Domain Adaptation of Learning-to-Rank Models with No Target Domain Data

DOI

Bibliographic Information

Other Title
  • ターゲットドメインのデータが不要なランキング学習モデルのドメイン適応

Search this article

Description

This paper proposes a weight regression model for domain adaptation of Learning-to-Rank (LtR) models when target domain query and relevance judgment data are unavailable. The model estimates the optimal ranking weights in the target domain using only domain features, which can be estimated by search engine engineers and others based on their domain knowledge. The weight regression model is trained using source domains prepared by dividing a single large dataset into multiple domains with different characteristics. Experimental results suggest that our proposed model outperformed the generic model trained on a large amount of data without considering domain differences.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top