问题
- RASA version : 0.10.5
- spaCy version : 1.9.0
- Installed models : en, en_core_web_sm
I am creating training data using dialogflow export data and successfully finished training but when I request and give some text it gives wrong intent result. It always gives the same intent in result and also every time same intent_ranking.
Please let me know how can I get proper intent results as well as entities result.
回答1:
The general recommendation when Rasa NLU seems to be functioning correctly, but incorrectly classifying intents and entities is, add more training data.
So try that, but also try these other suggestions:
- Rasa recently released the tensorflow embeddings pipeline. This pipeline creates a small language model with the training data set and word vectors. It can, in general, can operate on a more modest data set. Set your pipeline in config.yml to
"tensorflow_embedding"to give it a try. - Use the evaluate script to see where your training data is the weakest. The script will generate f1 scores and a chart for showing which intents are being misclassified the most.
- Increase the size of your spaCy model. Without information on the language you are using or the spacy model you have installed, this is just a shot in the dark, but especially if you have installed the small spacy model for English (en_core_web_sm), you should try the medium or large size models which include word vectors.
With all that said, my experience with Dialogflow and Rasa NLU both suggest more training data is required for Rasa's implementation. Also if you continue to have problems join us on the Rasa NLU Gitter and we can help you there.
来源：https://stackoverflow.com/questions/47862971/rasa-nlu-parse-not-give-currect-intent-give-same-intent-in-result