Negotiation Based Resource Allocation to Control Information Diffusion

Nudurupati, Sai Sravanthi
Major Professor
Samik Basu
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Computer Science
Organizational Unit
Journal Issue
Computer Science

Study of diffusion or propagation of information over a network of connected entities play a vital role in understanding and analyzing the impact of such diffusion, in particular, in the context of epidemiology, and social and market sciences. Typical concerns addressed by these studies are to control the diffusion such that influence is maximally (in case of opinion propagation) or minimally (in case of infectious disease) felt across the network. Controlling diffusion requires deployment of resources and often availability of resources are socio-economically constrained. In this context, we propose an agent-based framework for resource allocation, where agents operate in a cooperative environment and each agent is responsible for identifying and validating control strategies in a network under its control. The framework considers the presence of a central controller that is responsible for negotiating with the agents and allocate resources among the agents. Such assumptions replicates real-world scenarios, particularly in controlling infection spread, where the resources are distributed by a central agency (federal govt.) and the deployment of resources are managed by a local agency (state govt.).

If there exists an allocation that meets the requirements of all the agents, our framework is guaranteed to find one such allocation. While such allocation can be obtained in a blind search methods (such as checking the minimum number of resources required by each agent or by checking allocations between each pairs), we show that considering the responses from each agent and considering allocation among all the agents results in a “negotiation” based technique that converges to a solution faster than the brute force methods. We evaluated our framework using data publicly available from Stanford Network Analysis Project to simulate different types of networks for each agents.