Yunjie (Roya) He

Ongoing Work Under Review

HybridQE: Hybrid Query Answering over Incomplete Text-Labeled Graphs

Yunjie He, Bo Xiong, Daniel Hernández, Yuqicheng Zhu, Yi Wang, Evgeny Kharlamov, Steffen Staab

Introduces the task of hybrid query answering (HybridQA) over incomplete text-labeled graph databases, which requires joint reasoning over both symbolic logical constraints and free-form textual descriptions under information incompleteness. Proposes HybridQE, a hybrid query embedding framework that aligns symbolic and textual constraints in a unified embedding space via query-conditioned entity similarity, and constructs two new benchmarks from e-commerce and biological domains.

paper (coming soon)

Published TMLR

Counting Still Counts: Understanding Neural Complex Query Answering Through Query Relaxation

Yannick Brunink, Daniel Daza, Yunjie He, Michael Cochez

Neural methods for Complex Query Answering (CQA) over knowledge graphs (KGs) are widely believed to learn patterns that generalize beyond explicit graph structure, allowing them to infer answers that are unreachable through symbolic query processing. In this work, we critically examine this assumption through a systematic analysis comparing neural CQA models with an alternative, training-free query relaxation strategy that retrieves possible answers by relaxing query constraints and counting resulting paths.

arxiv

Published The Web Conference 2025

DAGE: DAG Query Answering via Relational Combinator with Logical Constraints

Yunjie He, Bo Xiong, Daniel Hernández, Yuqicheng Zhu, Evgeny Kharlamov, Steffen Staab

Defines DAG queries, a more general class of queries formulated in the ALCOIR description logic that extends tree-form queries by allowing quantified variables to appear multiple times. Proposes DAGE, a plug-and-play relational combinator module that extends existing tree-form query embedding methods (Query2Box, BetaE, ConE) to handle DAG queries, with proper regularization terms encouraging tautologies including monotonicity and restricted conjunction preserving. Introduces six novel DAG query types and new benchmark datasets.

arxiv code

Published NAACL 2025

Conformalized Answer Set Prediction for Knowledge Graph Embedding

Yuqicheng Zhu, Nico Potyka, Jiarong Pan, Bo Xiong, Yunjie He, Evgeny Kharlamov, Steffen Staab

Applies conformal prediction to knowledge graph embeddings for link prediction, providing statistically guaranteed prediction sets with controlled coverage.

arxiv

Oral Talk ECAI 2024

Generating SROI⁻ Ontologies via Knowledge Graph Query Embedding Learning

Yunjie He, Daniel Hernández, Mojtaba Nayyeri, Bo Xiong, Yuqicheng Zhu, Evgeny Kharlamov, Steffen Staab

Proposes AConE, a novel query embedding method that explains knowledge learned from knowledge graphs in the form of SROI⁻ description logic axioms. Embeds each SROI⁻ concept as a cone in complex vector space and relations as rotations and scalings, establishing a one-to-one mapping between logical and geometrical operators. Achieves superior results with fewer parameters, particularly on WN18RR where accuracy improves 18.35% over baselines.

arxiv code

Published EMNLP 2024 Findings

Predictive Multiplicity of Knowledge Graph Embeddings in Link Prediction

Yuqicheng Zhu, Nico Potyka, Mojtaba Nayyeri, Bo Xiong, Yunjie He, Evgeny Kharlamov, Steffen Staab

Investigates predictive multiplicity in knowledge graph embeddings for link prediction, analyzing how multiple well-performing models can yield conflicting predictions.

arxiv

Published AAMAS 2024

Robust Knowledge Extraction from Large Language Models using Social Choice Theory

Nico Potyka, Yuqicheng Zhu, Yunjie He, Evgeny Kharlamov, Steffen Staab

Bridges social choice theory with LLM knowledge extraction, applying voting mechanisms to aggregate and robustify factual knowledge extracted from large language models.

arxiv

Preprint ArXiv Survey

Geometric Relational Embeddings: A Survey

Bo Xiong, Mojtaba Nayyeri, Ming Jin, Yunjie He, Michael Cochez, Shirui Pan, Steffen Staab

A comprehensive survey covering geometric approaches to relational embeddings, reviewing methods based on points, boxes, cones, distributions, and other geometric objects for knowledge graph representation learning.

arxiv

Published ISWC 2023 · Poster & Demo

Can Pattern Learning Enhance Complex Logical Query Answering?

Yunjie He, Mojtaba Nayyeri, Bo Xiong, Yuqicheng Zhu, Evgeny Kharlamov, Steffen Staab

An early investigation into how learning logical patterns (symmetry, inversion, composition, etc.) can improve complex query answering — a foundational study leading to the AConE method.

pdf

Preprint Work Done At HUAWEI Noah's Ark Lab

Graph Attention with Hierarchies for Multi-hop Question Answering

Yunjie He, Philip John Gorinski, Ieva Staliunaite, Pontus Stenetorp

Proposes GATH (Graph ATtention with Hierarchies), two extensions to Hierarchical Graph Networks for multi-hop QA on HotpotQA: (i) completing the hierarchical structure by introducing new edges between query and context sentence nodes, and (ii) a novel graph attention mechanism that leverages the hierarchy to update node representations sequentially. Work conducted during an internship at HUAWEI Noah's Ark Lab London NLP group.

arxiv