Applying sentiment analysis usefully

Sentiment analysis can be fun to apply to varying types of text, but the usefulness of the results, as Rachael Tatman argues, is often low:

[T]he places where it makes sense for a data scientist or NLP practitioner working in industry to use sentiment analysis are vanishingly rare. First, because it doesn’t work very well and second, because even when it does work it’s usually measuring the wrong thing.

Although it’s not a lost cause. Tatman also points out areas where sentiment analysis could provide value.

Tags: ,

Measuring the varied sentiments of good and bad words

There was a survey a while back that asked people to provide a 0 to 100 percent value to probabilistic words like “usually” and “likely”. YouGov did something similar for words describing good and bad sentiments.

Tags: ,

Emotional arcs for inaugural addresses

Inaugural addresses come in different flavors, with different messages and purpose. Periscopic passed video of the ten most recent speeches through the Microsoft Emotion API to estimate emotion from each speaker’s facial expressions. Then they used a feather metaphor to visualize the results.

Shown here in the form of collected emotion arcs, each “feather” represents an inaugural address. Each barb of the feather is a moment during the speech where the president displayed an emotion — positive emotions are drawn above the quill, negative emotions below. The length of each barb represents the intensity of the emotion. The curve of the feather itself indicates the overall positivity or negativity of the speech.

As you might expect, the feather for Donald Trump weighs predominantly downward in red and orange.

Tags: , ,

Sentiment analysis on Trump and Clinton faces during debate

Debate sentiment

For anyone who watched the presidential debates, I think it was fairly obvious what emotion each candidate projected at various moments. However, a group of graduate students from Columbia University applied computer vision and sentiment analysis to get a more quantitative gauge. Because, sure, why not. Sarah Slobin for Quartz explains the results.

Tags: , ,