University of North Texas
SEEKER: Scholarly Evaluation and Exploration of Knowledge via Agentic Retrieval
Abstract
dc:descriptionThe swift growth of scholarly publishing, now exceeding five million articles per year, has surpassed the capabilities of conventional search tools, leaving researchers with inefficient, opaque, and overwhelming information discovery processes. Current systems depend on keyword-based retrieval or black-box neural ranking, offering limited transparency into how results are selected and no mechanism for users to inspect or contest relevance decisions. These limitations are particularly consequential in academic library settings, where trust and interpretability shape technology adoption. This dissertation addresses these challenges through three sequential studies. The first developed NeuroQuest, a multi-agent framework that generates structured scholarly content across multiple modalities, establishing the viability of multi-agent architectures through a user evaluation (overall satisfaction 4.87/5, n = 17). The second surveyed 304 academic users and found that perceived usefulness (β = 0.487) and trust (β = 0.370) are the primary predictors of adoption, with trust operating differently across students, faculty, and librarians. The third designed and evaluated SEEKER, a multi-agent scholarly search system operating over 59,497 articles. At its core is Agentic Convergence Ranking, a protocol in which four evaluator agents produce transparent, per-dimension scores while an arbitration agent mediates disagreements. On a 130-query benchmark with dual-LLM judging (α = 0.744), ACR achieves statistical parity with neural reranking (nDCG@10 = 0.656 vs. 0.678, p = 0.092) while providing auditable score decompositions. These results establish that interpretable, multi-agent ranking can match black-box methods without significant degradation, offering a principled path toward trustworthy scholarly search in academic libraries.
Degree
thesis:*- Grantor dc:publisher
- University of North Texas
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Bevara, Ravi Varma Kumar, 1997-
- Contributors dc:contributor
-
- Nguyen, Thuan
- Wheeler, the late Stephen F.
- Lund, Brady D.
- Ryan, Sarah E.
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Public
- Bevara, Ravi Varma Kumar, 1997-
- Copyright
- Copyright is held by the author, unless otherwise noted. All rights Reserved.
- Language dc:language
- English
Identifiers
dc:identifier.*- Identifier
-
local-cont-no: submission_4803
ark: ark:/67531/metadc2588660