MPII at TREC CAsT 2019: Incoporating Query Context into a BERT Re-ranker
MPII at TREC CAsT 2019: Incoporating Query Context into a BERT Re-ranker
复制标题
TREC CAsT 2019 上的 MPII：将查询上下文合并到 BERT 重排序器中
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Andrew Yates
中科院分区：
文献类型：
--
作者：
Samarth Mehrotra;Andrew Yates
MPII参加了TREC 2019的会话协助跟踪（CAST）。我们的方法包括一个初始阶段的排名器，然后是一个基于BERT [3]的神经文档重新排名模型。具有基于外部知识的查询扩展的BM 25（即，Wikipedia和ConceptNet）作为第一阶段的排名方法，而神经模型使用BERT嵌入和基于内核的排名模块（KNRM）来预测文档查询相关性得分。我们从TREC网络跟踪的多样性任务中重新调整和修改子主题来训练神经模块。我们发现，神经重新排序模块大大改善了初始排序方法。
MPII participated in the Conversational Assistance Track (CAsT) at TREC 2019. Our approach consists of an initial stage ranker followed by a BERT-based [3] neural document re-ranking model. BM25 with query expansion based on external knowledge (i.e., Wikipedia and ConceptNet) serves as the first stage ranking method, while the neural model uses BERT embeddings and a kernel-based ranking module (KNRM) to predict a document-query relevance score. We repurpose and modify subtopics from the TREC Web Track’s diversity task to train the neural module. We find that the neural re-ranking module substantially improves upon the initial ranking approach.