Not sure how you wrote this paper without citing at least one thing I've done in this space :(

[Hellisotherpeople]

1. Query focused text summarizer based on "highlighting" is something I wrote and eventually wrote a paper about way back in 2020:

https://github.com/Hellisotherpeople/CX_DB8
There is also a paper that I eventually threw up on Arxiv and is citable here: https://arxiv.org/abs/2012.03942

2. A large scale queryable, word level (highlighting), text summarization dataset already exists:

https://paperswithcode.com/paper/debatesum-a-large-scale-argument-mining-and
https://github.com/Hellisotherpeople/DebateSum
https://github.com/Hellisotherpeople/debate2vec