Rasa无法检测到垃圾短语（“afdas afadfasdf”）并且无法触发fall_back_action。任何帮助将不胜感激。谢谢
## intent:say_thanks
- Thanks you so much
- helpful it, thanks
- appreciate it, congrats
- appreciate it, helpful
- appreciate it, thanks
- nice, apreciate
- nice, lol
- really, thnx
- thanku
- thank u
- nice, thank you
- nice, thnx
- nice
- Thanks you too !
- Thank you :)
## intent:isnt_happy
- terrible performance
- i am not happy with your performance
- you are so stupid
- you are so bothersome
- you are so silly
- you are so tiresome
- don't test my patience
- dont piss me off
- you are pissing me off
- you are so annoying
- I do not approve of this bot. It's absolutely terrible .
- I am not content about this service of yours
- I am sad
政策：
- name: KerasPolicy
epochs: 1000
max_history: 3
batch_size: 10
validation_split: 0.2
- name: MemoizationPolicy
max_history: 3
- name: FallbackPolicy
nlu_threshold: 0.4
core_threshold: 0.3
fallback_action_name: "utter_fallback_action"
intents:
- say_thanks
- isnt_happy
templates:
utter_say_thanks:
- text: "You're welcome :-)"
- text: "Just doing my job!"
- text: "Any time :-)"
utter_isnt_happy:
- text: "I am sorry :-("
- text: "Sorry I didn't mean to make you sad"
- text: "Sorry, i didn't met your expectations. My creators are working hard to improve me."
utter_fallback_action:
- text: "I didn't understood that. My creators are working hard to make me better."
actions:
- utter_say_thanks
- utter_isnt_happy
- utter_fallback_action
stories.面对
## talk_say_thanks
* say_thanks
- utter_say_thanks
## talk_isnt_happy
* isnt_happy
- utter_isnt_happy
由于你只有两个意图，所以预测它们的置信度要高于你有更多意图。这通常不应成为问题，但在设置回退的置信度阈值时必须牢记这一点。
为了更容易检测回退，特别是在训练数据和意图很少的情况下，我建议添加第三个意图。例如out_of_scope，你添加一些不应被你的任何其他意图拾取的随机事物：
## intent:out_of_scope
- fggsf dnek ofjkfn
- I am hungry
- where are we going
- tell me a joke
- i want to book a restaurant
- jfhr qsaasr fvndveti
现在，您可以使用回退添加故事来处理此意图。
* out_of_scope
- utter_fallback_action