Large Language Models for Generative Recommendation: A Survey and Visionary Discussions (2309.01157v2)

Abstract: LLMs (LLM) not only have revolutionized the field of NLP but also have the potential to reshape many other fields, e.g., recommender systems (RS). However, most of the related work treats an LLM as a component of the conventional recommendation pipeline (e.g., as a feature extractor), which may not be able to fully leverage the generative power of LLM. Instead of separating the recommendation process into multiple stages, such as score computation and re-ranking, this process can be simplified to one stage with LLM: directly generating recommendations from the complete pool of items. This survey reviews the progress, methods, and future directions of LLM-based generative recommendation by examining three questions: 1) What generative recommendation is, 2) Why RS should advance to generative recommendation, and 3) How to implement LLM-based generative recommendation for various RS tasks. We hope that this survey can provide the context and guidance needed to explore this interesting and emerging topic.

LLMs for Generative Recommendation: A Survey and Visionary Discussions

Introduction

The integration of LLMs into the domain of recommender systems (RS) presents a novel avenue for enhancing generative capabilities within recommendation frameworks. The surveyed research explores the potential of these models to transform traditional recommendation methods by leveraging their generative capabilities to directly produce recommendations from an entire pool of items, rather than relying on multi-stage filtering and ranking processes. This approach centralizes the generative power of LLMs to streamline and potentially democratize the recommendation process.

Generative Recommendation Conceptualization

The primary contribution of this survey is the elucidation of the concept of generative recommendation systems—defined as systems that produce recommendations without scoring individual items. Traditional systems typically involve multiple filtering stages, but LLM-based systems propose a singular stage where recommendations are generated directly using LLMs.

The paper defines essential terminologies, such as IDs in RS, which are conceived as sequences of tokens that uniquely identify users or items. These IDs integrate seamlessly with LLM architectures, thereby allowing LLMs to utilize their natural language understanding and generation capabilities for recommendation tasks.

Pipeline Transition from Multi-Stage to Single-Stage

The transition from conventional multi-stage pipelines to single-stage LLM-driven pipelines is depicted in an illustrative comparison (Figure 1).

Figure 1: Pipeline comparison between traditional recommender systems and LLM-based generative recommendation.

In a single-stage approach, an LLM operates as the entire recommendation pipeline by generating ranked recommendations in a streamlined manner. This represents a significant shift from the traditional RS approach and offers opportunities to apply advanced