We present DeepWalk, a novel approach for learning latent representations of vertices in network. These encode social relations continuous vector space, which is easily exploited by statistical models. DeepWalk generalizes recent advancements language modeling and unsupervised feature (or deep learning) from sequences words to graphs.

We propose a new computational approach for tracking and detecting statistically significant linguistic shifts in the meaning usage of words. Such are especially prevalent on Internet, where rapid exchange ideas can quickly change word's meaning. Our meta-analysis constructs property time series word usage, then uses sound point detection algorithms to identify shifts. consider analyze three approaches increasing complexity generate such series, culmination which distributional...

We present HARP, a novel method for learning low dimensional embeddings of graph’s nodes which preserves higher-order structural features. Our proposed achieves this by compressing the input graph prior to embedding it, effectively avoiding troublesome configurations (i.e. local minima) can pose problems non-convex optimization. HARP works finding smaller approximates global structure its input. This simplified is used learn set initial representations, serve as good initializations...

Graph clustering and graph outlier detection have been studied extensively on plain graphs, with various applications. Recently, algorithms extended to graphs attributes as often observed in the real-world. However, all of these techniques fail incorporate user preference into mining, thus, lack ability steer more interesting parts attributed graph. In this work, we overcome limitation introduce a novel user-oriented approach for mining graphs. The key aspect our is infer by so-called focus...

Existing popular methods for semi-supervised learning with Graph Neural Networks (such as the Convolutional Network) provably cannot learn a general class of neighborhood mixing relationships. To address this weakness, we propose new model, MixHop, that can these relationships, including difference operators, by repeatedly feature representations neighbors at various distances. Mixhop requires no additional memory or computational complexity, and outperforms on challenging baselines. In...

We present WALKLETS, a novel approach for learning multiscale representations of vertices in network. In contrast to previous works, these explicitly encode multi-scale vertex relationships way that is analytically derivable.

We present HARP, a novel method for learning low dimensional embeddings of graph's nodes which preserves higher-order structural features. Our proposed achieves this by compressing the input graph prior to embedding it, effectively avoiding troublesome configurations (i.e. local minima) can pose problems non-convex optimization. HARP works finding smaller approximates global structure its input. This simplified is used learn set initial representations, serve as good initializations...

In this work, we examine a novel forecasting approach for COVID-19 case prediction that uses Graph Neural Networks and mobility data. contrast to existing time series models, the proposed learns from single large-scale spatio-temporal graph, where nodes represent region-level human mobility, spatial edges based inter-region connectivity, temporal node features through time. We evaluate on US county level dataset, demonstrate rich information leveraged by graph neural network allows model...

There has been a surge of recent interest in learning representations for graph-structured data. Graph representation methods have generally fallen into three main categories, based on the availability labeled The first, network embedding (such as shallow graph or auto-encoders), focuses unsupervised relational structure. second, regularized neural networks, leverages graphs to augment losses with regularization objective semi-supervised learning. third, aims learn differentiable functions...

Graph Neural Networks (GNNs) have achieved state-of-the-art results on many graph analysis tasks such as node classification and link prediction. However, important unsupervised problems graphs, clustering, proved more resistant to advances in GNNs. clustering has the same overall goal pooling GNNs - does this mean that GNN methods do a good job at graphs? Surprisingly, answer is no current often fail recover cluster structure cases where simple baselines, k-means applied learned...

The increasing diversity of languages used on the web introduces a new level complexity to Information Retrieval (IR) systems. We can no longer assume that textual content is written in one language or even same family.In this paper, we demonstrate how build massive multilingual annotators with minimal human expertise and intervention. describe system builds Named Entity Recognition (NER) for 40 major using Wikipedia Freebase. Our approach does not require NER annotated datasets specific...

Recent interest in graph embedding methods has focused on learning a single representation for each node the graph. But can nodes really be best described by vector representation? In this work, we propose method multiple representations of (e.g., users social network). Based principled decomposition ego-network, encodes role different local community which participate. These allow improved reconstruction nuanced relationships that occur -- phenomenon illustrate through state-of-the-art...

Given a graph with node attributes, what neighborhoods are anomalous? To answer this question, one needs quality score that utilizes both structure and attributes. Popular existing measures either quantify the only ignore attributes (e.g., conductance), or consider connectedness of nodes inside neighborhood cross-edges at boundary density).In work we propose normality, new measure for attributed neighborhoods. Normality together to internal consistency external separability. It exhibits two...

Graph embedding methods represent nodes in a continuous vector space, preserving information from the graph (e.g. by sampling random walks). There are many hyper-parameters to these (such as walk length) which have be manually tuned for every graph. In this paper, we replace with trainable parameters that automatically learn via backpropagation. particular, novel attention model on power series of transition matrix, guides optimize an upstream objective. Unlike previous approaches models,...

Graph neural networks (GNNs) have emerged as a powerful approach for solving many network mining tasks. However, learning on large graphs remains challenge -- recently proposed scalable GNN approaches rely an expensive message-passing procedure to propagate information through the graph. We present PPRGo model which utilizes efficient approximation of diffusion in GNNs resulting significant speed gains while maintaining state-of-the-art prediction performance. In addition being faster, is...

In this paper we present a new computational technique to detect and analyze statistically significant geographic variation in language. While previous approaches have primarily focused on lexical between regions, our method identifies words that demonstrate semantic syntactic as well. We extend recently developed techniques for neural language models learn word representations which capture differing semantics across geographical regions. order quantify ensure robust detection of true...

We seek to better understand the difference in quality of several publicly released embeddings. propose tasks that help distinguish characteristics different Our evaluation sentiment polarity and synonym/antonym relations shows embeddings are able capture surprisingly nuanced semantics even absence sentence structure. Moreover, benchmarking great variance captured by tested Finally, we show impact varying number dimensions resolution each dimension on effective useful features embedding...

We propose a new method for embedding graphs while preserving directed edge information. Learning such continuous-space vector representations (or embeddings) of nodes in graph is an important first step using network information (from social networks, user-item graphs, knowledge bases, etc.) many machine learning tasks. Unlike previous work, we (1) explicitly model as function node embeddings, and (2) novel objective, the likelihood, which contrasts from sampled random walks with...