Role extraction, dynamics, and optimisation on networks

by Dr Yu Tian

Monday 31 Oct 2022, 15:45 → 16:45 Europe/Stockholm

Albano 3: 6228 - Mega (22 seats) (Albano Building 3)

zoom link : https://stockholmuniversity.zoom.us/j/622224375

There are many fundamental problems in industry that are also of great research value. With our industrial partner Tesco, we are particularly interested in the retail industry, where one primary problem would be to understand the intrinsic relationships between products, including complements and substitutes. Once such relationships have been estimated, a subsequent problem would be to understand the inter-related demand changes between products, and further to make decisions, e.g. on which products should we implement promotions in order to maximise the overall revenue at the end of the promotional period. 

 

In this talk, we will first formulate the problem of identifying product relationships as role extraction on networks, and propose a novel data-driven network-based approach for this purpose. Then we will model the propagation of demand change as information spreading on networks, and consider its optimisation as a variant of the classic influence maximisation problem, first on simple networks only of one relationship and then on signed networks of two competing types of relationships. To do so, we will introduce a general class of model, unifying the mechanisms underlying the classic spreading models while accounting for specific features of demand dynamics, and a framework for the corresponding influence maximisation, via a customised direct search method with local convergence.