New post-doctoral research fellow: Anais Ollagnier

I started this year as a research fellow supervised by Hywel Williams. I am working on the development of the Institute of Coding Career Coach. I am investigating how we can manage social data in order to leverage them efficiently. My topics of interest are machine learning and natural language processing (NLP).

Previously, I completed my PhD in Computer Science at the University of Aix-Marseille. Throughout this, I focused on information retrieval and NLP, in the context of a digital scientific library. I designed algorithms in order to exploit the possibilities of digital access to scientific information.

Christmas festivities

For our Christmas celebration, we attended an escape room. Split into two teams, we were tasked with saving the human race by recovering a sample of a non-aging serum and discovering the formula for the most dangerous toxin ever created. Fortunately, we all managed to save the world and escape in time for our dinner reservation.

We wish everyone all the best for a happy and successful 2019.

New PhD student: Josh Buxton

I’m a first year PhD student supervised by Tim Lenton, Hywel Williams and Chris Boulton. I will be working to quantify the impact of changing climate variance on ecosystem resilience. Initially I will be focusing on changes in vegetation patterns using remote sensed data in Google Earth Engine, with the aim of expanding this study by using ecological models.

Previously I studied BSc Mathematics at the University of Bristol, which I graduated from in 2017. Throughout this, I focused on applied mathematics, while also developing an understanding of wider environmental issues and sustainable development.

Following this, I undertook the MSc Climate Change Science and Policy course at the University of Bristol. For my research project I studied the marine carbon cycle with the model cGENIE, with a focus on the impact of rain ratio sensitivity and variation on the wider carbon cycle.

New paper: @choo: Tracking Pollen and Hayfever in the UK Using Social Media

New paper available online in Sensors.

@choo: Tracking Pollen and Hayfever in the UK Using Social Media

Allergic rhinitis (hayfever) affects a large proportion of the population in the United Kingdom. Although relatively easily treated with medication, symptoms nonetheless have a substantial adverse effect on wellbeing during the summer pollen season. Provision of accurate pollen forecasts can help sufferers to manage their condition and minimise adverse effects. Current pollen forecasts in the UK are based on a sparse network of pollen monitoring stations. Here, we explore the use of “social sensing” (analysis of unsolicited social media content) as an alternative source of pollen and hayfever observations. We use data from the Twitter platform to generate a dynamic spatial map of pollen levels based on user reports of hayfever symptoms. We show that social sensing alone creates a spatiotemporal pollen measurement with remarkable similarity to measurements taken from the established physical pollen monitoring network. This demonstrates that social sensing of pollen can be accurate, relative to current methods, and suggests a variety of future applications of this method to help hayfever sufferers manage their condition.

New paper studying edge weightings in the projection of bipartite networks

New paper available online now.

Cann T.J.B., Weaver I.S., Williams H.T.P. (2019) Is it Correct to Project and Detect? Assessing Performance of Community Detection on Unipartite Projections of Bipartite Networks. In: Aiello L., Cherifi C., Cherifi H., Lambiotte R., Lió P., Rocha L. (eds) Complex Networks and Their Applications VII. COMPLEX NETWORKS 2018. Studies in Computational Intelligence, vol 812. Springer, Cham

Many real-world systems can be represented by bipartite networks that link two classes of node. However, methods for analysing bipartite networks are not as well-developed as those for unipartite networks. In particular, community detection in bipartite networks is often approached by projecting the network onto a unipartite network incorporating just one of the bipartite node classes. Here we apply a simple model to generate bipartite networks with prescribed community structure and then test the performance of community detection using four different unipartite projection schemes. Several important performance issues emerge from this treatment, particularly when the original bipartite networks have a long-tailed degree distribution. We find that a “hyperbolic” projection scheme mitigates performance issues to some extent, but conclude that care must be taken when interpreting community detection algorithm performance on projected networks. These findings have implications for any scenario where a unipartite projection is analysed in place of a bipartite system, including common applications to online social networks, social media and scientific collaboration networks.