TEXT MINING AND TOPIC MODELLING FOR KNOWLEDGE DISCOVERY IN  DIGITAL COLLECTIONS

Munyao V.W., Ndung’u H., Wasike J., Tsinale H.

Munyao V.W., Ndung’u H., Wasike J., Tsinale H. Kirinyaga University, KENYA

Keywords: text mining, topic modelling, Latent Dirichlet Allocation (LDA), TF-IDF, knowledge discovery, digital libraries, Kenyan academic libraries

Abstract

Kenyan research and academic libraries have been experiencing challenges managing the growing collections of digital scholarly publications making it hectic for librarians to systematically organize
resources and users to access relevant content efficiently. Text mining has evolved as a strong method for obtaining insights from large textual sets of data, allowing for thematic analysis, automated
knowledge discovery, and enhanced organization of content. Nevertheless, its usage in the Kenyan academic and research library context remains limited. This study explored text mining and topic
modeling application on scholarly publications using open access digital libraries. The study sought to identify research patterns, establish latent themes, and express how these techniques could enhance
accuracy in knowledge access, management of time, and thematic organization in the context of Kenyan academic and research libraries. The study involved collecting and preprocessing a corpus of library
science and digital library articles. This included tokenization, stop word removal, and standardization. Using word cloud visualization, prominent terms were identified, while TF-IDF analysis displayed
distinctive keywords. Dominant topics along with related document groupings were done using Latent Dirichlet Allocation (LDA), permitting insight into thematic constructs across the corpus. Analysis
showed existence of three major topics that include digital library systems, open access and repository management, and scholarly publishing practices. We identified frequent terms like “library,” “digital,”
“metadata,” and “repositories”, stressing on the use of text mining technology-driven library systems. The techniques expressed that text mining could systematically support automated classification,
display latent trends, and facilitate resource discovery in libraries. This study showcases the capacity of text mining to improve content organization and knowledge discovery globally, hence should be
practiced in Kenyan academic libraries. Future studies need to explore integration of AI driven text mining with library systems to further improve thematic analysis and user accessibility.