Learning Edge Properties in Graphs from Path Aggregations

Abstract

Graph edges, along with their labels, can represent information of fundamental importance, such as links between web pages, friendship between users, the rating given by users to other users or items, and much more. We introduce LEAP, a trainable, general framework for predicting the presence and properties of edges on the basis of the local structure, topology, and labels of the graph. The LEAP framework is based on the exploration and machine-learning aggregation of the paths connecting nodes in a graph. We provide several methods for performing the aggregation phase by training path aggregators, and we demonstrate the flexibility and generality of the framework by applying it to the prediction of links and user ratings in social networks.We validate the LEAP framework on two problems: link prediction, and user rating prediction. On eight large datasets, among which the arXiv collaboration network, the Yeast protein-protein interaction, and the US airlines routes network, we show that the link prediction performance of LEAP is at least as good as the current state of the art methods, such as SEAL and WLNM. Next, we consider the problem of predicting user ratings on other users: this problem is known as the edge-weight prediction problem in weighted signed networks (WSN). On Bitcoin networks, and Wikipedia RfA, we show that LEAP performs consistently better than the Fairness & Goodness based regression models, varying the amount of training edges between 10 to 90%. These examples demonstrate that LEAP, in spite of its generality, can match or best the performance of approaches that have been especially crafted to solve very specific edge prediction problems.

Toward Edge-Centric Network Embeddings

“…Furthermore, a number of recent applications making usage of node embeddings can be also rethought of in the light of edge embeddings. As an example, we show how recent approaches for link prediction that leverage paths between a pair of nodes to establish the plausibility of a link between them (e.g., [1]), can benefit from edge embeddings. With our approach, instead of vectorizing paths as sequences of node embeddings, we can vectorize them as sequences of edge embeddings able to better capture the peculiarity of each link in a path ( §4).…”

“…We now focus our attention on the link prediction task (see [8] for a survey) and present an approach called ECNE-LP, which leverages edge embeddings. As the goal of this paper is not to specifically tackle link prediction, but to show the potential usage of edge embeddings in a downstream application, we consider the stateof-the-art LEAP system [1] and adapt it to use edge embeddings instead of node embeddings. The problem we face can be stated as follows: given a pair of nodes (𝑢, 𝑣) assess whether a link between them should hold.…”

“…To be processed by the link prediction learning model, paths found by the Path Finder are given a numerical representation. While state-theart approaches for link prediction, including our direct competitor LEAP [1], consider paths as sequences of node embeddings, ECNE-LP considers them as sequences of edge embeddings. Consider the paths 𝜋 1 =𝑛 0 -𝑛 7 -𝑛 1 and 𝜋 2 =𝑛 2 -𝑛 1 -𝑛 4 -𝑛 0 .…”

“…The datasets for edge classification and clustering are popular networks 5 used in the context of community detection [9]. As for link prediction, we considered the datasets used to evaluate the state-of-the-art LEAP systems [1], which is our direct competitor 6 . We are not aware of any approach that focuses on directly learning edge embeddings.…”

Effective Model Integration Algorithm for Improving Link and Sign Prediction in Complex Networks

“…For example, the signed network embedding, such as Beside, can only obtain the better performance on the sign prediction. And how to build a generalized model to perform link and sign prediction with multi-scene networks in the same framework [29] is also the motivation of this work.…”