I have a dataset of 42 features, out of which 2 columns contain text (participants' responses from a survey). The rest are numeric or categorial features. I am trying to do regression for 6 target features. I can do it successfully using Catboost by excluding text features. But Catboost doesn't support multi-target regression with text features. However, it does support univariate regression with text values. Other libraries such as XBoost and LightGBM also do not support multi-target regression with text values. What could be a better way to do it? What if I do univariate regression with text values to train 6 different models for each target feature/column? In that case, can we take the average of the 6 R-square and adjusted R-square values? Also, would it make sense?