Proof-of-concept of an intention embedding space to classify an incoming message with an intention without training a model.
Uses the embedding layer of a pre-trained lightweight transformer to encode sentence context into vector space where new messages are encoded then use k-nearest neighbors to infer intent from a pre-calculated intention space.
pip install -r requirements.txt
python main.py
python app.py