Issue with entity values

[ghost]

Hi,

I am trying to map synonyms and misspellings to predefined values (using spacy + sklearn). In demo-rasa.json, I add the following (plus several other possibilities):
{
"text": "show me tchineez restaurants",
"intent": ""restaurant_search"",
"entities": [
{
"start": 8,
"end": 16,
"value": ""chinese"",
"entity": "cuisine"
}
]
}
After training, when I send "show me tchineez restaurants" to parse, I get "tchineez" as the entity value instead of the "chinese".

[amn41]

Indeed this isn't currently implemented. See entity normalization here in the roadmap.

The simplest way to do this would just be to create a list of aliases in your code & reconcile them later. Support for doing this natively (& more generally) is coming, but I can't tell you exactly when.

[amn41]

My suggestion for how to implement this is to build up a map when we read through the training data. All entities are then mapped by their "value". So the map would look like :

{ 
  "tchineez" : "chinese",
  "chinese" : "chinese",
  "china" : "chinese"
  ...etc
}

built up from entities like

{  
   "text":"show me tchineez restaurants",
   "intent":""   restaurant_search"",
   "entities":[  
      {  
         "start":8,
         "end":16,
         "value":"chinese",
         "entity":"cuisine"
      }
   ]
}

We would persist this in the model dir, as index.json or something