Can someone explain the RegexFeaturizer/RegexEntityExtractor for me?

Rasa Open Source
1

I’m trying to use the RegexFeaturizer/RegexEntityExtractor to improve intent classification. But I can’t get it to work. I have:

pipeline:
  - name: WhitespaceTokenizer
  - name: RegexFeaturizer
  - name: RegexEntityExtractor
  - name: LexicalSyntacticFeaturizer
  - name: CountVectorsFeaturizer
  - name: CountVectorsFeaturizer
    analyzer: char_wb
    min_ngram: 1
    max_ngram: 4
  - name: DIETClassifier
    epochs: 100
    constrain_similarities: true
  - name: EntitySynonymMapper
  - name: ResponseSelector
    epochs: 100
    constrain_similarities: true
  - name: FallbackClassifier
    threshold: 0.3
    ambiguity_threshold: 0.1


- lookup: food
  examples: |
    - italian pancake
    - pizza
    - doughy stuff
    - tomato sauce

- lookup: furniture
  examples: |
    - something to sit on
    - butsupporter
    - stool

- intent: order_pizza
  examples: |
    - I would like a [pizza](food)
    - I am hungry for a [pizza](food)
    - can i get a [italian pancake](food)
    - please give me a [italian pancake](food)

- intent: order_furniture
  examples: |
    - I would like a [chair](furniture)
    - I am hungry for a [chair](furniture)
    - can i get a [stool](furniture)
    - please give me a [stool](furniture)

Although entities from the lookup table are recognized, the model doesn’t seem to use the entities/lookuptable to classify the intents when using values not listed in the intent examples.

for example: “can I get a butsupporter” recognizes butsupporter as entity furniture but it’s kinda random to which intent this gets classified. Whats going on?

Can someone explain this for me?

2

Sounds like you need to add more utterance examples for the intents. Shoot for 15-20 examples for each one.

3

Thanks for your answer, It still didn’t work as I would expect after adding extra examples. I would like a cheese platter still gets classified as order_furniture

version: "3.1"

nlu:
- intent: greet
  examples: |
    - hey
    - hello
    - hi
    - hello there
    - good morning
    - good evening
    - moin
    - hey there
    - let's go
    - hey dude
    - goodmorning
    - goodevening
    - good afternoon

- intent: goodbye
  examples: |
    - cu
    - good by
    - cee you later
    - good night
    - bye
    - goodbye
    - have a nice day
    - see you around
    - bye bye
    - see you later

- intent: affirm
  examples: |
    - yes
    - y
    - indeed
    - of course
    - that sounds good
    - correct

- intent: deny
  examples: |
    - no
    - n
    - never
    - I don't think so
    - don't like that
    - no way
    - not really

- intent: mood_great
  examples: |
    - perfect
    - great
    - amazing
    - feeling like a king
    - wonderful
    - I am feeling very good
    - I am great
    - I am amazing
    - I am going to save the world
    - super stoked
    - extremely good
    - so so perfect
    - so good
    - so perfect

- intent: mood_unhappy
  examples: |
    - my day was horrible
    - I am sad
    - I don't feel very well
    - I am disappointed
    - super sad
    - I'm so sad
    - sad
    - very sad
    - unhappy
    - not good
    - not very good
    - extremly sad
    - so saad
    - so sad

- intent: bot_challenge
  examples: |
    - are you a bot?
    - are you a human?
    - am I talking to a bot?
    - am I talking to a human?

- intent: order_pizza
  examples: |
    - I would like a [pizza](food)
    - I would like a [pizza](food)
    - I am hungry for a [pizza](food)
    - can i get a [italian pancake](food)
    - please give me a [italian pancake](food)
    - i would like some [italian pancake](food)
    - can you provide me with [doughy stuff](food)
    - please give me a lot of [doughy stuff](food)
    - i really need [doughy stuff](food)
    - I wish to have [pizza](food)
    - please entertain me with [pizza](food)
    - i wont some of your [pizza](food)
    - if you dont give me some [pizza](food)
    - if you wont give me some [pizza](food)
    - isnt it nice to have [pizza](food)
    - provide me with please  [pizza](food)
    - i would like to get some [pizza](food)
    - a [pizza](food) would be nice
    - some [pizza](food) for me would be great
    - i'm really missing some [pizza](food)

- intent: order_furniture
  examples: |
    - I would like a [chair](furniture)
    - I would like a [chair](furniture)
    - I am hungry for a [chair](furniture)
    - can i get a [chair](furniture)
    - please give me a [buttsupporter](furniture)
    - i would like some [buttsupporter](furniture)
    - can you provide me with [kruk](furniture)
    - please give me a lot of [kruk](furniture)
    - i really need [buttsupporter](furniture)
    - I wish to have [chair](furniture)
    - please entertain me with [chair](furniture)
    - i wont some of your [chair](furniture)
    - if you dont give me some [chair](furniture)
    - if you wont give me some [chair](furniture)
    - isnt it nice to have [chair](furniture)
    - provide me with please [chair](furniture)
    - i would like to get some [chair](furniture)
    - a [chair](furniture) would be nice
    - some [chair](furniture) for me would be great
    - i'm really missing some [chair](furniture)


- lookup: food
  examples: |
    - italian pancake
    - pizza
    - doughy stuff
    - tomato sauce
    - ananas
    - grapes
    - cheese platter

- lookup: furniture
  examples: |
    - buttsupporter
    - kruk
    - chair
    - stool
    - something to sit on
    - couch
    - table

version: "3.1"

intents:
  - greet
  - goodbye
  - affirm
  - deny
  - mood_great
  - mood_unhappy
  - bot_challenge
  - order_pizza
  - order_furniture

entities:
  - food
  - furniture

responses:
  utter_order_furniture:
  - text: "you ordered a furniture"

  utter_order_pizza:
  - text: "you ordered a pizza"
  utter_greet:
  - text: "Hey! How are you?"

  utter_cheer_up:
  - text: "Here is something to cheer you up:"
    image: "https://i.imgur.com/nGF1K8f.jpg"

  utter_did_that_help:
  - text: "Did that help you?"

  utter_happy:
  - text: "Great, carry on!"

  utter_goodbye:
  - text: "Bye"

  utter_iamabot:
  - text: "I am a bot, powered by Rasa."

session_config:
  session_expiration_time: 60
  carry_over_slots_to_new_session: true


version: "3.1"

rules:

- rule: Say goodbye anytime the user says goodbye
  steps:
  - intent: goodbye
  - action: utter_goodbye

- rule: Say 'I am a bot' anytime the user challenges
  steps:
  - intent: bot_challenge
  - action: utter_iamabot

- rule: order pizza
  steps:
  - intent: order_pizza
  - action: utter_order_pizza

- rule: order furniture
  steps:
  - intent: order_furniture
  - action: utter_order_furniture
4

Sorry, ignore my prior message. These two intents are the same except for the entity type. You should combine the intents into a single order intent and use a single entity, item.

- rule: order
  steps:
  - intent: order
  - action: utter_order

  utter_order:
  - text: "you ordered a {item}"
5

Thanks again, I know this would be the way for the use case I’m showing. But, I made this to see the influence of the Regexfeaturizer and the lookup tables. The model doesn’t seem to attribute a lot of weight to these features when training. This is why I would like to get a deeper understanding. In the example I gave everything is the same except for the entities. Therefore I would expect the model to classify the intents correctly by using the lookup tables. However, it only does this for some.

6

Ok, so you probably noticed this from the docs:

When using lookup tables with RegexFeaturizer , provide enough examples for the intent or entity you want to match so that the model can learn to use the generated regular expression as a feature. When using lookup tables with RegexEntityExtractor , provide at least two annotated examples of the entity so that the NLU model can register it as an entity at training time.

While it’s not explicitly stated here, I don’t think Rasa is generating additional utterance examples for all of your lookup table entries as you might expect. I haven’t looked at the source code for lookup tables, but based on these docs, it’s probably only using a RegEx for model.

7

Thanks, it looks like this is the case to me as well… Was just hoping for a different outcome. I don’t really see the use for lookup tables if they don’t work the way I thought they would work. Sure, they could improve intent qualification but you still have to use the different values in the intent examples.