Pola Bereta

  School of Business and Economics | Bachelor Business Analytics

The Shape of Consumer Behavior: A Symbolic and Topological Analysis of Time Series

Pola's elevator pitch
My thesis investigates how we can better understand, interpret and cluster noisy Google Trends time series. I analyzed five years of public-attention data across twenty consumer topics, asking a simple question: can we detect structure in this chaos? To explore this, I compared three representations of the data: two symbolic methods, SAX and eSAX, and a topological approach, Topological Data Analysis (TDA). The symbolic models convert each sliding window into a compact "alphabet", which works well when trends are stable and predictable. But many consumer-attention series are anything but stable: they spike suddenly, collapse just as fast, and behave in unpredictable ways. This is where TDA becomes powerful. Instead of looking at the series point by point, TDA captures the shape of the data: its loops, its rises and falls, and the persistence of these features over time. It reveals structure that conventional summaries overlook. When I clustered these representations, TDA consistently revealed deeper relationships among the most volatile, high-variance series, identifying groups that symbolic methods could not. The main message of my thesis is that when consumer attention becomes unpredictable, TDA offers a clearer lens. It helps us to understand not just the values of a time series, but the underlying geometry of how public attention evolves. While TDA handles chaotic data, which is most of what we see in consumer attention, SAX and eSAX handle the predictable. Together, they reveal a fuller picture of the time series structure.