Chatmate

A chatbot which understands user input by applying intent classification and NER, and generates a suggestion of possible response based on neural network. Current version has only been trained and tested in german language.

Chatbot Setup Instructions:

• virtualenv venv -p /usr/bin/python3
• source venv/bin/activate
• pip install -r requirements.txt
• python -m spacy download de
• python -m rasa_nlu.train -c configs/spacy_regex_mb_nlu_model_config.json
• python -m rasa_core.train -s data/usecases_basic.md -d data/domain.yml -o models/dialogue --epochs 300
• gunicorn communication_layer.application:api -c gunicorn.cfg
• curl -i -H 'Content-type:application/json; charset=UTF-8' -XPOST localhost:8000/api/parse -d '{"query":"Ich bin Tom Hanks", "sender_id":"12345"}'
• Demo UI is available under localhost:8000/ui/chat.html.

NLU Training Data Tool:

Read https://github.com/RasaHQ/rasa-nlu-trainer for details about setup, usage. rasa-nlu-trainer

Cross validate of nlu model:

python -m rasa_nlu.evaluate -d data/nlu_traindata.json -c configs/spacy_regex_mb_nlu_model_config.json --mode crossvalidation

INFO:__main__:CV evaluation (n=10)
INFO:__main__:F1-score: 0.837 (0.052)
INFO:__main__:Precision: 0.914 (0.000)
INFO:__main__:Accuracy: 0.849 (0.044)
INFO:__main__:Finished evaluation

Start nlu as server

python -m rasa_nlu.server -c configs/spacy_regex_mb_nlu_model_config.json

Visulaization of dialogue training data:

python -m rasa_core.visualize -d data/domain.yml -s data/usecases_basic.md -o usecases_basic.png

Run Chatbot in console

python -m rasa_core.run -d models/dialogue -u models/nlu/ner-de/spacy_regex_mb/

Run Chatbot as Server

python -m rasa_core.server -d models/dialogue -u models/nlu/ner-de/spacy_regex_mb/ --cors http://localhost:63342

Start Chatbot with falcon

gunicorn communication_layer.application:api -c gunicorn.cfg Visit localhost:8000/ui/chat.html