Extracting most similar training examples for user's query

Is there any way to find out what are examples in NLU training data that are the most similar for user’s query? As of now, from the training data, the model will be generated on which the user’s query is parsed to extract the intent of the query. My question is whether we can find out the examples in the training data that are most similar to the user’s query. This will help us to give auto suggestions when the user’s start typing in the query/message.

Similar problem was asked many times in the forum. I just want to give a thought around it.

Any idea around this problem statement. Thanks in advance. One of the high-level ideas that I have is by using DIET Classifier. In DIET classifier, while training model, the architecture is in such a way that it masks some of the token and tries to predict that token. Later Similarity between a predicted token and actual token is considered for entity loss. Can we leverage this masking idea to auto-suggest things while the user is typing query?

@Rasa Tagging rasa team for better reachability.