Modeling, Estimation, and Management of High Dimensional Asynchronous Events from Information Diffusion Networks
Abstract
Modeling, Estimation and Management of High Dimensional Asynchronous Events from Information Diffusion Networks a technical proposal in response to ONRBAA15-001 submitted by Georgia Tech Research Corporation on March 13, 2014. PROJECT SUMMARY Complex social networks, such as Twitter or Facebook, have become large-scale information diffusion networks where users share, discuss and search for information of personal and public interests. Data collected in such evolving information networks are often high dimensional event data, generated ``asynchronously and ``interdependently by nodes in the evolving networks, and they require new machine learning paradigms for their representation and analysis. The proposed works will be based on multivariate Hawkes processes, but they will go beyond traditional low dimensional and small scale setting. The proposed works will explicitly take into account latent variables, low intrinsic dimensionality and sparsity present in high dimensional and big event datasets, and support subsequent time-sensitive decision makings where the effects of the decisions are assessed within a time upper bound. The proposed research will center around three themes: (i) model high dimension asynchronous, recurrent and mutually exciting events from information diffusion networks using multivariate Hawkes processes, and (ii) estimate the structure and triggering kernels of these diffusion models, and (iii) then use the learned model for a variety of information diffusion management problems, such as minimum activity maximization and consensus maximization. The proposed framework avoids the unrealistic assumption that (social) networks are static, and allows the two processes, information diffusion and network evolution, unfold simultaneously and excise bidirectional influence over each other. The proposed models and algorithms can be used to address challenging time-sensitive prediction problems such as ``who will do what and when? The proposed research, though targeted initially towards social networks, can be applied to similar data and problems arising from computational sustainability, P2P finance and cyber-security. It can bring practical values to Internet industry by better understanding and modeling of user and network behaviors. Additionally, the research problems addressed in this project also represent an excellent testing bed that lies at the interface between machine learning/data mining and social and behavioral sciences, the solution of which is expected to contribute to the emerging area of social computing, cyber-security, computational sustainability.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512340
Entities
People
- Le Song
Organizations
- Georgia Tech Research Corporation
- Office of Naval Research
- United States Navy