Avoid Recommending Out-of-Domain Items: Constrained Generative Recommendation with LLMs

2025-05-06

AI Summary

This work addresses the out-of-domain (OOD) item recommendation problem in large language model (LLM)-based recommender systems. To ensure strict domain adherence, we propose two complementary approaches: RecLM-ret (retrieval-augmented) and RecLM-cgen (constraint-guided generation). Our core contribution is a lightweight, recommendation-specific constrained generation mechanism that integrates domain lexicon guidance with optimized decoding strategies—preserving LLMs’ general capabilities while simultaneously enhancing domain compliance and generation accuracy. The methods are plug-and-play and incur minimal computational overhead. Evaluated on three standard recommendation benchmarks, our approach eliminates cross-domain recommendations entirely (100% reduction) and achieves up to a 12.7% improvement in recommendation accuracy, significantly outperforming existing LLM-based recommendation methods.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation
• 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: Vertical and domain-specific search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large Language Models (LLMs) have shown promise for generative recommender systems due to their transformative capabilities in user interaction. However, ensuring they do not recommend out-of-domain (OOD) items remains a challenge. We study two distinct methods to address this issue: RecLM-ret, a retrieval-based method, and RecLM-cgen, a constrained generation method. Both methods integrate seamlessly with existing LLMs to ensure in-domain recommendations. Comprehensive experiments on three recommendation datasets demonstrate that RecLM-cgen consistently outperforms RecLM-ret and existing LLM-based recommender models in accuracy while eliminating OOD recommendations, making it the preferred method for adoption. Additionally, RecLM-cgen maintains strong generalist capabilities and is a lightweight plug-and-play module for easy integration into LLMs, offering valuable practical benefits for the community. Source code is available at https://github.com/microsoft/RecAI

Problem

• Preventing out-of-domain item recommendations in LLM-based systems
• Comparing retrieval vs constrained generation for in-domain accuracy
• Ensuring lightweight integration of domain-constrained recommendation modules

Innovation

• Retrieval-based method for in-domain recommendations
• Constrained generation method for accurate recommendations
• Lightweight plug-and-play module for LLM integration