Grouping phrases in intents

How should bot phrases be grouped for the intent recognition algorithm to work effectively?



We cannot put sentences/statements that mean exactly the same in two different processes.

Let's start with an example:

  • how long will/do I have to wait for you to process an application for X - intent X
  • how long does it take to consider the application X - intent Y

These sentences mean the same thing because:

  1. They have the same purpose (obtaining information on how long to wait for, for example, a credit decision),
  2. They have ACTIONS (words describing an action) that mean the same (if something lasts or takes time, you have to wait for it), they differ only in expression or construction,
  3. They have the same OBJECT/OBJECTS (usually the product) that the statement involves.

These two sentences are virtually indistinguishable for a human (from the point of view of communication and intent), let alone for a machine. If we put two very similar/identical sentences in two different INTENTS, we do not give the model a chance to learn the distinction between them.


Other example of the same problem - very similar sentences, different intents

If the contract says ... does it means that I have ...
I have it written in the contract ... what does that mean