I have a dataset of 2 million search queries relative to 7000 categories. same query could have multiple categories. Aim is to predict category/categories for query with confidence score. I tried using bert multi label with all categories in single column with text. so for queries with multiple categories the same query phrase is repeated in multiple rows with categories. accuracy of model is very low around 40%. i believe following could be issues
1.imbalanced data from 100 queries per category to 80k queries
2.maybe 7000 categories should be rearranged as column and each marked as yes/no for the respective query.
now 1st issues is a little manageable. will the second one be any better or there could be a approach to follow?