arXiv · 2608.16269
Domain-Agnostic Neural Topic Modeling with Contextual Token-Level Semantic Graph Representation
Abstract
Recent advances in neural topic models with pre-trained language models (PLMs) have achieved strong performance by leveraging general-domain pre-training, yet their topic interpretability often degrades on specialized corpora. This limitation primarily stems from the geometry of the embedding space, where domain-specific terms unseen during pre-training collapse into an indistinguishable region, and neither domain-specific re-training, word-level graph enrichment, nor parameter-efficient fine-tuning can restructure this space without inheriting the capacity ceiling of the underlying encoder. Our key insight is that a learnable graph layer operating on token-level PLM embeddings can acquire corpus-specific semantic structure that the frozen encoder lacks, because token-level graphs preserve document-local context that word-level representations discard and joint optimization with the topic objective reshapes embedding geometry directly from target-domain evidence. We instantiate this insight as DARTopic, a domain-agnostic framework that constructs token-level semantic graphs from frozen PLM embeddings and jointly trains a GNN encoder with topic inference. Across three benchmarks spanning general, biomedical, and legal domains, DARTopic consistently outperforms strong baselines in topic coherence and document clus- tering without any encoder fine-tuning, while demonstrating robustness to PLM choice and favorable runtime efficiency over fine-tuning based alternatives.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Seung-Won Seo, Won Ik Cho, Yongmin Yoo. 2026-08-17. Domain-Agnostic Neural Topic Modeling with Contextual Token-Level Semantic Graph Representation. https://arxiv.org/abs/2608.16269
Cite the original work for its findings. Save a collection to share your selection of sources.