Hi,
My nlu pipeline is -
language: “en”
pipeline:
name: “tokenizer_whitespace”
name: “intent_entity_featurizer_regex”
name: “ner_crf”
name: “ner_synonyms”
name: “intent_featurizer_count_vectors”
name: “intent_classifier_tensorflow_embedding”
intent_tokenization_flag: true
intent_split_symbol: “+”
I have some ticket number in this format - i.e SR12345698 , this works fine when entered this word within a sentence like - “ticket status for SR12345698” , but on entering it alone bot returns
{
“query”: “SR12345698”,
“topScoringIntent”: {
“intent”: null,
“score”: 0
},
“intents”: [
{
“intent”: null,
“score”: 0
}
],
“entities”: [
{
“entity”: “SR12345698”,
“type”: “IncidentID”,
“startIndex”: 0,
“endIndex”: 9,
“score”: 0.9694517848359389
}
]
}
Is this under the same intent or do you have a something like an /inform intent?
no i have not added it to a single intent /inform i have a different intent for this phrase
Oh okay. I’ve never tried the tensorflow pipeline. Not sure if that is causing the issue.
Juste
March 7, 2019, 1:47pm
6
Juste
March 7, 2019, 1:49pm
7
My first suggestion would be to add a few exaples of providing just the ticket number so that the model would learn that these kind of inputs can appear too, not just the ones surrounded with actual words
i have done that… i have provided with utterences but its of no use as you can see it is catching intent as null but picking up the entity
Juste
March 7, 2019, 2:35pm
9
Hmmm, that’s a bit odd. What’s the ration of examples with just ID vs the examples with ID+surrounding words in your training dataset?
yes i have provided enough amount of training data including just ID and ID + surrounding
and i tried installing mitie
getting this error -
The image shows a number of only id’s mentioned in my training data
