Examination of user response states to information center services using a data mining approach

In contrast to the conventional library service paradigm, the contemporary landscape has witnessed the emergence of multifaceted and extensive volumes of information that are readily accessible to actual users. This paradigm shift has engendered a myriad of challenges for libraries, whose imperative is to facilitate user access to information. Furthermore, the complexities faced by users in their quest for information often lead to premature abandonment or cessation of the online information access process. The capacity to forecast, at an incipient stage, the behaviors of users likely to disengage from the research process and their potential failures in information retrieval is vital for optimizing library utilization. Through the present study, data pertaining to the usage of library resources and the processes of information access among library users were meticulously analyzed and subsequently transformed into actionable insights. Drawing upon the insights gleaned from this analysis, a pragmatic model proposal was formulated with the objective of mitigating or alleviating the challenges users encounter in their pursuit of information access. Within the scope of this study, the university library automation system, where data related to users' research processes were collected, was used for research purposes. The data used in the study were obtained from large web log files accumulated in this automation system. Modeling the performance of users in terms of research success, predicated upon variables derived from the online environment, is of paramount importance for the early identification of individuals who exhibit a propensity to abandon their research endeavors and may experience potential failures at the culmination of the research process. The resultant models can serve as instrumental tools for the automatic categorization of users by assessing their levels of activity within adaptive online research environments or for implementing automated modifications. The study consists of user data covering an 18-calendar-month period between January 2022 and June 2023. The data were obtained from the Ferit Melen Central Library Automation System at Van Yüzüncü Yıl University. The log files, recorded daily by the automation provider, were collected in a total of 459 text files containing daily operations. After converting the log files into an Excel file, the data consisted of 328,100 rows in total. Each action was represented by a unique identification code specific to users (user IP), ensuring anonymous tracking at an individual level. A data analysis was conducted based on the total usage data to identify users at risk of leaving the search process. The activity variables of users were calculated. Additionally, various measurements were created by comparing user actions with those of other users in the dataset, and 18 variables reflecting search behavior were identified. Users' research performance was hypothetically coded as "passed" – "failed." vi Within the parameters of this investigation, the automation system of the university library, which facilitated the collection of data pertinent to users' research activities, was employed for scholarly inquiry. The data utilized in this analysis were derived from extensive web log files amassed within this automation framework. The modeling of users' research success performance, predicated upon variables extracted from the online domain, is of paramount importance in forecasting individuals who are likely to disengage from the research endeavor and identifying potential failures at an incipient stage. The resultant models are amenable to automatic classification of users within adaptive online research contexts or to facilitate automated modifications predicated on the intensity of activities present in the environment. In addressing the initial research inquiry, various classification algorithms and preprocessing techniques were systematically compared to ascertain the most efficacious algorithm and variables for predicting users' research success performance. In relation to another research challenge concerning success, an investigation was conducted to ascertain whether users' research performance could be anticipated in the earlier weeks leveraging these variables and the selected algorithm. For the concluding research inquiry, user cohorts exhibiting analogous behavioral patterns within the online learning milieu were delineated, and their correlation with research success performance was scrutinized. To unearth latent patterns and insights embedded within databases, widely recognized data mining methodologies were employed. Classification algorithms were deployed for predictive analytics, while clustering algorithms were utilized to identify akin user groups. In order to generalize the predictive outcomes and ascertain the optimal number of clusters within the clustering analysis, the cross-validation technique was instituted. The outcomes of the research elucidated that the online search interaction data of users could be adeptly leveraged to predict research success performance. The highest rate of accurate classification was accomplished when the data were discretized utilizing the equal-width binning methodology, and the top ten variables were selected based on the ReliefF scoring system. In this instance, the Support Vector Machine (SVM) algorithm successfully classified 99% of users. Upon examining the analyses pertaining to the predictability of users' final research success performance from earlier weeks, it was determined that such performance could be forecasted with 97% accuracy by the fourth week. Upon review of the clustering analyses results, it was discerned that users were optimally categorized into three distinct clusters based on the levels of activity they manifested within the online research setting: inactive, active, highly active. Subsequently, the interrelation between these clusters and research success performance was investigated. The findings revealed that: Users exhibiting minimal engagement throughout the process demonstrated low research performance (low search engagement), users displaying moderate engagement manifested moderate research performance (intermediate search Engagement), users with high engagement levels showcased high research performance (high search engagement). These results signify that data mining and learning analytics can be effectively harnessed to enhance user engagement and academic achievement within online research environments. Keywords: Artificial eural network, Classification, Clustering, Data mining, Decision trees, İnformation-seeking behavior, Learning Analytics, Library and İnformation Center, Library user, Personalization, User behavior