arianedlns / chatbot-presidentielle2022 Goto Github PK

Create custom action for survey data scraping from wikipedia

Adding graph and whole poll data for a specific poll

{
    "theme_1": "name",
    "candidate": {
        "candidate_name_1": "idea",
        "candidate_name_2": "idea"
    }, 
    "theme_2": "name",
    "candidate": {
        "candidate_name_1": "idea",
        "candidate_name_2": "idea"
    }, 
}

theme_names define a first draft with 12 themes

Référence

https://ttt.studio/blog/rasa-chatbot/

https://rasa.com/docs/rasa/policies#unexpected-intent-policy
Happening when demande_sondage is the first action after bonjour

Split the inital data folder:

-data
   -nlu.yml
   -stories.yml
   -rules.yml
   -candidates.json

Create an out_of_scope intent (while keeping the FallbackClassifier) for all questions not related to the elections.

Tune hyperparameters using :

• https://rasa.com/blog/rasa-nlu-in-depth-part-3-hyperparameters/
• https://github.com/RasaHQ/nlu-hyperopt

Hyperparameters to be tuned in config.yml :

• DIETClassifier :
  • epochs
  • embedding_dimension
  • hidden_layers_sizes

Ne renvoie pas de réponse

Info Marine le pen ?
Info Marine ?
Résultat : Anne Hidalgo

Fix badly written entities with embedding/classification

Actual

Input :
Quel est le parti de Melanchon ?

Output:
Je ne reconnais pas le nom de ce candidat. L'avez-vous bien écrit ?

Expected

Input :
Quel est le parti de Melanchon ?

Output:
Le parti de Jean-Luc Mélenchon est La France Insoumise.

Update test_stories with all possibility we want to check