Datasets:
metadata
language:
- en
size_categories:
- 10K<n<100K
Dataset Card for Dataset Name
Dataset Description
- Homepage:
- Repository:
- Paper:
- Leaderboard:
- Point of Contact:
Dataset Summary
Dataset used to train a language model to do classification on 50 different waste classes.
Languages
English
Dataset Structure
Data Instances
| Phrase | Class | Index |
|---|---|---|
| "I have this apple phone charger to throw, where should I put it ?" | PHONE CHARGER | 26 |
| "Should I recycle a disposable cup ?" | Plastic Cup | 32 |
| "I have a milk brick" | Tetrapack | 45 |
Data Fields
- Phrase
- Class
- Class_index
Data Splits
train: 12.5K rows
test: 5.38K rows
additional data: 7.24K rows (unseen_phrases.csv)
Dataset Creation
Manualy with objects and phrases templates.
Annotations
Annotation process
Each object was annotated and then the phrases were annotated according to the object according to its annnotation.
Who are the annotators?
Myself
Personal and Sensitive Information
None
Considerations for Using the Data
Social Impact of Dataset
None
Discussion of Biases
Some classes are more present than others but the dataset is balanced overall. Because it was created using patterns, might not be very robust.
Other Known Limitations
Repetition of phrase patterns, have to verify performances of model on external phrases for robustness.