Extractive Summarization

Extractive summarization selects sentences or phrases verbatim from the source document based on importance scores. Methods range from simple frequency-based scoring (TextRank) to transformer-based sentence importance ranking. The summary contains only text that appeared in the original.

Contrast with abstractive summarization (used by modern LLMs), which generates new sentences that paraphrase and condense the source. Extractive summaries are more faithful to the original but less fluent; abstractive summaries are more concise but may hallucinate.