking out the item (start → check-out item), accounting for 60.25% of initial interactions in 2022. Unlike in 2018, the library users’ check-in and check-out items were much more in number of items, as presented in Fig. 7. There was a strong connection between check-in and check-out items from the library (check-in item → check-out item; check-out item → check-in item). The proportion of e-mail reminders was two times lower than the proportion of reminder e-mails observed in 2018. Meanwhile, the number of renewed items was also higher.

Fig. 7

Overall activity sequence of user interaction with the online library system in 2022 (own data processing).