We here demonstrate how two types of NLP models - a topic model and a word2vec model - can be combined for exploring the content of a collection of Swedish Government Reports. We investigate if there are topics that frequently occur in paragraphs mentioning the word "democracy". Using the word2vec model, 530 clusters of semantically similar words were created, which were then applied in the pre-processing step when creating a topic model. This model detected 15 reoccurring topics among the paragraphs containing "democracy". Among these topics, 13 had closely associated paragraphs with a coherent content relating to some aspect of democracy.