Predictive ad ranking

예측형 광고 순위

Updates ad candidate ranking from behavior sequences and engagement signals.

···
html
<div class="ranking"><div class="signal">LATEST SIGNAL <b id="signal">READING</b></div><div class="rank" data-ad="0"><strong>A</strong><i></i><span>1</span></div><div class="rank" data-ad="1"><strong>B</strong><i></i><span>2</span></div><div class="rank" data-ad="2"><strong>C</strong><i></i><span>3</span></div><div class="footer">PREDICTED ORDER <b id="order">A → B → C</b></div></div>
css
.ranking{width:min(86%,460px);font:800 clamp(10px,max(2vmin,1vw),16px)/1.2 sans-serif}.signal{display:flex;justify-content:space-between;background:var(--fg);color:var(--bg);padding:clamp(7px,1.8vmin,13px)}.signal b{color:var(--bg)}.rank{display:grid;grid-template-columns:clamp(20px,5vmin,35px) 1fr auto;gap:clamp(5px,1.5vmin,11px);align-items:center;padding:clamp(6px,1.4vmin,10px);border:1px solid var(--line);background:var(--surface);margin-top:clamp(5px,1vmin,8px)}.rank i{display:block;height:clamp(7px,1.8vmin,13px);background:var(--accent);transition:width .5s}.rank span{font-size:clamp(13px,3vmin,23px);color:var(--accent)}.footer{border-top:2px solid var(--fg);display:flex;justify-content:space-between;margin-top:clamp(8px,2vmin,16px);padding-top:6px}.footer b{color:var(--accent)}
js
const cases=[['READING',[85,55,30]],['PRODUCT VIEW',[45,90,65]],['COMPARE ITEMS',[55,70,95]]];let step=0;function draw(){const [signal,scores]=cases[step++%3];document.getElementById('signal').textContent=signal;const order=[0,1,2].sort((a,b)=>scores[b]-scores[a]);document.querySelectorAll('[data-ad]').forEach((el,i)=>{el.querySelector('i').style.width=scores[i]+'%';el.querySelector('span').textContent=String(order.indexOf(i)+1)});document.getElementById('order').textContent=order.map(i=>'ABC'[i]).join(' → ')}draw();setInterval(draw,1350);

Meta's 2026 account of ad ranking improvements describes using longer behavior sequences and additional organic engagement signals. Candidate ads can change order as predicted response changes.

The demo adds fictional reading and shopping signals, changing three candidate score bars and their rank. It is a teaching diagram, not the actual model or a real person profile.

A predicted score is neither observed performance nor causal impact. Review signal bias and targeting constraints, then evaluate real outcomes.

When to use

Use it to explain how automated ad ranking responds to changing signals.

Open as page ↗