Learning Listwise Domain-Invariant Representations for Ranking

Ruicheng Xian, Honglei Zhuang, Zhen Qin, Hamed Zamani, Jing Lu, Ji Ma, Kai Hui, Han Zhao, Xuanhui Wang, Michael Bendersky

Keywords: learning to rank, domain adaptation, text ranking

TL;DR: We establish a domain adaptation generalization bound for ranking and propose a method based on learning listwise invariant representations.

Abstract: Domain adaptation aims to transfer models trained on data-rich domains to low-resource ones, for which a popular method is invariant representation learning. While they have been studied extensively for classification and regression problems, how they would apply to ranking problems, where the metrics and data follow a list structure, is not well understood. Theoretically, we establish a generalization bound for ranking problems under metrics including MRR and NDCG, leading to a method based on learning listwise invariant feature representations. The main novelty of our results is that they are tailored to the listwise approach of learning to rank: the invariant representations our method learns are for each list of items as a whole, instead of the individual items they contain. Our method is evaluated on the passage reranking task, where we adapt neural text rankers trained on a general domain to various specialized domains.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)