Temporal Graph Kernels for Classifying Dissemination Processes

Abstract

Many real-world graphs or networks are temporal, e.g., in a social network persons only interact at specific points in time. This information directs dissemination processes on the network, such as the spread of rumors, fake news, or diseases. However, the current state-of-the-art methods for supervised graph classification are designed mainly for static graphs and may not be able to capture temporal information. Hence, they are not powerful enough to distinguish between graphs modeling different dissemination processes. To address this, we introduce a framework to lift standard graph kernels to the temporal domain. Specifically, we explore three different approaches and investigate the trade-offs between loss of temporal information and efficiency. Moreover, to handle large-scale graphs, we propose stochastic variants of our kernels with provable approximation guarantees. We evaluate our methods on a wide range of real-world social networks. Our methods beat static kernels by a large margin in terms of accuracy while still being scalable to large graphs and data sets. Hence, we confirm that taking temporal information into account is crucial for the successful classification of dissemination processes.

Dynamic Graph Echo State Networks

“…We evaluate our model for dynamic graph binary classification on twelve datasets developed by Oettershagen et al [3], where dissemination processes based on the susceptible-infected (SI) epidemic model have been simulated on six different real-world social interaction datasets. In a SI model vertices are labelled either susceptible or infected, switching label from former to latter with fixed probability p at each time-step when they are directly linked to an infected vertex.…”

“…In the second six tasks (ct2) classifiers have to discriminate between two SI dissemination processes with contagion probability p = 0.2 and p = 0.8. We refer to [3] for further details.…”

“…Temporal graph kernels compute the Gram matrix for a SVM by first transforming the dynamic temporal graphs into static, and then by applying 'classic' graph kernels such as k-step random walk (RW-k) or Weisfeiler-Lehman sub-tree kernel of depth h (WL-h). In [3] three static transformations are proposed: reduced graph representation (RD), which loses temporal information, but does not increase graph dimension; direct line expansion (DL) and static expansion (SE), which preserve temporal information by providing a much larger static graph representation (Table 3). Temporal graph kernels APPR-S are also proposed to approximate the random-walk kernel on direct line expansion (i.e.…”

“…Many approaches have been recently proposed for learning representation of dynamic graphs [2]. In particular, temporal graph kernels [3] extend classic kernels employed for static graphs by transforming dynamic graphs into equivalent static ones; spatio-temporal graph convolutional networks [4] deal directly with dynamic graphs, but require end-to-end training. On the other hand, efficient reservoir approaches such as echo state networks (ESN), which embed data structures without requiring recurrent weights training, have been successfully applied to classification and regression tasks on static graphs [5,6].…”

LAGOON: An Analysis Tool for Open Source Communities

“…Successful OSS communities tend to be large, distributed and highly interconnected. Two issues further complicate matters: (1) the list of authors actively contributing to an OSS project changes over time, which needs to be taken into account to contextualize a community's dynamics [18,21,26]; and (2) many community members make significant contributions outside of the code base, such as ideations and discussions, requiring an analysis platform capable of ingesting and merging multiple, disparate data sources. Available tools to help with this analysis are limited.…”