ocs-formula-shen-2023

IN premise — summaries/2026/08/24/shen-2023-icl-not-gd-s6-related-work.md

Created 2026-08-25T02:58:33+00:00

Overlap Cosine Similarity (OCS) is computed as Σ_{t∈O} p₁(t)·p₂(t) / √[(Σ_{t∈O} p₁(t)²)(Σ_{t∈O} p₂(t)²)(K − |O|)] where O is the intersection of top-K token sets.

Summary

This defines a specific way to score how similar two items are by looking only at the tokens they share among their top-K picks, weighting those shared tokens by their individual importance, and then dividing by a term that penalizes the score the fewer tokens they actually have in common. It matters because it sets the baseline formula the system uses to decide whether two things are "the same" or "close enough" based on their top features, and the (K minus overlap) term in the denominator ensures that items with tiny overlap never get artificially high similarity scores.