A fork to be used with Botfront, an open source chatbot platform built with Rasa.
For more information visit the Botfront project on Github
This policy implements fallback and suggestion-based disambiguation.
It works with actions rasa_addons.core.actions.ActionBotfrontDisambiguation
, rasa_addons.core.actions.ActionBotfrontDisambiguationFollowup
and rasa_addons.core.actions.ActionBotfrontFallback
, and NLU pipeline component rasa_addons.nlu.components.intent_ranking_canonical_example_injector.IntentRankingCanonicalExampleInjector
.
policies:
...
- name: rasa_addons.core.policies.BotfrontDisambiguationPolicy
fallback_trigger: 0.30 # default value
disambiguation_trigger: '$0 < 2 * $1' # default value
disambiguation_template: 'utter_disambiguation' # default value
n_suggestions: 2 # default value
excluded_intents:
- ^chitchat\..* # default value
- ^basics\..*
...
Botfront introduces the notion of "canonical" training examples, which provide a canonical human-readable text for intent labels. For example, for an intent pay_bills
with examples "Pay bills", "I want to pay my bills", "How does one pay the bills on this website?", the first example may be selected as canonical. Canonical status serves as a cue to the bot designer, since intent labels can become untractable over time. It may also come to serve more Botfront-internal roles in the future.
The Botfront Disambiguation Policy uses canonical status to provide localized text for the suggestion buttons shown to users during disambiguation. In order to enable this feature, the NLU pipeline for each language model needs to be extended in the following way:
pipeline:
...
- name: rasa_addons.nlu.components.intent_ranking_canonical_example_injector.IntentRankingCanonicalExampleInjector
...
This NLU component enriches the intent_ranking
key of user messages in the tracker with canonical text, so that the Disambiguation Policy may pick it up. If the NLU component is not used, buttons will have the intent name as their title.
Float (default 0.30
): if confidence of top-ranking intent is below this threshold, fallback is triggered. Fallback is an action that utters the template utter_fallback
and returns to the previous conversation state.
String (default '$0 < 2 * $1'
): if this expression holds, disambiguation is triggered. (If it has already been triggered on the previous turn, fallback is triggered instead.) Here this expression resolves to "the score of the top-ranking intent is below twice the score of the second-ranking intent". Disambiguation is an action that lets the user to choose from the top-ranking intents using a button prompt.
String (default 'utter_disambiguation'
): a response name resolving to a template containing a text
field with a message, e.g. "I could not quite understand. Did you mean...". Any button included under the buttons
field of this template will also appear at the end of autogenerated suggestions, e.g. {"title": "None of the above", "type": "postback", "payload": "/deny_suggestions"}
.
Int (default 2
): the maximum number of suggestions to display (excluding the 'Other' options).
List (default ["^chitchat\..*", "^basics\..*"]
): any intent (exactly) matching one of these regular expressions will not be shown as a suggestion.
This policy implements regular expression-based direct mapping from intent to action.
policies:
...
- name: rasa_addons.core.policies.BotfrontMappingPolicy
triggers:
- trigger: '^map\..+'
action: 'action_botfront_mapping'
extra_actions:
- 'action_myaction'
...
The default action ActionBotfrontMapping takes the intent that triggered the mapping policy, e.g. map.my_intent
and tries to generate the template utter_map.my_intent
.
credentials:
...
rasa_addons.core.channels.webchat.WebchatInput:
session_persistence: true
base_url: {{rasa_url}}
socket_path: '/socket.io/'
...
Rest Input Channel with multilanguage and metadata support.
credentials:
...
rasa_addons.core.channels.rest.BotfrontRestInput:
# POST {{rasa_url}}/webhooks/rest/webhook/
...
Conversation testing channel. Simulates each user event in the input test stories as a message sent by a user, then compares the input story to the results from rasa. Returns a diff of the input story and output story with expected and actual events.
credentials:
...
rasa_addons.core.channels.bot_regression_test.BotRegressionTestInput: {}
# POST {{rasa_url}} /webhooks/bot_regression_test/run
...
Idential to Rasa's TemplatedNaturalLanguageGenerator
, except in handles templates with a language key.
The new standard way to connect to the Botfront NLG endpoint. Note that support for the legacy REST endpoint is maintained for the moment. This feature is accessed by supplying a URL that doesn't contain the substring "graphql".
endpoints:
...
nlg:
url: 'http://localhost:3000/graphql'
type: 'rasa_addons.core.nlg.GraphQLNaturalLanguageGenerator'
...