SimGR: Escaping the Pitfalls of Generative Decoding in LLM-based Recommendation

TL;DR

This paper introduces SimGR, a novel framework for recommendation systems that models item preferences directly at the item level, overcoming biases caused by token-level generative approximations in large language models.

Contribution

SimGR is the first approach to model item-level preference distributions directly, reducing bias and improving recommendation accuracy in LLM-based systems.

Findings

SimGR outperforms existing generative recommenders across multiple datasets.

Token-level generation introduces systematic bias in preference distribution estimation.

Direct item-level modeling aligns better with recommendation objectives.

Abstract

A core objective in recommender systems is to accurately model the distribution of user preferences over items to enable personalized recommendations. Recently, driven by the strong generative capabilities of large language models (LLMs), LLM-based generative recommendation has become increasingly popular. However, we observe that existing methods inevitably introduce systematic bias when estimating item-level preference distributions. Specifically, autoregressive generation suffers from incomplete coverage due to beam search pruning, while parallel generation distorts probabilities by assuming token independence. We attribute this issue to a fundamental modeling mismatch: these methods approximate item-level distributions via token-level generation, which inherently induces approximation errors. Through both theoretical analysis and empirical validation, we demonstrate that token-level generation cannot faithfully substitute item-level generation, leading to biased item distributions. To address this, we propose \textbf{Sim}ply \textbf{G}enerative \textbf{R}ecommendation (\textbf{SimGR}), a framework that directly models item-level preference distributions in a shared latent space and ranks items by similarity, thereby aligning the modeling objective with recommendation and mitigating distributional distortion. Extensive experiments across multiple datasets and LLM backbones show that SimGR consistently outperforms existing generative recommenders. Our code is available at https://anonymous.4open.science/r/SimGR-C408/