2018/06/19 by Stephen Bonner, Ibad Kureshi, Bonner, Stephen +12 · 1 citation
Biochemistry, Genetics and Molecular Biology · Computer Science · Mathematics · Physics and Astronomy · #Advanced Graph Neural Networks #Artificial intelligence #Bioinformatics and Genomic Networks #Complex Network Analysis Techniques #Computer science #Embedding #FOS: Computer and information sciences #Graph #Graph embedding #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Theoretical computer science #cs.LG #stat.ML
paper · pdf · doi:10.48550/arxiv.1806.07464
published in arXiv (Cornell University) (Cornell University)
arxiv created 2018/06/19 · openalex publication_date 2018/06/19 · arxiv updated 2018/06/21 · openalex created_date 2022/08/20 · openalex updated_date 2026/08/08
Graph embeddings have become a key and widely used technique within the field\nof graph mining, proving to be successful across a broad range of domains\nincluding social, citation, transportation and biological. Graph embedding\ntechniques aim to automatically create a low-dimensional representation of a\ngiven graph, which captures key structural elements in the resulting embedding\nspace. However, to date, there has been little work exploring exactly which\ntopological structures are being learned in the embeddings process. In this\npaper, we investigate if graph embeddings are approximating something analogous\nwith traditional vertex level graph features. If such a relationship can be\nfound, it could be used to provide a theoretical insight into how graph\nembedding approaches function. We perform this investigation by predicting\nknown topological features, using supervised and unsupervised methods, directly\nfrom the embedding space. If a mapping between the embeddings and topological\nfeatures can be found, then we argue that the structural information\nencapsulated by the features is represented in the embedding space. To explore\nthis, we present extensive experimental evaluation from five state-of-the-art\nunsupervised graph embedding techniques, across a range of empirical graph\ndatasets, measuring a selection of topological features. We demonstrate that\nseveral topological features are indeed being approximated by the embedding\nspace, allowing key insight into how graph embeddings create good\nrepresentations.\n