To better understand audience preferences, we develop measures of an audience member based on:

• Interests - Politics, Sports, Social Issues
• Psychographics - trust the mainstream news, enjoy opinionated news, enjoy human-side of a story, discuss politics with friends, believe the news is too negative
• Brand Consumption - Al Jazeera, Vox, Vice, Buzzfeed, Fox, CNN, Huffpost

We cluster on these measures to produce audience segments and analyze the segments to identify key differentiators of audience segments, a winning content strategy for each one.

Hierarchical clustering

• Minimizes intra-cluster distance, maximizes inter-cluster distance between points
• Each pair of points is progressively nested until one cluster remains
• Shows most divisive measures
• Doesn’t require advanced notion of number of segments

Euclidean Distance

• The shortest distance between two points

Ward Clustering Criterion

• At each stage distances between clusters are recomputed and dissimilarities are squared before cluster updating