This Master Thesis explores an original solution to analyze text document collections. This method is constructed with methods from Data Mining, Text Analytics and Social Network Analysis. The basic idea underlying it is to create a network in which words are entities and relationships represent how often these words are used together, with the aim to find topics as words communities. After a concept definition part and an analysis of the benchmark dataset with existing techniques, the development of the method is explained. First, the parsing of the document is presented. Then, the method of creation and selection of the links, which is based on frequent item set mining, is exposed. Finally, the results of the analysis with both visualization and community detection algorithms are compared with results obtained using existing techniques.