This study investigates the existence of positional biases in
Transformer-based models for text representation learning, particularly in the
context of web document retrieval. We build on previous research that
demonstrated loss of information in the middle of input sequences for causal
language models, extending it to the domain of representation learning. We
examine positional biases at various stages of training for an encoder-decoder
model, including language model pre-training, contrastive pre-training, and
contrastive fine-tuning. Experiments with the MS-MARCO document collection
reveal that after contrastive pre-training the model already generates
embeddings that better capture early contents of the input, with fine-tuning
further aggravating this effect.

这项研究调查了 Transformer-based 模型在文本表征学习中的位置偏差存在情况，特别是在网络文档检索的背景下。我们基于先前研究的基础上，拓展了关于因果性语言模型输入序列中信息丢失的问题，将其应用于表征学习领域。我们在编码器 - 解码器模型的各个训练阶段，包括语言模型预训练、对比度预训练和对比度微调中，研究了位置偏差。通过对 MS-MARCO 文档集的实验发现，在对比度预训练之后，该模型已经生成了能够更好地捕捉输入早期内容的嵌入向量，而微调进一步加剧了这种效果。