|
|
--- |
|
|
license: apache-2.0 |
|
|
datasets: |
|
|
- gexai/inquisitiveqg |
|
|
language: |
|
|
- en |
|
|
metrics: |
|
|
- accuracy |
|
|
base_model: |
|
|
- distilbert/distilbert-base-uncased |
|
|
pipeline_tag: text-classification |
|
|
--- |
|
|
|
|
|
## Model Details |
|
|
Text classification model for ambiguity in questions. Classifies questions as ambiguous or clear. |
|
|
Based on distilbert/distilbert-base-uncased. |
|
|
|
|
|
**Example:** |
|
|
|
|
|
"Did he do it?" {'label': 'AMBIG', 'score': 0.9029870629310608} |
|
|
|
|
|
"Did Peter win the game?" {'label': 'CLEAR', 'score': 0.8900136351585388} |
|
|
|
|
|
## Out-of-Scope Use |
|
|
The model was only trained to classify single questions. Other kinds of data are not tested. |
|
|
|
|
|
### Training Data |
|
|
I manually labeled a small part of the inquisitiveqg dataset mixed with a private dataset to train the model to recognize ambiguity in questions. A satisfactory model with 85.5% accuracy was created. |
|
|
|
|
|
#### Metrics |
|
|
"eval_accuracy": 0.8551401869158879, |
|
|
|
|
|
"eval_loss": 0.3658725619316101, |
