Popular Git platforms host large scale software projects, containing huge volumes of source code which are difficult to understand during maintenance tasks. The source code understanding is affected by the vocabulary used for naming entities such as directories and source code files. The topic models support users to understand large collections of documents when there is too much documents. This work presents an approach for extracting topics in Git repositories of source code, using the frequent regularities in respect of the naming and structure of source code directories without necessitate reading or parsing source files. Popular open source software projects in GitHub were analyzed, the results has been evaluated attending the topic coherence.