deep_detektor

Automated Factual-Claim Detection in Danish Broadcasting

This project is part of the Deep Learning course at Technical University of Denmark, taught by professor Ole Winther. The project revolves around an automated "claim"-detection system based on subtitles from the danish TV-show "Debatten". The found claims are used later by the fact-checking TV-program "Detektor".

Data is currently not publicly available. Contact one of the authors if you want access to it.

Check the wiki for references and other good stuff

Team members: phav, jepno, jehi and sfvn at dtu.dk

Running system

Rules for ensuring runs of files:

1. All paths are relative and stored in a file called project_paths.py.
2. The runable files are to be placed in the run_files-directory and can be run with the main folder (deep_detektor) as working directory.
3. Data is stored in a data-directory next to the repository (../data from working directory).
4. The two directories DeepFactData and DRDetektorAutomaticFactChecking are to be copied directly into the data-directory.
5. The final structure of the program-files is thus:

./
    deep_detektor/                          <- Repository
    data/
        DeepFactData/                       <- From data-service
        DRDetektorAutomaticFactChecking/    <- From data-service

Programs can be run from a command prompt in the main directory deep_detektor and running command

$ PYTHONPATH=. python run_files/single_train.py

alternatively

$ CUDA_VISIBLE_DEVICES=0 PYTHONPATH=. python run_files/single_train.py

where run_files/single_train.py can of cause be replaced by a different wanted script.