---
library_name: transformers
tags: []
widget:
- text: 'Thank you for approaching me about the collaboration. You can talk to my manager, Kritik at 9874512563 or kritik.jun@asdf.com'
  example_title: Email 1
- text: 'Call me on 9874569874'
  example_title: Email 2
- text: 'You can email me at adsf@gmail.com or call directly on 9999988888. The point of contact would be my manager Manish Neupane'
  example_title: Email 3
---
Overview:

The Model is fine-tuned for 3 class + "0" class.<br>
The Dataset is custom annotated and contains 400 texts and the model was trained on the split of 0.76, 0.12, and 0.12.

The validation classification report is as follows:

|Class| Precision   |      Recall      |  f1 |
|-----|----------|:-------------:|------:|
|  0 | 1.00 | 1.00 | 1.00 |
|  1 | 0.98 | 1.00 | 0.91 |
|  2 | 0.95 | 0.89 | 0.92 |
|  3 | 0.8  | 0.88 | 0.84 |
|  macro-avg | 0.93 | 0.94 | 0.94 |

The test classification report is as follows:

|Class| Precision   |      Recall      |  f1 |
|-----|----------|:-------------:|------:|
|  0 | 1.00 | 1.00 | 1.00 |
|  1 | 0.98 | 1.00 | 0.99 |
|  2 | 0.66 | 0.97 | 0.79 |
|  3 | 0.84 | 0.78 | 0.81 |
|  macro-avg | 0.87 | 0.94 | 0.90 |

Possible future direction:

1. Clean data to a good enough format as much as possible.
2. Increase the data as much as possible. (Make sure to have data that is seen in real use cases.)
3. Ponder: Is it possible to use sth like Grammarly to clean the sentences before tokenization such that proper nouns are Capital and the grammer is correct such that a pattern is formed?