Hey @udit.pandey, you’re seeing this behaviour because next action prediction isn’t influenced by the actual values of entities, only by whether a given entity is extracted from the user text or not. So, to condition on the value extracted (“like” vs “hate”), you’d have to have a slot that gets filled from the entity – e.g. a slot of type text, with influence_conversation: true.
Side note: Please, format your code like this so that it’s easier to read: