How to use Sequence labeling for queries with different context?

Question

i'm creating a sequence labeling program using pycrfsuite(BIO taging) and nltk. The program should be able to process queries with different context.

I've trained different models for each context and saved em separately,one model to process flight booking queries, one model to process queries to send sms etc.

I've an interface where user can enter queries from any context. Can anyone suggest me the best way to find and use respective model for that specific query other than iterating over each model ? Or am i completely wrong about using different models ?

Answer 1

The way this problem has traditionally been addressed is to use an "intent" classifier to determine the intent of a query. This classifier is trained to route queries to the appropriate sequence model. Then what you can do is send the query to the top 3 models as predicted by the intent classifier and see which of those give reasonable results.