This repository contains supplementary source code and analysis of the master thesis project. vagga is used to create a reproducible environment. Therefore, in order to start Jupyter with the correct environment run:

$ vagga jupyter

This command will build all necessary libraries for training, testing, and analysis of used neural networks. Since dataset employed in this research was provided by Norwegian News Agency (NTB), to reuse code for training networks on other datasets, it has to be slightly adjusted.

The notebooks folder contains Jupyter notebooks used for all training, testing and analysis parts.

• metadata.ipynb contain the whole pipeline of metadata parsing, translation and restructuring. This file is the most dataset-specific. However, it can serve as an example of how the final structure should look like to work with other parts of the project.
• database.ipynb contain general analysis of the used dataset.
• training folder has several notebooks. Each of them represents one training process for each combination of category and network selection.
• testing folder contain one testing notebook where all trained networks are analyzed and compared.
• ntb folder contain additional library tools that were written used in the process.