Yadav, DipendraTonkin, EmmaStoev, TeodorYordanova, KristinaKlein, MaikeKrupka, DanielWinter, CorneliaGergeleit, MartinMartin, Ludger2024-10-212024-10-212024978-3-88579-746-3https://dl.gi.de/handle/20.500.12116/45196The rapid increase in research publications necessitates effective methods for organizing and analyzing large volumes of textual data. This study evaluates various combinations of embedding models, dimensionality reduction techniques, and clustering algorithms applied to metadata from papers accepted at the ARDUOUS (Annotation of useR Data for UbiquitOUs Systems) workshop over a period of 7 years. The analysis encompasses different types of keywords, including All Keywords (a comprehensive set of all extracted keywords), Multi-word Keywords (phrases consisting of two or more words), Existing Keywords (keywords already present in the metadata), and Single-word Keywords (individual words). The study found that the highest silhouette scores were achieved with 3, 4, and 5 clusters across all keyword types. Principal Component Analysis (PCA) and Independent Component Analysis (ICA) were identified as the most effective dimensionality reduction techniques, while DistilBERT embeddings consistently yielded high scores. Clustering algorithms such as k-means, k-medoids, and Gaussian Mixture Models (GMM) demonstrated robustness in forming well-defined clusters. These findings provide valuable insights into the main topics covered in the workshop papers and suggest optimal methodologies for analyzing research metadata, thereby enhancing the understanding of semantic relationships in textual data.enKeyword ExtractionClustering TechniquesDimensionality ReductionARDUOUS WorkshopNatural Language ProcessingContextual EmbeddingsResearch Metadata AnalysisText/Conference Paper10.18420/inf2024_371617-5468