HireVue’s Data Science Team Takes You Inside the Presidential Debate Interview
HireVue’s data science team turned Insights on the video from the first Presidential debate. Here is what they found.
Clinton word cloud:
Trump word cloud:
Specific Metrics the Data Science Team Analyzed
The HireVue Insights deep learning engine analyzes video, audio, and leverages Affectiva's emotion-sensing analytics based on over 4 million faces and 75 billion microexpression data points with accuracy in the 90th percentile.
The X Axis Confidence or - “Expression Strength" is the mean* confidence that the expression was detected. Higher scores mean it was detected more often.
*The mean, commonly known as ‘average’ is the result of adding all of the values in the data set and then dividing by the number of values in the data set. mean = (sum of the data)÷ (count or sample size of the data)
A sentiment score is based on sentiment analysis, often tied to positive or negative outcomes. Sentiment is often used to measure the degree to which a tweet or review is positive or negative. So if the language is more negative someone will get a higher negative sentiment score. If the language is friendly, optimistic, they will get a higher positive sentiment score. Here we looked at negative sentiment, so higher scores indicate more negativity.
- Clinton: 65.36 %
- Trump: 79.94%
Trump’s language and emotions were more negative than Clinton, giving him a higher negative sentiment score.
The Spam score* takes the candidate language and runs it against a typical spam filter. In other words, evaluating the phrases and word choice against those that tend to land emails into a spam folder. The closer the score is to 100%, the greater likelihood of being classified as ‘spam’.
- Clinton: 33.7%
- Trump: 46.2%
*Please note this is not part of the normal Insights evaluation of job candidates, this is a data scientist's idea of fun.