Mapping ideas: combining text mining and networks
Mapping ideas and how they flow is one of my favorite topics. First I was captured by the concept of misperception of feedbacks. In a nutshell, it refers to how humans have difficulties to understand the complex and sometimes counterintuitive nature of reality. I created models that tried to capture non-linear dynamics of fishermen and lobsters, and see how incomplete understanding of reality often leads to sub-optimal non-sustainable scenarios both in terms of lobster population viability, fishermen profitability and cooperation, as well as traditional knowledge being eroded.
Second I jumped into mapping how scientist explain phenomena of my interest: regime shifts. I use causal loop diagrams to collect the current understanding of such dynamics. In other words, I collect our knowledge about their key drivers and feedbacks. Each link on my CLDs is an hypothesis with different levels of certainty and uneven level of attention on the academic literature. Yet, the diagram capture what I’ve found reported by scientist, and I think of them as the state of the art when it comes to regime shifts. It does not mean it’s right, complete or 100% correct. But it maps how much we know, and what are we thinking is explaining the phenomena we observe.
Lately I’ve been working on a parallel project but with a different perspective. This time I’m looking at ecosystem services: the benefits human receive from nature. I’ve been reading about regime shifts during the past 4 yrs, and it is still really hard to pin down what are their impacts on ecosystem services. Most of the papers on regime shifts do not state what are the potential impacts, and when they do, they often report the easy-to-study / measure short term consequences: e.g. there is less fish, or there is an x% expected decrease on tourism industry revenue. Then most of the interpretation about impacts is subjective and depends on the “reader’s logic”.
To partially circumvent this problem, I’ve been working with Robin Wikström applying topic models – a type of machine learning technique for text mining- in order to discover which ecosystem services are referred to in the scientific literature about regime shifts. The logic behind is more or less the same as the TED video above, where Eric Berlow and Sean Gourley analyse TED videos by translating speech to text and studying the similarity of topics across TED. I find it inspiring because this data rich environment allow them to reduce complexity and see patterns that otherwise couldn’t be pick up just by watching the videos. In the same way, I want to discover patterns of ecosystem services bundles that are hard to grasp in reality, and even harder to see when reading hundreds of papers about regime shifts. Robin and me are re-running our analysis and tuning up the algorithm, hopefully leading to some exciting results about an old question from a different approach: what are the main ecosystem services impacted by regime shifts?