III: Medium: Collaborative Research: An Extensible Heterogeneous Network Embedding Framework with Application Specific Adaptation

University Of California-Davis, Davis CA

Investigators

Abstract

Network data is ubiquitous in the real-world, and many online websites providing various kinds services can all be represented as networks, e.g., online social networks, e-commerce networks, and academic networks. Learning and mining of network structured data have been one of the most popular yet challenging research problems studied in recent years. This project will study the problem of how to find a simple, yet effective representation for each network node, which can capture its characteristics or role in the network based on its connections. This is referred to as the network embedding problem. As an effective tool to transform network data into classic feature-vector representations, network embedding aims at mapping the network data into a low-dimensional feature space, i.e., with a small number of features for each network node. With the embedding results, all these aforementioned networks will be benefited to improve their services provided for the public. This project focuses on developing a general network embedding framework, and investigating its extension to application-oriented, multi-network and dynamic-network scenarios. This project will help support female and minority students to participate in academic research about network embedding. Network embedding studied in this project is a challenging learning task due to many reasons. (1) Data perspective, the heterogeneity of real-world social network data renders existing homogeneous-network oriented embedding models failing to work; (2) Structure preserving perspective, many first-order proximity based embedding methods can hardly preserve the complex social network structure with heterogeneous node types; and (3) Task perspective, the detachment of embedding process with external tasks makes the learnt results ineffective for application tasks with specific objectives. This project aims at tackling these challenges by proposing a novel extensible heterogeneous social network embedding model, which can effectively incorporate the objectives of external tasks in the learning process. This project covers five main themes: (1) extensible heterogeneous network embedding foundation; (2) application oriented embedding of single heterogeneous network; (3) embedding over multiple heterogeneous network for network alignment; (4) dynamic heterogeneous network embedding for friend recommendation; and (5) advanced scalable heterogeneous network embedding technique exploration. This project will greatly enrich the fundamental principles and technologies of social network mining and data mining. In terms of the broader impact, advances in network embedding analysis have transformative potential for fundamental advances in understanding the behavior and activities of the social networks. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.