Integrate Transformer-Based Detectors

[meriemjebali]

Description of Problem:

While HF models can be efficiently and easily used in the Melusine scope, the current version combines only deterministic based-on-regex detectors.

Overview of the Solution:

1. Technical features

Adding a base class allowing to load models from hugging-face and incorporating them in its methods . Thus providing more accurate results.

• Adding a MelusineMlDetector base class with the following methods :
  • pre_detect
  • by_regex_detect
  • load_model
  • predict
  • by_ml_detect
  • post_detect

2. Functional implementations :

   • Adding a DissatisfactionDetector : detection of dissatisfaction emails
     /or
   • Enriching the thanksDetector with a transformer-based model for enhanced predictions

3. Documentation
   The documentation of this work would be necessary , to provide a how-to-use template of these kind of models in the Melusine environment.

Examples:
Models such as. : distil-camembert-base

Blockers: None

Definition of Done: New Melusine version