Does LLM Focus on the Right Words? Diagnosing Language Bias in LLM-based Recommenders

2025-10-12

AI Summary

This work addresses language bias in LLM-based recommender systems induced by supervised fine-tuning (SFT), wherein models over-rely on auxiliary tokens (e.g., task instructions) while neglecting user interaction tokens that encode genuine preferences. To mitigate this, we propose GDRT—a novel fine-tuning paradigm grounded in group-wise distributionally robust optimization (DRO). GDRT semantically partitions input tokens into user interaction groups and auxiliary text groups, then dynamically reweights their contributions to suppress interference from auxiliary tokens and strengthen user behavior modeling. Evaluated on three public benchmarks, GDRT achieves an average 24.29% improvement in NDCG@10, significantly enhancing both recommendation accuracy and fairness. To our knowledge, GDRT is the first framework to systematically integrate distributionally robust optimization for token-level bias mitigation in LLM recommenders.

Technology Category

• Natural Language Processing: Safety and Robustness
• 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
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Large language models (LLMs), owing to their extensive open-domain knowledge and semantic reasoning capabilities, have been increasingly integrated into recommender systems (RS). However, a substantial gap remains between the pre-training objectives of LLMs and the specific requirements of recommendation tasks. To address this gap, supervised fine-tuning (SFT) is commonly performed on specially curated recommendation datasets to further enhance their predictive ability. Despite its success, SFT exhibits a critical limitation: it induces Language Bias, whereby the model over-relies on auxiliary tokens-such as task descriptions and prefix-generated tokens-while underutilizing core user interaction tokens that encode user-specific preferences. This bias not only undermines recommendation accuracy but also raises unfairness concerns. To address this issue, we propose Group Distributionally Robust Optimization-based Tuning (GDRT), a novel fine-tuning paradigm that enforces consistent model performance across token groups with varying degrees of relevance to auxiliary tokens. By adaptively upweighting underperforming groups, typically those weakly correlated with auxiliary tokens, GDRT shifts the model's attention from superficial auxiliary cues to informative user interaction tokens, thereby mitigating language bias. Extensive experiments conducted on three public datasets demonstrate that GDRT effectively mitigates language bias, yielding substantial improvements in recommendation accuracy (with an average NDCG@10 gain of 24.29%) and significantly enhancing recommendation fairness.

Problem

• LLMs over-rely on auxiliary tokens in recommendation systems
• Fine-tuning causes underutilization of core user interaction tokens
• Language bias reduces recommendation accuracy and fairness

Innovation

• Group Distributionally Robust Optimization-based Tuning paradigm
• Enforces consistent performance across token groups
• Adaptively upweights underperforming token groups