WHAT: TWed Talk: "Discourse-Aware Scholarly Knowledge Graphs for the LLM Era"
WHO: Nipun D. Pathirage
WHEN: 4p, Weds, 18 Mar 2026
WHERE: Winslow 1140
WEBEX: https://rensselaer.webex.com/meet/erickj4
EVENT PAGE: https://bit.ly/4bogBHx
Please join us Weds, 18 Mar (4p) as Nipun Pathirage leads us in a discussion of his work to enable the construction of discourse-aware scholarly knowledge graphs designed to support both human knowledge exploration and LLM-based scholarly reasoning. Pizza et.al. arrives approx. 3:30p, Nipun's talk begins at 4p.
DESCRIPTION: Scholarly knowledge graphs have traditionally linked publication metadata and extracted relations to support discovery across large collections of academic literature. However, these representations primarily capture concept-level relationships and largely omit the discourse-level knowledge encoded in scholarly texts. At the same time, the emergence of large language models (LLMs) has fundamentally changed how scholarly knowledge is accessed and consumed, while the underlying knowledge infrastructures have remained largely unchanged. In this work, we introduce a new class of scholarly knowledge graphs that integrates concept-level structure with explicit representations of scholarly propositions. We model scholarly discourse as networks of interconnected propositions grounded in concept nodes, allowing propositions to function as semantically refined relations between concepts. This representation captures both the conceptual structure and argumentative structure of scientific knowledge, enabling reasoning across discrete symbolic representations and the semantic spaces leveraged by LLMs. To support this representation, we propose the Scholarly Upper Discourse Ontology (SUDO) for modeling scholarly discourse and develop a bidirectional neuro-symbolic knowledge graph construction pipeline that combines high-recall syntactic candidate extraction with high-precision semantic refinement. This approach enables the construction of discourse-aware scholarly knowledge graphs designed to support both human knowledge exploration and LLM-based scholarly reasoning.
BIO: Nipun D. Pathirage is a second-year PhD student in Computer Science at Rensselaer Polytechnic Institute (RPI). His research focuses on using semantic technologies and knowledge graphs to guide large language models (LLMs) in order to improve the trustworthiness and privacy of LLM-based systems. Previously working in computer vision and interpretable machine learning, he holds a B.Sc. from the University of Moratuwa and an M.S. in Computer Science from RPI. He is advised by Prof. Deborah McGuinness and co-advised by Prof. Oshani Seneviratne.