Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences

TL;DR

This paper demonstrates that large language models can effectively serve as near cold-start recommenders using language-based preferences, outperforming traditional methods in zero-shot and few-shot scenarios, with more explainable recommendations.

Contribution

The study introduces a new dataset and shows that LLMs can be competitive recommenders for language-based preferences without task-specific training.

Findings

LLMs perform well in zero-shot recommendation scenarios.

Language-based preferences are more explainable than traditional item-based methods.

LLMs outperform state-of-the-art collaborative filtering in cold-start cases.

Abstract

Traditional recommender systems leverage users' item preference history to recommend novel content that users may like. However, modern dialog interfaces that allow users to express language-based preferences offer a fundamentally different modality for preference input. Inspired by recent successes of prompting paradigms for large language models (LLMs), we study their use for making recommendations from both item-based and language-based preferences in comparison to state-of-the-art item-based collaborative filtering (CF) methods. To support this investigation, we collect a new dataset consisting of both item-based and language-based preferences elicited from users along with their ratings on a variety of (biased) recommended items and (unbiased) random items. Among numerous experimental results, we find that LLMs provide competitive recommendation performance for pure language-based preferences (no item preferences) in the near cold-start case in comparison to item-based CF methods, despite having no supervised training for this specific task (zero-shot) or only a few labels (few-shot). This is particularly promising as language-based preference representations are more explainable and scrutable than item-based or vector-based representations.