Article: Adaptive knowledge-driven query generation for zero-shot document retrieval Journal: International Journal of Intelligent Information and Database Systems (IJIIDS) 2026 Vol.18 No.9 pp.1 - 23 Abstract: Document retrieval models often rely on large amounts of labelled data, which are time-consuming and costly to collect and annotate manually. To address this, recent methods generate pseudo queries directly from documents to train retrieval models without manual supervision. However, most existing approaches depend solely on internal document content, overlooking external knowledge that could boost performance, especially in zero-shot settings. In this paper, we propose the adaptive knowledge-driven query expansion with implicit feedback (<i>ADDEND</i>), a method that samples representative keywords, expands them using a knowledge graph, and refines pseudo queries to reduce redundancy and semantic drift. Evaluated on 12 publicly available benchmarks, <i>ADDEND</i> consistently outperforms existing models, with ablation studies confirming the effectiveness of the keyword sampling and expansion modules. Inderscience Publishers - linking academia, business and industry through research

Title: Adaptive knowledge-driven query generation for zero-shot document retrieval

Authors: Weiping Zou; Chenjun Sun; Chao Zheng; Bin Huang

Addresses: State Grid Hebei Electric Power Co., Ltd., Shijiazhuang City, Hebei Province, 050031, China ' State Grid Hebei Electric Power Supply Co., Ltd., Information and Telecommunication Branch, Shijiazhuang City, Hebei Province, 050031, China ' State Grid Hebei Electric Power Co., Ltd., Shijiazhuang City, Hebei Province, 050031, China ' State Grid Hebei Electric Power Co., Ltd., Shijiazhuang City, Hebei Province, 050031, China

Abstract: Document retrieval models often rely on large amounts of labelled data, which are time-consuming and costly to collect and annotate manually. To address this, recent methods generate pseudo queries directly from documents to train retrieval models without manual supervision. However, most existing approaches depend solely on internal document content, overlooking external knowledge that could boost performance, especially in zero-shot settings. In this paper, we propose the adaptive knowledge-driven query expansion with implicit feedback (ADDEND), a method that samples representative keywords, expands them using a knowledge graph, and refines pseudo queries to reduce redundancy and semantic drift. Evaluated on 12 publicly available benchmarks, ADDEND consistently outperforms existing models, with ablation studies confirming the effectiveness of the keyword sampling and expansion modules.

Keywords: document retrieval; zero-shot learning; knowledge graph; internal documents; external documents.

DOI: 10.1504/IJIIDS.2026.157456

International Journal of Intelligent Information and Database Systems, 2026 Vol.18 No.9, pp.1 - 23

Received: 10 Aug 2025
Accepted: 29 Apr 2026
Published online: 04 Nov 2026 *

Title: Adaptive knowledge-driven query generation for zero-shot document retrieval

Keep up-to-date