ContextFlowGNN: A Novel Graph Neural Network for Dynamic Contextual Flow Analysis in NLP

DOI

10.2991/978-94-6239-707-1_23How to use a DOI?

Keywords

Contextual Flow; Discourse Coherence; Dynamic Graphs; Graph Neural Networks; Natural Language Processing

Abstract

Discourse coherence prediction, essential for automated essay scoring, dialogue systems, and multi-document summarization, is hindered by the inability of existing Graph Neural Network (GNN)-based Natural Language Processing (NLP) models to capture dynamic, multi-granular contextual dependencies. We propose ContextFlowGNN, a pioneering GNN framework that constructs a dynamic Context Flow Graph (CFG) integrating tokens, phrases, and discourse segments, enhanced by a physics-inspired flow-based attention mechanism, adaptive graph rewiring, hierarchical flow regularization, cross-granular message passing, temporal context decay, semantic flow modulation, discourse-aware node clustering, and attention guided edge pruning. ContextFlowGNN outperforms BERT with a significant increase in accuracy and a decrease in MSE. ContextFlowGNN demonstrates an accuracy improvement of 12.4% and a 31.7% drop in MSE as compared to BERT over curated dataset of 20000 essays, 10000 Reddit comments and 50000 news articles. Our extensive set of experiments includes ablation studies, cross-dataset experiments, error analysis, and qualitative visualizations. The datasets and code have been made publicly available.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Bikki Kumar
AU  - Amrendra Singh
AU  - Aditya Kanaujiya
AU  - Aanjneya Nayak
AU  - Aryan Singh
AU  - Aditya Singh Sikarwar
PY  - 2026
DA  - 2026/06/18
TI  - ContextFlowGNN: A Novel Graph Neural Network for Dynamic Contextual Flow Analysis in NLP
BT  - Proceedings of the International Conference on Recent Advances in Intelligent and Sustainable Technologies (RAIST 2026)
PB  - Atlantis Press
SP  - 264
EP  - 277
SN  - 2589-4919
UR  - https://doi.org/10.2991/978-94-6239-707-1_23
DO  - 10.2991/978-94-6239-707-1_23
ID  - Kumar2026
ER  -