metadata
license: mit
language: en
tags:
- text-classification
- distilbert
- news-framing
- conflict-detection
- indian-elections
- computational-social-science
- websci2025
datasets:
- custom
model-index:
- name: Framing the Fray - Conflict Frame Classifier
results:
- task:
type: text-classification
name: Text Classification
dataset:
type: custom
name: Indian Election News Headlines
metrics:
- type: f1
value: 0.87
name: Macro F1-Score
args:
average: macro
Conflict Frame Classifier for Indian Election News
This is a distilbert-base-uncased-finetuned-sst-2-english model fine-tuned to classify news headlines into one of two frames: conflict or non-conflict (miscellaneous).
This model is the official artifact for the research paper:
"Framing the Fray: Conflict Framing in Indian Election News Coverage" accepted at the 17th ACM Web Science Conference (WebSci '25).
Citation
If you use this model or the associated code, please cite our paper:
@inproceedings{chebrolu2025framing,
title={Framing the Fray: Conflict Framing in Indian Election News Coverage},
author={Chebrolu, Tejasvi and Chowdhary, Rohan and Vardhan, N Harsha and Kumaraguru, Ponnurangam and Rajadesingan, Ashwin},
booktitle={Proceedings of the 17th ACM Web Science Conference 2025 (Websci '25)},
year={2025},
month={May},
address={New Brunswick, NJ, USA},
publisher={ACM},
doi={10.1145/3717867.3717900}
}
How to Use
You can use this model directly with the pipeline function from the transformers library:
from transformers import pipeline
# Replace with your actual model repo ID after uploading
classifier = pipeline("text-classification", model="tejasvichebrolu/conflict-frame-classifier")
headlines = [
"Days Before Polls, Kamal Haasan Meets Mamata Banerjee In Kolkata",
"SP-BSP alliance led to wave of happiness; BJP worried, says Akhilesh Yadav",
"Narendra Modi invoking Army for votes, says Tejashwi Yadav"
]
results = classifier(headlines)
for result in results:
print(f"Label: {result['label']}, Score: {result['score']:.4f}")
Training Procedure
The model was fine-tuned on a dataset of 860 headlines annotated for the presence of a conflict frame.
Hyperparameters
| Hyperparameter | Value |
|---|---|
| Conflict Class Weight | 1.69 |
| Non-Conflict Class Weight | 9.01 |
| Learning Rate | 6.008 × 10⁻⁵ |
| Epochs | 9 |
Evaluation Results
The model's performance was evaluated using 5-fold cross-validation. The average metrics are reported below:
| Class | Precision | Recall | F1-Score |
|---|---|---|---|
| Conflict | 0.92 | 0.91 | 0.92 |
| Non-Conflict | 0.81 | 0.82 | 0.81 |
| Macro avg | 0.87 | 0.87 | 0.87 |