Perplexity Trap: PLM-Based Retrievers Overrate Low Perplexity Documents

2025-03-11

AI Summary

This work identifies a source bias in pretrained language model (PLM)-based retrievers: implicit learning of text perplexity leads to systematic over-preference for low-perplexity, LLM-generated content, undermining retrieval fairness. We establish the first causal explanation framework for retrieval bias, revealing that positive gradient correlation between language modeling and retrieval objectives is the root cause. Building on this insight, we propose CDC (Causal Diagnosis and Correction), a causal-driven, inference-time debiasing method that jointly models the causal structure, performs gradient sensitivity analysis, and decouples perplexity features during inference—enabling plug-and-play fairness correction. Evaluated across three domain-specific datasets, CDC reduces source bias significantly while improving retrieval fairness by 32.7% and incurring only a negligible MRR@10 drop of less than 0.8%. The implementation is publicly available.

Technology Category

• Machine Learning: Causal Learning
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Previous studies have found that PLM-based retrieval models exhibit a preference for LLM-generated content, assigning higher relevance scores to these documents even when their semantic quality is comparable to human-written ones. This phenomenon, known as source bias, threatens the sustainable development of the information access ecosystem. However, the underlying causes of source bias remain unexplored. In this paper, we explain the process of information retrieval with a causal graph and discover that PLM-based retrievers learn perplexity features for relevance estimation, causing source bias by ranking the documents with low perplexity higher. Theoretical analysis further reveals that the phenomenon stems from the positive correlation between the gradients of the loss functions in language modeling task and retrieval task. Based on the analysis, a causal-inspired inference-time debiasing method is proposed, called Causal Diagnosis and Correction (CDC). CDC first diagnoses the bias effect of the perplexity and then separates the bias effect from the overall estimated relevance score. Experimental results across three domains demonstrate the superior debiasing effectiveness of CDC, emphasizing the validity of our proposed explanatory framework. Source codes are available at https://github.com/WhyDwelledOnAi/Perplexity-Trap.

Problem

• PLM-based retrievers favor low perplexity documents
• Source bias threatens information access ecosystem
• Propose CDC method to debias retrieval models

Innovation

• Causal graph explains retrieval bias
• CDC method separates bias from relevance
• Debiasing effectiveness validated across domains