Xueli Pan

Xueli Pan

she/her

A PhD candidate at UCDS group at VU Amsterdam

Talks and Posters

Enhancing QA over Scholarly Knowledge Graphs: Addressing Semantic and Structural Challenges

February 27, 2025

Poster, ELLIS workshop on Representation Learning and Generative Models for Structured Data, Amsterdam, NL

Poster Download
ELLIS workshop website
Abstract: Scholarly knowledge graphs (SKGs) capture bibliographic metadata and scientific elements such as research problems, theories, and evaluations. Question answering (QA) over SKGs is challenging due to the complexity of scholarly data and the intricate structure of SKGs. This task involves generating SPARQL queries from natural language questions (NLQs), but large language models (LLMs) face limitations in this task, struggling with correct entity and relation linking due to their lack of knowledge about the content of SKGs and insufficient understanding of their ontological schema, particularly in low-resource SKGs like the Open Research Knowledge Graph (ORKG).

Creating a Fine-grained Scholarly Knowledge Graph from Abstracts

April 11, 2024

Poster, ICT.OPEN2024, Utrecht, NL

Poster Download
ICT.OPEN website
Abstract: Scholarly knowledge graphs (SKGs) have been proposed as a promising solution to describe research papers in a structural way to improve scientific discovery. However, most current SKGs focus on describing metadata-level information such as title, abstract, keywords, author, venue, affiliation, references and citation, while less attention are paid to structuring the content of the papers. To bridge this gap, we design an ontology to describe the content of abstracts and then use this ontology to construct a fine-grained SKG from vast of abstracts. To construct this SKG, we first extract different types of sentence from abstracts, including research question, research method and research result. We compare two different approaches to achieve sentence type extraction: traditional-ML/DL based and LLMs-based approaches. We finally evaluate the fine-grained SKGs with several down stream tasks.