Does the component order in a pipeline impact the intent classifcation?


(Libindavis) #1


Does anyone know if the order in the pipeline would impact the intent classification result?

Here is my pipeline configurations. Note that “ner_duckling_http” is after “ner_crf”.

- name: "nlp_spacy"
- name: "tokenizer_spacy"
- name: "intent_featurizer_spacy"
- name: "intent_entity_featurizer_regex"
- name: "ner_crf"
- name: "ner_duckling_http"
  url: "http://locahost:8000"
  dimensions: ["time", "number", "amount-of-money", "distance", "duration", "ordinal"]
  locale: "en_US"
  timezone: "US/Pacific"
- name: "ner_synonyms"
- name: "intent_classifier_sklearn"

If I switch the order of “ner_duckling_http” and “ner_crf”, would the intent classification lead to different result for a same input text?
I am asking this because when I enter input text like ‘how many bugs in last 4 weeks’, would the ‘in last 4 weeks’ be extracted out by ‘ner_duckling_http’, then the remaining string ‘how many bugs’ is passed “ner_crf”. Is that working so ?

(Souvik Ghosh) #2

Both extractors are not dependent on each other and consume the input text in full to retrieve entities.

Have you noticed something different by switching the order in the pipeline?