atotalnoob
(Atotalnoob)
1
Hi guys,
I have a bot that is returning the values of entities inconsistently as “//”.
My pipeline :
Example of the entities:
intent: “topic”,
text: “tech support”
entities[
start: 0,
end: 7,
value: “support”,
entity:“topic”
]
I sometimes see from debug, "inconsistent BILOU tagging found, B- tag not closed by L- tag, IE [B-a, L-a, O] instead of [B-a, L-a, O].
Assuming last tag is L-
Could this be something?
wrathagom
(Caleb M Keller)
2
Assuming this is what you’re saying your training data looks like:
{
"intent": "topic",
"text": "tech support",
"entities": [
"start": 0,
"end": 7,
"value": "support",
"entity": "topic"
]
}
Then you have a problem in your training data. The word support
starts at position 5 and ends at position 12.
atotalnoob
(Atotalnoob)
3
Sorry I read from the wrong example.
For that one is it actually 0 and 11.
wrathagom
(Caleb M Keller)
4
That still wouldn’t work since it isn’t capturing whole tokens. 0 and 12 would work if you wanted the whole utterance.
You can use the online trained to validate your data format: https://rasahq.github.io/rasa-nlu-trainer/