Rasa NLU version: 0.13.8
Operating system (windows, osx, …): Windows
Content of model configuration file:
language: "en" pipeline: "spacy_sklearn"
Issue: I am creating a chatbot for the internal purpose of an industry which should be able to extract the group name, subgroup name, version, date and type of document from a given text.
Eg the user inputs - CASE 1 : I want details for Columbia, Fiji, dated 18/01/2019, version 1.2.3 and .txt file CASE 2 : May i know details for Canteen, Floor1, 12/01/2019, 5.5.5 and .xml file
The chatbot should be able to extract :
CASE 1 - “Columbia” as group_name, “Fiji” as subgroup_name, “18/01/2019” as date, “1.2.3” as version and “txt” as the doc_type. CASE 2 - “Canteen” as group_name, “Floor1” as subgroup_name, “12/01/2019” as date, “5.5.5” as version and “xml” as the doc_type.
Now the problem is if i use lookup table it works fine, but becomes restricted as the model does not evolve to extract unseen values that are not present in the lookup table. Is there any other way to do it by training the RASA NLU model because these names are all from varied contexts with different intents according to spacy. Also that the user might input values in any random order and might forget providing some values. I would want them to be correctly classified into the respective slots. But when i am simply training the model on examples and without lookup tables, the entities are NOT getting extracted except for the ones which i mentioned in the training examples. Is there a proper way to achieve this with training? And if there is, please may i know the config or the required change?