End-to-end Training for Recommendation with Language-based User Profiles

2024-10-24

arXiv.org

AI Summary

Existing studies rely on zero-shot large language model (LLM) generation of natural language user profiles, but produce low-quality outputs that hinder both recommendation performance and interpretability. To address this, we propose LangPTune—the first end-to-end LLM fine-tuning framework explicitly designed for recommendation tasks, enabling joint optimization of natural language profile generation and recommendation objectives. Methodologically, LangPTune employs a recommendation-aligned loss function to guide fine-tuning and integrates GPT-4–assisted evaluation with human expert validation. Experiments across multiple benchmarks demonstrate that LangPTune significantly outperforms zero-shot baselines; its recommendation accuracy matches state-of-the-art embedding-based methods, while simultaneously generating highly readable, credible, and interpretable natural language profiles. Thus, LangPTune achieves a principled balance between predictive performance and model transparency.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

There is a growing interest in natural language-based user profiles for recommender systems, which aims to enhance transparency and scrutability compared with embedding-based methods. Existing studies primarily generate these profiles using zero-shot inference from large language models (LLMs), but their quality remains insufficient, leading to suboptimal recommendation performance. In this paper, we introduce LangPTune, the first end-to-end training framework to optimize LLM-generated user profiles. Our method significantly outperforms zero-shot approaches by explicitly training the LLM for the recommendation objective. Through extensive evaluations across diverse training configurations and benchmarks, we demonstrate that LangPTune not only surpasses zero-shot baselines but can also matches the performance of state-of-the-art embedding-based methods. Finally, we investigate whether the training procedure preserves the interpretability of these profiles compared to zero-shot inference through both GPT-4 simulations and crowdworker user studies. Implementation of LangPTune can be found at https://github.com/ZhaolinGao/LangPTune.

Problem

• Optimizes LLM-generated user profiles
• Enhances recommendation system performance
• Preserves interpretability of user profiles

Innovation

• End-to-end training framework
• Optimizes LLM-generated profiles
• Matches embedding-based methods