Machine Learning Approaches to Predicting Citation Dynamics in Academic Literature on “Library Anxiety”

Abstract

Library anxiety is considered a psychological problem that affects the proper use of library resources by students. As research in this area has increased, the prediction of citation trends has become important. In this study, citation patterns in library anxiety research were explored by using Machine Learning (ML) methods. Citation data were collected from the Web of Science Core Collection for articles published between 2014 and 2023. Different ML models, such as linear regression, Support Vector Machine (SVM), decision tree, neural network, and Long Short-Term Memory (LSTM), were tested. Two predictor variables (article age and abstract sentiment) were used to predict the target variable (citation count). It was expected that the emotional tone of an abstract might affect citation growth. However, the results showed very poor prediction performance in all models. A very weak negative relationship (-0.05) was found between abstract sentiment and citation growth. Even advanced models like LSTM produced large prediction errors. The findings showed that citation prediction could not be done accurately with only simple features. Important factors like author reputation, journal quality, and social discussions were not included in the models. Therefore, it was concluded that richer and more dynamic data are needed for better citation prediction in future studies.