We apply methods of topological analysis to the attention graphs, calculated on the attention heads of the BERT model ( arXiv:1810.04805v2 ). Our research shows that the classifier built upon basic persistent topological features (namely, Betti numbers) of the trained neural network can achieve classification results on par with the conventional classification method. We show the relevance of such topological text representation on three text classification benchmarks. For the best of our knowledge, it is the first attempt to analyze the topology of an attention-based neural network, widely used for Natural Language Processing.
@article{arxiv.2207.01903,
title = {Betti numbers of attention graphs is all you really need},
author = {Laida Kushnareva and Dmitri Piontkovski and Irina Piontkovskaya},
journal= {arXiv preprint arXiv:2207.01903},
year = {2022}
}
Comments
This short paper was submitted to "Topological Data Analysis and Beyond" Workshop at NeurIPS 2020 at July 2020, but wasn't accepted. Later the ideas from this short paper found a rich development in arXiv:2109.04825 and arXiv:2205.09630