How to extract features from sparse Tensors

Hello everyone,
I have around 5 fields of ordinal data that have to be one-hot-encoded. These fields essentially describe the context in which a sentence was made. The one-hot-encoded would be sparse tensors. Although I am not sure, my understanding is that due to sparsity traditional layers applied such data would yield meaningless results. I was wondering if there is any way to extract features from these sparse tensors such that I would be able to “include the Context” in the sentence.