What can BERTopic do for our cities?
*method_nerd_alert*
Local Facebook groups are like local residents’ online billboards containing a large variety of posts related to anything from missing keys to discussions on urban development, local politics and new cultural initiatives…
As such they are valuable repositories of local knowledge about urban areas.
That is why it makes great sense to analyze their content, if you want to tune-in with the life of a neighborhood, before building a new housing block or developing your next infrastructural plan.
In our report Everyday life in the Capital Region (available in a few weeks from now) you can read the results of our analysis of over 60 local Facebook groups, which will in fact lay the foundation for Fingerplan 2.0.
In this article we take you behind the scenes and give you a sneak peek into our method.
In 1940, the first version of the world-wide known Fingerplan, which set the trajectory for the future development of the Copenhagen Capital Region, was drawn. Back then, the process took place in a closed room with only architects and urban planners.
70 years later, the wish is to develop Fingerplan 2.0, including the perspectives of the people who live in the Region already from the beginning of the process. But to carry out a physical citizens engagement at regional scale would be almost impossible — or extremely resource intensive.
What we can do instead, is to tap into the local knowledge of each of the 32 municipalities in the Capital Region, by “listening” digitally to residents’ conversations on their local Facebook groups (*).
This means analysing the content of over 60 Facebook groups for the past two years. It amounts to 1.191.605 posts — an enormous number, that would take forever to only analyze manually -. That’s why we need BERTopic (more on this later).
After the “digital listening”-phase, we engage representatives of each municipality in a physical process. We use the preliminary results of our analysis as a starting point for an informed discussion, where citizens -and architects- relate critically to the preliminary findings, question them, confirm some hypotheses, discard others, add nuances and build upon them.
Get Sara Nardi’s stories in your inbox
Join Medium for free to get updates from this writer.
So…back to the original question What can BERTopic do for our cities?
In Fingerplan 2.0 as well as in many other projects where we analyse massive amounts of qualitative data about people’s relation to their city, we are helped by BERTopic, a state-of-the-art technology within the field of Natural Language Processing (**).
First of all, it provides an overview over the topics that citizens in a big geographical area talk about on a daily basis. In our case, this would be the 1.191.605 posts from local Facebook groups. BERTopic summarizes the most important themes present in the massive data material and makes it easier for us to understand the overall context of the data as well as interpret it through the lenses of social science.
Secondly, it helps us identify the posts related to topics that are very relevant for urban development. In the case of Fingerplan 2.0, this could be the connectivity of public transport, the noise pollution of streets or missing green areas in the neighborhood.
BERTopic groups together Facebook posts that cover similar topics. This helps us identify and cluster posts into themes and subthemes.
Thirdly, it improves our ability to visualise the data and make it available for many people — e.g. citizens or other urban stakeholders — to interpret and relate to it. By grouping similar text data together and summarizing the key topics, BERTopic enables the data to be visualised in a way that is easy to understand (see a visualisation example below).
In short, BERTopic helps us navigate in the massive amount of textual data, open the process up for more people to contribute to the analytical process, hereby allowing us to extract valuable insights that would otherwise remain hidden.
About Volcano. We are a consultancy specialized in data-driven urban analysis and citizen engagement. If you want to read more about our work you can find us on LinkedIn or on our website.
Alternatively, drop us an email here sara@volcano.nu
(*) Data privacy.
Collecting data through social media, and especially Facebook brings the question of data privacy to the forefront. Therefore we pay exceptional attention to not only be GDPR compliant with our data handing but also be ethically considerate with our data collection and anonymisation processes.
In the Fingerplan 2.0 case, most of the local Facebook groups we analysed, are open. When they are not, we asked administrators for permission to enter, always revealing our intentions. Only when we got accepted in the group, we started doing so-called n-etnographical work. Basically ethnography, just on the net.
(**) How does BERTopic work?
BERTopic is a specific type of language model called a topic model. Topic modeling is the process of discovering the underlying themes or topics from a large collection of text documents. The muscle power of BERTopic comes from the BERT architecture for language modeling. To create a BERT language model, you feed the model with enormous amounts of data. By reading through millions of sentences, the model then learns to “comprehend” words and the contexts they appear in. With this knowledge the model can then e.g. be used to find texts that are similar to each other and group them together in a topic.
