shadowavenger9/garbage-text-classifier
Model Card for Garbage-Text-Classifier
A simple fine-tuning of the distilbert-base-uncased model for binary text-classification as Garbage/Nonsense or Not-Garbage.
- Developed by: shadowavenger9
- Model type: text-classifier
- Finetuned from model : distilbert-base-uncased
Uses
For determining if the results of Optical-Character-Recognition are "meaningful" and removing garbage sentences.
Training Details
Training Data
Trained and evaluated on a small, hand-labelled dataset of 500 text messages, obtained via OCR on public screenshots of digital chats.
Data sample:
... | 76| "fair bit of money burrgured you're worth it. what's a good time for a date, beautiful?"| False | 77| Ummmm| False | 78| Who are you| False | 79| "Actually I don't wanna know, please never text me again"| False | 80| "Wow. I went through all this trouble, and you just reject me? I went through 40 dollars just to get your phone number, you UNGRATEFUL BITCH! I'II still go out with you if you change your mind but don't think for one second that it will be to anywhere nice. In some parts of the world a woman could be stoned to death for talking to a generous, loving man like you just did. Consider yourself on a thin line."| False | 81| Can take you out to dinner tonite?| False | 82| Ignore me??| False | 83| Been 20 minutes and I'm with family. Gosh.| False | 84| Well it said your were online...| False | 85| And it's 8pm already| False
Training Hyperparameters
10 Epochs at Learn-Rate: 2e^-5 Batch-Size: 16
Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
Testing Data
<!-- This should link to a Dataset Card if possible. -->
Metrics
Final Validation Loss: 0.369251 Final Accuracy: 0.883333