The Power of Online Sentiment Data: Article Review

An article in the April 2012 issue of Human Communication Research entitled, “Emotions, public opinion, and US presidential approval rates” argues that emotions in online discussion forums can be directly linked to approval ratings of the president. Not only does this article find an application for aggregated data from online sources, but it also uses the emotional content of the forums to create conclusions about the state of public opinion. If people, especially youth, are becoming more cynical and skeptical of political activities, they pull away from the system, but still need an outlet to express their emotions. Gonzalez-Bailon, Banchs, Kaltenbrunner (2012) separate out emotionally-charged words from online discussion forms labeled “politics” using the ANEW (Affective Norms for English Language Words) list. These words are rated based on three types of emotional factors: valence, arousal, and dominance. Valence indicates strong feelings, whether positive or negative (e.g., happiness or sadness), arousal indicates physical changes (e.g., excitement, anger), and dominance indicates response or actions (e.g., submission, awe). The conclusions were that the invasion of Iraq used the most words associated with a high arousal and that this use of arousal words helped emotions “crystallize into generalized sentiment” (Gonzalez-Bailon, et al., 2012). Thus, important events and the effects on public opinion can be assessed through online forums where “public officials can use it to respond faster to issues of public concern, and ultimately improve the channels for democratic governance” (Gonzalez-Bailon, et al., 2012).

This article has important implications for researchers that are attempting to create coherent narratives for public opinion from online spaces such as discussion forums and social media sites. The current issues regarding such data is how to code and make sense of the information and how (if possible) to generalize this data to the general public as these populations tend to be extremely narrow and race, gender, and age specific. The Gonzales-Bailon et al. article is an important step in understanding online information by using emotions as the tool to glean political views and opinions and also the validity of using such skewed online populations to make conclusions about the public. Using emotions as a standard is complicated, but the ANEW list creates a formalized standard for the comparison. It would be interesting to see if these emotions corresponded to other important political events besides presidential approval ratings and what other emotion-charged online discussions could be analyzed using this same method.

A current example of this type of data analysis is being performed in the Annenberg Innovation Lab on the Twitter Sentiment Team to determine how tweets regarding republican primary candidates can predict success and overall popularity among the candidates. Take a look at their real time dashboard to learn more: AIL Twitter Sentiment Dashboard.

As people become more withdrawn from traditional methods of participating in politics, perhaps the solution is to assess opinions online instead of hoping for a reversion to traditional practices.