Rasa: Rasa nlu - Multiple intents

[rishikesh1205]

Rasa NLU version: 0.12.3
Operating system : linux
Issue:
i'm not able to classify the conversations with multiple intents while training in Rasa-nlu and is it even possible??
the config file:
pipeline:

name: "intent_featurizer_count_vectors"
name: "intent_classifier_tensorflow_embedding"
intent_tokenization_flag: true
intent_split_symbol: "+"

[AbdurRub]

Please post how you define your multi intent? and how many examples you have for multi intent.

[wrathagom]

If that is your entire pipeline you are missing a tokenizer like tokenizer_whitespace, which should come at the beginning of your pipeline.

[Ghostvv]

@wrathagom if you don't use ner tokenizer is optional

[rishikesh1205]

@AbdurRub thank you for the reply. here i don't want to define multi intents. if i have seperate intents like greet and question and when i parse a sentence say "hello walter, do you know about java?" the intent should be like greet + question. so is there any pipeline or components that can be included so that i can get multi intents without training the data in that way

[AbdurRub]

No you have to train it with new intent having greet+question
https://github.com/RasaHQ/rasa_nlu/issues/1072

[prasgaut]

hi @AbdurRub ,
If i have to define a new intent (greet+question) with the training data, it seems like a new intent having name 'greet+question'. Than what is the benefit of multi intent functionality.

[mzq20180601]

@AbdurRub i have the same issue raised by @prasgaut

[akelad]

The benefit is that you can handle multi intents differently, you can take a look at the blogpost from Juste here: https://blog.rasa.com/how-to-handle-multiple-intents-per-input-using-rasa-nlu-tensorflow-pipeline/
I'll close this issue now as it's more of a discussion than an issue with rasa_nlu. If you would like to discuss this further, please create a post on our forum