When two intents were trained by quite similar data, the intent classification doesn't work so well

How to solve this problem? i come up with two solutions: 1.add more training data 2.merge these two intents and use entities to classify them. Are there anymore ways to solve this problem? And with the second solution i don’t quite know how to do it.

@blackforever i also face this problem, if u found any solution then share here.Thanks in advance

I’m looking into that, but also hopping someone could answer this question :slight_smile:

yes @blackforever i also think second option its a good solution.

but the first one is much easier under my circumstance lol.

Can you elaborate? What makes these two intents different?
In general, obviously adding more training data and getting a more precise model is the solution.

but @lgrinberg how we differentiate same more data on diffrent intent? Here i show you one example: i have 2 intent 1st is get_email and 2 is get_mobileno. then i have a data like if user say give me mobile no and give me email then that time sometime it’s confuse and given a wrong output.like sometime if user say give mobile no and its give you a emailid even also if they have more data on that particulate intent.

so how we solve this issue??

Thanks .

So first you should not have ambiguous examples in your training data, i.e don’t have “give me emails and mobile number” there. That way when user asks for mobile number, he won’t get email. That will still leave you with a problem when user asks for both. The way to workaround that is to have a general “get_info” intent, merge your training data for “get_mobileno” and “get_email” and then use ner_crf to pick out mobile numbers and emails addresses as entities.

1 Like

yes thanks @lgrinberg, but can you tell me how to use ner_crf ? If you have any resource ya any link or any example then can you share here please? we want to know how to do this. Thanks again.