论文信息 - UCAS at TREC-2019 Deep Learning Track

UCAS at TREC-2019 Deep Learning Track

This paper describes the experiment conducted for our participation in the TREC-2019 Deep Learning track [1]. We test the effectiveness of two pre-trained language models, BERT [2] and XLNet [3], for the re-ranking subtask of the document ranking task, with an adoption of the passage-level document ranking approach as proposed in [4]. Our preliminary results indicate that the uses of BERT and XLNet lead to comparable performance.

[1] Jimmy J. Lin,et al. Simple Applications of BERT for Ad Hoc Document Retrieval , 2019, ArXiv.

[2] Kyunghyun Cho,et al. Passage Re-ranking with BERT , 2019, ArXiv.

[3] W. Bruce Croft,et al. Investigating the Successes and Failures of BERT for Passage Re-Ranking , 2019, ArXiv.

[4] Jamie Callan,et al. Deeper Text Understanding for IR with Contextual Neural Language Modeling , 2019, SIGIR.

[5] Bhaskar Mitra,et al. Overview of the TREC 2019 deep learning track , 2020, ArXiv.

[6] Yiming Yang,et al. XLNet: Generalized Autoregressive Pretraining for Language Understanding , 2019, NeurIPS.

[7] Zhiyuan Liu,et al. Understanding the Behaviors of BERT in Ranking , 2019, ArXiv.

[8] Ming-Wei Chang,et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , 2019, NAACL.