Deep Reinforcement Learning for Task-Driven Discovery of Incomplete Networks

Abstract

Complex networks are often either too large for full exploration, partially accessible, or partially observed. Downstream learning tasks on these incomplete networks can produce low quality results. In addition, reducing the incompleteness of the network can be costly and nontrivial. As a result, network discovery algorithms optimized for specific downstream learning tasks given resource collection constraints are of great interest. In this paper, we formulate the task-specific network discovery problem in an incomplete network setting as a sequential decision making problem. Our downstream task is selective harvesting, the optimal collection of vertices with a particular attribute. We propose a framework, called Network Actor Critic (NAC), which learns a policy and notion of future reward in an offline setting via a deep reinforcement learning algorithm. A quantitative study is presented on several synthetic and real benchmarks. We show that offline models of reward and network discovery policies lead to significantly improved performance when compared to competitive online discovery algorithms.

Topology-Guided Sampling for Fast and Accurate Community Detection

“…Reducing the size of graphs before storing or processing is a common technique, including agglomerative approaches that combine multiple vertices into a single super-node [44], selective edge deletion [1], task-driven subgraph discovery [29], and various sampling techniques [1,22,27,48]. In this work, we focus on vertex-sampling techniques for their relative simplicity and because removing a vertex from a graph also removes the edges connected to it, thus reducing the size of the graph along both the vertex and edge dimensions.…”

Estimating the Number of Induced Subgraphs from Incomplete Data and Neighborhood Queries

“…Similar network augmentation problems have been studied in social networks (see e.g., (Koskinen et al 2013) and the references therein). Our model and approach bear a resemblance to exploring a partially visible network through node probing (Soundarajan et al 2015(Soundarajan et al , 2017LaRock et al 2018;Morales, Caceres, and Eliassi-Rad 2019). However, research in this direction does not assume anything about the visible part of the network, considers different query models, and uses online learning approaches, following an exploration-exploitation pattern.…”

Graph Neural Network for Crawling Target Nodes in Social Networks

“…However, the access to its data is often severely limited by API or bandwidth restrictions. For instance, Twitter API rate limit for follows lookup are 15 requests per 15 minutes 1 . A serious problem when searching for target vertices in real social graphs is the lack of knowledge about the entire graph.…”

“…In the work [1] a crawling agent is learned to detect tightly connected subgraphs. A deep actor-critic network is pretrained on a synthetic graph with dense target subgraphs and then is applied to a real graph, where it is fine-tuned during crawling.…”

“…In our case we consider three types of target set topology. The first case is one dense subgraph, like is considered in 1 https://developer.twitter.com/en/docs/twitter-api/rate-limits Fig. 1.…”