arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2104.08253cs.CLcs.IR
收藏

Condenser:一种用于稠密检索的预训练结构

Condenser: a Pre-training Architecture for Dense Retrieval

Language Technologies Institute(语言技术研究所);Carnegie Mellon University(卡内基梅隆大学)

Luyu Gao, Jamie Callan

摘要:

预训练的Transformer语言模型(LM)已经成为文本表示编码器。先前的研究微调了深度LM,将句子和段落等文本序列编码为单个密集向量表示,以实现高效的文本比较和检索。然而,密集编码器需要大量的数据和复杂的技术来有效地训练,并在低数据情况下受到影响。本文发现一个关键的原因是,标准的LM的内部注意力结构是不准备使用的密集编码器,这需要聚合文本信息到密集表示。我们建议使用一种新的Transformer架构Condenser对密集编码器进行预训练,其中LM预测条件基于DENSE表示。我们的实验表明,Condenser在各种文本检索和相似性任务上比标准LM提高了很大的幅度。

英文摘要:

Pre-trained Transformer language models (LM) have become go-to text representation encoders. Prior research fine-tunes deep LMs to encode text sequences such as sentences and passages into single dense vector representations for efficient text comparison and retrieval. However, dense encoders require a lot of data and sophisticated techniques to effectively train and suffer in low data situations. This paper finds a key reason is that standard LMs' internal attention structure is not ready-to-use for dense encoders, which needs to aggregate text information into the dense representation. We propose to pre-train towards dense encoder with a novel Transformer architecture, Condenser, where LM prediction CONditions on DENSE Representation. Our experiments show Condenser improves over standard LM by large margins on various text retrieval and similarity tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑