Combining Features for NLP tasks

Not a fast.ai subject per se, but does anyone in this community have experience in how to combine features as inputs to a classifier? For instance text might be modelled as tf-idf over lemmatized n-grams, or as dense vectors - but either of these are a single set of data. How could I also feed a classifier with other features alongside these - things such as LDA topic distributions, sentiment features, structural features. I have read several papers alluding to this approach (e.g. Social media mining for identification and exploration of health related information from pregnant women, but no descriptions about how it might be done.

1 Like