Text mining, a subfield of artificial intelligence, uses natural language processing to turn non-structured text into structured text suitable for machine learning algorithm analysis. The goal of text mining is to discover new and extract most important information from the large digital corpus in a quick and automated way. The thesis describes the most important steps in data mining and natural language processing. Text preprocessing is done through transformation, tokenization, filtering, and normalization which proved to be extremely important steps. Descriptions of algorithms for stemming and lemmatization provide information to understand why lemmatization is a better approach than stemming. Some other algorithms that are described are associated with a bag of words and why the TF-IDF factor is better than word weight at determining key words. The two most used algorithms for theme modelling are LDA and LSA. LDA and LSA mostly give very similar results which were shown to be true in the practical part of the thesis as well. The most effective standard process model for text mining is CRISP-DM, which was also used in practical part of the thesis to make a model for deep text analysis. The goal of the practical part of the master's thesis was to study how effectively a large and extensive analysis can be made by using text mining on a small and morphologically complex language such as Croatian. The analysis was made on a corpus composed of news articles from the five most read news portals. The model was created by using the "Orange Data Mining" tool for text mining. The foundation of the analysis is the key term "cijepljenje" (Croatian word for vaccination) which is often controversial in society and is thus an often and recurring theme in news articles. Analysis has shown that articles discussing vaccinations were written with negative sentiment and often more in political, rather than medical context.