A Boosting Approach to Learning Graph Representations

Learning the right graph representation from noisy, multisource data has garnered significant interest in recent years. A central tenet of this problem is relational learning. Here the objective is to incorporate the partial information each data source gives us in a way that captures the true underlying relationships. To address this challenge, we present a general, boosting-inspired framework for combining weak evidence of entity associations into a robust similarity metric. We explore the extent to which different quality measurements yield graph representations that are suitable for community detection. We then present empirical results on both synthetic and real datasets demonstrating the utility of this framework. Our framework leads to suitable global graph representations from quality measurements local to each edge. Finally, we discuss future extensions and theoretical considerations of learning useful graph representations from weak feedback in general application settings.

💡 Research Summary

The paper tackles the problem of learning a high‑quality graph representation from multiple noisy data sources, a task that is central to relational learning and downstream network analysis such as community detection. The authors observe that each source can be viewed as a “weak learner” that provides partial, possibly unreliable evidence about whether an edge should exist. Inspired by boosting and bandit learning, they propose a novel framework called Locally Boosted Graph Aggregation (LBGA).

LBGA maintains a weight for every possible edge‑source pair (w_{u,v,i}) across the m input graphs H₁,…,Hₘ, which are assumed to be unweighted, undirected, and defined on the same vertex set. In each iteration t the algorithm proceeds in four steps:

1. Candidate graph construction – For each edge (u,v) the algorithm normalizes the current weights over the m sources to obtain a probability distribution, samples a source Hᵢ according to this distribution, and includes (u,v) in the candidate graph Gₜ if the sampled source contains the edge.

2. Clustering – A clustering algorithm A (the authors use Walktrap, but any clustering or graph‑based operation can be substituted) is run on Gₜ, yielding a partition A(Gₜ).

3. Local quality evaluation – A local quality function q(Gₜ, e, A(Gₜ)) ∈