Hi, I have a list of Hindi (local language) FAQs split into different intents on which I want to create a model to predict the intent.To give you an idea, the texts between two intents can be quite close and need to be matched exactly as they are. The model should do two things:
- Split sentences into chunks of words and figure out if the words match with any of the entity examples in the lookup tables exactly
- Match the exact sentence as it is to figure out the intent where no entities exist I am using DIET Classifier and EntitySynonymMapper to achieve this primarily with a WhitespaceTokenizer at the start - but am running into issues of over-fitting if I increase examples of a certain entity in intent, and if I don’t, am under-fitting.
Any other model I should be trying? Anything else I should be doing? Have attached my config file below. Thanks a lot in advance. config.yml (1.5 KB)