Task: Token-level binary classification for Effect spans.
Architecture: roberta-large + PEFT LoRA (RobertaForTokenClassification).
This repo: Single-marker model for Effect only.

Intended use

Research and evaluation on SemEval 2026 marker extraction.

How to load

python

from transformers import RobertaForTokenClassification, RobertaTokenizerFast
from peft import PeftModel

repo_id = "MMHusnain/semeval26-psysemev-extraction-Effect-2026-05-17-1834"
base_model = "roberta-large"

tokenizer = RobertaTokenizerFast.from_pretrained(repo_id, add_prefix_space=True)
model = RobertaForTokenClassification.from_pretrained(base_model, num_labels=2)
model = PeftModel.from_pretrained(model, repo_id)

Training

Trained under Markers-Extraction/ with LoRA.

Limitations

English Reddit submission statements only.
One marker type per model.

Citation

SemEval-2026 Task 10: Psycholinguistic Markers of Conspiracy Theories in Social Media Conversations.

Model metadata

Field	Value
Hub repo	`MMHusnain/semeval26-psysemev-extraction-Effect-2026-05-17-1834`
Marker type	Effect
Upload batch	`2026-05-17-1834`
Base model	`roberta-large`