Warning: This document is for an old version of rasa NLU. The latest version is 0.15.1.

Training Data Format

The training data for rasa NLU is structured into different parts, common_examples, entity_synonyms and regex_features. The most important one is common_examples.

{
    "rasa_nlu_data": {
        "common_examples": [],
        "regex_features" : [],
        "entity_synonyms": []
    }
}

The common_examples are used to train both the entity and the intent models. You should put all of your training examples in the common_examples array. The next section describes in detail how an example looks like. Regex features are a tool to help the classifier detect entities or intents and improve the performance.

You can use Chatito , a tool for generating training datasets in rasa’s format using a simple DSL.

Common Examples

Common examples have three components: text, intent, and entities. The first two are strings while the last one is an array.

  • The text is the search query; An example of what would be submitted for parsing. [required]
  • The intent is the intent that should be associated with the text. [optional]
  • The entities are specific parts of the text which need to be identified. [optional]

Entities are specified with a start and end value, which together make a python style range to apply to the string, e.g. in the example below, with text="show me chinese restaurants", then text[8:15] == 'chinese'. Entities can span multiple words, and in fact the value field does not have to correspond exactly to the substring in your example. That way you can map syonyms, or misspellings, to the same value.

{
  "text": "show me chinese restaurants",
  "intent": "restaurant_search",
  "entities": [
    {
      "start": 8,
      "end": 15,
      "value": "chinese",
      "entity": "cuisine"
    }
  ]
}

Entity Synonyms

If you define entities as having the same value they will be treated as synonyms. Here is an example of that:

[
  {
    "text": "in the center of NYC",
    "intent": "search",
    "entities": [
      {
        "start": 17,
        "end": 20,
        "value": "New York City",
        "entity": "city"
      }
    ]
  },
  {
    "text": "in the centre of New York City",
    "intent": "search",
    "entities": [
      {
        "start": 17,
        "end": 30,
        "value": "New York City",
        "entity": "city"
      }
    ]
  }
]

as you can see, the entity city has the value New York City in both examples, even though the text in the first example states NYC. By defining the value attribute to be different from the value found in the text between start and end index of the entity, you can define a synonym. Whenever the same text will be found, the value will use the synonym instead of the actual text in the message.

To use the synonyms defined in your training data, you need to make sure the pipeline contains the ner_synonyms component (see Processing Pipeline).

Alternatively, you can add an “entity_synonyms” array to define several synonyms to one entity value. Here is an example of that:

{
  "rasa_nlu_data": {
    "entity_synonyms": [
      {
        "value": "New York City",
        "synonyms": ["NYC", "nyc", "the big apple"]
      }
    ]
  }
}

Note

Please note that adding synonyms using the above format does not improve the model’s classification of those entities. Entities must be properly classified before they can be replaced with the synonym value.

Regular Expression Features

Regular expressions can be used to support the intent classification and entity extraction. E.g. if your entity has a certain structure as in a zipcode, you can use a regular expression to ease detection of that entity. For the zipcode example it might look like this:

{
    "rasa_nlu_data": {
        "regex_features": [
            {
                "name": "zipcode",
                "pattern": "[0-9]{5}"
            },
            {
                "name": "greet",
                "pattern": "hey[^\s]*"
            },
        ]
    }
}

The name doesn’t define the entity nor the intent, it is just a human readable description for you to remember what this regex is used for. As you can see in the above example, you can also use the regex features to improve the intent classification performance.

Try to create your regular expressions in a way that they match as few words as possible. E.g. using hey[^\s]* instead of hey.*, as the later one might match the whole message whereas the first one only matches a single word.

Regex features for entity extraction are currently only supported by the ner_crf component! Hence, other entity extractors, like ner_mitie won’t use the generated features and their presence will not improve entity recognition for these extractors. Currently, all intent classifiers make use of available regex features.

Note

Regex features don’t define entities nor intents! They simply provide patterns to help the classifier recognize entities and related intents. Hence, you still need to provide intent & entity examples as part of your training data!

Markdown Format

Alternatively training data can be used in the following markdown format (Regex features not supported yet). Examples are listed using the unordered list syntax, e.g. minus - or asterisk *:

## intent:check_balance
- what is my balance <!-- no entity -->
- how much do I have on my [savings](source_account) <!-- entity "source_account" has value "savings" -->
- how much do I have on my [my savings account](source_account:savings) <!-- synonyms, method 1-->

## intent:greet
- hey
- hello

## synonym:savings   <!-- synonyms, method 2 -->
- pink pig

Organization

The training data can either be stored in a single file or split into multiple files. For larger training examples splitting the training data into multiple files (e.g. one per intent) might be beneficial in terms of maintainability.

Storing files with different file formats (e.g. markdown and JSON) in one directory is currently not supported.

Note

Splitting the training data into multiple files currently only words for markdown and JSON data. For other file formats you have to use the single-file approach.