The more tags you use in a classifier, the more complex the model is. This means you’ll need to tag more data to achieve an accurate model as each additional category requires data to learn from. You’ll also need to:
spend more time fixing confusions,
finding tagging mistakes and inconsistencies,
adding stopwords to your classifier.
Experimenting with advanced settings and parameters
In other words, it is much harder to train a classifier with 30 tags than one with just 5-10 tags.
Because of this, we strongly recommend removing tags that are too small, specific, or niche. We also recommend eliminating tags that are not essential to your business needs. This way, you can focus on training and improving a less complicated classifier, which will be easier to achieve accurate predictions. This will enable you to use the classifier to analyze new data much faster and get value sooner.
Remember, you can always add tags later, but it’s better to start small and then go big.