Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper investigates the impact of training data's positional distribution of relevant information on positional bias in dense retrievers. They fine-tuned eight diverse pretrained models on synthetic datasets where query-relevant evidence was skewed to the beginning, middle, or end of documents. The results show that skewed training data strongly influences the retriever's positional bias, while position-balanced training significantly reduces positional sensitivity.
Training data's bias toward where relevant information appears in documents is a major driver of positional bias in dense retrievers, dwarfing architectural factors.
Dense retrievers exhibit positional bias, favoring documents whose query-relevant information appears near the beginning and degrading retrieval performance when the information appears later. While prior work on positional bias in dense retrievers has largely focused on architectural explanations, we study how the positional distribution of evidence in training data affects retrieval-level bias direction. To test this, we construct synthetic position-targeted training sets in which query-relevant evidence appears at the beginning, middle, or end of documents, and fine-tune eight architecturally diverse pretrained models under position-skewed and balanced training distributions. At the ranking level, we observe a strong directional pattern across the examined models: skewed training distributions favor evidence at the corresponding positions. Position-balanced training reduces positional sensitivity by 57--87\% on position-aware benchmarks, with competitive mean retrieval performance in our controlled setting. Representation-level analyses further suggest that fine-tuning often reshapes learned positional preferences, although pre-existing architectural or pretraining-specific tendencies persist in some models. These results identify training-position distribution as a major controllable factor in retrieval-level position bias and suggest balanced data curation as a practical mitigation strategy.