Quantifying Positional Biases in Text Embedding Models

TL;DR

This paper investigates positional biases in text embedding models, revealing a strong preference for initial content that affects retrieval robustness, regardless of encoding methods, with implications for improving model design.

Contribution

It provides the first comprehensive analysis of positional biases in embedding models, highlighting their impact on similarity and proposing considerations for robustness.

Findings

Embedding models prioritize beginning of input over other positions.

Ablation of initial content reduces cosine similarity more than end ablations.

Bias persists across different positional encoding techniques.

Abstract

Embedding models are crucial for tasks in Information Retrieval (IR) and semantic similarity measurement, yet their handling of longer texts and associated positional biases remains underexplored. In this study, we investigate the impact of content position and input size on text embeddings. Our experiments reveal that embedding models, irrespective of their positional encoding mechanisms, disproportionately prioritize the beginning of an input. Ablation studies demonstrate that insertion of irrelevant text or removal at the start of a document reduces cosine similarity between altered and original embeddings by up to 12.3% more than ablations at the end. Regression analysis further confirms this bias, with sentence importance declining as position moves further from the start, even with with content-agnosticity. We hypothesize that this effect arises from pre-processing strategies and chosen positional encoding techniques. These findings quantify the sensitivity of retrieval systems and suggest a new lens towards embedding model robustness.