Decay Ranker
Public Preview

This chapter discusses Decay Rankers, which use dynamic ranking methods based on the idea that similarity scores for certain entities should decrease according to values in specific numeric fields, helping others stand out.

Decay Ranker Overview [READ MORE]

In traditional vector search, results are ranked purely by vector similarity—how closely vectors match in mathematical space. But in real-world applications, what makes content truly relevant often depends on more than just semantic similarity.

Gaussian Decay [READ MORE]

Gaussian decay, also known as normal decay, creates the most natural-feeling adjustment to your search results. Like human vision that gradually blurs with distance, Gaussian decay creates a smooth, bell-shaped curve that gently reduces relevance as items move away from your ideal point. This approach is ideal when you want a balanced decay that doesn't harshly penalize items just outside your preferred range but still significantly reduces the relevance of distant items.

Exponential Decay [READ MORE]

Exponential decay creates a steep initial drop followed by a long tail in your search results. Like a breaking news cycle where relevance diminishes rapidly at first but some stories retain importance over time, exponential decay applies a sharp penalty to items just beyond your ideal range while still keeping distant items discoverable. This approach is ideal when you want to heavily prioritize proximity or recency but don't want to completely eliminate more distant options.

Linear Decay [READ MORE]

Linear decay creates a straight-line decline that terminates at an absolute zero point in your search results. Like an upcoming event countdown where relevance gradually fades until the event has passed, linear decay applies a predictable, steady reduction in relevance as items move away from your ideal point until they completely disappear. This approach is ideal when you want a consistent decay rate with a clear cutoff, ensuring that items beyond a certain boundary are completely excluded from results.

Tutorial: Implement Time-based Ranking [READ MORE]

In many search applications, the freshness of content is just as important as its relevance. News articles, product listings, social media posts, and research papers all benefit from ranking systems that balance semantic relevance with recency. This tutorial demonstrates how to implement time-based ranking in Zilliz Cloud using decay rankers.