Message Passing Attention Networks for Document Understanding

17 Aug 2019  ·  Giannis Nikolentzos, Antoine J. -P. Tixier, Michalis Vazirgiannis ·

Graph neural networks have recently emerged as a very effective framework for processing graph-structured data. These models have achieved state-of-the-art performance in many tasks. Most graph neural networks can be described in terms of message passing, vertex update, and readout functions. In this paper, we represent documents as word co-occurrence networks and propose an application of the message passing framework to NLP, the Message Passing Attention network for Document understanding (MPAD). We also propose several hierarchical variants of MPAD. Experiments conducted on 10 standard text classification datasets show that our architectures are competitive with the state-of-the-art. Ablation studies reveal further insights about the impact of the different components on performance. Code is publicly available at: .

PDF Abstract

Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Document Classification BBCSport MPAD-path Accuracy 99.59 # 1
Text Classification IMDb MPAD-path Accuracy (2 classes) 91.84 # 12
Accuracy (10 classes) - # 3
Document Classification MPQA MPAD-path Accuracy 89.81 # 1
Document Classification Reuters-21578 MPAD-path Accuracy 97.44 # 1
Sentiment Analysis SST-2 Binary classification MPAD-path Accuracy 87.75 # 66
Sentiment Analysis SST-5 Fine-grained classification MPAD-path Accuracy 49.68 # 18
Text Classification TREC-6 MPAD-path Error 6.2 # 11


No methods listed for this paper. Add relevant methods here