This paper explores advanced Natural Language Processing (NLP) techniques for modelling semantic relationships through the processing of unstructured text data. Traditional methods such as co-occurrence graphs and Word2Vec embeddings have been fundamental in this area. However, the unique linking structure of Wikipedia offers an alternative approach through the generation of linkage graphs. This study investigates the efficiency and quality of linkage graphs compared to traditional methods to improve text classification and semantic representation. Preliminary results indicate that linkage graphs, although more time-consuming to generate, offer superior classification accuracy compared to co-occurrence graphs, especially for large datasets. Compared to Word2Vec embeddings, linkage graphs provide fairly precise semantic relations on small datasets, while the higher applicability of Word2Vec embeddings is associated with a much higher need for corpus material. This research contributes to the advancement of NLP by demonstrating the potential of knowledge-based linking structures for improving semantic analysis and text representation.