GraphCycle-CLTM: A cycle–consistent graph contrastive framework for neural topic modeling

doi:10.1016/j.knosys.2026.115815

GraphCycle-CLTM: A cycle–consistent graph contrastive framework for neural topic modeling

2026-03-29

0

PRE

AI

A

Aytuğ Onan *

DOI：10.1016/j.knosys.2026.115815

Original

Original request for help

Save

Abstract

En 中文

• Proposes GraphCycle-CLTM, a cycle-consistent graph contrastive neural topic modeling framework. • Introduces multi-hop graph-guided contrastive sampling for semantically unbiased negative selection. • Incorporates topic-aware semantic graph propagation over a three-partite document–topic–word graph. • Establishes strong improvements in topic coherence, diversity, clustering, classification, and robustness. • Provides an interpretable and knowledge-grounded approach to robust semantic modeling and reasoning.

Keywords:

GraphCycle-CLTM

neural topic modeling

graph contrastive learning

topic coherence

semantic propagation

GraphCycle-CLTM: A cycle–consistent graph contrastive framework for neural topic modeling

Abstract

Journal

Organization