ba-claim/electra

Model Details

Fine-tuned ELECTRA Model for Claim Relevance Identification

Based on this model: https://huggingface.co/bhadresh-savani/electra-base-emotion

Model Description

This Hugging Face model is a fine-tuned ELECTRA model specifically developed for identifying relevant claims in the context of combating fake news. The model was trained as part of a bachelor thesis project that aimed to automate the process of fact-checking by automatically identifying claims of interest.

The project participated in the CheckThat!2023 competition, focusing on task 1B, organized by the Conference and Labs of the Evaluation Forum (CLEF). The CheckThat! lab provided relevant training data for predicting the checkworthiness of claims. The data was analyzed, and various transformer models, including BERT and ELECTRA, were experimented with to identify the most effective architecture.

Overall, this fine-tuned ELECTRA model serves as a valuable tool in automating the identification of relevant claims, reducing the need for manual fact-checking and contributing to the efforts in mitigating the challenges posed by the widespread dissemination of fake news.

Examples

37440 There's no way they would give it up. No

37463 They're able to charge women more for the same exact procedure a man gets. Yes

Training Details

Hyperparameters
Learning Rate 2.650e-05
Weight Decay 91.942e-04
Batch Size 32
Number of Epochs 5