GraphConnect 2018

Random walk algorithms help better model real-world scenarios and when applied to graphs can significantly improve machine learning. DeepWalk is a supervised learning algorithm developed to analyze graphs for classification, clustering, similarity search, and representations for statistical models. Developed by Dr. Skiena and his team, its wide applicability in data mining and information retrieval, DeepWalk has become extremely popular, having been cited by over 700 research papers since its publication at KDD 2014. In this talk, we’ll look how this graph algorithm exploits an appealing analogy between sentences as sequences of words and random walks as sequences of vertices to transfer deep learning (unsupervised feature learning) techniques from natural language processing to network analysis. Dr. Skiena will introduce the notion of graph embeddings, explain how DeepWalk constructs them, and demonstrate why they make such powerful features for machine learning applications.

In this session, you’ll learn about the motivations for graph enhanced machine learning and how vector representations are learned using walks (usually random walks) on the vertices in the graph. We’ll also look at the challenges of using other matrix factorization techniques like principal component analysis (PCA) and singular value decomposition (SVD). Mark Needham will review how random walk algorithms can be implemented within Neo4j.