TL;DR Led project to improve matching processes for a leading dating platform in Denmark using message metadata and Dynamic Time Warping (DTW) Clustering. Created 8 distinct clusters of dating behaviours and recommended matching users based on these clusters. Used Exploratory Data Analysis (EDA) to bring clusters to life as personas with the help of a communication agency. Implemented DTW Clustering algorithm in Python and applied it to drive project success.
Excerpt from analysis
To analyze dating habits, we used the frequency of messages as a proxy for dating habits. We began counting messages from the moment a match occurred between two users. The counting process would only stop under three conditions: if the couple unmatched, if one of the profiles went inactive, or if the communication between them had been idle for more than 30 days. This methodology allowed us to capture the natural “lifecycle” of dating habits, from initial interest to either sustained communication or gradual disengagement.
Dynamic Time Warping (DTW) played a crucial role in our analysis by allowing us to align time series of different lengths. This capability was essential because dating conversations can vary significantly in duration and pace, but still show the same overall trends. DTW enabled us to compare trends (which we interpreted as dating habits) more effectively by finding the optimal alignment between two time series, even if they were of different lengths or slightly out of phase. This alignment made it possible to cluster the time series into distinct dating patterns, revealing common behaviors and communication styles among users.
Through this analysis, we were able to identify 8 distinct clusters of dating behaviors. These clusters provided valuable insights into user communication patterns, allowing the platform to make more informed matching recommendations. By understanding these patterns, the dating platform could potentially improve user experience and increase the likelihood of successful matches.