#summarization
salience (no capitalization) is an "extractive summarization" tool.
"Extractive summarization should be preferred over abstractive summarization when nuance is essential and when the summary is meant as a companion to the source text. LLMs are effective at abstractive summarization, but they can also be leveraged in extractive summarization. Rather than solve this with a prompt, LLM embeddings can be combined with the TextRank algorithm to reliably yield high-quality extractive summaries."
"In other words, embeddings can be used to automatically generate highlights. The internal representation is an affinity matrix between sentences that can be used to find the most salient sentences in a text."