Collapse of Dense Retrievers: Short, Early, and Literal Biases Outranking Factual Evidence
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916773595447296 |
|---|---|
| author | Fayyaz, Mohsen Modarressi, Ali Schuetze, Hinrich Peng, Nanyun |
| author_facet | Fayyaz, Mohsen Modarressi, Ali Schuetze, Hinrich Peng, Nanyun |
| contents | Dense retrieval models are commonly used in Information Retrieval (IR) applications, such as Retrieval-Augmented Generation (RAG). Since they often serve as the first step in these systems, their robustness is critical to avoid downstream failures. In this work, we repurpose a relation extraction dataset (e.g., Re-DocRED) to design controlled experiments that quantify the impact of heuristic biases, such as a preference for shorter documents, on retrievers like Dragon+ and Contriever. We uncover major vulnerabilities, showing retrievers favor shorter documents, early positions, repeated entities, and literal matches, all while ignoring the answer's presence! Notably, when multiple biases combine, models exhibit catastrophic performance degradation, selecting the answer-containing document in less than 10% of cases over a synthetic biased document without the answer. Furthermore, we show that these biases have direct consequences for downstream applications like RAG, where retrieval-preferred documents can mislead LLMs, resulting in a 34% performance drop than providing no documents at all. https://huggingface.co/datasets/mohsenfayyaz/ColDeR |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_05037 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Collapse of Dense Retrievers: Short, Early, and Literal Biases Outranking Factual Evidence Fayyaz, Mohsen Modarressi, Ali Schuetze, Hinrich Peng, Nanyun Computation and Language Information Retrieval Dense retrieval models are commonly used in Information Retrieval (IR) applications, such as Retrieval-Augmented Generation (RAG). Since they often serve as the first step in these systems, their robustness is critical to avoid downstream failures. In this work, we repurpose a relation extraction dataset (e.g., Re-DocRED) to design controlled experiments that quantify the impact of heuristic biases, such as a preference for shorter documents, on retrievers like Dragon+ and Contriever. We uncover major vulnerabilities, showing retrievers favor shorter documents, early positions, repeated entities, and literal matches, all while ignoring the answer's presence! Notably, when multiple biases combine, models exhibit catastrophic performance degradation, selecting the answer-containing document in less than 10% of cases over a synthetic biased document without the answer. Furthermore, we show that these biases have direct consequences for downstream applications like RAG, where retrieval-preferred documents can mislead LLMs, resulting in a 34% performance drop than providing no documents at all. https://huggingface.co/datasets/mohsenfayyaz/ColDeR |
| title | Collapse of Dense Retrievers: Short, Early, and Literal Biases Outranking Factual Evidence |
| topic | Computation and Language Information Retrieval |
| url | https://arxiv.org/abs/2503.05037 |