by Veronica Giannotta
The comments section is one of the most emotionally dense places on the internet.
Every website offers a place where users may leave a comment, from social media sites, to retailers, to news outlets, to personal blogs.
One never knows what to expect when they venture into the comments section, as they may well be met with love, hatred, or anything in between.
As user comments have grown in popularity to become a staple of internet behavior, the visibility of consumer content has only increased. This stands to reason - the more people who engage in conversation on a given page, the more eyes on the main content. Given the reputation of the comments section as an emotionally fraught place, what are the emotional impacts of the most popular internet content?
I was curious to find out if NLP could be used to describe the emotional make-up of such comments.
From a business perspective, there is a great deal of value in understanding how users react to content; it allows businesses to increase their exposure to new consumers, and to keep existing consumers engaged with their brand.
Virality is, after all, the “it” factor, in that there is no specific formula out there for making content appeal to a given audience, so any insights about the content’s reception is helpful in guiding future marketing efforts. Personally, as someone with a background in psychology, I was also interested to explore the emotional impacts of viral media content, and to apply machine learning in order to map an “emotional footprint” of the media we collectively consume as a culture.
I set out to do this by collecting comments from viral social media posts, and modeling the data based on specific words that have been shown to have associations with certain emotions. I took the results of these models and built out a profile of emotions for each post, based on it’s corresponding comments.
Youtube & Reddit - I collected data on viral posts from Reddit & Youtube, along with stats for each post, and the first 50 comments in each post’s comment thread. The first 30 replies, if any, were also collected.
NRC Emotion Lexicon Dataset - The National Research Council Canada is the Government of Canada's largest research organization supporting industrial innovation, the advancement of knowledge and technology development.