LBR: Towards Mitigating Length Bias in Large Language Models for Recommendation

2026-07-05

AI Summary

This work addresses the performance and fairness degradation in large language model (LLM)-based recommender systems caused by input-side attention bias and output-side decoding bias due to varying item text lengths. To mitigate these issues, the authors propose LBR, a lightweight and model-agnostic framework that calibrates attention through length-aware mechanisms and normalizes autoregressive decoding scores using an information-theoretic effective length estimate derived from a trie-based structure. This approach avoids the performance deterioration commonly observed with conventional normalization methods. As the first method to systematically tackle both input and output length biases simultaneously, LBR achieves an average 16.82% improvement in NDCG@5 across three Amazon datasets, significantly enhancing both recommendation accuracy and fairness while introducing negligible overhead in training and inference.

Abstract

Large language models (LLMs) have recently emerged as powerful backbones for recommender systems by reformulating recommendation as a token-level generation task. Despite their promise, we identify a pervasive yet underexplored issue: $\textit{Length Bias}$. Because items are represented by textual descriptions of varying lengths, LLM-based recommenders can be systematically biased in two ways. On the input side, longer item descriptions occupy more tokens in the context and thus receive disproportionately large aggregate attention mass during user preference modeling. On the output side, decoding based on summed autoregressive log-likelihood score inherently disfavors long items. Worse still, conventional length normalization can introduce an additional bias and even degrade recommendation performance. To address this problem, we propose $\textbf{LBR}$ ($\textbf{L}$ength $\textbf{B}$ias $\textbf{R}$eduction), a lightweight and model-agnostic framework for mitigating length bias in LLM-based recommendation. LBR mitigates input-side bias via Length-Aware Attention Calibration, which incorporates a length-dependent offset into attention logits to neutralize attention skew. For the output side, LBR introduces Effective Information Length Normalization, replacing naive token count with an information-theoretic length surrogate derived from the branching structure of the prefix tree. Extensive experiments on three real-world Amazon datasets and two representative LLM-based recommenders demonstrate that LBR substantially alleviates length bias while consistently improving recommendation accuracy and fairness, with negligible additional training and inference overhead (with an average NDCG@5 gain of 16.82%). The code is available at https://github.com/Void-JackLee/LBR.

Problem

• Length Bias
• Large Language Models
• Recommender Systems
• Attention Mechanism
• Token-level Generation

Innovation

• Length Bias
• Large Language Models
• Recommendation Systems
• Attention Calibration
• Information-Theoretic Normalization