Analysing Library and Information Science Articles Using Topic Modeling Approaches

Identifying trends in research through co-citation or content analysis of journal contents is quite a common practice in LIS research. In this study, however, we proposed the Latent Dirichlet Allocation (LDA), a popular topic-modeling approach for identifying research trends of published articles in three Scopus-indexed Indian LIS journals. A total of 1213 titles & their abstracts published between 2011 and 2022 have been considered. From these data, a corpus of frequently used 15 key phrases was identified from each journal using Count Vectorizer and then ten topics having higher coherence scores were extracted from each journal corpus using LDA techniques to understand to what extent these topics are different in these journals. The analysis of the study indicates that ‘Library users’ studies’ especially in academic libraries; and ‘bibliometric indicators for measuring research growth are a few common topics in these journals and, technological innovation; utilisation of electronic and print information resources; library management; or network analysis are some of the topics that are journal specific. From the t-SNE visualisation and pyLDAvis diagram, it was seen that the topics of DJLIT are significantly unique with discrete distributions than the other two journals. On analysing the growth of the top ten topics longitudinally, it was seen that research on digital libraries, analysing the global output, online search strategy, ranking universities, etc. are concurrent interests of research among researchers while academic library resources, including electronic resources and its use, open access are among diminishing research interests of authors. Since the topic-modeling approach can provide results devoid of bias, it can be used to identify research land scape longitudinally as well as obsolescence of topic in a domain.