Fact from Fiction

This repo accompanies our paper to distinguish feuilleton fiction in Danish newspapers.

📝 In notesyou will find the annotation scheme for the fiction/nonfiction categorization

🧭 Directions for code:

In scripts you'll find the code, including:

• get_features.pyto get MFWs, TF-IDF, and stylistic/syntactic/affective features, the functions of which are defined in scripts/feature_utils.py.
• classify.py which employs a random forest model across our 4 different feature sets (MFW100, TF-IDF, selected features, and embeddings)
• descriptives.py which visualizes and test differences between the classes of fiction/nonfiction
• a clustering_task.py which tests embeddings for clustering feuilleton series (note that these need to be precomputed and are not available here because of size-issues)

Note that the script for creating embeddings (various) is at this anonymized repo

And that the script to benchmark SA models on the Fiction4 corpus is in this anonymized repo