← Index
Instrument N° 22 · AI
Summarizer.
TextRank-style sentence scoring — the summary is always quoted from the source. Drag the slider to lengthen or tighten.
Source · 143 words · 9 sentences
Summary · 60 words · 3 kept
Extractive summarization selects actual sentences from a document to form the summary. Unlike abstractive methods, it never generates new text — it only ranks what's already there. The most widely used algorithm is a variant of PageRank called TextRank. In TextRank, each sentence becomes a node in a graph. Edges between sentences are weighted by how much vocabulary they share, after filtering common words. Running PageRank over this graph gives each sentence a score; the top-scoring sentences are then kept in their original order, producing a summary that reads coherently. The advantage is honesty: the summary is quoted verbatim from the source. The disadvantage is that the summary can feel choppy, because it is a collage of the most-central sentences rather than a rewrite. For tasks where trust matters — legal, medical, journalistic — extractive summarization is often preferred for exactly this reason.
Length · 30%
Extractive: sentences are scored by how central they are to the whole text, and the highest-scoring ones are kept in original order. You always see the source — kept sentences are bright, dropped ones dim.