Issue in extracting Date/Time entities

(Varun) #1

Hello RASA people,

I have an issue in extracting the Date/Time entities using spacy_sklearn pipeline. I am not sure why it can not extract entities such as last week, 2 days ago, or 3rd Dec 2018? After a few searches, I found that I need to use Duckling which performed well in my case. But, the returned JSON will include entities that are extracted by ner_crf and ner_duckling. So, I need to parse the JSON and detect which library has a higher confidence in order to use the extracted entity from that library. My question is there a certain way to extract the date/time entities ONLY using duckling? or is there a way to improve the ner_crf in extracting the dates (I use the DATE prebuilt entity)? The reason for using ner_crf is because I have custom entities in my agent.

Here is the nlu_config.yml:

language: "en"

- name: "nlp_spacy"
- name: "tokenizer_spacy"
- name: "intent_entity_featurizer_regex"
- name: "intent_featurizer_spacy"
- name: "ner_crf"
- name: "ner_synonyms"
- name: "intent_classifier_sklearn"
- name: "ner_duckling_http"
  url: ""
  dimensions: ["time"] 

Looking forward for your answers :slight_smile:

(Libindavis) #2

I am encountering the same issue. The doc of Rasa-NLU does not explain how to use this ner_duckling_http component in detail. A lot of confusions to me. But to me, the nlu_config.yml should be something like:

Hope someone could help here.

(Varun) #3

Thanks for your reply. I think we need to run the dockling docker container.