Publications.
Figure from the tabular representation study

How can attribute semantics adapt to heterogeneous table contexts? We jointly learn from neighboring attributes and column values, routing objectives by entropy to preserve domain coherence or entity distinctiveness.

ICML 26

Segment-driven Structural Induction and Semantic Alignment for Heterogeneous
Tabular Representation

Woojun Jung, Susik Yoon

Figure from the massive document stream evaluation study

Can structural cues help LLMs understand massive, evolving document streams? We evaluate them across temporal QA, topic clustering, and summarization under conflicting and continuously accumulating information.

KDD 26

Can Structural Cues Save LLMs? Evaluating Language Models in Massive Document Streams

Yukyung Lee, Yebin Lim*, Woojun Jung*, Wonjun Choi, Susik Yoon

Figure from the in-domain table pretraining study

How can header semantics remain stable across heterogeneous tables? We propose header anchoring, using context-independent anchors to regularize contextualized header embeddings across rows and schema variants.

SIGIR 25 (Short)

HAETAE: In-domain Table Pretraining with Header Anchoring

Woojun Jung, Susik Yoon