Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PALR: Personalization Aware LLMs for Recommendation

Fan Yang, Zheng Chen, Ziyan Jiang, Eunah Cho, Xiaojiang Huang, Yanbin Lu

TL;DR

PALR introduces a personalization-aware framework that couples user history with retrieval-grounded LLM reasoning to generate recommendations. By fine-tuning a 7B LLaMA model on instruction-based tasks (Recommend and Recommend_Retrieval) and using a three-part natural language prompt plus a retrieval module, PALR achieves state-of-the-art performance on sequential recommendation benchmarks. The approach demonstrates the importance of explicit retrieval grounding and prompt design for LLM-based ranking, with strong inductive generalization from limited fine-tuning. The work suggests promising avenues for integrating LLMs into practical recommender systems with potential for explainable and conversational recommendations, while highlighting latency considerations for real-world deployment.

Abstract

Large language models (LLMs) have recently received significant attention for their exceptional capabilities. Despite extensive efforts in developing general-purpose LLMs that can be utilized in various natural language processing (NLP) tasks, there has been less research exploring their potential in recommender systems. In this paper, we propose a novel framework, named PALR, which aiming to combine user history behaviors (such as clicks, purchases, ratings, etc.) with LLMs to generate user preferred items. Specifically, we first use user/item interactions as guidance for candidate retrieval. Then we adopt a LLM-based ranking model to generate recommended items. Unlike existing approaches that typically adopt general-purpose LLMs for zero/few-shot recommendation testing or training on small-sized language models (with less than 1 billion parameters), which cannot fully elicit LLMs' reasoning abilities and leverage rich item side parametric knowledge, we fine-tune a 7 billion parameters LLM for the ranking purpose. This model takes retrieval candidates in natural language format as input, with instruction which explicitly asking to select results from input candidates during inference. Our experimental results demonstrate that our solution outperforms state-of-the-art models on various sequential recommendation tasks.

PALR: Personalization Aware LLMs for Recommendation

TL;DR

PALR introduces a personalization-aware framework that couples user history with retrieval-grounded LLM reasoning to generate recommendations. By fine-tuning a 7B LLaMA model on instruction-based tasks (Recommend and Recommend_Retrieval) and using a three-part natural language prompt plus a retrieval module, PALR achieves state-of-the-art performance on sequential recommendation benchmarks. The approach demonstrates the importance of explicit retrieval grounding and prompt design for LLM-based ranking, with strong inductive generalization from limited fine-tuning. The work suggests promising avenues for integrating LLMs into practical recommender systems with potential for explainable and conversational recommendations, while highlighting latency considerations for real-world deployment.

Abstract

Large language models (LLMs) have recently received significant attention for their exceptional capabilities. Despite extensive efforts in developing general-purpose LLMs that can be utilized in various natural language processing (NLP) tasks, there has been less research exploring their potential in recommender systems. In this paper, we propose a novel framework, named PALR, which aiming to combine user history behaviors (such as clicks, purchases, ratings, etc.) with LLMs to generate user preferred items. Specifically, we first use user/item interactions as guidance for candidate retrieval. Then we adopt a LLM-based ranking model to generate recommended items. Unlike existing approaches that typically adopt general-purpose LLMs for zero/few-shot recommendation testing or training on small-sized language models (with less than 1 billion parameters), which cannot fully elicit LLMs' reasoning abilities and leverage rich item side parametric knowledge, we fine-tune a 7 billion parameters LLM for the ranking purpose. This model takes retrieval candidates in natural language format as input, with instruction which explicitly asking to select results from input candidates during inference. Our experimental results demonstrate that our solution outperforms state-of-the-art models on various sequential recommendation tasks.
Paper Structure (11 sections, 2 figures, 2 tables)

This paper contains 11 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. PALR Framework
  4. Fine-Tuning
  5. Experiments
  6. Experiments Settings
  7. Datasets
  8. Evaluation
  9. Baselines
  10. Overall Performance Comparison
  11. Conclusion

Figures (2)

  • Figure 1: Here is an overview of our proposed PALR architecture. The "Natural Language Prompt" for the "LLM for recommendation" comprises three components: the "Interaction History Sequence," the "Natural Language User Profile," and the "Candidates for Recommendation". The "Interaction History Sequence" is created by simply concatenating the items that the user has interacted with. The "Natural Language User Profile" is a high-level summarization of the user's preferences, generated using an LLM based on user-item interactions, item attributes, or even user information if possible. The "Candidates for Recommendation" are the output of a retrieval model, and in our design, we have the flexibility to use any retrieval model for this purpose. We have included an example in Figure \ref{['fig:prompt example']}.
  • Figure 2: Here is an example for movie recommendation. (Above) The "Natural Language User Profile" leverages an LLM to summarize the user's preferences by taking into account the movies the user has watched and the movie keywords. (Below) The "LLM for Recommendation" takes the "Natural Language Prompt" as its input, which is composed of three parts: the "Interaction History Sequence," the "Natural Language User Profile," and the "Candidates for Recommendation."