Underpinning Data Analytics with Agent Based Models

Lead Research Organisation: University of Leeds
Department Name: Applied Mathematics

Abstract

What drives human behaviour? This question is of interest to both science and industry, but a persistent difficulty is the lack of individual level data to validate putative answers. However, this situation is changing due to the availability of new and emerging data sources, including social media, smart phones, loyalty cards and travel smart cards. Consequently we now have access to people's opinions, their locations and mobility patterns, their consumer behaviour and their friendship networks. These data sources open up new opportunities, but also pose new methodological challenges: the nature of big data means that many existing statistical techniques for data analysis are limited, e.g. due to non-random sampling or nonlinearities in individual behaviour.

These issues are directly relevant to our project partner Jaywing Intelligence. Jaywing Intelligence analyse social media data in real-time and make inferences about people's behaviour. These insights are used to develop a range of APIs, software, and managed services for their clients. This research project will address the following issues:
1) Jaywing Intelligence buy data from a range of sources and would like to reduce this cost where possible.
2) Combining inferences from multiple data sources presents a methodological challenge.
3) Jaywing Intelligence would like to understand how people's online social behaviour is connected to their consumer behaviour.
Progress towards these goals could have transformative impact for Jaywing Intelligence and the marketing sector.

In response to these challenges, the central goal of this project is to determine a strategic approach to data collection and analysis from multiple data sources, informed by agent-based models (ABMs) of social and consumer behaviour.

ABMs are simulation models in which individual agents interact within an environment according to a prescribed set of rules. A distinctive feature of this project is that we will use ABMs to create a range of ground truth scenarios on which to test the efficacy of new data analysis methods. These methods will use a Bayesian approach to infer what combination of the multiple data sources produced by the models is needed in order to differentiate between agent behaviours. We will combine this with data assimilation techniques so that this approach can be applied iteratively in real-time. The outcomes of this research will help to identify strategies that determine the what, where and when of data collection. This is a novel use of ABMs, but requires the simultaneous development of mathematical and numerical tools to characterise ABM dynamics and quantify the uncertainty of underlying model mechanisms.

This research project will develop new data analysis, statistical and modelling methodologies. It is therefore aligned with the EPSRC's research areas in Statistics and Applied Probability (growth) and Complexity Science (maintain). The methods developed in the project will be shaped by the issues faced in the digital media sector. This research therefore sits at the nexus between the EPSRC's Mathematical Sciences and Digital Economy themes.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/N509681/1 01/10/2016 30/09/2021
1999107 Studentship EP/N509681/1 01/10/2017 31/03/2021 Alice Louise Tapper