When people say, "the algorithm showed me this," they are usually talking about a recommendation system. This is software that tries to predict what you are likely to watch, read, buy, or click next by using patterns from past behavior.

The important thing for beginners is that recommendation systems are not mind-reading machines. They are ranking systems. They take many possible items, score them using signals, and place a few near the top where you are most likely to notice them.

What signals recommendation systems use

A platform cannot directly measure your true interests, so it uses observable signals. Typical signals include what you clicked, how long you stayed, whether you skipped quickly, what you saved, and what similar users interacted with.

Signals have different quality. A five-second accidental click is weaker than finishing a full video and then sharing it. Systems often combine many weak clues to produce one score that estimates relevance.

How scoring and ranking work in practice

A recommendation engine usually starts with a candidate pool, such as recent posts, items from topics you follow, or products purchased by similar users. It then computes a score for each item and sorts from highest to lowest.

The simple illustrative JavaScript below shows this idea. It multiplies each signal by a weight, sums the result, and sorts the feed by score. Real systems are far more complex, but this explains the core ranking logic.

javascriptIllustrative JavaScript: simple weighted ranking for recommendations
const candidates = [
  { id: "video-1", watchTime: 0.9, liked: 1, hidden: 0 },
  { id: "video-2", watchTime: 0.4, liked: 0, hidden: 0 },
  { id: "video-3", watchTime: 0.7, liked: 1, hidden: 1 }
]

const weights = {
  watchTime: 0.6,
  liked: 0.5,
  hidden: -0.8
}

function score(item) {
  return (
    item.watchTime * weights.watchTime +
    item.liked * weights.liked +
    item.hidden * weights.hidden
  )
}

const ranked = [...candidates]
  .map((item) => ({ ...item, score: score(item) }))
  .sort((a, b) => b.score - a.score)

console.log(ranked.map(({ id, score }) => ({ id, score })))

Why your feed can become narrow

If you repeatedly engage with one topic, the model learns that topic is a strong predictor of future engagement. Over time, it may show more similar content and less variety because that strategy often increases immediate interaction metrics.

This is why intentional behavior matters. Searching for different topics, pausing on diverse sources, and using controls like "not interested" can broaden your future recommendations by giving the system new evidence.

Recommendation systems are prediction tools that rank options using behavior signals. Understanding that ranking process helps you use platforms more deliberately instead of assuming the feed is neutral.

Topics
Back to all writing