L2 founder and resident provocateur Professor Scott Galloway has a way with words. In the Winners & Losers series, his matryoshka-like use of language impacts viewer engagement in a manner that vanilla algorithms can’t easily comprehend.
This week, we tried to see if our beta video content analysis tool could do the job. Could we “measure the immeasurable,” bridging the last mile between standard YouTube Analytics dashboard data and actual video content?
Methodology: To answer our question, we built a tool that extracts language from YouTube videos and leverages Naive Bayes Classification (a custom implementation of the NLTK sentiment analysis package) to categorize content types across 17,347 seconds of Winners & Losers YouTube video content.
We then joined these clusters with the appropriate YouTube performance metrics (views, likes, comments, and shares), adjusting for promotional ad spend. This allowed us to bucket and isolate Professor Galloway’s brilliant but sentiment-neutral insights (which account for roughly 75% of the channel’s content) from his hot takes on the Four (Amazon, Apple, Facebook, and Google) as well as his other favorite topics and rhetorical devices (sarcasm, shirtless-ness, drugs, and sex).
Last but not least, we performed sentiment and engagement impact analysis on this non-neutral content to determine the associated effect on channel performance.
Results: In terms of behavioral content, Professor Galloway’s deadpan, sarcastic humor had the greatest impact on engagement, followed closely by shirtless-ness and wearing a blazer. Apparently, our viewers are more engaged by Professor Galloway either completely covered up or exposed at the torso. This may, in part, be the result of the way Winners & Losers shots are framed and their relative use of grayscale versus color, drawing the eye to Professor Galloway’s torso. However, we fully support the bare-chested blazer look.
*Due to limited data we were not able to track engagement impact against Professor Galloway pants-less.
We also wanted to measure the relative impact of brand-related content. Videos containing content related to Amazon and Apple impacted overall channel engagement more than videos referencing Facebook or Google. We included Microsoft here because, curiously, videos with references to Microsoft had a greater impact on overall channel engagement than any of the Four.
After extracting the sentiment of Professor Galloway’s targeted statements about the Four, we determined which firms he speaks of more positively and which more negatively.
The overall sentiment score of videos containing Google- or Apple-related Scott-isms is far more positive than those containing statements about Facebook and Amazon. However, after clustering Professor Galloway’s language that specifically targets each of the Four Horsemen and processing for sentiment, Amazon is spoken about with language that is 10 times more positive than Facebook.
Looking at the bigger picture, Professor Galloway’s language in 2017 Winners & Losers videos was approximately 32% more negative than it was in 2016. As we illustrated in chart 1, the Professor’s rhetorical use of humor and calling brands out on their bullshit (classified reliably as rants) continue to resonate strongly with the Winners & Losers audience. Indeed, Winners & Losers’ audience engagement grew an exceptional 71% in 2017.
Next steps: This beta run of our tool suggests that we can extract and join feature-engineered content from YouTube videos to learn what content resonates most with a channel’s audience. We will continue to refine our methodology in 2018 and perform analysis on both influencer and brand content.