Our team explored the emotional dimension of reader comments. We were interested in which articles tended to trigger joy, fear, anger, sadness, or disgust in readers; we were also curious whether we could build an "emotional profile" of users and articles. Harnessing emotional data from comments could start to answer useful questions: do certain types of stories tend to elicit a particular tone in the comments? Is an angry comment more likely to be abusive, or an emotionally balanced comment more insightful? Do readers adopt consistent emotional tones, and what does it mean if they veer away from their usual emotional character? Are readers reacting to the content of the story, the quality of the coverage, or a previous comment?
We started by gathering all the comments from two sample stories, and running them through AlchemyAPI's Emotion Analysis service. With this emotionally-enhanced comment dataset, we created some visualizations and explorers, which are screenshotted and pasted below. We also began work on an exploratory tool for analyzing the relationship between keywords and emotions, which we'll share here when finished.
These outputs were interesting and useful right away, but we also found that different emotions had different numeric baselines, so that for instance the "anger" meter was always significantly higher than "joy." We're not sure if this is a bias in AlchemyAPI in general or our dataset in particular; but it speaks to a need to normalize emotional data relative to a desired baseline (e.g. an emotionally-balanced, average news article). Future explorations would want to compare comments' emotional tone to one another rather than analyzing the emotions in the absolute.
Tool for comparing most joyful and angriest comments
by EJ Fox, source http://ejfox.github.io/coral-sentimentviz/
Static visualizations of comment data
by Sean Mussenden