Experimental Study of Clustering Approaches for Text Embeddings

With the rapid growth of text data, effective automated analysis methods are essential. Text clustering helps uncover patterns and structure large datasets, but existing algorithms face challenges due to high-dimensional text representations from language models. This study compares clustering algorithms to evaluate their strengths, weaknesses and applicability. The research examines partition-based, hierarchical and density-based clustering, analyzing k-means, agglomerative clustering and DBSCAN on the 20Newsgroups dataset. Performance is assessed using key clustering quality metrics. Results show that k-means effectively captures data structure, while DBSCAN struggles with high-dimensional text data. The study highlights the impact of distance metrics on clustering outcomes and provides recommendations for improving algorithm performance in text analysis. This research is valuable for professionals in automated text processing and machine learning. It offers practical insights into selecting suitable clustering methods based on task requirements and data characteristics, enhancing text analysis in fields such as IT, marketing and analytics.