Today I want to understand the distribution of the characteristics of TED talks – are they rated by users as beautiful, ingenious or long-winded?
To do so I tapped on thelatest dataset with TED including all talks with ratings from the start until year-to-date, parsed the json column into a dataframe containing 14 characteristics both positive and negative, ranging from persuasive to ingenious to obnoxious to jaw-dropping, then normalized the characteristics by the biggest value of the talk (otherwise the most watched would also show up as the most long-winded just because of the bigger viewer base), and plotted them on parallel coordinates to get an overview of their distribution.
Parallel coordinates might not be the most digestible view at first glance, but it has an advantage of clearly layout the distribution of multivariate dataset. Here each line represents a talk and their intersection with the multiple axis represents the difference characteristics.
After normalizing each talk by its biggest attribute, I zoomed into the most beautiful, courageous, ingenuous, persuasive and obnoxious talks by focusing on those in the 0.95 to 1 range of these attributes.

The most beautiful talks are seldomly not inspiring or fascinating.

The most courageous talks are often inspiring, but not so fascinating.

The most ingenious talks are often inspiring, fascinating, informative. Among them one is rated unconvincing, which is a beatjazz performance.

The most persuasive talks are often also considered highly informative and inspiring.

There are a few obnoxious ones too, including a musician’s performance with breath and music (maybe too avant-garde) .
This is #day61 of my #100dayprojects on Data Science and visual storytelling. Full code on my github. Thanks for reading. If you like it, please share it. Suggestions of new topics and feedbacks are always welcomed.