MindWeb

Search works well when you know what to search for. Traditional literature search is built around queries. That is efficient when you already know the authors or subfield you need, and much less useful when you're trying to understand the structure of an unfamiliar research landscape.

MindWeb approached the problem spatially instead. We built an interactive map of a year of Imperial's open-access research, turning individual publications into a network that could be explored rather than a list that had to be searched. Publications were connected through shared topics, grouped into communities and made visually legible as parts of a wider research ecosystem.

The project combined a data pipeline using the OpenAlex API with analysis and visualisation in R and an interactive interface built in Shiny. Louvain community detection was used to surface clusters within the network rather than imposing disciplinary categories from the top down. Articles could then be represented as parts of those emerging clusters, with visual prominence used to communicate relative impact.

The interesting design question was not simply how to display more research. It was how to transform an abstract dataset into something a person could reason through visually. A list tells you what exists. A network can also show what sits together, what bridges different areas, and where relatively empty territory might exist between established clusters.

MindWeb sits between data analysis and interface design, which is why it still belongs in the portfolio. The algorithmic output was not the final product. Clustering only became useful once its structure could be interpreted by a person.