Deal with long-context in machine reading comprehension

AI Discussions
1

Hi everyone,
I am doing research in NLP field, and my task is question answering from a large scale collection of document. These document is very long (such as textbook or novel). There is a way to do it is truncation. But when truncate it, the document definitely lost a lot of information. So is there any way better to deal with long-context ?

2

How about RAG ?

3

You mean like split the long document into array of sentences and then find the most k relevant sentences and feed these as contexts into the encoder ?

4

Correct.