Vector representations of textual data play a pivotal role in modern natural language processing tasks, enabling the transformation of words, sentences, and documents into dense numerical vectors that preserve semantic relationships. This article provides an overview of the evolution of vectorization methods, from classical models such as Bag-of-Words and TF-IDF to advanced neural approaches like Word2Vec, GloVe, and contextual embeddings from BERT. Particular attention is given to the challenges of sparsity in vector spaces inherent to traditional methods and strategies to overcome them through training on large text corpora. The authors analyze the effectiveness of various word vector aggregation techniques for whole-document representation, including averaging, Doc2Vec, and attention mechanisms.