A Channel Coding Perspective of Recommendation Systems

TL;DR

This paper models recommendation systems as a channel coding problem, deriving bounds on error probability for matrix estimation under noise and erasures, and identifies thresholds for successful recovery based on cluster sizes.

Contribution

It introduces a novel channel coding perspective to analyze matrix estimation in recommendation systems, establishing bounds and thresholds for error probability.

Findings

Error probability approaches one for small cluster sizes

Polynomial time algorithm achieves vanishing error for larger clusters

Thresholds depend logarithmically on matrix dimensions

Abstract

Motivated by recommendation systems, we consider the problem of estimating block constant binary matrices (of size $m \times n$) from sparse and noisy observations. The observations are obtained from the underlying block constant matrix after unknown row and column permutations, erasures, and errors. We derive upper and lower bounds on the achievable probability of error. For fixed erasure and error probability, we show that there exists a constant $C_1$ such that if the cluster sizes are less than $C_1 \ln(mn)$, then for any algorithm the probability of error approaches one as $m, n \tends \infty$. On the other hand, we show that a simple polynomial time algorithm gives probability of error diminishing to zero provided the cluster sizes are greater than $C_2 \ln(mn)$ for a suitable constant $C_2$.