Zero-Shot Recommendation as Language Modeling

TL;DR

This paper introduces a novel recommendation approach leveraging pretrained language models to generate personalized item rankings from unstructured text prompts, eliminating the need for structured training data.

Contribution

It proposes a framework that uses off-the-shelf language models with textual prompts for recommendation, bypassing traditional data-dependent methods.

Findings

LM-based recommendation performs comparably to matrix factorization.

Prompt structure significantly impacts recommendation quality.

The approach works across different data regimes.

Abstract

Recommendation is the task of ranking items (e.g. movies or products) according to individual user needs. Current systems rely on collaborative filtering and content-based techniques, which both require structured training data. We propose a framework for recommendation with off-the-shelf pretrained language models (LM) that only used unstructured text corpora as training data. If a user $u$ liked \textit{Matrix} and \textit{Inception}, we construct a textual prompt, e.g. \textit{"Movies like Matrix, Inception, ${<}m{>}$"} to estimate the affinity between $u$ and $m$ with LM likelihood. We motivate our idea with a corpus analysis, evaluate several prompt structures, and we compare LM-based recommendation with standard matrix factorization trained on different data regimes. The code for our experiments is publicly available (https://colab.research.google.com/drive/1f1mlZ-FGaLGdo5rPzxf3vemKllbh2esT?usp=sharing).