threads-of-subtlety

Code repository for the paper: "Threads of Subtlety: Detecting Machine-Generated Texts Through Discourse Motifs" (ACL 2024)

Some preprocessing updates were made, which may have contributed to performance improvements on the OOD and OOD-Para test sets.

	HC3	MAGE-Test	MAGE-OOD	MAGE-OOD-Para
Longformer-Plain	96.74%	88.64%	69.48%	55.95%
Longformer-Motif	97.42%	91.83%	82.63%	76.15%

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
data/motifs		data/motifs
tos		tos
.gitignore		.gitignore
0_prepare_and_parse_datasets.py		0_prepare_and_parse_datasets.py
1_add_graphs_to_datasets.py		1_add_graphs_to_datasets.py
2_extract_single_triads.py		2_extract_single_triads.py
3_extract_double_triads.py		3_extract_double_triads.py
4_extract_triple_triads.py		4_extract_triple_triads.py
5_add_motif_dists_to_datasets.py		5_add_motif_dists_to_datasets.py
6_split_hc3_dataset.py		6_split_hc3_dataset.py
7_train_longformer.py		7_train_longformer.py
8_evaluate_longformer.py		8_evaluate_longformer.py
9_prepare_and_parse_ood_sets.py		9_prepare_and_parse_ood_sets.py
LICENSE		LICENSE
README.md		README.md
adjust_json_file.py		adjust_json_file.py
merge_motifs.py		merge_motifs.py
multi_gpu.yaml		multi_gpu.yaml
requirements.txt		requirements.txt

Provide feedback