Document chunking

문서 청킹

Split long documents into smaller passages for retrieval and model context.

···
html
<div class="scene"><div class="heading">CHUNK BOUNDARIES</div><div class="document"><span>첫 문장</span><span>연결 문장</span><span>핵심 문장</span><span>끝 문장</span></div><div class="chunks"><div id="chunk1">조각 1</div><div id="chunk2">조각 2</div></div><div class="muted">겹치는 문장은 두 조각에 남습니다</div></div>
css
.scene{width:min(94vw,800px);height:min(88vh,326px);padding:clamp(10px,2.7vmin,19px);border:1px solid var(--line);border-radius:13px;background:var(--surface);font:500 clamp(13px,2.4vmin,17px)/1.3 var(--font-sans,sans-serif);position:relative;overflow:hidden}.scene .mono{font-family:ui-monospace,SFMono-Regular,monospace}.scene .muted{color:var(--muted)}.scene .accent{color:var(--accent)}.scene .heading{font-weight:750;color:var(--accent);margin-bottom:clamp(5px,1.6vmin,12px)}.document{display:flex;gap:4px;margin:4px 0 7px}.document span{flex:1;min-width:0;border:1px solid var(--line);border-radius:5px;padding:5px 1px;text-align:center;background:var(--bg)}.document span.on{border-color:var(--accent);color:var(--accent)}.chunks{display:flex;gap:8px;height:35%;margin-bottom:4px}.chunks div{flex:1;display:grid;place-items:center;text-align:center;border-radius:7px;background:color-mix(in srgb,var(--accent) 14%,var(--surface));border:1px solid var(--accent);transition:.3s}
js
let overlap=false;function draw(){overlap=!overlap;document.querySelectorAll('.document span').forEach((x,i)=>x.classList.toggle('on',overlap&&i===1));document.getElementById('chunk1').textContent=overlap?'첫 + 연결':'첫 + 연결';document.getElementById('chunk2').textContent=overlap?'연결 + 핵심 + 마지막':'핵심 + 마지막'}draw();setInterval(draw,2400)

Chunking divides long text by size or semantic boundary and stores each passage as a retrievable unit. Tiny chunks lose context, while large ones can mix irrelevant material into a result.

The demo moves a boundary and shows how overlap preserves nearby context. Tune chunk size and overlap for the documents and measured retrieval quality.

When to use

Use it to index long documents for RAG or fit relevant passages into context.

Open as page ↗